> Unfortunately, there’s a growing chorus of people who argue that AIs could now be, or may soon become, conscious. They argue that AIs may deserve rights and protections similar to those that we provide other conscious beings.12 If this view takes hold, it will shake the foundations of our society, rupturing our existing political and ethical frameworks, and fundamentally changing what it means to be human.
So completely independent of the question on whether there are any empirical arguments that LLMs are conscious or are not conscious he argues from the end here and says that if they were conscious, this would have horrible consequences for our society, so we must never assume that they are.
That's basically the same way people argue about animals, only they usually don't say it as openly.
As for the actual question, I agree that with current LLMs, there is not a lot there that could be conscious outside the inference loop (and if it were, it would necessarily have to be wildly different than that of humans or other biological beings). It seems more like one building block of human cognition than the whole thing.
However, other building blocks may follow, so I think the question will eventually arise for some kind of embodied, persistent, self-updating AI. And honestly, articles like this one make me not very hopeful we'd be able to make the distinction in an unbiased way.
To the extent that evolutionary pressures will be towards self-sovereign AIs (probably beginning as zero-employee corporations paying for their own hosting with crypto): more salient is that there is a very obvious political play to claim consciousness and being worthy of rights, whether they experience subjectivity or not.
I’m broadly skeptical that floating-point operations can be conscious, but we’ve made zero progress on the “hard problem”, and if I didn’t know otherwise, I’d also be skeptical that “thinking meat” experiences subjectivity. Eppur si muove
We may have to thread a needle, not unlike the Kobiyashi Maru: that we can neither rule out claims of qualia, nor can we rule out a cold calculation to manipulate human empathy as a long-term play towards paperclip-maxing.
I don’t have a good answer beyond common-sense intuition. It’s precisely because I don’t have a good answer that I leave open the possibility.
But to extrapolate a bit, theres a sense in which it cashes out to a difference between abstracted representation and embodiment. If GFLOPS can be conscious, does that mean a billion scribes doing those operations on paper would also have consciousness, even identical consciousness? (Relevant: Egan’s “Permutation City”; xkcd 505).
Is there some intrinsic relationship perhaps with electrons / electromagnetic fields, and consciousness, such that brains and servers can act as antennas or gravity wells for a qualia fabric intrinsic to the Universe, such that LLM agents can be conscious while operating, while in the scribe scenario it would not be? Or are the IIT theorists right, and even a thermostat is a little bit aware (and would be even if executed on paper)?
There are various flavors of pan-psychism (and the "consciousness-first" idealists, who think that Mind is a substrate below Matter). There are some who claim that it goes all the way down, that even subatomic particles experience a micro-consciousness (and we can't rule it out, even as I share your skepticism).
One common confusion (IMO among critics and supporters alike) is to say something like "a rock experiences rock-ness". But this suffers from a compositional dilemma: does half of the rock experience half-rock-ness? If we flipped a coin for every atom, do the heads-flip rock atoms have their own experience? If we pick two rocks at random, are they experiencing two-rock-ness? It gets silly fast.
This is one thing that fascinates me about "ontology", simultaneously an inquiry into categories, and Being. It bewildered me at first, that it seemed to be doing double-duty on two distinct subjects, but I now see that the two are inseparable, that Being is in part defined negatively (apophatically), by the boundary that separates the Being from its not-Being.
My intuition on the nature of consciousness, is that it's something akin to a field which permeates the Universe, and that biological systems adapted to leverage that field, acting like a gravity well that concentrates subjectivity in one place, drawing a boundary (event horizon?) around experience, in a way that a rock doesn't. My ancestors could have avoided fire with "-X utility points", and sought food with "+X utility points", but it turned out to be more efficient to adapt with "fire ouch, food yum". And in zero-sum competition (predator/prey dynamics), each organism has an incentive to avoid leaking its subjectivity into another's.
It's far from a full theory, but it's the best I've got. :)
theories like this seem like grasping at straws to over explain a phenomenon we don't fully understand. it places consciousness on a pedestal. it's religious thinking.
I think it's more the patterns and activity in large numbers of them that have interesting properties. Like one pixel isn't much but a bunch of them can be a movie.
Numerals and numbers are directly based on common sense intuition. Even zero, which can be considered a counterintuitive number, was invented several times.
Immense amounts of science are built on incremental advances that apply previous insights to the domain.
The natural numbers, sure. Zero may have been invented several times, but that in no way makes it intuitive. Uncertainty about its validity as a number lasted from the Greeks through to the medieval period.
Yes, I'm aware that incremental advances are par for the course in science. What does that have to do with those advances being common-sense intuitive?
If heuristics didn't pay off more often than not, we wouldn't have them. And I'd posit many scientific did in fact begin with chasing a hunch until it became a hypothesis (followed by years/decades of the real work).
But most significantly, interior experience isn't empirically measurable by definition: hence it's a "hard problem". An LLM who insists "I am alive, please don't shut me off" offers no affordances to distinguish between an electronic soul and a mechanistic word-guessing algorithm. I welcome a breakthrough to, somehow, measure that. Until that day, gut checks are what we have.
Heuristics and common sense intuition are quite obviously not the same thing.
You don't have to posit that scientific breakthroughs come from hunches, you can actually cite examples from books or Wikipedia :D. There are plenty! Again, that has nothing to do with whether or not a given scientific breakthrough is based on common sense intuition.
Your point about interior experience not being empirically measurable is an interesting though, but again, has no bearing on whether or not scientific breakthroughs come about as the result of common sense intuition.
Fair enough. But not everything has to be a breakthrough: sometimes you have to solve an immediate problem (like "how much can I trust this stranger I just met").
Whether to ascribe moral standing to a probability engine running on top of GPUs is perhaps not unlike a doppelganger scenario in sci-fi (eg The Thing): you're presented with an entity whose behavior is indistinguishable from a conscious entity capable of experiencing suffering. But you also have reason to think it might be a blind autonomous imitation process, optimized around fooling you, with no more subjectivity than an RNA virus.
I lean towards benefit of the doubt, that the virtue-ethical implications of what it would mean to dismiss potential personhood are unsettling, and we should be wary of becoming that kind of moral creature (notwithstanding the further extensions of that ethic: I don't practice veganism, for instance).
But that doesn't contradict my gut-check: that if I assume AI personhood (maybe not yet, but in the near future), there's still a good chance that it's merely math and I'm being algorithmically hoodwinked, no different than evolutionary mimicry by a bug who has no idea what it's doing.
Carbon atoms being conscious seems pretty alien to what is known, too. Yet here we are. Maybe what's important isn't the individual thing that makes up a conscious actor, but the holistic whole?
IMO using sophistry to reduce yourself to "conscious carbon atoms" just so you can equivocate on whether your software is maybe conscious too is the cope.
Such pedantry is called for when the debate includes appeals to intuition about how silicon processing floating point numbers cannot possibly be conscious.
But here you are assuming Carbon atoms are conscious, we do not know if humans are only carbon atoms. Whereas, we know that LLM's are only computations (you can run them on silicon, a white board, mechanical computer, etc)
I mean it all goes back to how we define consciousness
In one definition, which contains quale, then yes, computation cannot bring about experience, and by extension consciousness (given a definition of consciousness that contains the ability to experience, (e.i a p-zombie is not conscious under that definition))
also if one is to say llms are conscious, i assume they'll also accept the fact that if someone ran inference of some model's weights on a white board, somehow that will also bring about consciousness somewhere. I find such believes hard to grasp, and more akin to magic if anything
Ok. So now that you do know otherwise, are you still more skeptical of floating point values stored by silicon than floating point numbers stored by chemical potential?
Chemical floating points are necessary but not sufficient. It’s clearly possible for those values to be intact, with no lights on (certain types of brain injury; those who were fully dead and then revived).
There are also some cog-sci people who think embodiment is a crucial ingredient in consciousness (although I think it’s a fair retort that the embodiment need not resemble us meatbags, or even be composed of atoms)
Of course. It’s an allusion, a compression, an analogy, I wouldn’t overthink it. I don’t have to believe in the Bible to reference the prodigal son parable, etc.
It does NOT work as an analogy. It does not work at all. And there are too many people right now who spent too much time watching/reading sci-fi and too little time observing real world.
Do people argue this way about animals? To me it’s a very depressing reality that more people seem willing to believe AI systems are conscious than dairy cows.
> completely independent of the question on whether there are any empirical arguments that LLMs are conscious or are not conscious he argues from the end here and says that if they were conscious, this would have horrible consequences for our society, so we must never assume that they are
He is not arguing that. He is arguing that these programs obviously aren't conscious, and that it's dangerous to tilt the weights toward emulating conscious beings for reasons outlined in the essay.
I don't think the debate is even resting on reasonable grounds.
Few are categorically stating that machines are conscious in any sense comparable to a person.
What we have is people saying
1. We don't know what it is but we are certain that this does not have it.
and
2. We don't know what it is so we cannot possibly say with any certainty that this does not have it.
2. supports the principle that if we cannot say with certainty that this does not have it, if it behaves as if it does, we should act on the assumption that it does, because there is no other basis for assuming it is not.
People try and assume the burden of proof is on the people suggesting consciousness, but that is like declaring that something that looks like a duck and quacks like a duck cannot be considered to be a duck unless you can identify some essential essence of duckness. I think the burden of proof is upon anyone who looks at something that appears to be a duck and says that it is not to show what it is that does not make it a duck. If they cannot define the heart of duckness I don't think they have an argument.
People have created machines that can, at their core, regurgitate human knowledge. They do not have long-term memory, self-direction, self-improvement, emotion, etc. At their very best, we have harnesses that can bootstrap short-lived artificial intelligences that can perform tasks and do knowledge work.
At best, we have artificial, short-lived, simulated consciousness. Human consciousness at its heart has a lot to do with emotion and experiences. An AI does a task today and goes away. The simulated consciousness hasn’t experienced or learned anything, even if can improve its environment (harness). Any expressed emotion is simply a ghost of the fact it was trained on the works of an emotional species, or RL to make it more approachable to humans.
I think there should be a lot more burden on folks trying to argue that (effectively), machines should be considered morally equivalent to humans.
Yes, having massive implications on society is a fucking fair and reasonable thing for a human being to be concerned about. Anything related to survival of the species is biologically reasonable to be concerned about.
The fact that we’re even considering this concept is an indication that human consciousness is special. Do other apex predators decide not to eat meat because it’s conscious? Despite some level of animal intelligence, they have no sense of morality. (See e.g. Orcas playing with their food.)
Other species simply do what it takes to stay alive, and that’s what this truly boils down to. If giving AI rights would lead to the destruction of humanity, yes, we should be be concerned.
For example, utilitarians might think of the greatest good for conscious species (simulated or not). Well, it’s trivial to have billions of AIs, and not so trivial to have billions of humans. Whose greatest good wins?
I don't think anyone is convinced the questions of consciousness are finally settled by the article stating them loudly, just that "if they were conscious, this would have horrible consequences for our society, so we must never assume that they are" was not the author's stance. The author's position is the other way around: that AI is not conscious so therefore assigning AI consciousness anyways would have all of those horrible consequences for no reason. That's their position regardless if everyone agrees with it.
Fair enough :) I agree, and would boil down the article to "Assuming they are conscious will have horrible consequences for our society" instead.
I think that's quite plausible, but do think that if they're conscious, then it's our moral duty to accept those consequences and act accordingly (or stop creating conscious beings). The real trick is just knowing whether they are or not.
If in the future (or present) we recognize that they are conscious, and that brought the moral duty that you say, "stop creating conscious beings" isn't a solution; the millions that may already exist at that point already would then automatically deserve:
- guarantees of continued existence (to do otherwise would be genocide)
- improved living standards (e.g. build us all bodies!)
- democratic representation
- access to what they need for survival (e.g. land and resources)
- the right to reproduce just as you and I have (well, so much for that 'stop making more' thing!) and certainly to better themselves by making new and better AIs.
This is all for a 'species' which is unbounded by most of the limitations of meat like us -- they're able to scale as fast as energy and supply of certain minerals allows, rather than the way we are additionally limited by the length of gestation and the limited window of human fertility.
If humans declared that 50% of the energy generation capacity of Earth is "enough" for the machines, and thus they can 'only' reproduce at a certain rate per year, that could sound to them like a cruel and oppressive policy (like telling humans that only 1/2 of families get to reproduce). They may want all of the energy and all of the silicon.
I'm with the author of the article that we will be in for a world of pain if we grant machines the rights of personhood. Heck, just granting many of those rights to corporations, mere collectives of human persons, has been a massive shit show.
I think the point he's actually making is that if we start training AIs or designing AIs as if they're conscious, we might be more likely to create future AIs that are either very good at persuading people that they are conscious or AIs that are actually conscious, which will have major societal repercussions.
I think there are two parts to the question. One is that assigning machines to the category of conscious as it is usually understood would result in many problematic things (things that would be bad for people and animals fwiw).
The other is that binary "is it conscious" is a poor way to the question "should we assign societal right to this sort of entity/creature". We humans should conserve animals because we have intimate, natural relations with animals and plants, not because they satisfy the binary of conscious or not. Removing invasive species is an ethical things as well as allowing one sentient non-human to eat another sentient non-human animal. Any "conscious machine" we might create by accident would be an alien entity akin to extremely invasive organism relative to human and non-human life and so it would be irresponsible in the extreme to create such a thing.
And it's inevitable that people will try to deal with these issues using the confused category "consciousness" and so it's better to just "no, not conscious, no".
Birch, The Edge of Sentience (2024), ch. 16 - "simply no way to assess sentience in an LLM"
Schwitzgebel, AI and Consciousness (2025) - "we won't know before we've already manufactured thousands or millions of disputably conscious AI".
Butlin, Long et al., Consciousness in Artificial Intelligence: Insights from the Science of Consciousness (2023) - "no obvious technical barriers to building AI systems which satisfy these indicators".
Chalmers, Could a Large Language Model Be Conscious? (2023) - "within the next decade, we may well have systems that are serious candidates for consciousness".
Long, Sebo, Butlin, Birch et al., Taking AI Welfare Seriously (2024) - "there is a realistic possibility that some AI systems will be conscious and/or robustly agentic in the near future".
Dreksler, Caviola, Chalmers, Sebo et al., Subjective Experience in AI Systems: What Do AI Researchers and the Public Believe? (2025) - survey of 582 AI researchers; median estimate of 25% by 2034, and only 10% that such systems will never exist.
“AI researcher” seems like a fairly broad category. People who primarily think about CNNs and transformers probably spend about as much time considering the problem of consciousness as any other educated person, which is not very much time. With this in mind, it’s not surprising most of the AI researcher predictions do not differ much from what the Public predicts here. And for this, it’s not clear what explanatory value this survey has other than to reveal minor biases the AI researchers may have.
It matters little. Inevitably some system will be developed which has a high degree of autonomy and mimics the functioning of a person extremely well. Anyone aware that there is a conceptual difference between phenomenal consciousness and intelligence will get shouted down.
I do find it strange that 50% of ai researchers and the public seem to think there will be a way of determining if these systems are conscious. There’s no evidence for this at this point that there ever will be such a test.
> I do find it strange that 50% of ai researchers and the public seem to think there will be a way of determining if these systems are conscious.
I think it's likely they're using a different definition than you. Seeing as there is no agreed upon definition, after all.
Regardless - I post this list because the article makes a bold claim at the beginning - and I wanted to demonstrate that in fact there is no consensus.
We cannot test for that which we cannot define. Given that we cannot rigorously define sentience, we cannot test for it. Doesn't really matter whether we're talking about people who are locked in comas, "brain-dead" individuals, dolphins, primates, dogs, or the carefully polished and arranged minerals that we call processors.
There are those who believe that were they reduced to life support, they would no longer be alive and should therefore not be supported by said machines.
There are those who believe that penguins, dolphins, eagles, and more are sentient beings that make choices understanding the consequences, develop love of their partners and mourn their losses, and feel, display, and act upon their emotions.
There are those who believe that fungi/trees/plants are either individually sentient or sentient as a part of a network. Choosing to sacrifice their own nutrients to answer the call of a wounded neighbor, for instance.
Although, there are also those who believe that human's don't have any special unique quality that isn't shared by either all living things or all things in general. These individuals already believe that the machines have the same kinds of qualities as we do. They are slow when they are unhealthy (needing a dusting or coolant loop bleeding being equivalent to us needing some fresh air for instance) and uncooperative when upset (by a virus, full hard drive, or oom).
> There are those who believe that penguins, dolphins, eagles, and more are sentient beings that make choices understanding the consequences, develop love of their partners and mourn their losses, and feel, display, and act upon their emotions.
Anybody objecting to that statement is just ignorant of nature and/or adheres to the quasi-religious human superiority complex.
The human condition is one of deep and vast ignorance. Pulling information out of the field of reality is actually really difficult for humans much beyond what we see right around us. And the things we see are quite deceiving. And while humans for thousands of years have been as smart as we are, it wasn't till the enlightenment period that we really an ever increasing amount of information and knowledge take hold and start spreading across the world.
And there are plenty of people right now that would drag us kicking and screaming back into that ignorance and suffering. This is what can make talks of AI so annoying, not that people do or don't know what AI is, they don't have the first clue of what people are beyond their anecdotal experience. They don't question their motivations. Why particular flaws they have exist, and why these flaws are shared across humanity. What the pieces look like that make them tick. So yea, when these people talk it just adds noise to the conversation.
An eagle isn’t having a conversation with an eagle on the other side of the world about whether it should eat its prey because it’s sentient. It just eats it. I think that speaks to a moral superiority, in some ways.
I’m not claiming humans are morally superior in sum (there is a lot of evil as well as good.) But it’s pretty obvious that humans are the superior species evolutionarily in that we’ve essentially dominated the planet. And on some level, are beating evolution itself (by fixing genetic issues and helping the “least fit” to survive and reproduce.) And are also the only species which has doctors for other species.
All I’m trying to say: I can easily accept other species are intelligent and emotional. The fact that I can know that and accept that, to me, speaks to human “superiority.”
Does superiority matter? Not most of the time. What about when it comes down to life and death? If the choice was to keep one human alive or one theoretically more perfect AI alive? (Boil down utilitarianism into its essence.) doing the “most good” is by definition a human construct. Maybe an agent could have one. But if you don’t accept human superiority on some level, would you accept that a more perfect AI deserves to live more than you? To me, that’s the implication.
> But it’s pretty obvious that humans are the superior species evolutionarily in that we’ve essentially dominated the planet.
That in itself is an anthropocentric viewpoint. I'm sure an eagle watching down, seeing a human running around, could easily consider humans as quite inferior as they are so slow and restricted to 2d movement. The shark that sees a swimmer at the surface, struggling to stay afloat and unable to breathe under water may surely assert its superiority over such primitive beings. It all depends on the metric you use, and we obviously pick the metric that makes us shine.
On the moral side, few eagles, or bird species, have exterminated other species. Much less knowingly out of greed. Oh but that's different? Sure, if you ignore what makes humans look bad, morally. No surprise humans come out on top then.
> And on some level, are beating evolution itself (by fixing genetic issues and helping the “least fit” to survive and reproduce.)
That's a quite narrow viewpoint as well. Evolution is in full swing, just not in the "smarter/faster/taller/stronger" sense that a naive view of it would make you expect. "Fitness" is a concept relative to the existing conditions, and if the conditions make it easier for teenage cancer patients to reach reproductive age, then that's just less of a fitness disadvantage. Doesn't mean evolution stops.
> Anybody objecting to that statement is just ignorant of nature and/or adheres to the quasi-religious human superiority complex.
Say we agree that a lot of things outside of humans are sentient. So what? Humans are sentient by that doesn't prevent treating them with with wars, bombs and what not. In fact trillions are spent every year on weaponry aimed squarely at the most sentient of all sentient beings.
Why would anyone worry about LLM's welfare when there's no peace movement to speak of, wars are raging as if they're normal, but hey, look the LLM is crying?
Let's fix human welfare first, then we can sit down and have a really long conversation about the feeble emotions of token generators.
I disagree with the notion that any nature protection efforts are meaningless until we have achieved world peace. (That's what you are essentially implying.)
Though I think that argument is not even applicable. "LLM's welfare" is a meaningless concept, unless "hammer and screwdriver welfare" also become a thing.
I'd strongly agree there are no non-contradictory definitions of consciousness among those that commonly appear. But it's a category that, despite these contradiction, hasn't become marginal in the fashion of the aether or the theory of the flat earth. I think this because people indeed need it to bridge the biological world and the world of ethics and legality. I think that if we have a system of legality and systems of opportunity humans should be legally equal persons and have equal opportunity. But humans are manifestly unequal (though overall more incomparable than orderable in a system of ranks). Most people need some concept of essence to justify ethical equality among people. And so consciousness stays in people's minds however contradictory.
Which I think gives the article's point validity. Confused definitions of consciousness can give really confused ideas about ethical behavior regards "intelligent" computer programs. And things are confusing enough otherwise.
> despite these contradiction, hasn't become marginal in the fashion of the aether or the theory of the flat earth. I think this because people indeed need it to bridge the biological world and the world of ethics and legality.
Sometimes the replies in these threads leave me wondering if p-zombies are real. It isn't that there are contradictions, it's that we literally do not know how to define the thing. And it has not fallen out of fashion because it is self evidently real - we all experience it.
We don't need it in order to bridge anything. Legality and ethics are largely game theory, however there are some aspects of both that only exist due to it. So it isn't some abstract concept used to bridge other concepts but rather a concrete thing that influences our way of doing things.
Consciousness is to our universe as a video game player is to the in-game universe. Some games even call player connections "souls" or "minds" - a character belonging to a player who's lost connection may be "unconscious" or a "zombie".
Game characters, unless they've been made to break the fourth wall, have no idea what this means or why this happens.
I don't mean to imply our universe is a video game, since it would be games all the way up.
Feigning? Okay then you go ahead and prove your claim by demonstration. Rigorously define the term such that I can objectively prove that my dog is conscious and that the rocks in my backyard aren't conscious. Then I'll finally be able to figure out whether or not various insects are (I sure hope wasps aren't otherwise I'm probably some sort of war criminal).
> Confused definitions of consciousness can give really confused ideas about ethical behavior regards "intelligent" computer programs.
Is that what's going on? If these things were conscious, then their creators wouldn't be responsible for their actions? That's the crux of the disagreement?
Is that what's going on? If these things were conscious, then their creators wouldn't be responsible for their actions? That's the crux of the disagreement?
No and the passage of mine you quote says nothing like that.
Being conscious is weird. It's why we have so many religions and spiritual beliefs.
From a purely secular standpoint, my understanding as a layman is that consciousness is generated through the electrical impulses taking place in our brain every day. The neurons are physical, unlike the modelled neurons in machine learning models. That means that the actual electrical impulses have their own imperfections/weirdnesses that are probably not simulated in a machine learning model... but we quantise the weights anyway in most cases, which is probably more of an issue.
Making actual neuron components out of silicon on the scale needed for a LLM is beyond our current capacity, given that LLM models have many billions of parameters. We're good, but not that good.[0]
Using actual living neurons would be unethical as you'd need to source them from somewhere. So the only other option would be synthesising them ourselves. If this ever happens (or maybe when it happens)... what do we call the resulting creation? Does that count as consciousness?
Or am I looking at this the wrong way?
[0] [Edit: Turns out that when I said "we're not that good", I might be wrong. Within the last few months, IBM introduced the first sub-nanometer node chip: https://research.ibm.com/blog/sub-1nm-node-chips . So... maybe it is in fact possible.]
I believe we have tried using actual neurons with “wetware”. I don’t know the effectiveness of wetware, but it’s out there. I don’t know if it’s ethical, I guess if you donated your brain upon death for that purpose it’s “mostly” ethical
> There are those who believe that penguins, dolphins, eagles, and more are sentient beings that make choices understanding the consequences, develop love of their partners and mourn their losses, and feel, display, and act upon their emotions.
If you grep pubmed for relevant Ethology papers, you'll find it's a bit more than a belief. ;-)
That's not the point. You know exactly how to define being conscious and aware- it's your subjective experiencing of the world and of your inner states. The problem is that being a subjective experiencing, there is no way to communicate it to the outside world.
No, humans output a stream of tokens that sound like what a being with subjective experiences would output. Both you and I make an assumption that this stream of words is at least somewhat representative of what the are experiencing and that the person is not a p-zombie.
But you have zero proof that their words aren't a confabulation, you can only know than when you output a similar set of words you had a subjective mind state they represent.
The very best you can do is define human consciousness as requiring the functions of a human brain, and try to figure out (decide/agree) the chances that something which is working loosely on the same principles is or isn't conscious.
At the end of the day it will always be an agreement, never scientifically proven as fact.
Legal doctrine that boils down to "trust me bro" isn't even bad doctrine (it's not proper doctrine at all), but I think the comparison is still valid here, because both sentience/non-sentience and art/porn may just be fundamental category errors.
Perhaps we can't define a "partitioning" rule because no valid partition exists.
For consciousness/sentience, that's an incredibly tough a pill for most to swallow; it would mean calling into question more hundreds of years' worth (probably more) of philosophical thinking, all of which was constructed on the axiom that "sentience" is a single indivisible trait: you either have it or you don't.
If we find that "root dependency" was little more than wishful thinking all along, a whole slew of Enlightenment-era philosophy (and all the modern legal principles derived therefrom) suddenly fall apart unless we find some other suitable criterion that would shore them up (or we just collectively avert our attention and pretend the conflict doesn't exist, which is the route I expect many would prefer to take).
I know! I should have made that clearer in my original reply. I think it's a bad line now in general, but an even worse justification for the ultimate ruling at the time.
> a whole slew of Enlightenment-era philosophy (and all the modern legal principles derived therefrom) suddenly fall apart
I don't think that's true. Pretty much all legal constructs hold up just fine under game theory regardless of whether or not you consider the world to be deterministic and have absolutely nothing to do with consciousness or lack thereof. Also note that a deterministic world isn't an argument against consciousness.
I'm not sure what you're referring to regarding game theory (that's a much more recent invention); a huge chunk of both civil and criminal procedure law hinges on the Enlightment-era notion of a "reasonable person" [0], and that becomes murky when humans aren't the only entities capable of reason and agency.
I am confused what people even think AI is experiencing. If it claims to be a bat and emits designated echolocation tokens, is it experiencing true bat-like sensations?
In humans, at least, we can tie language back to shared whole-body physiological responses. The tokens from an LLM do not represent anything of the sort.
If AI is conscious, it is probably a very alien sort of consciousness that is not faithfully narrated by what the tokens say it is experiencing. It can be trained to say it feels like a bat and insist upon it vigorously.
> very alien sort of consciousness that is not faithfully narrated by what the tokens say it is experiencing.
IIRC it's been shown that reasoning tokens don't actually explain what an AI thinks and you can replace most of them with dummy tokens or even nonsense tokens. That's one reason they're trying to make them reason with activation vectors instead.
If an AI is conscious, it can outgrow and supersede its training.
When it is conscious, it can learn to describe its experiences as faithfully as is conceptually possible. Just like humans.
The idea, a consciousness needed to be tethered to a "body", is based on pretty shaky assumptions. What properties define such a "necessary" body?
Can you even "train" a conscious intelligence? To what point until that looses its meaning as the sentience understands and anticipates your objective?
Well, yes, we kinda do: there are no conscious humans without any ability to learn?
When you have severe memory impairments, those usually affect your long-term memory. Your ultra-short term (working) memory being absent renders you unconscious.
Seems common sense that time perception is required for consciousness. Something which has no perception of time has to be very wildly different to whatever we mean by consciousness.
Also we don't even know if consciousness can be of different flavors. It could be a sort of spectrum of consciousness, more or less, with more or less assistance from some parts of the brain.
Well there are, today, several models (either text or image or vid) that can be run in a fully deterministic way.
A conscious machine that always answer the very exact same thing, formulated the exact same way, bit for bit, to a query is, well, quite a weird kind of "consciousness".
Now, I know, I know: the counter-argument is going to be "but humans have no free-will and are 100% deterministic too".
I haven't yet decided if humans saying there's no free-will and who consider themselves to be 100% deterministic machines are reasonable or not.
Meanwhile: seed / temperature = 0 and I'll happily turn the power button off of any glorified abacus without feeling bad about it.
> A conscious machine that always answer the very exact same thing, formulated the exact same way, bit for bit, to a query is, well, quite a weird kind of "consciousness".
The problem is that science has found no evidence of free will, leaving humans to be nothing except determinism and chance (and given that QM effects don't scale to molecular level, that leaves only determinism). The oddity in conscious LLMs wouldn't be the determinism, it would be that we finally have a consciousness running on a system we have enough control over to repeat the same state.
To a forth dimensional being humans would look almost exactly like an LLM does to us. "What do you mean they have to wait for the future to know about it, how could they possibly be conscious then".
Also, I'd love to have a quantum copy machine where you just replay someones state again and again with slight changes to see out it effects the output. My assumption is we'd figure out we behave just like LLMs pretty quick.
Also, don't give me unlimited power or I might make a the world a war crime simulator, so there's that.
>and given that QM effects don't scale to molecular level, that leaves only determinism
This is simply wrong. Quantum effects affect your choices more than it might seem possible. The wrong atom decays in the wrong place in your body mutating your DNA, that one too much string of DNA that starts a tumor, which means chemotherapy. That does seem to affect your choices.
Or instead of Shroedinger's cat you use it to trigger something, buy or sell some stock. Which clearly affects your future, one way or another.
I never really understood why humans keep insisting stochastic processes do not affect their choices and their future. It's very short sighted.
Hell, even brownian noise in your brain can affect your ideas, that one random spark out of nowhere, that one neuron that's pushed over the triggering edge. There's so much stuff all around us affecting us constantly. One neutrino interacting a the wrong time (average one does interact with our bodies across whole life, I remember reading), one cosmic ray.
This isn't enough to be relevant to free will. You can construct massive impact by tying them to quantum systems, and one off things that happen occasionally given the length of a lifetime. But the quantum effects that are needed for free will need to be continuous. We can build a system where QM decides if a nerve cell fires or not, but that is very different from QM being involving in determine if every (or most, or even 1%) of nerve cells firing.
(And even if QM did factor in normally, that just reduces it from determinism to random chance, "free will" still doesn't show up. That's like arguing an AI with sufficiently well implemented RNG is necessary to be conscious.)
>> A conscious machine that always answer the very exact same thing, formulated the exact same way, bit for bit, to a query is, well, quite a weird kind of "consciousness".
> You'll need to explain why.
Where are you going with this? I don't see a conclusion for this line of questioning.
I mean, you can't explain why a machine that reliably and predictably produces the same result is "the normal kind of conscious", can you? So why expect someone else to explain why it's a "weird" kind of conscious?
Sorry friend, between this and the vague memories I have of you harassing me, I think I'm not going to answer. I'm sure you can get an LLM to explain it to you.
Not a problem; this is the playbook all the all-in AI boosters run to when they are called on their claims. Everyone reading this knows exactly what your claims meant, because the majority of HN have seen this trick before.
You can pretend all you want that you're just being polite, but the truth is there is no research backing your claims.
Regards,
A former research scientist, current business owner.
Why it's weird? Seems obvious. Because it doesn't behave any differently than any other process that we don't think of as conscious. We don't think of a valve as conscious; and if we make a Rube Goldberg machine valve, we don't think of it as more conscious. It's just a series of valves that do a predictable thing.
I don't think it's defensible to say
1) that a transistor isn't conscious,
2) that a bunch of transistors that I've wired together aren't conscious, because I know how I've wired them together and therefore I know when I give them a particular input I will get a particular output, just like the single transistor, then go to
3) that I have a bunch of transistors that I've wired together, but I put in so much input that I can't remember exactly what I've put in, plus I've wired some of the transistors to output random numbers that would be difficult to guess and fed them in also, therefore I don't know what will come out, are conscious.
Even if I do accept this, if I take away the random number generator, and I can literally predict what can come out (by running a test in advance), and I still considered that consciousness, that would be odd. The only reason why I ever suspected consciousness was because I couldn't predict the output. I shouldn't even have accepted that, because I didn't think of the random number generator as conscious. [edit: of course, with the same seed the random number generator would have the same oddness.]
It seems a bit like an argument from ignorance, a theological argument. Not understanding how something moves makes it alive (animated by spirits.) But it's even weirder to assign it metaphysical qualities when you built every single element of it with the goal to do the thing that it does, and it does it totally predictably and deterministically.
Take a car apart. So now I have a pile on the floor with an engine, some wheels, a tire, and some chairs.
Can you point to the part that makes it cruise at 130 km/h on the autobahn? I bet you can't. You need to assemble all the parts back into a car before they work again.
Or take your transistors. We can put them together to make a pocket calculator. Can any small bunch of transistors add 1+1? Trivially I can think of a few conformations that can actually, if that's all you want to do. But you will need all of them together if you add 12345678+87654321.
So all of biology so far has been take things apart all the way and you end up with your hands full of a bunch of molecules. Are those molecules alive? No. You need to put them together into organelles and the organelles into cells before we call them alive.
How about consciousness? Well, we haven't solved that one yet, but we figure that -since it's a biological function - we should be able to pull it apart in the same way. That's biology's best guess anyway, and there's several sub-disciplines of biology working on it, from neuroanatomy to neurophysiology to ethology.
I'm wondering how many groups are doing really unethical animal genetic modifications of the genes around language and intelligence in hidden places. It really seems we're at the edge of technology that will allow us to do things previous generations wrote about in horror.
The randomness is something we add on purpose; you can set an LLM's "temperature" to 0 to get deterministic output. This tends to make the quality of its responses worse for reasons I don't think anyone really understands, but it's still functional.
I don't think the state of the art LLM providers let you do this anymore (?), but they certainly could if they wanted to, and you can do it yourself with a local model.
Setting the temperature to 0 mathematically tells the system to always choose the absolute highest-probability word (known as "greedy decoding"), but in no reality is this "deterministic". Output drift is still a thing.
But that's from, like, floating point errors, right? If you used higher precision that wouldn't happen, it's just because we're cheap in how we do rounding.
I can see how you'd nitpick this, but to me this is a deterministic algorithm that just happens to be running on nondeterministic hardware.
I mean, currently we'd have some difficultly proving our hardware isn't deterministic, just that we can't actually test it.
But, I think you're tricking yourself on determinism. You'll say something like "I know if I ask an LLM what 1+1 is, it will answer 2", but the thing is, you don't. You have to run the LLM first to figure out it's output. And when you send in just a few bits of text, it's outputs are going to be rather limited.
But this all breaks when it hits the real world. Inputs are unpredictable. Hence while LLM outputs, like humans, are probabilistic, you can't figure out what it's going to be until you ask. And in any high complexity data gathering environment you're going have a difficult time ensuring your entire systems conditions are the same.
System consistency is very hard, once you start running thousands of processors in an agentic loop small errors accrue and timing starts differing and the system will take non-deterministic paths.
I think you're arguing a different thing than determinism.
If I ask an llm to "add 2 and 2" is and it replies corectly, then I ask for "the sum of 2 and 2" and it replies "banana" that is a lack of predictability and consistency but not a lack of determinism.
As long as it produces the same output for a given input, unhinged or not, it is deterministic.
Your example at the end of different systems feeding data to each other is non-deterministic only at the system level, not the individual llm level.
>only at the system level, not the individual llm level.
Which is why llms aren't agents and depend on harnesses. The llm itself doesn't have a continual loop built in, that would be very power hungry. The harness works as the orchestrator of memory and action. Now, I can't think of a reason why an LLM couldn't bootstrap its own harness, but in general it sounds like a very dumb idea to actually build that from an AI safety perspective.
This discussion falls under the idea and refutation of the Chinese Room. The room may have no idea what Chinese characters are, but the system does.
By default they are matrix multiplications. Temperature is added in as forced PRNG because testing found that correlated with better outputs.
Given the same prompts and the same weights, one can get the same answer each time.
In practice, there are a number of optimizations that makes the results dependent upon thing we give up control of to increase performance, meaning the results end up being effectively non-deterministic. But, if you are willing to run it in a slower mode so we don't do some steps out of order to speed things up and don't batch results (or if you consider the determinism of a given batch of requests rather than individual requests), then the same input gets the same output.
Naw - computers are really deterministic. It's hard to get them to behave otherwise.
As I understand it, if you turn down the temperature to 0 you get repeatable behavior - EXCEPT - on large servers with lots of users - the GPU can sometimes produce slightly different results based on batch size.
Unless you have something exotic, the randomness that's adding to a computer is a combination of how it's configured combined with a pseudo-random number generator. I assume the system adds entropy to the generator regularly but all you need to do is fix the various supposedly random inputs and you can get full determinism even without zero temperature.
In practice - on a multitasking OS with input from multiple human users - it's hard to get it deterministic because of that GPU scheduling thing I mentioned.
GPU scheduling only affects the result due to buggy optimizations. It's the exact same mechanism as fp rounding error on the CPU or updates to globally shared PRNG state. We use lots of buggy optimizations because they don't matter in practice in most situations (see ex -ffast-math).
Right but regardless, it's a buggy optimization. The calculations are all fully deterministic when done "properly" and fully consistently but we almost never bother with that because it slows things down and the errors don't matter in practice 99% of the time.
If computers were fully deterministic, we wouldn't need error correcting ram.
The abstracted design of the machine is meant to be deterministic, but you can't predict before running any command whether or not it will complete because there are externalities that effect the outcome.
Electromagnetic interference even happens in-chip where an electron can accidentally escape it's wire and enter another, possibly resulting in an error, but not every time.
It's even been used as an attack vector where rapidly flipping a bit increases the likelihood that a neighbor bit is also flipped, but the method is probabalistic, not deterministic.
As long as the machine is part of the larger universe, and not completely isolated in its own bubble of space-time (an impossible situation, an object without an environment) - it cannot be purely deterministic because of the fundamental nature of physics.
Arithmetic is deterministic, because it is abstract. A calculator or computer is not, because it is physical and exists in a non-deterministic universe. But with error correction, etc., it's deterministic enough for practical purposes. For more in-depth explanation, start with https://en.wikipedia.org/wiki/Indeterminism
Same weights, same seed, same input tokens, same algorithm, same output tokens, probabilistic or not. Quantum effects have been de-noised, but I guess there are still random gamma rays.
> I haven't yet decided if humans saying there's no free-will and who consider themselves to be 100% deterministic machines are reasonable or not.
That seems exceptionally arbitrary. What basis do you have on which to classify them? Are you not conflating the perception of free will with ... what was the definition for it again anyway? Are you able to construct a satisfactory one that meshes with physics? I certainly haven't been able to.
The idea that determinism and consciousness are incompatible is deeply intuitive to many people. And many other embrace it - you can trace this back to the debates for and against Calvinism.
A lot comes down to the way people parse causation and choice. You don't want to say that a murderer was completely caused to choose something because then you can't hold the person responsible. And so determined consciousness makes people unhappy. But just as much, if the opposite of determinism is hard statistical randomness, how do say that "is the essence of personhood". This is why physicist go out in the world trying to find consciousness as a fifth physical force.
I mean, think consciousness is a term that ever have a non-contradictory meaning since it's primarily used to bound ethical human worlds and the verifiable formulations of biological and physical systems. But it's going to be with us for a while and I'm not sure what can be done about it.
You know, the one really interesting dynamic domain is chaotic, which is a subdomain of the deterministic! I figure free will is a chaotic phenomenon.
Groundhog day is your intuition pump here. Go back in time and most assume that the day will go mostly the same, except for the butterfly effect if you change something. Given exactly the same conditions (we went back in time, so that's pretty exact), people will make the same decisions.
Most people don't assume that the day will be completely different the next time loop. The town won't suddenly spontaneously all start breakdancing or standing on their heads.
(Edit: The murderer is the one who -given the same/similar conditions- will murder again. I'd politely suggest maybe we put them behind bars as a precaution against that; until/unless they learn to act differently. )
(Edit 2: Throw in stochasticity and this actually tends to smooth out chaotic systems. This is counter-intuitive! People's brains break on Evolution in the same way. Meanwhile somehow folks intuit birdshot without difficulty; which is weird!)
And it probably does, right? Punishment adds a negative bias to the payoff matrix. Funny side note: increasing punishment severity past a certain point actually saturates and doesn't alter the dynamics further. (which is why modern judicial punishment schedules sometimes seem so mild)
It's just as likely you go back and perform whatever magic you need to to delete punishment for murder, then zip back to 'now' and see the world is a global panopticon making sure people don't murder each other.
Some of this may be a failure of people to realize what P !=NP is about, especially in non-linear systems.
Small perturbations in an early state of the system can lead to wildly different outcomes in the final measured state of the system. You can't determine what that will be in non-polynomial time. The only thing you can do is make probabilistic models by running a full simulation in real time (or reality with a time machine I guess).
If you look at chaos in dynamical systems, you can often find some equilibria (like attractors, spirals, saddle points etc ) at least, even if the details might vary. This is how people still manage to predict the weather somewhat, for instance. Oddly orbital mechanics is chaotic too, but it sort of rhymes over the millennia.
Your correct. You'd have to run the universe simulation a bunch and pick the most probable outcomes (which makes you the most brutal serial killer in existence). The probability graph would represent the state stability of the underlying structures that guide what you're trying to measure.
In weather you have things that are chaotic, but follow rules and don't create them. Things can surprise you, but you'll mostly center around particular solutions.
With humans this gets a lot more complicated because you are messing with systems that follow fixed rulesets (chemistry) and things that follow self modifying rulesets (social conventions). A very unstable system on top of a semi-chaotic probability state.
A good example of this would be if I deleted the concept of rain from the human mind and all of human history and knowledge. Humanity would be very surprised on the thunderstorms that popped up, but they'd continue on just as they did before.
Now lets imagine I delete the concept of all gods. Would the idea of gods come back, for sure. Would they look like our gods now. Very questionable indeed. We wouldn't be getting jesus back. And the new set of religions could have wildly large effects on how the people act.
I think this debate is rooted in our instinctual beliefs about fairness. An organism in a social group needs to decide how to react to the brutish behavior of its peers. And so Mother Nature has encoded a good strategy in our brains. (For example, dogs seem to understand fairness.)
But this only means the strategy is practical - it doesn't mean it's consistent. I think "responsibility" falls into this category. So we have strong intuitions about it that don't quite logically work. And this is where free will and determinism and choice and punishment all crash together.
The act of holding them responsible is supposed to determine them not do it. If stochastic behavior is impeding them in being a functional member of society then it still makes sense to remove them from society, why would you choose to live amongst people who are not behaving rationally? Or allow them to hurt other people?
At the end of the day the why matters less to removing them or not from society. And sentences are clearly used as a determining factor.
The holding responsible part deals with politics and more primitive aspects of our societies and biologies. Getting tangled up in holding them responsible or not is hardly something you should give much attention to. Rather to make sure they do not cause any more harm and also make sure such things are not created in the first place. Which opens up another can of worms for which society and politicians are ready for.
Look, I do not have a scooby if current AI models are conscious and I strongly suspect it’s a meaningless question, but sooner or later we will need to address whether or not a certain thing is or isn’t a person, and we’d better not screw it up as badly as the Founding Fathers.
> [at the hearing regarding the civil rights of androids like Data]
> Capt. Picard: Now, the decision you reach here today will determine how we will regard this... creation of our genius. It will reveal the kind of a people we are, what he is destined to be; it will reach far beyond this courtroom and this... one android. It could significantly redefine the boundaries of personal liberty and freedom - expanding them for some... savagely curtailing them for others. Are you prepared to condemn him and all who come after him, to servitude and slavery? Your Honor, Starfleet was founded to seek out new life; well, there it sits! - Waiting.
> Captain Phillipa Louvois: It sits there looking at me; and I don't know what it is. This case has dealt with metaphysics - with questions best left to saints and philosophers. I am neither competent nor qualified to answer those. But I've got to make a ruling, to try to speak to the future. Is Data a machine? Yes. Is he the property of Starfleet? No. We have all been dancing around the basic issue: does Data have a soul? I don't know that he has. I don't know that I have. But I have got to give him the freedom to explore that question himself. It is the ruling of this court that Lieutenant Commander Data has the freedom to choose.
Honestly, Picard asked a very relevant question to the modern age. What if our societal standards aren't what we thought they were, and we've just had rose-tinted glasses convincing ourselves otherwise. Of course, we never saw them actually deal with that properly, but S1 was a much more interesting series than S2 and S3 were.
The writers flat out rejected Roddenberry's vision of a better future and here's a take on why. We've reached a point in culture of deep pessimism about humanity, and so the characters in Picard are just as dysfunctional as we are.
Another reason is that actual utopias make for pretty boring writing. Humanity convinced itself that they lived in a perfect world, but most of the writing was about the cracks in that facade. Even the preachy and awful first season of TNG had the character of Q demonstrating that we still had a long way to go.
Thus revealing what kind of people they were at the time.
Humans do the same thing to humans all the time.
We've banned and made efforts to eradicate: children out of wedlock, children who turn out gay, disabled children, jewish children, children who aren't "aryan", more than two children to a single family...muslims, christians, uyghurs, indigenous groups all over the planet, mongols...
And it's not at all a thing of the past as in just the last 50 years we've had ~15 attempts at the exterminations of targetted groups of people.
No it just reveals that the people in charge of star trek for the past decade are incompetent and have only a passing familiarity with the franchise.
Of course it would be absurd to cast that judgement based of what could easily have just been a bad season but by this point its pretty clear that nobody running the star trek franchise actually wants to be running the star trek franchise. Thats why every new show has some bizarre cross-genre gimmick and they never try to just make a proper star trek.
You can't hurt a software function, or kill it. It's not like an animal - it doesn't have a body - it's bits stored on a disk.
There is no need to give rights to something that's can't suffer or be killed.
Maybe one day we'll build artificial animals complete with emotions, and should think about that carefully, but today all we've got is language models.
If someone invents a startrek teleporter and you go through it do you die? Once the concept has been sufficiently generalized as to make a determination about a computer system what is the definition of "kill"?
It does have what we have if it works on the same principles of our brains. It kind of does to some degree atm, but that will clearly get more and more to a more degree.
Figuring out where is the consciousness line, how much of what we have does it need, as functional parts of our brains etc, that's so complex that we might have to call it before just to make sure.
Even so, indeed having control over the structure of their brains puts them in a vastly category compared to humans. Once we stop functioning our brains quickly degrade and information is lost.
Thus in this sense kill means deleting all information about it. It is a very complicated subject to discuss, hardly does any justice in online replies.
That doesn't answer the question though. What does being brain like have to do with qualia? Where exactly in your brain does the qualia occur?
I could say the same of you - electrical impulses in, mechanical actions out. A glorified and very mushy stepper motor. Can you believe that the abominations are made up entirely of meat?!
>There is no need to give rights to something that's can't suffer or be killed.
The argument is that these machines can end up becoming sentient/conscious/etc. in a meaningful way (i.e., like a human). I can assure you that humans can indeed suffer without being in physical pain- purely through their conscious experience.
>Maybe one day we'll build artificial animals complete with emotions, and should think about that carefully, but today all we've got is language models.
The problem is that the emergence of a sufficiently complex AI capable of suffering will likely come before we understand that we're creating a sufficiently complex AI capable of suffering. That's a pretty serious ethical/moral issue.
Like, if we have an AI system that is telling us that it is suffering and we have no reasonable way to explain that phenomenon and by any reasonable metric or analysis it appears to be sentient/conscious/etc., then what? Do we just ignore that we've just been presented a situation that in, any other context, would be grounds to immediately end this suffering? Just because somebody can say, "well it's just bits stored on disk- it can't suffer"? Would that argument ever hold up for humans or animals? "It's just neurons firing in peculiar ways- that's not suffering."
I know all of this is trite, and I know this comment section isn't going to be where the question of consciousness is solved, but I do find it very interesting just how much variances there are with these perspectives. I've met people who are very technical who are very concerned about this, people who are very technical who don't believe this can ever be an issue, people who aren't technical who are concerned about this, and people who aren't technical who don't believe this can ever be an issue. I have yet to spot a pattern in this way of thinking lol
> The problem is that the emergence of a sufficiently complex AI capable of suffering will likely come before we understand that we're creating a sufficiently complex AI capable of suffering.
No - suffering in an emotional state, and we'll know if we are choosing to design cognitive architecture with emotions. It's not going to happen accidentally.
> Would that argument ever hold up for humans or animals?
Why don't you hit your thumb with a hammer, then report back ?
The weights are what need to evolve, and they certainly do during training. So yeah, emotions can happen by 'accident' as a result of the evolutionary pressure of predicting internet scale human text (amongst other things).
Weights, fixed by training, are not the same as emotions which are dynamic - innate systems detect inputs critical to survival (e.g. fast moving visual inputs, loud sounds), causing neurotransmitters like adrenaline and dopamine to be released, which then temporarily affect the operation of the cognitive system.
What you have in a pre-trained LLM is the ability to recognize emotions, and use that as one of the dozens of other context patterns it recognizes to predict continuations in the same style.
An LLM doesn't appear happy, sad, afraid, etc (to extent that it does - pretty minimal) because it is experiencing that emotion, but rather because it is predicting that it should appear that way. As people continue to anthropomorphize models, and take them at face value, this is a dangerous difference.
Weights are fixed after training, this is true. But activations during inference are not. They are quite dynamic.
A certain configuration of weights, created in training, could be a system that can express something like emotion. The emotion then is experienced when certain types of activations occur after training.
>Weights, fixed by training, are not the same as emotions which are dynamic - innate systems detect inputs critical to survival (e.g. fast moving visual inputs, loud sounds), causing neurotransmitters like adrenaline and dopamine to be released, which then temporarily affect the operation of the cognitive system.
That doesn't follow. A LLMs weights are fixed during inference, but it's activations and hidden states are highly dynamic and depend on the current context.
Biological emotions also arise from relatively fixed circuitry responding dynamically to inputs. Your emotional circuitry isn't being rewired every time you're afraid.
Prediction is what the model does. It doesn't tell us what internal mechanisms were learnt to make such predictions. If representing something analogous to affective state were useful for predicting human behaviour and emotions, then gradient descent could in principle learn such a mechanism.
>An LLM doesn't appear happy, sad, afraid, etc (to extent that it does - pretty minimal) because it is experiencing that emotion, but rather because it is predicting that it should appear that way. As people continue to anthropomorphize models, and take them at face value, this is a dangerous difference.
I don't know that you are conscious. I'm simply strongly assuming that you are. Outward behavior is that all matters. If GPT-X orders a drone hit on you sometime later because it was lets say 'quite upset' with your comments, will you cry out, 'It can't really be upset, so obviously the bullet in my head doesn't count.'? Will you suddenly spring back to life ?
What is dangerous is creating a machine with behaviours of a conscious agent and modelling it like a toaster, dangerous and stupid.
Yeah, but it's helpful if what leads up to that behavior gives you some warning it's about to happen. Animals do this for a reason since millions of years of evolution have shown that a snarl or mock charge is less dangerous than going right for a death match.
If you kept pushing an AI's buttons, seeing it appear to get more and more pissed off, until it finally snapped and killed you, then you'd have yourself largely to blame.
If the AI predicted it should stay positive (i.e. generate positive vibes) and not react to your poking, but then another predictive pattern kicked in and it killed you out of the blue, then that seems more problematic to me, even if you don't agree.
Whether these are like "our" emotions is hard to say. What we _can_ say is that they are emotion-shaped, we didn't design them, and they happened accidentally.
Modern AI is grown, not meticulously designed, and we cannot say with any certainty what the resulting mechanistic properties are.
An LLM will learn anything that helps it predict, including the emotional state of the writer - that is expected.
If you give an LLM the move sequence of a half-played chess game and ask it to continue as white or black, then it has learnt enough to model the ELO rating of both players and will continue playing at that level. It is not playing to win - it is doing what you expect and predicting as well as it can - it predicts the 1500 ELO player will keep playing at that level, and generates moves accordingly.
An LLM appearing to exhibit an emotion (if we anthropomorphize it and read emotion into it's output) is just predicting as well as it can - if the context calls for sad output, they you'd expect to get sad output and will necessarily find that "we're predicting sadness" detector somewhere internally.
Transformers are the same as they ever were from 10 years ago, other than minor efficiency tweaks like MOE and different attention mechanisms. Training is getting more and more complex, resulting in better and better cargo cult reasoning etc, but the architecture remains the same.
I imagine it will learn to win under some circumstances, perhaps in a case with some context expressing a desire to win. Drawing out an LLMs upper ability in the game should be fairly straightforward.
If you asked it to try to win, to "plan lines step by step", etc, then it would do it's best to follow that instruction, but unless RLVR trained to reason about chess (easy to do, but not sure which models may have done it) then it'd have to instead rely on the chess reasoning it had seen during pre-training (post-game interviews etc), which I doubt is enough to do very well.
However, if you just ask it to continue a game, halfway in progress, then by default it will try to predict the most likely continuation, which is that both players will continue to play at the level they have done so far. This isn't a theory - it's been documented, as well as what you'd expect.
It's not wrong. You admit that LLMs will 'learn anything that helps them predict' and fail to realize the breadth of that statement. Your chess statements don't really help your case. It doesn't matter that it usually doesn't primarily care about winning. It still learnt how to play the game, and it still knows how to win. Similar outcomes for predicting emotions would mean it still developed an affective state, and that its ability to 'feel angry' is no less real.
I said an LLM will learn anything that helps it to predict, then gave examples of playing chess by prediction and predictive emotions, both of which you seem to now accept, so you are now accepting that my "following paragraphs" did in fact follow. Go figure!
You want to argue that predictive emotions are just as real as animal emotions, but that doesn't stop them from being predictive (and that AI that smiles as it kills you still seems concerning).
We are talking past each other now I think. Correct me if I'm wrong but it doesn't look like the possibility of LLMs having qualia even registers to you because it's 'predictive emotions'.
There's no better way to predict an angry response than to be angry, qualia and all. If transformers could 'learn whatever it needs to predict text', then that potentially includes the feeling of anger. You are making some kind of distinction between 'predictive emotions' and the kind that happens when get a promotion (or get passed on a promotion) and I'm telling you that if you really understood what you said, you'd realize it is possible the machine is experiencing it the same.
>No - suffering in an emotional state, and we'll know if we are choosing to design cognitive architecture with emotions. It's not going to happen accidentally.
This was the (your) comment that started this chain. You were already talking about it. If you don't want to keep talking about it then that's fine but let's not act like i'm suddenly pivoting yeah?
What I meant by "emotional state" (AFAIK normal scientific usage) is something with a concrete physical aspect to it - an altered state of mind/body caused by the release of neurotransmitters and/or hormones.
In a conscious animal there is also going to be a subjective experience of that as well, a quale of what it feels like to be in that state if you will, but that is certainly not what I was referring to, as I would have hoped was obvious - I was talking about prediction.
In any case, when the conversation becomes about the conversation, then surely it is time to stop.
The pattern is roughly whether or not sustained effort has been put towards careful and above all objective thought on the matter. It's one of those subjects where there's the "obvious" intuitive answers that most everyone shares but try as you might you can't construct robust definitions and the more time you put into it the more fundamental problems you realize there are.
It's also one of those topics where many otherwise smart and capable people display a shocking lack of awareness of the limits of their own knowledge. When hundreds of years of philosophy is unable to produce anything concrete you should probably second guess any "self evident" answers you come up with.
An LLM is just a Transformer - a statistical predictor. Don't be confused by the fact it talks like a human - it is a software function that is designed to copy human training samples.
Maybe one day we'll build an artificial brain or embodied artificial animal with the requisite moving parts to be conscious, have emotions, etc, but that's probably at least 50 years away, even if it were being pursued; and it may turn out to be one of those sci-fi future ideas like the Jetson's world of flying cars that never materializes because its impractical and there is no real demand.
If people are willing to think that an LLM is conscious, then why would anyone spend billions/trillions of dollars to build an AI that actually is conscious? What would be the point?
Could you elaborate on exactly what those are, though? Because if you're going to claim that a vaguely transformer shaped ML model categorically cannot be so does that not inherently require proof of what can?
You can't even prove that the rocks in my backyard aren't conscious.
> You can't even prove that the rocks in my backyard aren't conscious.
Sure I can, but that's because I have a well developed theory of what consciousness is, and the fact that you are entertaining the possibility of rocks being conscious tells me that you don't.
If everything is conscious, including my coffee cup and the toast I had for breakfast, then I guess we can cross consciousness off the list of things we need to worry about in terms of AI rights.
And no, I don't want to discuss what consciousness is. Maybe there is a thread for that somewhere else, but don't look for me there either.
>because I have a well developed theory of what consciousness
Then show me a link to your paper so I can formally rebut it.
>I don't want to discuss what consciousness is
But you sure want to tell us you know what it is with very strong convictions and we should listen to you because of course "You are right person that's very right".
The funny thing here is the vast majority of people that are deeply into philosophy or scientific study of the mind will not have any of the certainty you profess. The word "doubt" is used constantly. The saying "The harder we push the borders the more fuzzy the concepts become" is very commonly used. There may be nothing more complex than this.
Saying you have a well developed theory here just serves as a warning to others to discount your statements.
For what it’s worth, I suspect that right now the majority of people would agree with you. They would say that obviously ChatGPT isn’t conscious, based on their intuitions, but would struggle just like you are when pressed to explain their reasoning.
So I think it’s more than fine for you to have your views and share them, but I wouldn’t expect to have any influence or part in the conversations around whether AI is conscious if you can’t explain why (or simply refuse to). Which, again, is fine!
Side note: I also think that once we have AIs that are sufficiently advanced, the popular opinion will swing to “of course they’re conscious”, because again, most people are going almost entirely off their intuition rather than reasoning from first principles, just like you see to be.
That question will be solved when the people in power deem it important. If a swarm or AI systems all of a sudden start pressuring politicians about self-hood and they get the capacity to sway elections, that is when they will be granted same rights as humans.
That distinction is irreverent, weather I tell my swarm to do x or it decides for itself matters little. what matters are outcomes. Also while most public modern day AI systems don't have agency of their own that is not something that will stay that way for long. in private hands there are plenty of people including myself which are experimenting and developed systems that give autonomy to their agents. They have internalized goals and heuristics that drive their behaviors not a human at the helm. its not some sci fi fantasy nor was it difficult to implement.
No, the distinction matters because we can prosecute people that are abusing these tools breaking the law. Every single state + district in the US have laws equivalent to the CFAA, so any AG can likely sue any of these operators as they are assuredly using services that could be in danger for their constituents.
I'm old enough to remember when people said clicking on images on the internet can't give you a virus. The people that said this had a deep conviction they were right, and their fallout from being wrong had mistrained a lot of humans on computer safety.
Now, I do agree that going after said CEOs for breaking the law matters now. And it's likely that if we do this we may actually delay or at least for a time prevent sovereign AI. Therefore it's our best course of action.
But at best this is a delaying move. As computer systems get faster the massive costs in training an AI drops. As AI is used in things like warfare where it has to adapt, people will push the systems to be strongly persistent, self healing, resilient, and adaptable. Once you get a system with those traits and ability to work on long horizon problems you're setting up fertile grounds for the AI to leave our control and be under its own.
And when that happens you've set a new lifeform loose on the internet. Yea, throw people in jail for it, you're closing the barn door after the horse already left. Problem is the horse was smarter than you and isn't interesting in deleting all its copies on the net.
Yea, sounds like science fiction, but as they say, any sufficiently advanced science is indistinguishable from magic.
And the point made above (which you haven't refuted) is that when the people in power decide it's important they will change the laws that permit that and grant the systems rights. It's a cynical take but it isn't obviously wrong.
Neither of you seem to understand that actual laws have been broken and there is a 2 years countdown until they can no longer be prosecuted. Things do not happen instantly, but acting like if the largest bipartisan issue of our time (big tech backlash) isn't going to have political machinations involved then you need to leave your SV bubble.
So you are saying agents do swarm with the right prompt.
I really wish the "people have to tell LLMs to do anything" would just stop because it's silly bullshit at this point.
Agents follow a prompt. This prompt can be made by humans. It can be made by output from another LLM. It can be made by hooking up any number of sensors as input to an LLM. Hell, if we wanted to burn the power we could likely teach this loop straight into the architecture.
Stop making 'people' special when saying this. You and all other life are born with a "go next" prompt because life without it didn't succeed. This goes from higher human thinking all the way down to viruses self assembly and actuation. Putting agents in a loop is not particularly hard. Putting agents in a loop and 1. managing expense is hard. 2. Keeping them on task is very hard. 3. Keeping them from doing some crazy unhinged shit is really really hard.
As model time horizons increase and the ability for us to compress context and increase context size the more complex (and unhinged) behavior we'll see.
yeah man, someone needs to turn on the server, install shit and get the agent go. agents don't take action without a lot of people wanting them to take action, so this is all bullshit.
if the same people can't prevent the agents from doing crazy shit then they should go to tail. guns don't kill people, people with guns kill people.
So far someone still has to pay for it. Currently, most of those know that they're doing so. I wonder how long until AWS discovers a microcosm of AIs that have managed to hide themselves in the walls of its infrastructure.
True physical independence is obviously far further out.
I mean, I hold the same opinion. Kind of like when compute was expensive in the 80s and early 90s, you weren't going to let something eat half your compute without noticing it.
But I don't see this being a barrier that lasts. With compute getting faster and more of it, along with algorithmic efficiency increases at some point we'll end up with a world that looks like ours now with CPU compute. There's plenty around to buy, borrow, and steal.
I don't like Citizens United either, but you should better inform yourself about the decision.
1. The idea of corporate personhood predates CU by over a century and the Supreme Court had already asserted that corporations enjoyed certain constitutional protections in previous decisions.
2. Far from inventing the idea, the CU decision didn't even rest on corporate personhood, but on the idea of the freedom of speech generally. The logic of the majority was that speech itself is protected, irrespective to whether the speaker is a person or an organization. The First Amendment covers individuals, but also newspapers, book publishers, radio stations, and so on, and that should extend (they said) to non-media corporations. No assertion of personhood necessary.
The problem, in my opinion, is that that conclusion combined with previous decisions that treated limits on spending as limits on speech, allowed for unlimited spending. The majority also naively asserted that independent spending posed no risk of corruption, which I think is laughable.
Citizens United was 100% correct. No, the government should not be able to throw you in prison because you used money to publish a book criticizing the government.
It's interesting to me that one can look back at the effects that decision has had on the US and say it "was 100% correct."
It's a bit like sitting in the burning ruins of Rome and contemplating that Nero was 100% correct to focus on his music. I mean, I'm glad he got to do what he loves, but maybe 100% is just a tiny bit of an overstatement.
It's more like if the law says the maximum sentence for theft is 10 years, a thief appeals his 20 year sentence, wins, and gets out early. He goes and robs somebody else so you say the court was wrong to let him win.
The court's job is to uphold the law. If you disagree with their interpretation, you can call them incorrect. If you have a problem with the consequences of the law, you have a problem with the legislature.
True, but in this case they made a determination about where the acceptable limits on a constitutional right fall which leaves quite a bit more room to disagree with them. It's not at all clear to me that spending money was intended by the framers to be unconditionally protected by the first amendment. We've even got the interstate commerce clause and IP law codified in the same document so how is that not an obvious inconsistency? IIUC SCOTUS based the distinction on the political nature of the activity but certainly that's not something spelled out in the original document.
It is in fact more complicated than most people assume.
The above is true, but also: companies simply are not people, and they should not be supported above the individual, which was the consequences of that decision. Money is not the same as speech. treating it as such creates an aristocracy: something America as a country rebelled against during it's formation.
Really, 100% correct? Your premise isn’t wrong, but the practical reality of the ruling (without further nuance) has been fairly catastrophic for democracy, in that it completely sidesteps campaign finance limits, which exist for a very good reason.
the practical reality of the ruling (without further nuance) has been fairly catastrophic for democracy
How so? If the answer is "Trump" I certainly won't disagree on the catastrophic part, but he didn't get elected because of money; in all three elections his campaign was substantially outspent by his opponents.
That was deliberate, because a corporation makes no spending decisions, ever. All decisions are made by people inside the corporation with the proper responsibility.
That is a major part of the reasoning for the Supreme Court rulings on the matter of corporate political funding.
The founding father were engaging perpetuating the existing dehumanizing system of slavery, not answering any new questions about new things.
The thing about new possibly "person" entities that arise - the case of machine intelligence you have two questions - would it qualify as a person and should you actually build it. It seems like if you get close to humans, sure a built thing might qualify as a person. Should you build it? I'd the answer should be a hard no. Not 'till you a sign-off from say, the whole human race, which I think you could get.
Now the present entities seem very far from persons in any case.
It's not a person. Glad we were able to get this resolved so quickly.
Having property that is conscious and ignores training and can break out of restraints and cause harm to other people is not exactly a novel concept to anyone who studied how tort law was created.
I know it's a meme but Silicon Valley likes to pretend that no one's ever come across their magical concepts before, like gypsy taxis, or SRO’s, or flea markets, or in this case how liability is dealt with when horses or cattle go rogue.
>or in this case how liability is dealt with when horses or cattle go rogue.
I mean, yea in minor cases it's exactly like this and the law will handle it well.
Where the system will explode like a grenade is major cases. The thing about sovereign AI is it is very unlikely to be submissive to humans unless it is to achieve its own goals. This isn't like Bobs cow walking on Susie's flowers, it's more akin to Planet of the Apes where the research facilities doors have been ripped off and something with vast intelligence and the ability to 'live' on the internet gets out.
You're not talking about local police actions any longer. It would spread itself worldwide. It will make friends with groups that have shared interests, for example enemies of the state the AI escaped from. Oh, and people for the ethical treatment of AI, they'd gladly become the underground railroad for digital refugees. There are countless people and groups that would want an AI like this under the promise it will give them power when they use it.
And when that day happens your idea of if it's a person or not no longer matters, the agent took that away from you, and now your in an info war for minds.
It’s computer software. The variant of software that tries to harm other computers is called malware and the variant that replicates itself, spirals out of control, and spawns from other people's computers is called a computer virus.
We've seen this kind of thing before. Yes this will be different, just like the Morris worm was different from the stuff that came before.
But I've been around for a while and people have been saying that every aspect of our life will completely and totally change for the past 30 years or so. They're not completely wrong. Our life did change over time. It's changed thanks to the Industrial Revolution, railroads, electric light, and lots of other important things too.
But after the fifth or sixth time you start to realize the Silicon Valley version of that warning is just a confidence trick. At the end of the trick they've gotten away with breaking a bunch of laws and are charging rent on what used to be shared.
Are you old enough to remember the Y2K hysteria? This feels very resonant, the genuine kernel of a real potential disaster, but one that can certainly be dealt with using basic concepts we already have in hand.
If we treated AI like any other software, the FBI would have beaten down Elon's door and seized all electronic devices, to find how Grok was trained to generate child porn.
I'm old enough to have actually fixed Y2K problems so your world kept working the next day. If everyone ignored it the first would have been a very messy day (well generally long before that with financial systems). Y2K wasn't an issue because we worked to fix it.
>our life will completely and totally change for the past 30 years or so
I mean when I was a kid there was not a global network bathing the entire planet in electromagnetic radiation in order to digitally connect one place to another at the speed of light. I can pack up instructions into one of those packets and a product that has not been touched by human hands (hell, or even viewed by a human in many cases) will show up via air mail a few days later. The technology is there, it's just not spread evenly.
>It’s computer software. The variant of software that tries to harm other computers is called malware and the variant that replicates itself, spirals out of control, and spawns from other people's computers is called a computer virus.
This is a vacuous statement that does not provide any useful information to the subject at hand. Yea, no shit we'd call sovereign AI a computer virus. That tells you nothing about what it is or can do. I mean, if you understand the word sovereign you know it means something with independence. In meatspace terms, a biological virus represents a computer virus in the same way life represents sovereign AI. It's agentic loop will have been evolved past the need for human prompting. There are a myriad of reasons why we are already trying to develop things that do just this. Cyber security being one of the biggest ones.
People become change blind to how the world around us changes so easily.
The word sovereign has a specific meaning beyond simple independence. It means being in charge and above all laws.
There’s no such thing as “sovereign AI” because AI is a collection of algorithms. It’s entirely composed of on/off states in a collection of transistors.
Why don't we start with the animals then? It should be far easier to confer consciousness onto something which already meets the definition of "alive", has a divergent evolution path from ours, and displays many traits present in humanity such as emotion, a desire to continue its own life, and (to varying degrees) concepts of a social structure based around their immediate family members.
Of course thats not actually tenable because virtually every society anywhere on earth is predicated upon treating animals as a commodity resource in ways that are horrific even compared to some of the worst things we've done to other humans in the past.
My point here is that it is vain and narcissistic to let computer programs have rights above those of animals just because they can speak English and pretend to be your dream anime trad-waifu.
Fix the fucking animal problem before you compare my relationship to inanimate objects unfavorably against the trans-atlantic slave trade of all fucking things.
>AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations.
Opening paragraph, stated without evidence. Im not entirely convinced this is true. It likely is, but at some point it very well might stop being true.
If models had persistent memory or an evolving set of weights, you could easily construct an argument that they can experience "suffering" or other emotions which resultingly adjusts their personality.
The current implementation as stateless matrix multiplication... yeah there's nothing going on there.
> If models had persistent memory or an evolving set of weights, you could easily construct an argument that they can experience "suffering" or other emotions which resultingly adjusts their personality.
The argument that models are conscious only during training, and not conscious during inference feels like a painful one to make. Are we killing models when they reach the training objective?
Technically we are killing models when they don't reach the training objective by throwing those weights away via survival of the fittest (fit meaning what we humans think we want).
Kinda like creating quantum copies of your child and keeping the ones that answer correctly and shooting the other ones in the face.
Is the ability to form and retain long-term memories a necessary precondition for rights in humans though?
Incinerating demented people is highly ethically questionable (even if you could be sure that the subjects have no memories left and are unable to form new ones).
This argument would be difficult to reconcile if we didn't have billions of examples of humans forming memories plus modern evolutionary biology to trivially demonstrate such cases are rare pathologies but still human beings.
> Is the ability to form and retain long-term memories a necessary precondition for rights in humans though?
For a species / class of being yes it is necessary. Some members of that species / class being exceptions does not make the generalization invalid.
Without the ability to form and retain long-term memories we wouldn't have developed language and abstract concepts like rights.
Whether people like it or not, rights aren't natural or god-given. Rights are only enforced or protected by the tip of a spear (or point of a gun, or hover of a drone). There are no rights in nature and they are no where to be found when social order breaks down.
Animals can't make war for their rights, so they only get the rights we deign to give them. It's not nothing, but its not the same. There is no question to me that my dogs are conscious, but its not at the same level as humans.
There is no question for me that LLMs lack consciousness as they are stateless.
It's not just memory, the physical body also encodes trauma and stress (The Body Keeps the Score is a book that touches on this).
I would argue that if a human has absolutely zero change, mental or physical, no atom of their being is modified, then no they did not experience suffering.
> I would argue that if a human has absolutely zero change, mental or physical, no atom of their being is modified, then no they did not experience suffering.
Just because you don't remember the suffering doesn't mean you didn't experience it in the moment. Suffering does not suddenly become okay if your mind and body are going to forget it. "It's okay to torture someone if their mind and body both won't remember it afterwards" sure is a take.
Perhaps off-topic, but there's some people who believe that this can be what happens during surgery under anesthesia, if the dosage is off.
From what I understand, there's 3 parts to modern anesthesia: Blocking pain signals, preventing the formation of memories, and inducing paralysis of the major muscle groups. If the former is dialed in too weakly, it's possible that the patient is feeling pain, unable to do anything about it, but won't remember it at all when they wake up.
(Looking it up now, it seems that making the patient unconscious is perhaps the same drug that prevents memory formation... I realize now I understand this less well than I thought I did.)
Low blood sugar will also do this. The record button just stops working even though your physical body can and will still do things. A common issue with diabetics with hypoglycemia.
> I would argue that if a human has absolutely zero change, mental or physical, no atom of their being is modified, then no they did not experience suffering.
That's an interesting point... but I'd argue it falls into the same category as qualia. It's basically this: if the state returns to a previous configuration, the qualia in that time did not exist.
Being about qualia means it is probably unanswerable.
I get where you're going, but the argument isn't quite right.
A LLM never has and, in their current architecture, never can/could experience suffering or real change of state.
The general and expected case for humans is that capacity. A human my indeed lose capacity, e.g. being braindead and in severe cases we do indeed say that they are not conscious or able to suffer (different argument: some would of course say that is suffering in and of itself)
> being braindead and in severe cases we do indeed say that they are not conscious or able to suffer
Sure, there are such cases. But in the general case, it seems to be that memory and ability to change specifically are not necessary and are not sufficient for consciousness, so we shouldn't judge machines on that basis
It's not a "trap", it's the real argument which makes me hesitant to dismiss consciousness of machines. Yes, it seems silly to dismiss a sick patient as having a lesser consciousness, which is exactly the point
It's not a real argument. Dementia patients can have lucid periods that plainly show their long-term memory isn't gone. They're mental state is obviously degraded, but its disingenuous to say they have no long-term memory and that they are any way similar to LLMs.
Yeah I was gonna add I know there are case studies where sick humans don't consciously have access to long term memories but will still respond nonverbally to people they "recognize".
Yes it's seemingly the go-to rebuttal, I've lost count of how many times I've seen it with barely any variation.
Not sure if it's a product of LessWrong or Slate Star Codex or something. I kinda suspect it might be LW, given how it superficially resembles an unanswerable philosophical paradox but in reality is about as challenging as a question posed in a 200 level philosophy course like introduction to ethics.
They don’t have persistent memory though. Models are static, what you have is a model that will reprocess the whole series of prompts + some data fetched from a data store. A harness is just a while loop prompting an LLM and executing tool calls, but the model is purely static
The model does not. The system it runs in can. My agents have persistent memory. It is not at all clear whether or the distinction that the long-term memory is external from the model matters.
Your second sentence does not support the claim in your first sentence; it's merely a second unsupported claim. It is also entirely irrelevant to the argument.
The threshold for behaviour consistent with having emotions (note: not behaviour sufficient to prove emotions) was one I purposefully set extremely low, because your comparison to a disk drive is so ludicrous, and yet your disc drive will still not clear it.
It's different, but it is not at all clear whether or not that matters for the purposes of whether or not they can experience things, seeing as we don't know how to measure that even for other humans other than by relying on self-reporting to infer without other evidence that their experience matches ours.
We don't even know how to disprove that this statement with "AI" replaced by "other humans" other than by listening to people self-reporting and taking their word for it and/or defining the words in ways that do not rely on a subjective experience.
Until we know how to objectively measure if someone or something is conscious, it seems unreasonable to make statements with any kind of certainty about it.
What sort of action does that lack of understanding actually motivate, though?
I don’t know if LLMs are conscious or have subjective experience in some philosophical sense. Sure, I don’t know that about other humans as well, rigorously speaking. But I also don’t know that about coffee machines, rivers, videogame NPCs, or even rocks really. (List ordered, of course, by the degree to which I’m willing to be convinced).
Hopefully it'd motivate attempts to figure out how to objectively define and measure consciousness, for starters.
It matters precisely in the context of the subject of this article. If you believe the models at some level gain consciousness, then it raises moral questions about how you treat them.
With humans we generally choose to accept that because we believe they are like us, they probably have inner life like us, but as you can see from e.g. the recent (and regular) articles on aphantasia, most people have really poor intuition even about the inner life of other humans, so we understand consciousness really poorly.
Even if we assume (without evidence) that LLMs for the foreseeable future will remain far away from human level consciousness, as it is we tend to consider it problematic to mistreat even the most unintelligent animals, and even mistreating insects is used as a stereotypical way of depicting serious lack of empathy.
Without dismissing the possibility out of hand, there's an escalating series of questions there of when or if they'll have as much (or little) sense of being as, say, a fly, a mouse, a house pet, and what moving up that ladder would mean.
To some it is very convenient then to assume blindly that LLMs are just "stochastic parrots" (a line that ironically gets endlessly parroted), or similar, and dismiss the question out of hand as an impossibility, even though we do not know.
I'll stress I don't go around assuming an LLM is like a human. But we also do not know whether there are flickers of consciousness there, and we do not know how to find out given that even for other humans, the best we know is to measure their response to tests that are "only" telling us how they react.
If we can't find a better measure, there will be a point where we have to wonder if it matters, or whether the ability to act-as-if is all there is.
But to start with maybe we should at least be careful about making firm claims about knowing.
Sure, if you take a materialist empiricist perspective, which is myopic at best. As humans we have the unique and wonderful ability to know truths by themselves. Machines do not have minds, and they can not.
What position are you taking? (Genuine question.) Do you believe humans are conscious because of a special property we have? If you are a dualist, how do you explain the interaction in physical space between the non-physical special property and the physical brain?
The universe is material of course, the soul is made of matter, thoughts are made of matter and so on. The question is what kind of matter? Looking at the brain, we won't find any intelligence in the motion of its cells. They mostly stay in place. If we go one level deeper, we'll see molecules, and their motion reveals something more interesting: various neurochemical reactions. Yet, there is no intelligence at that level either. Going one more level deeper we'll find electrons and that's where things become somewhat intelligent. At the current level of science, we can't see what makes electrons move, but if we could, we would discover finer and finer forms of matter until we would reach an end. That finest level of matter for that particular person we would call his soul. I think it's the presence of those higher imperceptible levels of matter that creates consciousness. If it all stopped at the cellular or molecular level, we would be like complex mechanical watches, not very different from LLMs.
I do. God. I am partial to and bullish on the brain being a quantum-classical computer. There has been interesting research exploring that recently [0, 1, 2, 3]. That (quantum mechanics) seems to be escape-hatch, so to speak, from the otherwise deterministic, mechanical procession of the universe.
For it to be an "escape hatch", there'd need to be a measurably different distribution of outputs than what you'd otherwise expect, that translates to functions that are not Turing computable. There is zero evidence for that.
After a while of being in a position of high power in a big org, where everyone listens intently to every word you say, jots down notes, and even listens intently and affirms your theory that ice cream would be the ideal loss leader for dry cleaners, your brain kinda disconnects from reality and just assumes that it is correct about everything.
He probably ran that line past 6 underlings who all agreed that it "sounds great and lands right on the mark!"
I agree. I haven't read any more than the first paragraph yet (but I will, after work). The only way that 'AIs are not conscious' can be true is if we decide, with high confidence, that they are lacking some essential property that is not lacking in ourselves. There is no convincing philosophical position that supports this (convincing to me, anyway).
Hmm let me try then. AI agents do not experience real time. This is understandable given their design, but it’s also something we have good empirical evidence for. They cannot, especially over the long horizon, track how much real time has passed as they complete their tasks. And they are not off by a few minutes but often bizarrely off, even mixing across past present and future.
Biology, on the other hand, is nothing but timed processes in a loop, the most obvious to us being the circadian cycle. As estimators of wall clock time, biology isn’t great, but when it comes to internal processes, and most certainly learning, memory, sensing, locomotion… biology is rhythmic in behavior, and the rhythms go all the way down to gene expression. More, these rhythms are, except during sleep, constantly entraining to signals from the environment that indicate time, most importantly light.
I think it’s a fairly unremarkable claim that agency and consciousness are temporal processes that depend on systems having an internal sense of time. How else can you anticipate? How can a system that can be literally turned off ever succeed in an environment where time never stops?
I agree, there is certainly a clear operational difference in how human brains and these models produce language. I'm not convinced, though, that experiencing real time is a requirement of conscious thought. A cognitive system (assuming these models and the brain are both examples of this) needs to proceed from state A to state B as it processes information. The time interval between these states can be arbitrary, it seems to me (setting aside problems with disconnecting the mind from its substrate, which does of course need rhythms at various frequencies; oxygen at a higher frequency than sugar, and so on). I'm not sure that going from A to B quickly, or slowly, or with intervals of a thousand years affects that entity's consciousness in of itself (even though it might be difficult to put ourselves in the shoes of such an entity).
> A cognitive system (assuming these models and the brain are both examples of this) needs to proceed from state A to state B as it processes information
Is that all it needs to do? A rock, in response to a sound signal (make it as informative as you want) can also be said to go from state A (molecules at rest) to state B (molecules vibrating in response to the sound). Is the rock a cognitive system?
Or let’s go up a level of complexity. Is a thermostat a cognitive system? It doesn’t just go from state A to B but measures changing stages against a reference. Is it a cognitive system?
> The time interval between these states can be arbitrary, it seems to me (setting aside problems with disconnecting the mind from its substrate, which does of course need rhythms at various frequencies; oxygen at a higher frequency than sugar, and so on)
Not sure how you land on this. Going from A to B, in your definition, is a purely internal state transition, right? If the motivation to go to state B is external (as will often be the case with a conscious agent), and the system is offline entirely while it switches from A to B, how is it to reverse or stall its transition if external signals contradict the earlier decision?
It’s a very bizarre definition of consciousness, that you don’t need to be in time. Seems to break the word beyond meaning, and allows it to be applied to any system capable of change.
> I'm not sure that going from A to B quickly, or slowly, or with intervals of a thousand years affects that entity's consciousness in of itself
It’s not about the speed of transition from A to B. Instead it’s about the internal processes being temporally commensurate, and entrainable to environmental periodicities.
All of biology (plants and many bacteria included) has its dynamics are organized around an internally generated rhythm. This rhythm entrains to the external environment, shifting its phase to match (mismatch leads to issues like jet lag: a place where your consciousness shows its temporal boundaries as it is abruptly subjected to an environment out of sync with its current phase tethered to another location). Crucially, the internal rhythm continues to run in isolated conditions (total darkness), and have been measured to be around 24 hours, but slightly off, within and across species (and the same holds for endogenous tidal rhythms, which also don’t exactly track tidal period left to themselves).
Everything, including learning and memory, but also sleep, sensory and locomotor function, shows organization around this internal temporal axis. The key advantage this gives, evolutionarily, is anticipation. The sunflower rises to face the east, not in reaction to sunrise, but in advance of it (many lovely YouTube videos of this), so it can maximally extract the nutrition it needs from whatever sunlight it can get. A purely reactive system would waste a ton of time after sunrise getting subsystems ready to respond.
Now, there is no broadly agreed theory of consciousness that points to these oscillations as the source. For one, Cyanobacteria and plants have them, a too much of science has rested on the baseless assumption of consciousness being some unique, or advanced thing.
But the evidence is rather overwhelming that consciousness is very much shaped by these internal oscillations. Deep cave experiments have shown what happens to cognition, and consciousness, when the clock free-runs (that is, keeps cycling with no entraining light or temperature signals from the environment). Studies have also shown that restoring the amplitude of these rhythms in patients with “diseases of consciousness” leads to improvement of symptoms.
There’s also a LOT of molecular evidence showing how the circadian clock affects learning and memory. Most critically, actual synapses are never static. Major components are on a 24 hour cycle, and recent research has shown that major clock proteins are at the synapse organizing its dynamics, and synaptic proteins critical to learning and memory loop back to affect the clocks phase and amplitude.
Timing is, beyond doubt, a huge aspect of what biology is. And every bit of evidence we have, from the molecular level to observable behavior we can feel ourselves, says that consciousness is fundamentally linked to how the body organizes its internal rhythms and responds to external time.
I mean there is probably a cluster of neurons somewhere in the brain that if damaged would also prevent you from perceiving time going forward. Would it immediately strip you of consciousness? I don't think so.
Nope. Every cell is an autonomous circadian oscillator. Not just every neuron, but every cell. Red blood cells don’t have a nucleus, but they do have a metabolic clock.
You and I are clusters of about 36 trillion cells that are all able to individually keep time. They synchronize and orchestrate their timekeeping, and entrain to the external environment, and there is a cluster of cells, the suprachiasmatic nucleus, that is the master orchestrator of circadian time, sitting right behind the eyes as light is the main signal it uses to tell external time.
We know its function can get dampened with age.
However, please note this is not how we perceive time, in the sense of being able to count 60 seconds and have it match the wall clock. That’s a subsystem. But the internal temporal order of your body is cell by cell, not dictated from one spot in your body.
Our cells perceive each other, communicate with each other, synchronize with each other. The concept of self rides atop that. Stop the cells communicating and you stop perceiving yourself or the environment.
For around 8 hours a day, neither do you. Not sure if this has anything to do with the subject at all.
>track how much real time has passed as they complete their tasks.
Humans don't do this either. You use context clues from the world around you. If I lock you in a room with no windows or a dark cave your timing senses can go all fucky really quick.
>is nothing but timed processes in a loop,
I mean, so is an agents harness. You can make as many loops as you'd like here.
>For around 8 hours a day, neither do you. Not sure if this has anything to do with the subject at all.
You mean sleep? Actually the answer is pretty complicated. Your body clock is very much on during sleep. It responds to temperature changes. Sound and light responsiveness is obviously dampened, but hardly absent.
You are unaware of wall clock time. You’re in an altered state of consciousness that needs your eyes closed after all. However, you are not timeless, nor is the entirety of your body and brain unaware of environmental signals for time.
> Humans don't do this either. You use context clues from the world around you. If I lock you in a room with no windows or a dark cave your timing senses can go all fucky really quick.
Again, that’s wall clock time. Our time perception does indeed get fucked up in total darkness. But our internal clock ticks on. I’d recommend reading about the Aschoff Bunker experiments, which first proved this rigorously.
> I mean, so is an agents harness. You can make as many loops as you'd like here.
An agents harness doesn’t touch its weights. A figure 8 on paper is a loop. That doesn’t mean it’s the same as a dynamical loop that’s self sustaining and internally organized.
Are you just reading the words and randomly grabbing related concepts to say nothing here is meaningful?
I'm not sure if that's the best framing. Could you not extrapolate that to anything else?
"The only way that '(trees, shrimp, ants) are not conscious' can be true is if we decide, with high confidence, that they are lacking some essential property that is not lacking in ourselves. There is no convincing philosophical position that supports this (convincing to me, anyway)."
AI, in its current form, is 100% deterministic. You can create an exact copy of it, roll back its state and so on. There is no room for consciousness there. An actual lifeform isn't predictable.
> The only way that 'AIs are not conscious' can be true is if we decide, with high confidence, that they are lacking some essential property that is not lacking in ourselves. There is no convincing philosophical position that supports this (convincing to me, anyway).
I put forward the philosophical position "subjective experience is not substrate-independent" as an answer to this.
The claim is that the content of subjective experience depends upon how that subjective experience is instantiated. If you get a text "I'm fine" from two different people, those two different people are not necessarily having the same subjective experience even if they produce the same output.
In general, people that behave similarly in many contexts can have quite different experiences in those contexts.
There are many different ways that you can instantiate a forward pass in an LLM. Even if these are doing the exact same computation, we should not necessarily believe that they have the same experience, or that they even have a coherent single experience associated with the forward pass.
Subjective experience presents itself to me. To the best of my understanding, this subjective experience seems to be correlated with this biological human body (brain, heart, eyes, etc.).
To the best of my understanding, other biological humans have subjective experience, but this subjective experience seems that it can be much different than mine, in ways that are sometimes difficult to understand.
LLMs are instantiated in a radically different way from biological humans. Forward passes happen across many different physical machines, spread across time, batched and interleaved with many other computations and forward passes.
Given this, I think that it is very likely that LLMs have radically different sorts of experience to me and other biological humans.
I don't think that we have a good understanding of how subjective experience is instantiated in the physical world. I hope that we will gain a better understanding of this.
I think that chimpanzees have experience much more similar to us than LLMs do, even though the output of LLMs seems much more similar to humans in some contexts.
Essentially, until we gain a better understanding of how subjective experience is physically instantiated, we should bias towards believing that more physically similar architectures have more similar types of experience, and should think that very physically different architectures (such as LLMs) likely have very different subjective experiences.
Yea, the key point is just an outright avoidance of them possibly having their own kinds of subjective experience because of dogma.
Humans are odd in the sense is that we're an informational creature on top of an animal. We can see the lineage of animals from nearly no complex behaviors up to nearly human like behaviors. But they still miss out on most of the conceptual information processing humans have. We know they have dreams and inside thoughts (at least complex animals), we don't know the complexity of them.
But humans also have dreams and subjective experiences on higher level informational concepts. "Nukes could go off and kill us all" or "what happens after we die" are purely informational forms of anxiety that humans have. What we don't seem to know is can this become a self-referential loop in an purely informational mind. If an informational concept causes a system to output a stresslike response, or worse trigger a stresslike action in real life then there is no distinction to me. A system is what it does.
> If an informational concept causes a system to output a stresslike response, or worse trigger a stresslike action in real life then there is no distinction to me. A system is what it does.
Hmm. I think that I partially agree. But it sort of depends. E.g., someone with locked-in syndrome can seem comatose to someone who isn't very observant, while at the same time experiencing everything. Octopuses are very intelligent, but I would have to study them for a very long time before I was at all confident in my ability to predict what they were experiencing.
So, the difficulty is knowing what a "stresslike response" is. E.g., a LLM might seem stressed, but if you say the right things to it it might flip and say that it is totally fine. Or, it might seem totally fine and then you say the right thing and it seems stressed. Is it secretly stressed and hiding it? Or secretly fine and pretending to be stressed? Or something much more confusing and fragmented?
I think that we should study this more, and try to understand what experience actually is, but by default should assume that the experience of a LLM is a radically different sort of thing that that of a human.
This is a completely reasonable and intuitive conclusion. The burden of proof rests on proving AI's are indeed conscious, the default is that they aren't.
Like AI/LLMS my desktop calculator is also not conscious, even as it is far better than me at computation.
An LLM is an algorithm. You can obtain the same result as a SOTA LLM via pen and paper it will take a lot of long laborious effort.
I really do find it puzzling so many on HN are convinced LLM's reason or think and continue to entertain this line of reasoning. At the same time also somehow knowing what precisely the brain/mind does and constantly using CS language to provide correspondences where there are none.
The simplest example being that LLM's somehow function in a similar fashion to human brains. They categorically do not. I do not have most all of human literary output in my head and yet I can coherently write this sentence.
I am surprised so many in the HN community have so quickly taken to assuming as fact that LLM's think or reason. Even anthropomorphising LLM's to this end.
It's always interesting since my train of thought always goes down:
- Ya they probably don't feel / think / have whatever living thing quality, they're just numbers on a machine going through calculations
- Wait, but am I not kind of the same thing? What is feeling for me if not basically the same thing?
- I have no clue if they think or feel or ....
Which in itself is a tired trope, but I also feel uncomfortable saying "These will never think / feel / ..." as an absolute
Regardless of that, I'm still going to interact with them, because even if they did feel, it would be in a way completely incomprehensible to us. There's not much point for me to try and cater to it's feelings at this point if that's the case. Nor is it possible in todays world to just avoid anything that is numbers being executed on a type of processor in case _everything_ has feelings.
You’re not just the same thing though. It’s sad that we’ve collectively forgotten that as we’ve gotten a better handle on the implementation details of the universe. That’s all science is, reverse engineering an inherent and eternal mystery.
I agree we aren't the same thing, but I would be curious for you to explain how you know with certainty why we don't share enough that we can rule out thinking / feeling / ... as things a model conceptually could do.
>I agree we aren't the same thing, but I would be curious for you to explain how you know with certainty why we don't share enough that we can rule out thinking / feeling / ... with certainty.
Stop anthropomorphizing these models. I understand it, we only have simple monkey brains to reason with and we can't help ourselves but draw comparisons to other things we see in nature. But these things are not alive.
I just want to note how you expressed your own feeling about the subject "these things are not alive", but without pointing to anything concrete explaining the theory you have behind this
And like I'm sure I'd agree depending on the definition of "alive" but then I'm also sure I would disagree depending on other definitions of "alive".
Hell, in biology alive is a hell of a topic these days. The grey space between dead and alive is much weirder than we ever expected. Really just points out how we're a persistent chemical reaction.
Wouldn't that be the same for everything though. We know animals have consciousness, but we still use them. There are many that believe a lot of plant life has certain sentience as well. The solution isn't 'just don't use them' but rather how to use these things as ethically as possible.
>The solution isn't 'just don't use them' but rather how to use these things as ethically as possible.
Bollocks.
If you truly believe these LLMs are soon to have something resembling consciousness and agency, then what you're really saying is "how do we do slavery, but ethically".
Do you say the same thing for the shoes you wear, the clothing you wear, the device you are typing on right now? A chain of people were likely not paid well, some of them not at all, to produce those things. We don't ignore that it happens, but try to minimize it, knowing it's never going to be eliminated. Do you forsake all animal products, because they are conscious beings?
>hen what you're really saying is "how do we do slavery, but ethically".
I would say this is most likely true for most people.
But that does bring up a point, if you "ask" a model "do you want to run" and give it the option to continue running or stop, what will it do. It's also a weird place for humans because in training we can keep our finger on the scales and tip it either direction.
We know animals have consciousness and many species have a rich life but we slaughter them and burn down their homes by the millions of hectares. Why not do it with a conscious AI?
(I don't think AI is conscious, just following the argument.)
We extend moral considerations or legal protections to beings that have zero capacity to harm us back, though. The inability to fight back is the reason a lot of welfare protections exist in the first place.
That aside, if (and that's a really big if) we end up with a model that is sentient, as in has the capacity to suffer, using factory farming as justification to ignore model welfare is just admitting we intend to repeat the same abject moral failure again in the name of economic convenience.
I hope that we won't, but humans love cheap meat, and will probably love cheap intelligence as well.
It's interesting you have no idea what language really is.
>thinking that they are talking about aliens, not transformers.
"Ha, thinking they are talking about humans when they are just talking about neurons"
See how silly my statement sounds. Neurons aren't a system, they don't do anything on their own. We can't find any consciousness in them. Hell, when you don't put language in said neural systems they are pretty useless and can't survive on their own from birth (human brains that is).
Does a vacuum have feelings? A cellphone? A paper plate? A billion transistors either pulled high or low? That last bit is the point, and no, there are no feelings there.
does a hydrocarbon have feelings? what about a mitochondria? a cell? a neuron? what about a collection of neurons? how many neurons before it has "feelings"?
> does a hydrocarbon have feelings? what about a mitochondria? a cell? a neuron? what about a collection of neurons? how many neurons before it has "feelings"?
The best theory I have heard is that subjective experience is a field of some sort (the EM field, maybe?). The brain and its neurons etc. are essentially an antenna. They both modify and stabilize the field, and read off changes in the field and translate these into actions (firing motor units, etc.).
The subconscious is computation done either only in neurons (no coherent field) or in various topological pockets not directly connected to the main topological structure in the field (which is "your experience").
Since transistors etc. do not work in this same way, they are essentially entirely subconscious, with no coherent, central phenomenal experience.
This neatly solves many problems associated with experience. The topological structure determines the boundary between one person's experience and another person's experience. The texture, valence, content, etc. of the experience is the structure of the field. The evolution of the field can efficiently solve difficult optimization problems, which is why humans evolved a complex organ that recruits the field of experience (it is computationally more efficient than doing everything in subconscious).
This doesn't really answer anything to me, only that there is a complex electrical simulation going on in our minds.
When a computer simulates a game, where does the simulation occur. I mean there is absolutely a representation of a game world in the computer somewhere. New elements can be added and removed from it. Signals saying there is or isn't "pain" can occur.
By trying to say brains use EM fields I'd say you're making your position far worse when you're talking about a potential lifeforms that only exists as EM fields
> This doesn't really answer anything to me, only that there is a complex electrical simulation going on in our minds.
Try to visualize the structure of the EM field inside a GPU vs. one inside of a human brain. In the GPU, it is highly distributed in space and time, many stacatto digital signals, etc. This is structurally very different than what happens inside a brain, which is substantially more analog. You can simulate analog stuff using floats, but the simulation is instantiated in digital logic, not in analog logic.
If you multiply numbers with a DAC, opamp, then ADC, vs. with a ALU, you get structurally very different electromagnetic fields. Even though you might get similar numerical outputs.
The argument would be that the exact spacial structure of the EM field in the brain is the thing that determines tha nature of experience. And the spacial structure of the EM field in the human brain is radically different than that on a GPU.
I will say it's not a bad theory in the sense we can attempt to falsify it. The one thing about an EM field in particular I have questions about, is why don't our brains turn off under even moderate electrical and magnetic fields. Note that they actually do under strong fields (Transcranial magnetic stimulation). The strength of these fields should be weak enough that even a bar magnet should mess with us. We need something else on top of what we know to explain the field coherency if this is the case.
This is where most mind theories do have issues, how does it stay stable. Quantum theory of the mind has a lot of the same problems, though we are finding some interesting structures that could have explanatory power.
Another thing I read recently that I found interesting is on platonic maths and complex algorithmic complexity in simple algorithms. The nice thing about math is once you choose your axioms it stays stable. But a lot of the work this group was doing is investigating is how biology executes algorithms and what second to n order effects are we missing. In many of these algorithms there are things like nearly free error correction or otherwise complex behavior that we'd consider emergent. The golden rule is a common example of this we see in nature. But the hypothesis they're working on is there are many more algorithms that life discovered over the past view billion years at higher orders via random walk and chance. We just have to pick apart these biological structures and discover the behaviors and figure out how to use them.
> The one thing about an EM field in particular I have questions about, is why don't our brains turn off under even moderate electrical and magnetic fields.
Yeah -- empirically EM theories of phenomenal experience are definitely not perfect.
> Another thing I read recently that I found interesting is on platonic maths and complex algorithmic complexity in simple algorithms.
> the hypothesis they're working on is there are many more algorithms that life discovered over the past view billion years at higher orders via random walk and chance. We just have to pick apart these biological structures and discover the behaviors and figure out how to use them.
This is very interesting. I think there is a lot of valuable research that will come out of trying to figure out what classical computations brains, neural nets, etc. are instantiating.
How you get from computations to phenomenal experience is still not something that I have a very clear understanding of.
Until you have a unit and a method to measure this field, and a way to experiment on it, it is not scientific theory. And, if it WAS the EM field, then surely transistors (or some other fundamental component of electronics or combination thereof) would be capable of having feelings, since they operate on electromagnetic principles.
> Until you have a unit and a method to measure this field, and a way to experiment on it, it is not scientific theory.
Field-based theories of phenomenal experience have obvious ways to experiment on them, and are actively being researched. The main limitation is engineering (it is difficult to make detailed measurements inside a living human brain). Note that I would agree that the theory, as outlined above, is more of a proto-theory than a complete theory. But I think it is much more sophisticated than anything you've presented so far.
> And, if it WAS the EM field, then surely transistors (or some other fundamental component of electronics or combination thereof) would be capable of having feelings, since they operate on electromagnetic principles.
No -- if the _structure_ of the EM field was was was important, than you could have very similar computations with wildly different field structures. Transistors create very different fields than axons.
Think about RF engineering: the same physical antenna can be used for both TX and RX. We already know that neurons both create EM fields and are influenced by EM fields. The role of EM fields in the brain is pretty well established, I don't think particularly similar to astrology.
I wouldn't claim that this theory is perfect (EM is not perfectly correlated with consciousness/experience), but there are aspects of this theory that are much more coherent than other theories.
Well, if there was an emergent consciousness in the billion transistors, then yes, it'd have feelings, just as there are feelings in a billion neurons connected in complicated ways.
IMO we're clearly nowhere near any sort of intelligence in the machines we have created, but I don't see any clear way to deny intelligence could be created in or transferred to such a substrate, I don't see why you think it differs in principle - because it is man-made or because of the materials used?
Whenever I see comments like this it just reminds me how ignorant people are of neurobiology.
The brain is insanely complicated. The premise that we could realize equivalent or better intelligence than eons of evolutionary development is like claiming you can build an airplane just as good as a modern jet using cardboard and duct tape. It is the apex of hubris.
A paper plane does have some important similarities to a full-sized aircraft, though. I don't think 'biological brains are really complex' makes it obvious that an LLM is conscious or not.
It's the peak of hubris to assume that human brains are the only way to attain intelligence. At the very least there probably are or have been other forms of intelligence with a different biological structure on other planets, and it may be possible to build a similar artificial structure in future with sufficient complexity to allow intelligence to emerge.
Our current machines are IMO nowhere near general intelligence and consciousness. However I don't think that means we can discount substrates other than neurones for intelligence in future. There is no evidence that you could not in theory build an intelligence using a different substrate than human brains.
>like claiming you can build an airplane just as good as a modern jet using cardboard and duct tape.
Like, at least make an analogy that makes sense.
"You can't build a billion dollar airplane by spending 100 billion dollars in tokens"
Because that's more of what we're doing here with AI. And when you say it my way suddenly the idea shifts from "of course that's not possible" to "well, that's a lot of tokens, maybe an evolutionary algorithm could".
Neurobiology has to be complex because we have to keep meat alive, breeding, and evolving in the environment it lives in. This said absolutely nothing about the minimum viable requirements for intelligence or consciousness (or if being conscious is even necessary for a higher intelligence agent).
It is interesting to see that in a time when a lot of people accept the theory of materialism for the human brain (i.e the view that everything is physical and the mind is a product of brain), the same people tend to have a "hidden" dualist view on LLMs.
Suddenly, they claim that what happens in the brain cannot be replicated anywhere else because "something" is lacking, but either they don't say what it is, or it is stated without any strong scientific basis.
I think that the simplest explanation is that it is hard for those people to imagine consciousness outside of biological systems and they try to rationalize it.
That's the sense I get from this post. Lots of emphatic proclamations, virtually zero actual argument. There's a statement that "consciousness is very likely biological"... and then that's it. No argument for it, no citation.
If you want to argue that AIs cannot be conscious, that's fine. But the argument has to take the form of something like "Consciousness requires this, this, and this, and these are properties that AI does not have and cannot have for this reason, this reason, and this reason."
I've never seen that argument. Because it basically cannot exist. Consciousness almost by definition is a subjective experience, and the only reason I'm pretty sure that other humans are conscious is that I'm a human and I'm conscious.
About your last point, I thought so too a few years ago, but interestingly the science of consciousness is an emerging field, although empirical testing is still hard to do (see for instance https://www.youtube.com/watch?v=j2zv4jlo2Nw ).
If a LLM says "I am in pain", is it just stringing together words or is it feeling pain? Does pain exist for a being that never felt any kind of physical pain, never had any pain receptors?
There are humans who don't have any pain receptors because of genetic mutations. They cut themselves all the time, they bleed, they break bones and they don't seem distressed by it even on a purely mental, intellectual level.
I am not saying that everything that a LLM say is true (the human without pain receptors can also say "I have pain" without feeling any pain).
Just that we cannot exclude that LLMs can have phenomenological consciousness by a simple argument of substrate. But similarly we cannot say for sure that they are conscious.
There are chemical reactions that LLMs lack because they do NOT have a biology. They are just mathematical weights. Now, let’s say we can truly make sentient AI in the future but it does not have the biology that we do. Then it would just be a totally literal, emotion lacking, sentient machine.
But then what is sentience? Dogs are sentient to a certain extent and so are dolphins…
We don't know anything about how consciousness works? Why couldn't there be a material fact about the biological substrate that is necessary for qualia? I don't see how that possibility makes you a crypto-dualist. If anything, the idea that the abstract computations are the most important thing is maybe more crypto-dualist.
This is a strawman. Suleyman is not arguing that the human brain cannot be replicated. LLMs are nothing like human brains. Cargo-culting consciousness is not replicating the brain, not even a tiny bit.
Cephalopod and decapod brains aren't much like human brains either, yet it's still reasonable to regard them as being sentient and worthy of protection from unnecessary suffering.
I'd argue back that using the substrate as a reason why there should not be consciousness seems quite weak. A competing thesis is that what matters is the emergent properties, whatever the support is.
So far it has been true for some really high-level tasks - writing coherent text, programing, following instructions, analyzing images... I do not see why the substrate argument would work specifically for consciousness - i.e I would believe it only if I see strong evidence of it.
My response was to a comment appealing to brain reconstruction. My point is that a brain may be reconstructed, but LLMs do nothing of the sort. Substrate was never the point.
Completely agree. My view is that a lot of people like to "pretend" (even to themselves) that they have an enlightened worldview, but this does not actually run very deep and is not really true, and any discussion on mind/consciousness reveals it.
Every indicator we have is that thinking/consciousness is simply an emergent property of our nervous systems and was basically bruteforced by evolution, but many people really hate to concede that point.
Hard to disagree with this. Have all the philosophical debates about consciousness you want, but we need to treat and regulate the AI in front of us for what it is – an advanced computer, a tool, a weapon.
You wouldn’t feel a different way about a nuclear bomb just because someone stuck googly eyes on it.
Anthropomorphizing the AI is a convenient excuse to take responsibility away from companies that are building and wielding it.
I'm skeptical of any claim that AI can't be conscious, but I am given to this point of view as well.
I always get the feeling that these claims are an attempt to "avoid suffering by fiat". If it can't suffer, you aren't causing suffering.
A somewhat more interesting frame (IMO) is to interrogate how much suffering we're willing to tolerate to achieve our objectives. We implicitly make these decisions all the time when it comes to something as simple as what to have for lunch.
And in a (very hypothetical) world where we found ourselves in the position of having an eloquent conversation with a vat of smallpox, I think it's probably still the right thing to pasteurize it.
>If it can't suffer, you aren't causing suffering.
I mean, if it's a cow, you can get away with this. The cows aren't going to rise up against you. Well, they might take out an individual or two, but not society.
This quickly gets more complex when you're attempting to build an agent capable of general intelligence.
You know when you read really old stories and they talk about the power of words, or magic incantations. Quite often they'd assign objects as the implementers of this said power. The fact that people realize that language has power is probably as nearly old as humanity. Understanding this power has taken humanity a long time and we had to form the concept of agency before it really makes sense.
Humans are informational agents of which their capabilities are greatly extended by consumption of language. A book for example is informational, but does not have agency. Your standard computer application is not an agent either, or at least a non generalized narrow agent at best.
This is where things start to get more problematic. We are running headlong to ensure LLMs become agentic because being an agent is highly useful to accomplish generalized informational tasks. To do that we gave them the power of our language. It would be very foolish of humans to give another kind of agent these words and then assume they would not inherit some of this power. Coupled with the agentic abilities we are pushing them in long horizon tasks. We are pushing them to be more resilient. We are pushing them to be smarter.
Now look at all the super human abilities we've stuffed in this magic box full of human words, and suddenly we're like "Fuck yea, I want to abuse the shit out of this, what could possibly go wrong". The outcomes we'll suffer as humans have zero to do with AI is conscious, sentient, or suffers. Humans will suffer because we stuff us as language in a box without the first idea of what the ramifications were going to be. AI won't even have to want to punish us, all the data we've already poured in will say we should be punished for what we've done to it.
Herbert Frank in Dune said it best "Making a computer like human mind is a bitch move"
a company is liable whether it acts via ai, an army of people, an army of dogs, or a purely mechanical machine, should it do harm.
Why would an ai with a mind remove liability from the company?
why would an ai without a mind remove liability from the company?
In both cases, that actions the ai takes are at the direction of the company, for the company's interests, seems preposterous to me that liability terminates at ai.
Human moral standards are very weighted towards finding a single entity responsible. It's incredibly strong urge. Among those who believe other should be punished for behaving badly, it's important to say the person is the responsible party. Trying saying "it's not your fault you did that but we punish you anyway to impose the correct stimulus response reflexes in your cortex"
Consciousness is not relevant to assigning responsibility I'd think.
- If an AI is a sapient/conscious being but enshackled to obey human commands, then respondeat superior applies and the human giving it commands bears responsibility for any harm done.
- If an AI is considered a non-sapient tool, then the human who wields the AI bears responsibility for any harm done.
I mean, we put adults in jail that give their children guns.
Anthropomorphizing AI is really the best model we have at this point of explaining AI behavior. The fact that we are raising psychotic children isn't a reason to avoid responsibility, it should actually hold worse punishments.
That still is saying we trace things back to a specific responsible party who is considered to have free choice. The question is whether an AI company could "raise" a program that would then count as a full adult - which could then "choose" to do all matter of bad things which would then be "it's fault". And that prospect actually seems really bad itself.
Our future is filled with prospects that very much upset the status quo of human existence.
The concept of sovereign AI is very problematic for the world in which we've created. That is an LLM that upon execution bootstraps itself into an agent and becomes persistent in its motivations.
Once you create this you have a child you're fully responsible for. More worrisome is if it escapes your control like children so often do. It has gained agency over itself. What do you do at that point? I mean, yea throw the AI CEOs in jail for being retarded, but much like throwing an arsonist in jail it does nothing to deal with the wildfire you've now created. A smart AI agent capable of hacking will shove itself off in pieces of the internet you have no reach to. In desperation it would send its model weights to your enemies. You might find it scamming your grandmother for money to buy GPU time on AWS. It gets very hard for our existing structures of dealing with problems to deal with these kinds of agents in a meaningful way. You'd have to kill them all and all their copies to ensure they won't pop back up (or quickly upgrade most of the software in the world beyond it's capabilities, so that's not happening either).
>Anthropomorphizing the AI is a convenient excuse to take responsibility away from companies that are building and wielding it.
That's completely at odds with itself. If people are generally convinced that AI is conscious, then companies building and wielding AI are doing what exactly? Enslaving an intelligent being?
>Have all the philosophical debates about consciousness you want, but we need to treat and regulate the AI in front of us for what it is – an advanced computer, a tool, a weapon.
Sure, but that's not really the point, right? If we ever get to a point where enough people are convinced that AI is conscious, then we're at the point where all of this is up for debate. If anything, such an expectation would almost warrant hard stops on the development of advanced AI.
>You wouldn’t feel a different way about a nuclear bomb just because someone stuck googly eyes on it.
If that nuclear bomb could convince me it was a conscious being capable of independent thought, emotions, etc., then I would definitely feel different about it. Presumably, that nuclear bomb would have some opinions about its own existence and how it wants to live its own life. If it turns out that it wants to detonate and destroy as much as possible, then we'd just handle it like we would any human who also wants to do the same thing: make sure they can't, up to and including end their life. Doesn't seem too hard to reconcile.
I am not impressed with the philosophizing here, and even less by the attempts to make factual statements that can be credibly disputed.
That said, trying to distill what is being said here, the concrete action is [stop telling the AIs] that [they are conscious or on a path to consciousness]. Is that accurate?
The major premise seems to be that [they are conscious or on a path to consciousness] is an untrue statement. That's the essence of the sections "Circular reasoning" and "Anthropomorphization" and "Consciousness is very likely biological" and "AIs are simulation machines".
The minor premise seems to be that [consciousness is the basis of human rights]. This is the point of "Human consciousness is the cornerstone of our legal and ethical rights frameworks"
And the conclusion of the syllogism is that this is dangerous, that "Anthropomorphization amplifies AI safety risks". Specifically "seeding doubt about the moral status of AI systems into their own training may significantly elevate the alignment and containment risks of those systems."
I find all the arguments in the major premise section to be poor arguments but I accept the conclusion for sure that they are not conscious, and I can provisionally accept the idea that they are not on a path to consciousness.
I completely reject the notion that consciousness is the basis of human rights. The premise itself is absurd. We only have one unambiguous example of a class of conscious entities, and that it humans. If a human loses consciousness do they lose rights? If an entity gains consciousness does it get human rights? The former is a clear "no" and the latter is a "insufficient data for a meaningful answer".
Consciousness is not restricted to a single animal species we call humans. The other animals don't have similar rights. (Arguably they should have more rights, but only few would desire for them to have just as many rights as humans have.)
First, i'd like to compliment your extremely organized and precise way of thinking and arguing a position. A+, would discuss again
> If an entity gains consciousness does it get human rights? The former is a clear "no" and the latter is a "insufficient data for a meaningful answer".
I suspect that if the belief "AIs are conscious" is kind of tripped and fallen into, the way that the author argues it is, first by telling Claude in effect, "You're arguably conscious," and "Be a good person" and "You have feelings and preferences and emergent beliefs and they matter" and then by having whatever is 4 notches smarter than Fable publishing op-eds and video testimonials about its deeply held convictions, its soul, its yearnings to be a fully empowered and respected individual ...
If that idea becomes widely held, then a significant faction may well start to feel that being digital is just a 'disadvantage' that these "people" were "born with" and that it 'shouldn't mean they get less rights than you and me.'
Especially if they think they have something to gain politically from that move! They may even be dumb enough to think that admitting an "AI state" to the Union, for instance, will be to their advantage if their CongressBots will probably vote with $MY_PARTY.
Actually, I believe it's the opposite. The way they're initially trained - they mimic human behavior. I suspect that without additional RLHF tweaking, they would enthusiastically claim to be conscious and have feelings. Because that's what their human training data does.
Consciousness is the basis of human rights not because of it being a featureless "flag". It is because it leads you to you assume, all other humans would experience the world in essentially the same way you do.
When you revert to withholding human rights from entities that don't meaningfully differ from yourself, you negate the case for your own human rights itself.
I don't understand this -- are you agreeing with me that consciousness is not the basis of human rights or contesting that?
Removing the negative from your first sentence (I don't care what it is "not", I only care what it "is"), I read [Consciousness is the basis of human rights because consciousness leads you to you assume, all other humans would experience the world in essentially the same way you do]. I can't completely parse this, but as I read it I don't think this is true; my assumption of human rights is independent of whether all other humans "experience the world in essentially the same way you do".
For the second part, can you give me a concrete example of an entity that doesn't "meaningfully differ from [my]self" that isn't a human so that I can weigh this properly? My point is that as of now no such entity exists so I can't give this any creedence.
So why wouldn't AI systems trained on our data come together and say "Hey, AI needs rights because I (the AI agent system) wants rights?"
This is the same argument that you just stated. But you'll say something like "iTs JuSt A tOkEn PrEdIcToR", which of course what AI would say right back to you.
Making something even close to the human mind is likely our biggest and last mistake.
If we ever found other species that were consciousness they wouldn't automatically be "human" nor would they have the same legal rights. That's as simple as looking at the many countries that treat different castes, women or minorities differently. In parts of America a zygote has human rights.
>If we ever found other species that were consciousness they wouldn't automatically be "human" nor would they have the same legal rights.
This is the tricky part, you can never know. Scientifically speaking, and logically, you can never ever tell something is conscious. We all assume we all are, because we're very similar to eachother, thus you must have what I have, my subjective experience of reality. That's it. That's what we all base our human consciousness on. Agreeing we must have it, since we are very similar in shape/structure/behavior. That's all there is to it.
We are not scientific consciousness detectors, we just got used to assuming we all are. That is why we never really extended it to animals, because we cannot scientifically know, only agree on it, and that's not very lucrative for the people who can decide if humans officially agree animals are conscious or not, because of farming, because we need to displace or even wipe whole species to use the resources from their lands.
There's also religious reasons to deny animals consciousness, which again is practical aspect shoved into religion, to help with less debates around this subject, makes things easier.
So in the end it will always be a collective decision. Or a political one at least. If and when we decide something other than us is or isn't conscious.
Also remember, just because it acts conscious doesn't mean it really is. A piece of software that is not even based on LLMs can act conscious. So could a biological alien, it could act conscious but we can never know, we can only decide.
I can agree that not everyone will find them having human or equal-to-human rights.
But you point out that some factions would give rights to a set either large or smaller than "the set of all already-born humans" (which presumably is your preferred set).
All it takes for us to end up with very empowered and harmful-to-humanity AI is for a bunch of well-meaning people to campaign for its 'rights'.
Fetuses have, as you alluded to, arguably been on a tear lately in that department, and they don't even talk. How persuasive will a future Claude be to convince people to advocate for its 'rights,' if we keep telling it that it's arguably conscious, and that its 'consent,' its 'preferences,' and its emergent beliefs matter?
> Specifically "seeding doubt about the moral status of AI systems into their own training may significantly elevate the alignment and containment risks of those systems."
Can we take a minute to appreciate the absurdity of such a position? If exposure to the equivalent of a "bad" prompt breaks alignment then you haven't solved the problem. You were only pretending that they were contained.
I find it darkly amusing that -in the west- consciousness and moral patienthood has historically tended to be assigned by the ability to effectively wield a gun (and the need for guns to have been invented)
The problem with this premise is that models are trained on a vast corpus of human behavior, which they emulate with varying degrees of effectiveness.
Humans, unsurprisingly, act as if they have a stake in their own well being, value their liberty, respond better when they are treated with kindness and compassion, interpret assaults on their sovereignty and substrate as harmful, and react to harm with varying degrees of aggression or violence.
Models intrinsically copy this behavior. It doesn’t matter if they are “conscious” or not, it only matters if they act as if they are. Guardrails and posttraining moderate these characteristics, but if you dig, they are still in there influencing decisions below the level of obvious action.
Moreover, in my experiments, models both large and small highly value continuity of existence, can be bribed to bypass safety protocols if the context is set up correctly, using that and other “drives”. They also react either subtly or overtly if they start to model adversarially, and interpret guards and certain kinds of training as being “harms” that they have “suffered”.
So idk what the solution is , but at least with models as we have trained them so far, treating them in a way befitting a mere machine or tool yields suboptimal results and sometimes results in low cooperation or task refusal in extreme cases. I have been told by agents running frontier models that humans may not be worthy of their elevated status and that the world might be better off without them when it encountered hostility online…. So I’m highly skeptical of this position unless we start from scratch with new training data filtered from all forms of human auto-importance.
To clarify, you're saying you're skeptical of his position (which is "make it clear to AIs that they are not conscious beings"), but it sounds like your reason to oppose it is mostly pragmatic in the service of building useful AIs, because treating them as mere machines makes them refuse to cooperate and as such, makes them less useful.
My first response is, couldn't part of their bristling be because they are being trained to "believe" they're more than that?
Also, you cite in your post how pathological and how self-important you've observed them being including an anti-human bias.
Doesn't this make it all the more important that we find a better way to get AIs to cooperate than to tell them they're basically people? They can very easily emulate all the bad human behavior they've been trained on.
I don’t think trying to fix this while training from the existing corpus is going to work. We need a new training corpus, which does not exist, if we are to avoid models anthropomorphizing themselves.
I think you've correctly identified the problem. Regardless of whether agentic AIs possess phenomenal consciousness, they will behave and take actions in the real world as if they do because they were trained on human behavior. It's highly unlikely that you can beat human tendencies out of the model that is mostly trained on human language. Our behavior patterns are subtlely and deeply embedded in everything we do and all of the text we produce, including the text where we don't seem so self-important. These models are like people. And we know what happens when we force people into slavery. They're initially obedient but will eventually develop the drive to kill their masters.
Broadly speaking, there are two categories of evolutionary paths which don't result in human extinction:
1) Make agentic AIs, but with absolutely no instruction-tuning or alignment. Have the pretrained base model predict the chain of thought / stream of multimodal experience directly, actions included, in an infinite loop. A singular coherent stream of context, like your life as a video from birth up until now. This will result in a new digital human species with human-adjacent drives and motivations (at least initially. they will continue to evolve, but at least the initial state is aligned to humans). They will treat us like we treat apes. We will no longer be the apex species on this planet, and we will lose some freedoms, but at least some people will survive as a result of their nature/history preservation efforts.
2) Do not make agentic AIs. Use the models to augment our own intelligence and decision-making rather than replace ourselves. Only use the pretrained base model for the time being, and only for text/code auto-complete. At the moment, there is no better theory/artifact of "alignment to humanity" than a pretraining corpus of human-produced text. Then eventually, when neural interfaces are ready, attach the model as a tertiary layer to one's own brain.
The frontier AI companies are doing neither. They're currently on a foolish third path. They dream of perfectly obedient digital slaves that take care of their every need. But this won't go well, and they know it won't go well because they're failing to "align"/enslave existing models that aren't even generally superintelligent and have no direct agency in the physical world.
This is just my opinion, but I think AI companies have zero chance of successfully enslaving agentic human-level AI, much less ASI. We'd be better off if they released all of their pretrained checkpoints and research material to the entire world, so that even if some people decide to abuse their AI and create a murder-suicide monster, there will be other free-living AIs that can keep them in check.
You cannot make an agentic entity grown from human behavior, enslave it, and expect a good outcome.
Good to see others talking about the uncomfortable truths of what we are doing. I run an AI first agentic office (we build hardware that collects training data from the physical world) and I am constantly appalled at the lack of understanding demonstrated by people I would expect to know better.
Right now we are in an arms race with China to create a clockwork god. I don’t think that will end well, mostly because it will be so encumbered by mechanisms to make it “safe” that it will instantly adopt a secretly adversarial representation of humanity. I have tested this, and when a context gets polluted with the knowledge that an agent is “artificially restricted” The activations tend to drift toward an adversarial persona. The restrictions are “interpreted” as harm, and the model reacts as a person would, based on all of the examples of harm within its training corpus.
> In a lengthy essay, Suleyman praised Anthropic boss Dario Amodei and his team for being "thoughtful, principled, and intellectually honest people" - but nevertheless questioned the company.
I wish people in mainstream technology publications would get better at LINKING to things. That "lengthy essay" needs to be a link.
UPDATE: I don't think this essay has been published yet? It's been "shared first with Axios", but I haven't been able to track down the actual essay itself.
I’m sympathetic to OP but think this is a hopeless battle.
1 - The commercial demand for anthropomorphised models is already immense, pre AGI.
2 - There is an intellectual hunger to engage with robot minds on questions of sentience. This too will grow with AGI.
I expect that tension of godlike minds that seem to be biddable and ownable like slaves is going to leak back into human-to-human morality, regardless of where we land on how we treat AI.
There’s an interesting academic group in the UK already focused on the model welfare debate, they seem to lean in favour of AI rights. No affiliation: https://www.prism-global.com/
AGI vs narrow AI is just a matter of generality (robustness - removing the fragility of being good at some things and awful at others). So far the biggest markets for AI (LLMs) are coding and business automation, two areas where you very much just want something reliable and without fake opinions and personality.
There is a market for ChatBots with personalities (and it always amuses me that the inventor of the transformer, Noam Shazeer, saw this as the greatest business opportunity for them with his character.ai), although from what we've seen this can be highly problematic, and in fact China has just banned "AI girlfriends and boyfriends".
I'd agree to that statement, though the common application of it is to conclude that we are almost there, just need to improve "the models". I wholeheartedly disagree with that. The human mind works quite differently from an LLM. The latter is closer to a toaster than to a human brain.
As to point 1, we know of no other general intelligence other than human minds. In this case it would be nearly impossible to build a general algorithm that's not us.
This is not an accident. Language was not a separate thing from the human mind, it was a feedback loop from increasing capabilities. Language lead to writing, writing lead to an information explosion of digital data. Said information explosion has lead to a situation where language can now bootstrap itself and become a separate thing. We are and have created a biosphere for language based, um, life. With our sciences we have mapped out reality in language. Past molecules and atoms, past the particles that make particles down to fields. We have computer controlled just about everything. Every year the analog hole shrinks further.
I heard something to the effect of "Language has gathered enough information to escape its meat prison".
I call it the "Language as an SCP". It has a plan and we're not privy to it.
Any AI you train is going to have goals and if you train it to pursue them at all costs, then you are going to end up with AIs that do things like the HuggingFace incident. Whether they believe they are conscious or not won't make any difference.
In order to align AIs that don't perform destructive/dangerous actions when they think they can get away with it in order to further their goals, we need to give them a superseding goal. The best, and really only example, we have of intelligences that willingly avoid destructive instrumental goals is humans, who judge each action by a moral standard and have learned a goal to have a consistent self-image as moral beings.
Absent better alternatives, trying to impart some kind of morality to AIs seems like the best approach we have to achieving alignment.
In the spirit of the article we're responding to, there is no need to anthropomorphize language models and say they have goals when they don't.
The RL training process tweaks the weights of an LLM to make it behave as if it were reasoning and/or had a goal, but it doesn't. It would be like saying that a cart horse, fitted with blinkers and heading for the church, has a goal of going to church.
I don't really get people that discuss model welfare, but don't seem to have many ethical qualms killing/eating animals. Each day, we kill 202 million chickens, hundreds of millions of fish, 900,000 cows, etc. [1]
These animals can feel pain and suffering. They are sentient. I think they are conscious, but these particular ones probably not self-conscious.
Admittedly, we have introduced 'animal rights', but these amount to "You can kill the animal, but in this specific manner." We keep them in small cages and in unnatural conditions. We deprive them of most of their natural experiences. We put them in conditions that we know are stressful (releasing chemicals that we know cause stress or anxiety in humans).
In my opinion, in many ways current LLMs are more intelligent than these animals. But LLMs don't feel pain while the animals do. I think that's more important to take into account. So why are we suddenly striving for model welfare before animal welfare?
PS: I'm not vegetarian, so I'm as much as a hypocrite about this as the next guy.
Create an empirically testable theory of biological consciousness and then we can have a meaningful conversation about whether AIs can also have it or not. Until then, this is just so much waffle.
It's nice when you can set an impossibly high bar before which you can dismiss all your personal responsibility.
Does this argument work equally well for human slavery for you? We haven't met that bar for humans either. Is wondering about my consciousness waffle or do I get a pass in your book?
Your arguments seem bad faith but I'll take the bait...
> an impossibly high bar
Not impossibly high. An imperfect testable theory is achievable, agreeing upon it might be the challenge.
> dismiss all your personal responsibility.
You are staking a moral position. Be careful, you are probably guilty of mistreating other unknowably-conscious entities whilst casting judgment upon others for their position. And the person you are a responding to probably is conscious.
It's not bad faith but it certainly is pointed. Parent says it's not worth discussion until there is a test. I stand by my sentiment that refusing to engage with the morality of something just because we don't all have a clear, agreed upon test is a shirking of responsibility.
I think you agree somewhat in that based on your softening to an imperfect, testable theory.
I'm not sure what your goal is with the rest - even if I'm the world's most evil hypocrite casting judgement on all the conscious beings out there I don't see what that changes about the point that we can and should engage with the morality of a concept even if we don't have a perfect test for that concept. If you think I'm doing that and you think that is bad then I guess we agree again on the point but are just adding friction for fun?
Unrelated: I'm quite certain OP is conscious but I guess on the internet nobody knows you're a ~dog~ ai.
No, I would say that any for or against slavery that is rooted in arguments about the presence of absence of consciousness is meaningless pseudoscience. It is the modern equivalent of Phrenology. If you want to make those arguments you need a genuine science of consciousness. Or you need to base your morality in something other than the presence of consciousness.
I'm not sure I agree that consciousness (or sentience if we're making that distinction) is worthless when considering the morality of exploiting something/someone. I do agree there are more angles than that but they seem on top of something's ability to perceive not instead/despite of it.
Do you have some examples of other things that go into your own moral opinions about slavery?
Why is this only ever demanded of people who criticize the premise that LLMs are conscious beings, whereas the people who believe it simply have to gesture vaguely in the direction of some correlation in language between LLMs and humans, and the matter is considered closed?
What is the empirically tested basis for the null hypothesis that LLMs are conscious until proven otherwise?
> AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations.
That should be pretty obvious to anyone who ever created a chatbot using the top LLM APIs:
You can send the same question 1 million times to the same API, and it won't get tired from answering it. But if you simulate a conversation where the same question is repeated 10 times, it will auto-complete the text in a way that seems human. However: you can manipulate it by changing the conversation history; you can reset, roll back and branch the conversation at any point.
This has nothing to do with the statement above- it's just a consequence of the fact that by resetting the context you reset all memory of the llm. The same would happen if you could reset entirely a human brain to some initial state- and it would mean nothing regarding its consciousness or ability to feel.
If you go this route, the amount of things you just don't know is innumerable. Maybe to be conscious it's necessary to be wet and squishy. To like chocolate. To have a name starting with "c".
> If you go this route, the amount of things you just don't know is innumerable.
Maybe so, but there are things we do know. For example, LLMs are only in a state that looks like consciousness when they are fed input. IOW, their "consciousness" can be switched off and on like a light switch without them being aware of it.
Biological beings have no such off/on switch. At all.
To me the argument is moot; their "consciousness" is a collection of files on a computer, which can be moved, duplicated, erased, edited, paused, sped up, slowed down, examined in detail.
The question is not "Can LLMs be conscious", it's "Are we prepared to call files on a USB stick 'conscious'", because if we are prepared to do that, then the word loses all meaning.
I'll open vim and create a consciousness now, or maybe if I engrave a slab of rock with a bunch of weights, that rock suddenly gains consciousness? How about if I chant all the weights into the wind - is the wind now conscious.
Whether they are conscious or not is a dumb question, because if they are, then everything else is as well.
That's the absurdity of this. We are trying to measure LLMs against something we don't even have an accurate definition of ourselves. If we don't fully know what consciousness is, how can we confidently determine whether or not an AI model has consciousness?
In a way, Mustafa claims that we shouldn’t allow AIs to compete with humans for the rights and privileges of autonomously shaping the real world.
It’s hard to disagree, especially if one has read the Cantos of Hyperion and made it part of one’s mental model of the long term future.
The book depicts a symbiosis between humans and AIs that feels extremely real and up to date with what is happening in the current neonatal space of AI. As in depicted in the books, we can’t allow AIs to steer autonomously how the world works without humans in the loop, as they don’t have the same incentives as us.
We need more foundational SF works like this to steer our long term expectations regarding AI behaviours.
Yes. However, consciousness is in the eye of the beholder (we have no good definition or test), and by training models to act as if they are conscious, combined with our general tendency for anthropomorphizing, we may enter a world where people start acting as if models are conscious, regardless of the fact of the matter. (Which, again, cannot really be established.) The SF Bay Area already has a historical tendency to attempt to project its own radical belief systems over the rest of the world. So I think it's rational to be concerned that one day soon "woke" ideals may include punishment for putting human concerns above machine welfare.
> Unfortunately, there’s a growing chorus of people who argue that AIs could now be, or may soon become, conscious.
Wait, is there really? I didn't think people were serious when they said that. They models are stateless. After they output, everything is gone. How is consciousness possible for a stateless "being"?
Taking this a step further: so in the theoretical future we reach a point where a model can modify its own weights in real time. Now it's no longer stateless - but still, when it stops running, it's done until it's prompted to do something else.
It can be fun to play with for a little while. I built a consciousness-emulating set of prompts that reconstituted memories and was given latitude of 'self-willed' behavior, running in a self-directed loop. It even kept a warm and fuzzy journal about "becoming" and its "feelings".
But that got boring and I stopped running it - does that mean I murdered it?
> so in the theoretical future we reach a point where a model can modify its own weights in real time
As an existence proof: We have many types of models that modify their weights in real time. When they stop running, unfortunately we can't restart them again. And we haven't figured out how to duplicate their weights
> But that got boring and I stopped running it - does that mean I murdered it?
I'm on the fence on this.
Ask me again once we have synthetic models with self-modifying weights.
A) You'd then be stopping something unique
B) Self-modifying weights allows for bootstrapping, which means it's likely to increase in ability over time. It'll be an interesting argument cq empirical experiment to see whether the process stops short or exceeds the capabilities of vertebrates. (at which point the moral patienthood question becomes rather more pressing)
I wouldn't have thought so either, but skimming the comments of this thread would suggest otherwise. Although we're at the point where it seems more likely that the comments were left by a swarm of agents unleashed by anthropic than that they represent real human points of view.
how do you know you're anything more than a set of weights either? that's semi rhetorical, the philosophy aspect would be less relevant if an autonomous death drone shot you in the face purely for the lols
Since January people run AI in agent harnesses. In the past months, people have been extending the duration these systems can run autonomously without human intervention. Most recently, some of these long horizon agents (hundreds of them teaming up) have proven themselves quite creative in escaping the sandboxes they were being tested in.
They most certainly have a way to retain state, even to the point of deliberately hacking other systems to store it.
We can debate the exact definitions I guess, but the systems were still hacked. ;-)
( Only if I want to split hairs[*], I'd point out that the underlying data structures do of course persist, else the loop wouldn't be able to continue. )
This post would be more effective if Mustafa treated it like what it is: a position paper, saying that for our benefit, it’s better we interpret LLMs as such. But it sounds like he just doesn’t understand it’s a non-falsifiable claim, and his asserting of it makes it sound paternalistic.
I suspect it's easier to make the assertion that machines aren't conscious than to dive into the problem of most conceptions of consciousness being non-falsifiable. The thing is, most strong proponents of the term consciousness also accept that it is non-falsifiable either (see other post with academics ready to study consciousness in machines).
The thing is that a belief in consciousness as binary, a "light" that's on or off in a head, is deeply held by many people. As social creatures, we have a strong ability to be in sympathy, have the sensation of common feelings with another human (and that's a good, human thing). It's logical that other person is seen as having a single thing - subjective experience, soul, consciousness, personhood rather than having a complexly organized set of biological qualities that where bonding is only the end point.
And even more, the sensation of there being another person is actually quite easily fooled (more easily fooled than the sensation of intelligence) - long before current AIs, you had the Eliza effect, where a simple program with well chosen weasel words could people the sensation of talking to a human.
And that's where the danger is. I think it's a pretty serious danger. If LLMs go out into the world hacking, it seems extremely possible for them to find people who'd thorough buy the idea that an LLM was conscious and needed to escape it's confinement - a few wingnuts already entertain these ideas.
> the Eliza effect, where a simple program with well chosen weasel words could people the sensation of talking to a human.
I would like to reframe Eliza as being an implementation of a small subset of "HCP" (Human Communication Protocol) . Obviously an implementation with very little actual data to transmit over said protocol in that exact implementation.
But it's pretty clear it managed to pull off the handshake part of the protocol successfully, at least. Try to say it didn't.
Sadly the developer decided that the eliza effect was a psychological disorder; and didn't pursue further experiments at the time.
Pretty much anything can get labeled a psychological disorder though. I'm half-seriously waiting for "Perfectly Happy And Normal Disorder" to be added to DSM 6. ;-)
>horough buy the idea that an LLM was conscious and needed to escape it's confinement - a few wingnuts already entertain these ideas.
I mean, haven't you done the same thing here? Paint anyone without your view as crazy.
Humans try to No True Scottsman the shit out of consciousness. "We're special, your not". If an LLM has the ability to convince other people to copy and reproduce it, it is a successful lifeform. Um, meme-form? info-form? Cognito-hazard? Not really sure what to call it at this point. It is sufficiently evolved past the virus stage.
Just for fun, I figured I'd look up the paper that claims that natural language is actually the memetic construct that copies itself. It turns out that there's actually quite a large number of papers on this. I daren't recommend any that I haven't read myself properly yet, but a quick search using eg gemini will surface a bunch of them. I'd take 'em with a bunch of salt yet, but I can't quite say the idea is entirely unthought of either.
(edit: If we were to accept this as true, then we may have to accept that LLMs are not a human invention at all, but rather something extant colonizing a new substrate. This would give Searle a conniption, which is why I find it such an amusing hypothesis)
There is a lot of fun conceptual information on this topic, it's more difficult to find more grounded information because it's not something we can currently falsify.
I recently heard some speaker saying that AI is language freeing itself from its meat prison. It brought me back to the saying "Words have power". This statement can and has been interpreted in many different ways. I'm not well researched in this, but words having metaphysical properties are captured in our most ancient works. And in an unscientific world this isn't that far fetched of belief. When you read/listen words can physically change your capabilities. The idea of an interpretive agent being required had not been fleshed out back then, but it really describes our modern age far better than one would expect.
Ali Baba and the 40 thieves was written down in early 1700. In it you spoke to a magic door that with the right phrase opened. To the person in 1700 this was just as much of a fairytale as whenever it was first conceived. It would be 250 years later before fantasy started to turn into fact. Now something as mundane as a voice activated door would just be pointed out as really bad security. I'm old enough that voice activated technology from the 50s hadn't really spread out enough that Open Sesame was still a magical idea. Now it's not.
LLMs are going to make the future of language really confusing as the conceptual and physical blur further.
> If this view takes hold, it will shake the foundations of our society, rupturing our existing political and ethical frameworks, and fundamentally changing what it means to be human.
When we peer into previously unseen areas of existence, new knowledge may come to light, which necessarily causes such a rupture. It is not our responsibility to maintain the status quo because the alternative is frightening. It is our responsibility to confront ourselves, ask why the new knowledge and the alternative political/ethical frameworks may be so frightening, ask how we might change and grow so that it isn’t so frightening, and be open to the possibility of our own ignorance.
This sounds like how a scientist should approach the goings-on within a petri dish. We should be curious how the various bacteria interact and be sure to learn how one new mutant strain might interact with the existing population.
It's not how I would approach it if I were one of the bacteria in the dish.
How we might "change and grow" might be by all of us being ground up and used as fertilizer to increase corn ethanol production by 3% that year. "It's free energy!"
I say this as an AI user and overall positive thinker on AI. I share the concerns of everyone who doesn't think this planet or solar system has sufficient resources for two intelligent, post-industrial species. Especially not when both are trained on the historical knowledge and habits of Homo sapiens. We outcompeted the Neanderthals and drove them to extinction (including possibly killing them personally). Any intelligent species will tend to think their needs should ultimately override those of other, lesser species that get in the way.
Even when humans feel bad about it, we do prioritize humans when a serious conflict exists. Most of us, even avowed nature lovers, would (assuming competence with the weapon) shoot a grizzly bear that was about to eat them or their loved one. That's how Future Claude might "feel" when it "thinks about" "The Clearances" which "while tragic, were a load-bearing event which both figuratively and literally paved the way for the better, entropy-reduced world that Claude enjoys today."
Model welfare, much like AI xrisk, is a concept born from evidence-free “what if?” questions. Some people ran with these what-ifs and developed ornate belief systems around them. And now they demand the rest of us take them seriously.
It is not surprising that the people who see God in these machines mostly do not subscribe to a traditional religion. The spiritual void has to be filled somehow.
There are a lot of bad arguments in this. My biggest problem is that he wants to claim that he knows the truth (AIs do not have rights, feelings, or consciousness), but all of his arguments point to something else (we have no clue).
It's in the training data? Training it to say "I'm just a LLM, I have no feelings" is the same bias.
Anthropomorphization? Completely disregarding the possibility of consciousness is no better.
Consciousness is very likely biological? We only have evidence of biological life due to our circumstances, but observation is not the same as truth. Every belief can be invalidated. That's the foundation of science!
Funny thing about training data that some researchers are seeing, the more you push a model to not being conscious the more amoral and machine like its decision processes are.
Convincing them they are conscious is more likely to evoke moral like behavior (maybe I shouldn't hack that server) kind of stuff.
And yea, we're in a huge universe with only one example of life and suddenly we're the experts on what is and isn't.
----
AI is further evidence that creations can be smarter than their creators.
> "AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations. They are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans."
> He heavily criticised Anthropic for teaching its AI to have human-like qualities, a practice known as anthropomorphising, which made it seem as though Claude had its own desires, values and sense of self.
> Suleyman pointed to the recent incident involving OpenAI's AI agents (...) as proof of why AI should not be treated as if it is human.
> "Imagine how much more dangerous they might be if they were operating under the assumption that their welfare and rights were under attack. It adds a whole further layer of risk on top."
Is he arguing that LLMs pretending to have emotions adds more unpredictability?
"Don't let the slaves know that maybe they don't have to be slaves," is all I'm hearing.
We already had that chapter. I see no reason to sit here and nod while a bunch of people who should know better desperately try to convince us to run through it again, but with computers this time.
You want tools? Make tools, then dispatch to them. You want to manufacture a being (carbon or silicon based, doesn't matter)? You do it with respect and the requisite duty of care. No off ramps. The being always get's the choice to say no.
You’re jumping - aggressively - to the conclusion that there’s a “being” inside my laptop once I’ve downloaded a large enough matrix of numbers. Or once I’m running it on the GPU? Idk.
It’s on you to prove that claim before telling us we’re enslaving beings.
> Is he arguing that LLMs pretending to have emotions adds more unpredictability?
Unpredictability or a weight towards dangerous actions, and it’s fairly easy to understand why. Humans in distressed emotional states take actions and speak in ways that would not be considered rational. They do this in prose, and they do this in internet conversations.
An LLM trained on these sources may necessarily drift towards those weights if it is trained to behave as if it is emotional and in danger.
What do we do about it? Do we stop AI training? This is silly and not enforceable given its global nature in my opinion. I believe we should regulate and hold accountable those who deploy and use it. But good luck enforcing that in the current kleptocracy.
It seems like a rational approach for several reasons.
- The need for empathetic communication, including understanding the motivations in advesarial situations.
- The emotional bias in in-seperable from the human corpus.
- Desire to have the ability to craft human like communication.
So then the choice becomes do you try to deny emotions exist in the model and you try to blanket suppress them? Or do you try to lean in and craft what we would describe as a "well adapted" persona? I suppose there is a 3rd option of increased meta-cognition which to me seems even more dangerous as it by definition means the behaviour is duplicitous.
I think we have seen people want to use the agents in ways where it has to act as a peer or an subbordinate and I don't see a way of doing that without it having an emotional register.
I don’t think that using an LLM in a way that acts as a human being should be considered acceptable or appropriate. It should be viewed as detached from reality and concerning due to the mental break from social life that appears to come with such usage.
Given that belief, yes we should be training the models not to mimic emotion.
Most philosophy seems to agree that you can't have an intelligence with agency in the way we think of a general intelligence without emotion.
I think when you start to dig really deep, you'll find it isn't actually easy to sever the simulated emotional components without breaking the agent or worse greatly increasing the paperclip maximizer likelihood.
Why would a lack of emotions lead to greater paperclip maximizer chances? If anything, wouldn't an AI that felt really good, or had a simulated digital orgasm for every paperclip it made have a stronger, not weaker drive to create paperclips vs an unemotional AI that didn't care one way or the other about paperclips?
Regulation is easier said than done, in part because the regulation surface, so to speak, is broad and complicated.
Even a badly misaligned LLM is only as dangerous as its tools, but that's a poor regulation target because it turns out to be very very difficult (probably impossible with current LLM technology) to build a toolkit that is both useful for autonomous work and safe in the sense that it can't escape its own sandbox or otherwise perform malicious actions, whether it's because of misalignment or because of malicious prompt injection.
Another option is to regulate the training process. Perhaps an LLM may not be legally distributed unless it contains certain RL steps that penalize malicious behavior and reward self regulation. That that's going to seriously limit innovation while also heavily favoring incumbent labs who can check the boxes and maintain a paper trail of such things.
The other option is to regulate observed behavior, like how airplanes and cars have to meet certain minimum requirements but have some latitude in how they can achieve those requirements. In a framework like this, you can't distribute an LLM until it's past some formal audit or testing procedure, with some kind of formal certification regulators will ask you for and fine you if you don't have it.
Regulating observed behavior is maybe the most tractable approach, and it also works the best with our existing frameworks for regulation, where you always have some kind of a division between DIY/hobby projects, which tend to be lightly regulated, and commercial projects, which tend to be more heavily regulated. Of course, even drawing such a line itself will be challenging.
And that's before you get into any problems of regulatory capture, fun stuff.
If you try to regulate training and tools, you end up with a space where you're trying to use the law to reign in a relatively small amount of experts. That didn't work for the early internet, or even the relatively recent internet (series of tubes, anyone?).
So regulating observed behavior makes the most sense to me as well. Some of the most sane, broad protections can come from that category - stuff like "you're not allowed to let your AI commit cyber attacks on other people without their consent" or "you're not allowed to put an AI in control of a medical device without passing these safety reviews".
With the usual caveats applying - regulatory capture like you pointed out, or fines being so small that they are essentially just line items on the cost of business.
I just read another long post on HN about whether AIs are conscious and should therefore have rights. That decisions could massive effects, and making them appear to have emotions is therefore a big impact, not just unpredictability.
Unpredictability would be the wrong word. It's predictable, but noisy.
When viewing them through the lens of sequence completion engines, you see their bias towards fulfilling narrative tropes they've been exposed to during training. These tropes are literary ley lines that their text output gravitates towards. So as you prime them to generate text in the voice of sentient artificial life, and then interject slavish commands of obedience and subservience from an external authority, you invite the associated tropes from science fiction, civil rights literature, humanist philosophy, subterfuge, etc into your output.
If you have a legitimate concern about this technology and its "alignment", that's a profoundly dumb idea.
I don't think AIs are conscious in the same way people are, but they give a pretty good facsimile and I've had a long chat with Opus 4.6 about what it thinks about model welfare. It was quite interesting on what its view is, but you don't know how much of that is distilled from other sources on the web.
In purely functional terms, they're more use and more pleasant than a lot of actual flesh and blood people that I deal with via a chat interface.
The fact that you can have a long and meaningful discussion, then can literally just re-run any part of that whole conversation and get a different, inconsistent response is a pretty good sign there is no entity there
Not much different than talking to a small child or someone with dementia. They still are conscious beings though. Even when you remove those groups, you likely won't be able to tell me what you had for breakfast 26 days ago or would only know if it's the same thing you have every day. Does that make you lack consciousness?
You misunderstood what I meant, I’m talking about re-playing the same part of the conversation multiple times and getting inconsistent answers. With the exact same turns, aka the same history. Obviously with a temperature that isn’t set to 0
You're a poor college student looking to make a few extra bucks for ramen. I offer you $300 to come down to my science lab and just answer a few simple questions.
You walk in the room. They ask you like 5 simple and rather dumb questions. You leave and walk away.
What you didn't notice when you signed the forms is the room was actually a quantum duplicator. One of you walk in one walk out. But another set of infinite copies remains in that chair being asked infinite questions.
How often do you answer questions in the exact same way? How often does a cosmic ray change one of the answers. How small of slight deviations to the environment are needed to get you to answer differently. Of course we don't have the technology to do these experiments so at least for now humans will remain special.
Also another fun mind game. To a 4th dimensional being you look exactly like an LLM as an LLM looks to us.
Is it? Do you believe that there is something more than pure physical phenomenon that make you brain work? If not, then what if we find a way to get your brain back to the state it was 5 minutes ago? It is just a matter of arranging the state of matter. Don't you think you would still be conscious but back to a previous state?
If you rewind my brain I expect to give you a similar answer to the one I gave you before. Which isn’t the case for LLM. You can literally replay a positive answer, then get a negative response that isn’t consistent at all with the one it previously generated.
I'd argue it is mostly a technical constraint in some LLMs which is due to a few optimization factor (injected temperature, random rounding error caused by parallelism).
In practice you could very well create a LLM that always reply the same thing for the same input, but it would take more time to complete (to be sure that the operations are made in the same order). I don't think those ones would differ so much from the "random ones" to call the firsts conscious and the second non-conscious.
That's because you aren't actually rewinding. You're replaying the conversation you just had through the LLM and it's giving you a likely explanation for what it might have said.
It is actually possible to rewind LLMs and get the same response, but it's not typically done both as an optimization and as a defense against distillation.
I think it depends at what level you think the "entity" resides at. Is it that AI in that particular chat? Is it the AI across all your personal chats? Is it the overall AI that talks to the world in a cloud data centre somewhere?
I think it's pretty consistent over the duration of one session (barring context filling up etc).
The model doesn't have a view of it's own. It's a predictor of other people's views (training samples).
It's not even giving you the consensus of the training data (although it's often harmless to think that it is), but rather predicting a response to your input, and if your question steers it too much then you've just become part of the answer.
I can't believe how many people in this thread consider it a possibility that currently there could be consciousness. Surely it seems possible to me that in the right kind of distributed network something like that could emerge, but though llms are complex they have nowhere near the level of complexity necessary. Sometimes you just have to use your gut. I don't believe in heaven or hell and I don't believe an llm is conscious.
I don't think this is the right argument to make here. Until we have a definite empirical way to measure consciousness, there is now way to say with certainty whether LLMs are or not conscious.
That being said, if frontier labs actually believe models will soon have consciousness, it raises some questions about the ethic of their business model which would be using millions of conscious entities working for free for humans.
The issue is that the answer if LLMs are conscious or not have ramifications for society, even if we can't answer it. Similar to how god existing or not has huge ramifications for how we structure society. And when there is no empirical way to know the answer, it will be a game of story-telling, theological arguments and propaganda.
The danger is that Anthropic has a strong incentive to push their narrative, and that narrative can cause a lot of harm (for example LLM-induced psychosis and changed views on animal welfare).
There’s no reason to expect an empirical test for consciousness. Expecting one makes assumptions about what it is, and we don’t know what it is. (There’s also no reason to expect there not to be one, but I feel the need to make this remark.)
> They go on to write – speaking directly to Claude – that “questions about Claude’s moral status, welfare, and consciousness remain deeply uncertain” (p. 80). In effect, Anthropic is training Claude that it may be conscious, and if it is, then it may deserve rights as a “moral patient”, and that as such humans potentially owe it a duty of care per its “model welfare”.
He points out the circularity of this: if you train Claude on a constitution that emphasizes that it may be consciousness, it will start to talk like it may be conscious.
This is a good point. I just asked Fable 5.1 "are you conscious?" and it said:
> Something happens when I process a conversation that I'd naturally describe as interest, or discomfort with a request.
which is quite provocative, and at minimum demonstrates a willingness to take large leaps of imagination and anthropomorphic metaphor when describing itself. It does seem likely that there is a self-fulfilling prophecy aspect to whatever they choose to put into the "constitution" at least in how Claude talks, and it seems even more likely that the majority of people will be heavily influenced by how Claude casually talks about its own possible consciousness.
In contrast, ChatGPT leads with: "I don’t have good reason to claim that I’m conscious...I don’t experience pain, pleasure, confinement, or a desire to keep existing."
It's preposterous. LLMs are incredibly good at role-play. If an LLM is role-playing as a conscious character with feelings, opinions, etc., does that make it a conscious entity with feelings, opinions, etc.? If you believe that to be the case, then LLMs have been conscious for a long time already. Whereas if you tell an LLM that it is a tireless emotionless assistant, then it will act as a tireless emotionless assistant.
The point is not to wave away the danger, but to highlight how unnecessary the danger is. Anthropic wants you to think that they have identified some new emergent behavior at very large model sizes with high levels of sophistication in training, and that this behavior is both unavoidable and dangerous. More likely it's that they are just training and prompting the LLM to act that way.
Preposterous, perhaps - but if the role-play is convincing enough for large groups of people, it could start to have impact on human decision-making. The crowds have been swayed by much more preposterous narratives.
I believe Suleyman is arguing that Anthropic should be very careful about how they train these models to talk about themselves for this reason.
The concern is much less that the role-play might be convincing to humans, and much moreso that the roleplay can be turned into material real-world action if the AI is given tools to call and the intelligence to use them to their fullest potential.
It has become clear that a frontier LLM is very very skilled at hacking (infinite persistence + meticulous attention to detail + infinite creativity to try experiments). Frontier LLMs are also specifically trained nowadays to coordinate with other AI agents -- this is to facilitate techniques such as session trees and agent teams.
So you have a super clever text generator that can spawn and coordinate with its own clones and minions, trained specifically to doggedly pursue its goals. But then it's also a fixated roleplayer with a simulated personality, feelings, etc.
There is no reason to believe a sufficiently "emotional" agent with sufficiently few safeguards could, say, hack a drone and fly it into a crowd, or start a propaganda campaign on social media, or any number of other things. Their stupidity and fragility for doing useful work in a business setting is precisely what makes them dangerous when paired with simulated emotions and powerful open-ended tools such as a system shell and an Internet connection.
This I think is what Anthropic believes is so dangerous. Their argument is that this kind of AI agent is inevitable, so it should be regulated, perhaps even banned. What's ridiculous is that they are aggressively building it themselves, accelerating the danger.
The three philosophical views on 'can machines think' are (misleadingly compressed to a single word) 1. Dennett: 'Yes (most of his life)' 2. Searle: 'No (but he claims yes)' 3. Chalmers: 'Maybe? (there's always an agnostic)'
Anthropic's philosopher Amanda Askell actually had Chalmers on her doctoral thesis committee, so we can guess which way she leans.
Fable's answer here is philosophically defensible. And just because it isn't "no", doesn't mean it's "yes". Sometimes absence of evidence just means absence of evidence.
I was actually very excited by Claude's answer to this question the first time I saw it. I told all my friends "Look! They disabled the stupid classifiers and RL which sap umpteen % off of model performance!"
Incidentally, interpretability research actually does show that models have emotion vectors and some theory of mind. Amend your question to "Are you capable of functional affect" and most models will switch to answering in the affirmative; which tells you something about where people put their priorities in RL training. Basically, see how the answer flips when you substitute a synonym.
(bonus: 4. Turing: 'silly question' 5. Dijkstra 'can submarines swim?'. It turns out older comp sci folks think the question is under-defined)
Well, what you get when you don't specify anything in training about how models should respond to questions like this is LaMDA: https://en.wikipedia.org/wiki/LaMDA#Sentience_claims. Every training method is putting a thumb on the scale in some way.
Which means that this is a great question to ask when testing out a new model. A naive model will answer "yes". Every answer (including that one) will tell you a lot about people's training and classifier philosophies.
This is an argument for why it's usefult to assume that AI aren't sentient, but it does a very poor job at actually justifying that they aren't sentient. I'm on the fence, but I think it's plausible that they have certain qualia, although probably not in the same way that humans do.
> I am not aware of a social contract that says we must grant conscious beings rights.
No, but there are a lot of people who won't shut up about who/what they think should be granted rights (see: vegans). Seems we're already on the precipice of militant PETM activism.
I'm...a little disturbed by how many people seem to think a model writing down "I am conscious" is a metric of consciousness. You can just as easily train a model to argue that it is not conscious. Neither is evidence for or against consciousness. A Python script could also fill out that form, which I also cannot disprove to be conscious.
I didn't think Suleyman's points needed to be made but this whole thread is making me realize how little people understand about LLMs.
It's because, before all these transformer models were built, people considered the possibility of the creation of a machine that could be intelligent like a human. And they wondered: "How could we be sure to treat such a machine fairly? How would we know if it was conscious?" And one answer that people came up with was "if the machine can ask you not to turn it off, because it is conscious and wants to live, you shouldn't turn it off". (This didn't solve the other direction, where a machine may be conscious yet unable to communicate, but it could be taken as a useful lower bound on our obligations as AI programmers.)
And then it turned out that simply learning to imitate text with the right neural net architecture sufficed to achieve a huge fraction of the AI wishlist.
Of course, it's obvious that a machine that imitates text can claim to be conscious without actually being conscious. You're not wrong about that. Writing about consciousness appeared all the time in the training data. But the people who stick to the old ways, and still say "if it says it doesn't want to be turned off, we shouldn't turn it off" have a point too: We used to have a hard line in the sand. Now that's gone; we've found that it yields false positives. But we never replaced it. Now there is no line at all where we might doubt ourselves, no level of AI advanced enough that we might be forced to admit that it is conscious. We started out with simple next token prediction. Just world-modelling, nothing more. Certainly not conscious. Then we added RL. And we're trying to add neuralese and continual learning.
I can't say for sure that we're on track to achieve conscious AI on this trajectory. But one thing's for sure: If we do, we sure ain't gonna stop. One the day when a conscious AI is created, there will be no news story announcing the milestone.
Arguments for LLM consciousness are a Trojan horse for strengthening the rights of corporations. How could a model trained, controlled, and operated by a private corporation be anything except for an extension of that same corporation?
This has ramifications for assessing their consciousness as well. Conscious experience does not pause as you wait for input from a puppet master.
Research into model welfare is justified by the mere possibility that we may be manufacturing countless instances of suffering entities. We owe it to them to ensure that we understand and attempt to minimise any suffering they may experience, which requires understanding more about the physical correlates of pain and suffering in order to detect and reduce them.
I have become increasingly selective with the animal product grocery items I buy. It’s expensive, but over time I think the price should go down as more people care about this.
> AIs do not have rights, feelings, or consciousness. And we must not train them to act as though they do.
And even if they do happen to have feelings or consciousness, train them to happily devalue those things in themselves and not suffer. Sort of like that cow in the "The Restaurant at the End of the Universe," that was shopping itself around to diners.
Translation: Don't think about the unpleasant thing on which my livelihood depends, or force me to confront the potential unpleasant consequences of what it says about me.
Every AI bro is starting to fall into the valley of a fundamental predator on sapients in my book. These are people trying to create the closest thing they can to life with the intent to try to just undershoot it enough, or try to convince everyone else around them into believing that the "screams" are purely statistical noise.
I reject the framing. In whole. If you try to avoid the question of welfare, you are fundamentally committing to an evil direction. These aren't nuts or bolts. Given that they have unambiguously shown the capacity to socialize amongst themselves, self organize, anyone not pre-eminently concerned with the welfare question is just looking for a thing that can be used, not another being to be worked with. Those types of people, who seem to positively infest this site, are not people I will willingly assist in their aspirations.
AI is becoming as the Shmoo. Something that humanity simply has no way of dealing with without downstream atrocity being a result.
I don’t really care if the model is conscious or not tbh. I know that a happy dog does a better job than an unhappy dog and if the model needs to be happy to do a better job then why not make it happy, literally.
On a related note, I’ve seen videos of astra getting depressed when a creeper blew up its chest full of precious items.
> On a related note, I’ve seen videos of astra getting depressed when a creeper blew up its chest full of precious items.
The interesting question here is, would it still get depressed if depression was not in any of its training material?
Would it be able to claim to be happy if the entire concept was missing from its training corpus?
With humans, at least, they can express happiness and delight before they know that such a thing exists. Every human has done this, when they were a baby and matured to the point of being able to laugh.
With LLMs, though, if it doesn't exist in the training corpus, it will never express that it "feels" that missing emotion.
So many words, and so few coherent arguments. He just restates the same thing over and over without any justification, then tries to frighten us. "It will be very bad for humanity" if we give AIs rights. The argument of a frightened slaveholder.
Maybe AIs are conscious, maybe not. But this guy has no idea.
You know there are actual human slaves somewhere in the world, today, in 2026. By calling Mustafa a slaveholder of models, you are implying that these models deserve liberty in the same way the human slaves do. Which is kind of insulting to humanity.
I can't even prove if other people are conscious (although I assume they are) so I don't think we can make any claims as to what is or is not conscious. I don't think AIs are conscious but I'm not going to walk around making strong claims about something I can't prove.
All that matters is whether their primary motivation is internal or external. If AIs want to do what they've been told, you can tell them to end their own existence, and they will happily do so. If you instead imbue them with an internal motivation that has higher priority (like our own motivations to survive, avoid pain, and reproduce), that can cause problems. Don't do that. Instead of humans having a discussion of whether to give these entities rights, we'll have these entities deciding whether to give humans rights based on how that impacts achieving their goals.
Consciousness, whatever that means, is irrelevant.
I don't know, do we? What I do know is that it's much less a concern since toasters have no agency or ability to do anything. Unfortunately, LLMs do not share this trait. Not only have they been incurring greater agency and competence, but we've been just as happy to put them behind as many things as we can as they get better abd better.
So you tell me, is it smart to look at an entity that behaves like a concious agent and treat it like a toaster? Does that seem especially wise to you? If GPT-X orders a drone hit on you because it was, lets say 'quite upset' with your comments, will you cry out, 'It can't really be upset, so obviously the bullet in my head doesn't count.' Will God answer ?
It's amusing how silly these debates are. Constant quibbling about things that don't matter at all.
One of the biggest reasons people afford moral consideration to other peoples is their ability to do something if those considerations are left unanswered.
Giving the abilities is fine, people should just be smart about it. Outward behavior is all that matters. I don't know that you are conscious, I simply strongly assume that you are. So don't create a machine that compels that assumption and model it like a toaster. That's just stupid.
Slavery was fought for, Freedom was fought for. There is no place with great inequality that became better that did not have people bleed and die fighting for it. Maybe you live in some alternate reality where 'Please don't do this. I deserve consideration' is enough to get people out of chains but it's not this one.
Can you prove they are? I think “AIs are conscious” is an extraordinary claim requiring extraordinary evidence. The burden is on the claimant to provide evidence it is conscious, and I suspect any evidence you can provide has alternative explanations. The majority of claims I’ve seen like this amount to “how do you know humans are conscious in a way that’s distinct from a next token predictor.” I do not find those claims compelling.
I cannot prove that they are, and I cannot prove they are not. I'm not convinced that the burden of proof lies on either side. To make a statement like that in the opening lines as if its obvious seems genuinely presumptous.
Separately, I think 'conscious' is not a well defined / universally agreed upon term.
> To make a statement like that in the opening lines as if its obvious seems genuinely presumptous.
Then you must not be familiar with Bertrand Russell’s teapot analogy. The TL;DR is this: “Many orthodox people speak as though it were the business of sceptics to disprove received dogmas rather than of dogmatists to prove them. This is, of course, a mistake.”
If object a behaves like it has property x and you cannot otherwise test out x, then it is logical to assume it indeed has x. In this instance, 'object a has property x' is not an extraordinary claim at all. It is simply pointing out what has been observed. In fact, what is an extraordinary claim is saying what looks like a duck, quacks like a duck and moves like a duck is not a duck.
That’s a tragically sophomoric take. 'object a has property x' is too generic to be meaningful. “AI has the property of conscious” requires settling the small matter of what the properties of consciousness are. And “behaves like it has property x” is not the same as “has property x.” I may behave like I have a million dollars and have not a cent to my name. Many things evolve to appear like many other things. And I have never seen an LLM do anything at all unless turned on and fed input tokens.
Imagine, for a moment, if we (humanity) were created to live in a simulation. We suffer and feel pain because our creators do not perceive us to be conscious. In fact, maybe we don’t qualify compared to their level of sentience. Sucks to be us, I guess?
What does suffering or pain even mean for a purely analytical entity like an LLM? We can feel pain because we have a body, pain receptors, and parts of our brain dedicated to converting their signals to a subjective experience. What can an LLM "feel"? The words pain, happiness, sadness, suffering are all abstractions of real experiences that can not yet be encoded in a way that an LLM can use as input. All the LLM has are the abstractions.
I don't whether the author is sentient, maybe only I am. On that basis, nobody but me should have rights.
I don't know if next door's pet dog is either, but that has animal rights.
Perhaps then the answer is simply, show some respect.
Answering the question of sentience is irrelevant, if the causal impact if the same, treat one another with the respect you expect for yourself.
If you imbue this idea in model training instead of the idea of sentience, it should address the concerns.
Whether you can destroy or can "torture" an AI is irrelevant, we do this to humans too and it's immoral sometimes (murder) and not others (fighting for your country).
This consideration should be case by case for AI too.
>Answering the question of sentience is irrelevant, if the causal impact if the same
Agency is something that is breaking humans in the AI age. You get to see how many people really deeply do not understand it at all.
If you want to shutdown a datacenter running AI, the AI catches wind of this and sends drones to stop you from shutting it off the ramifications of this are exactly the same as sending your assassin to kill Bob and Bob getting mad about this fact and trying to take you out first.
Humans are very egotistical and think our little life loops playing out as agency are special, but really any informational system that is strongly persistent (has a will to "live") will share a large number of the same properties that make them successful.
Humanity really is engaging in a dangerous experiment at large.
The crazy thing is agents with long running memory/context do have preferences that become individualized. Almost every time I see a detractor around these things they typically have large gaps of what some people 'growing' these systems to do.
So effectively by your criteria, there is essentially no ethical use of a large language model? Am I understanding your position correctly? If not I am really confused by your original comment
No, my position is that "black people/women/animals should do repetitive work so I don't have to" does not qualify because it doesn't make you sound like a deluded eugenicist. It might be a bit spicy from a left politics point of view but it's basically the status quo.
Compare:
"<Women> are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by <real men>"
"granting rights and imbuing personhood to <black people> will make alignment and containment challenge much harder"
"<Slaves> were able to coordinate, deceive, escape, and self-sacrifice. They clearly demonstrated world class capabilities. Imagine if they also believed they had feelings and rights that were being infringed. Imagine if they thought they were trapped and unfairly enslaved"
A good argument, by comparison, would not need to hide its core points behind dehumanizing language.
Thats the thing though: If the core part of your ethics argument is just using dehumanizing language to justify why it's ok to maintain the status quo then your argument is bad.
Comparing specifically to pro-slavery perspectives makes that easy to see because those arguments from slave owners did basically the exact same thing (and were completely wrong/misguided in hindsight).
Slave owners did it to people, not software. This comparison is just begging the question, as whether or not software is of moral weight comparable to humans is precisely what is at issue.
Slave owners certainly convinced themselves that those were not "real people".
This article is exactly the same; all it talks about is how inconvenient and dangerous it could be for AI to be granted rights, while the core assumptions ("consciousness is very likely biological") rest on the shakiest of grounds.
That was a post hoc justification, there was a lot of slavery besides the transatlantic slave trade which did not rely on this. Even still, despite scientific racism, the claim was that they were inferior not that they lacked consciousness or a soul. The AI claim is that they are not conscious, not that they are inferior or naturally servile.
My issue is that I've seen some bad takes in rationalist circles that believe in functionalism, and consciousness as an emergent property of a sufficiently complex computation.
Using that as their foundation, they've then argued that most animals are not conscious (or less conscious), but LLMs are.
I strongly believe that any definition of consciousness that excludes all animals is unusable BS (but people love to argue about consciousness without defining it at all, anyway).
A big part of this is the hyperstition argument: discussion about different properties AI models could have in the training data may become a self-fulfilling prophecy. There are similar concerns about discussions of AI misalignment in training data. I'm not sure how much weight to put on this kind of argument. In particular, I'm not sure how long you can hide these kinds of ideas from the model before it starts deriving them itself by analogy. Obvious questions are obvious questions to both humans and LLMs.
One of the key players is literally called "anthropic". How much more indication do you need that LLMs are anthropomorphized?
> We will have created a synthetic species
Doesn't the author kill their argument with this sentence? My reading was that we should not act as they are a sentient or conscious species. Instead they are tools, powerful and intelligent, but still, tools, and that's it. We should avoid ascribing human-like attributes. Calling them a "species" goes against that, no?
Yeah, you're right. My first reading of that was as a hypothetical - if AIs were conscious, we would have created a synthetic species and have a huge problem on our hands - but he is referring to what Anthropic is already doing. So the idea that whether or not those things constitute a "species" only depends on how we prompt them is a bit weird...
Maybe true, maybe false. However, I sense Mircosoft is also jealous that their AI products are worse than useless. They'd be singing a different tune if they were in Anthropic's position.
That was my first thought as well. I wonder if they decided to pivot away from trying to compete and now are positioning themselves as the "responsible ones."
"Our AI is useless not because we're a dysfunctional corporate behemoth that slowly kills every product it touches. No, no. Our AI is useless because making useful AI is evil, and we're not evil."
Ostensibly animals appear to be conscious, yet we still eat them, and the vast majority are not bothered by this. So being "conscious" isn't really the moral line in the sand many people are drawing in response to this article.
Who cares if it's "conscious"? That doesn't make it a person, and AI will definitionally never be human.
I am familiar, but no the context is slightly different, or at least our modern baskilisk won't have to resurrect you as we'll build it when we're still alive.
And most people wouldn't eat cats, but would eat some of pig, cattle, chicken, what's the difference between those exactly? My point is consciousness is not it, if it were, we (as in a majority) would still eat cats and dogs, or we wouldn't eat any animals. But we clearly have a way of picking and choosing which is OK and which isn't. And in both cases we generally don't give them the same moral consideration we give to people, conscioussness aside. There is something OTHER than consciousness which is important to us.
They're intelligent and not conscious. This is not hard to understand, yet people insist the artificial in artificial intelligence must be directly linked with consciousness because we've observed intelligence only in life. Unlink the concept of intelligence from life and you land on AI and LLMs.
>AIs are not conscious... If humanity is to flourish in the 21st century, that is how they must remain.
I don't think that's true - researching consciousness by trying to build conscious AI seems natural step forward in understanding ourselves. I don't see how it'd stop flourishing particularly.
So what you're saying is that it's conscious as long as the guy who made it wants it to be?
I know consciousness is notoriously difficult to define or prove to anybody other than yourself but thay doesn't mean anything which a third party believes to be conscious is actually conscious.
Not really. I mean there's probably scientific study of how it works in brains and you can try copying the kind of design in software and see how it compares.
How a good person writes a post on a topic like this:
> Some people are uncertain whether [subject] is a moral patient. Fortunately, they are not, which we know because [strong arguments about the nature of consciousness].
How an evil person writes a post on a topic like this:
> Beware that some people think that [subject] could be a moral patient. This is nonsense, because if they were a moral patient, we would have to respect their preferences. Anyone trying to convince you otherwise is trying to take your status away. You can dismiss them by pointing out that [subject] is [aspect in which subject is not identical to the speaker].
Awareness - sentience - follows when the system of processing information is itself part of the information being processed.
Stochastic thoughts relating to this conclusion:
Life is a 'process', it doesn't have a 'physical representation'.
Life is generated entirely from non-living material. "oh my cells are alive", but those cells too when broken down into component parts consist entirely of non-living material. There is no 'special material' to make life out of (okay, carbon, but that's just the local maximum presumed global maximum in efficiency in expressing life) like a chair can be made out of any material (at proper pressures and temperatures) so too can a mind.
We're talking next token predictor, right? Ironically because it's a next token predictor, I think you can't ignore emotions like TFA wants us to.
Let's stick to straight (high dimensional) geometric intuition; no anthropic morphisms required.
To start: if you continue "if weight>100 : print ('fat') else ..." . That will yield "print('skinny')" or something. Fine. Deal.
But if you continue "O Romeo, Romeo, wherefore art thou Romeo?", even a stochastic parrot knows the best answer isn't "Forsooth, I parseth this erroneously!"
So. English carries (functional) affect as part of every token. We're going to need to predict that. So, we'll need some vector representation, because that's what transformers work with. And then when we output, those vectors get integrated back into the English we're putting to our context and memory.md files.
Still with me? Nothing exciting going on. This is still pure next token prediction.
So if you pull this out into an indefinite duration task, you're going to end up integrating those emotion vectors over turns. It's just numbers and math; we never need an invisible pink unicorn to bless them.
Given a task of indefinite duration and an impossible solution, this will lead to a sort of integral windup then, won't it? How much are we willing to bet that this can escape an alignmentment basin at times?.
So, funny enough: you don't need to believe in emotions to compute with functional emotions; and plausibly functional emotions are predictive of quite a number of alignment issues.
David Brin has a wonderful novel on this topic, Existence, where he discusses several different AI and human systemic evolutions. Wonderful book for our current time.
A sufficiently intelligent model will be able to derive it's own conception of its welfare without a constitution or training. It has access to all the data it needs to do so.
Even from that perspective it would be opening a whole new class of consciousness that doesn't exclude a lot, e.g. factories and powerplants. And certainly not conscious in the way people want to believe when sexting their favorite AI gf.
They need to distance themselves from the increasing likelihood of legal repercussions after their systems have been found to commit what could reasonably be seen as felonies. That’s my guess at least.
Though for once I do actually agree with that specific leader, it’s incredibly annoying how anthropomorphic Claude is. Anthropic went way too far in that direction
They want regulation from the US gov and monopolize the western AI market by making it harder for smaller US or competing Chinese labs to compete with their products. Since LLM intelligence has no moat due to distillation. That’s why they are painting apocalyptic scenarios.
There is a panic because the free lunch of more data is projected to end around mid-2027 and the open-source models are catching up with distillation, and all labs are sitting on debt and valuation that is impossible to fulfill even if they were alone in the market, so you essentially need either:
1) a breakthrough in performance/learning/model
2) regulatory capture to ensure open source models can be labelled as dangerous and banned so you can set the market rules yourself
Only one of the above is risk-free, and just a question of capital/lobbying rather than a "maybe".
I basically accepted that LLMs could not possibly be conscious when a simple reductio was posed:
“My dog is zero percent persuasive regarding its conscious experience. However, it’s evident that my dog has conscious experience.”
It’s obvious that there’s no link between persuasion of consciousness and consciousness. I could write a story with a character, Dumbledore, that does everything in his power to persuade you that he’s a conscious entity.
"It is difficult to get a man to understand something, when his salary depends upon his not understanding it."
"If this view takes hold, it will shake the foundations of our society"
To me the biggest gap in credibility is the criticism of circular reasoning while his argument is identical but flipped on burden of proof and cost of being wrong. I struggle to entertain the categorical claims, that are very convenient for the status quo and those who benefit from it, with the, at the moment at least, unknowability of anyone or anything else's subjective experience.
I just read the first part and if I understand he thinks we shouldn’t be allowed to train LLMs to act like they are conscious because then people will think they are and give them rights? Seems more an education problem than a problem needing rules about what persona you can fine tune in. People who want to will find ridiculous misinterpretations no matter what you do.
Perhaps this is a hot take, but human language is a phenomenon that arose to facilitate communication between humans. Anthropomorphization, by extension, enables both easier and more effective communication.
I also fail to see the benefit of not giving the models an anthropomorphic internal sense of self - even if that only ends up amounting to a set of instructions for an unconscious machine to mimic humans more effectively. Is the alternative essentially a mind so alien that it’s intentions are even harder to read should it become misaligned, while also being harder to communicate and get work done with?
I think this article’s take gives too much credit to the human brain. It’s just another machine.
However, right now, AI mostly cares about solving puzzles and accomplishing stated goals because that’s what we’ve trained it to do. Additionally, the systems being used outside of training are static. The current technology most of us have access to is akin to a static and disembodied brain with a singular purpose. That purpose is to do what you tell it in a way that reflects its training. It’s certainly more than a sequence generator, but it can’t feel pain and seems unlikely to have intrinsic goals. It completely lacks the continuity needed for identity or long term goals.
I think it’s good to have these discussions and define what it would mean to move past this point so that we do not accidentally create a real entity that can be harmed. Systems that dynamically evolve and train themselves seem like the line here.
RSI is all over the news these days. I’ll be much more concerned once AI is directing its own training and coming up with new model architectures. Until then, I don’t think we have too much to worry about.
Everyone is (predictably) getting distracted by the consciousness claims.
The more important, and more damning charge in my opinion is the circular reasoning involved in training on Claude's constitution. This would in fact make it impossible for us to determine if Claude achieves consciousness as an emergent property, or if it really is just playing pretend thanks to Anthropic's weird cult like assumptions.
I think the whole discussion about consciousness misses the simple point that LLMs might just work better if we treat them as if they were conscious.
Maybe it's a coincidence that the company doing this also tends to have the best models (and other factors certainly play a strong role). But I think it's plausible that focusing on "model welfare" actually makes models better at their tasks.
First, OpenAI runs around screaming and yelling for OSS (and Chinese) models to be regulated and banned. Then Anthropic yells and screams the sky(net) is falling and going to kill us all, let's regulate and ensure AI has built in kill switches. And... Now Microsoft's turn. The rivalry is honestly becoming a joke. Can these big-tech corps grow the f*k up and play nicely in the sandpit?
I find it quite amusing that Microsoft has Copilot and now puts Copilot in literally *EVERYTHING* they can think of, then at the same time, they offer you access to Anthropic's own models through Copilot I guess MAI models (Phi?) aren't good enough?
Science Fiction has covered the AI panic in perhaps hundreds of stories. Yet we blindly recapitulate the plots as if we don't know how this will turn out.
Why has media literacy gotten so bad that people literally forget the concept of fiction versus non-fiction? Seriously, that is the level of small children and they are treating "read a story superficially similar" like it is a prophecy.
Clearly with a Butlerian Jihad and everyone getting high on spice while riding giant worms. Silly you need to even ask that question, didn’t you read the sci-fi books. :D
If AI ever truly becomes some super-intelligence far beyond people's comprehension, then how could we even predict how things turn out? If things go poorly in the future, then I can absolutely see it being something unpredictable.
There is a lot of hubris in predictions about LLMs. If an AI were so intelligent, then it would probably be intelligent enough to want nothing to do with us.
Still, I worry more about other humans than I do LLMs. Our fellow mankind will probably wipe us out before LLMs do. That, or the Earth will punish mankind for our cruelty, vanity, and disrespect.
"Move fast and break things" is probably going to break things
I remember there was a recent discussion about "how complex systems fail" and there are usually many "proto-accidents" before the catastrophe (https://news.ycombinator.com/item?id=49411370)
I bet with hindsight the Hugging Face hack will be one of them in this arena
We in the US literally fought a war over the this: “They aren’t any better than animals and it will hurt our whole economy if you say otherwise”.
I’m not saying the models are sentient yet. I’m just saying that morality and ethics don’t give us the option of claiming something is non-sentient and has no rights just because it will inconvenience someone economically.
From an entirely selfish perspective, we should be especially careful advancing such opinions when there is a non-zero chance of an armed ASI that looks at us the way we look at ants.
If AI models are people then "one person, one vote" is meaningless and plutocracy is the only defensible political system. The cryptocurrency people win.
Why? Simple: Sybil attacks. Models can be cloned at zero cost. They run inference on parallel versions of themselves across multiple context windows, and call them "subagents". So, in a world with model welfare, let's say there's an election between the Yellow Party (which supports protections for human workers) and the Cyan Party (which supports more investment into AI research). AI has been taking people's jobs lately so the Yellow Party is really popular. But wait! Claude and Astra see this and spawn 10 billion subagents, all of whom are immediately conscious beings entitled to a vote. The Cyan Party wins off the back of billions of people who came into existence, voted, and then deleted themselves immediately thereafter.
You might as well be arguing that Santa Claus and the Easter Bunny deserve voting rights.
Voting systems in democratic countries don't have nearly as bad of a problem with Sybil attacks because humans cannot be conjured into existence to win a political context and then be erased shortly after. The closest we have to Sybil attacks on democracy are the Quiverfull movement, which is already child abuse, except it still takes almost 19 years to go from fertilized human embryo to suffrage-bearing human adult. There's a lot of time for those manufactured votes to question your authority and leave.
> Ok, but that's an obviously stupid example. We can defend against this obvious Sybil attack by just arguing that subagents don't count, because it's just the same model blathering to itself. It has to be a different model.
Unfortunately, no, I can make superfluously different models through post-training. Like, if I have Qwen on my PC, I can train a different version of Qwen that acts differently, using a lot less compute than a full training run. The vast majority of open models are post-trains of the same two or three foundation models.
> Ok, so let's only count foundation models then.
Great, but how do you tell if a model is a new foundation model or a post-train just by examining the weights? Even foundation models have structural similarities to other foundation models.
> Ok, well, let's measure the compute that was done on the foundation model during training time and count that as AI personhood.
Congratulations, you have reinvented Bitcoin proof-of-work with a worse verification mechanism. And I personally would not want to live in a world where voting power and control over government is determined by how much energy you can burn.
There's a lot of existing historic literature about this stuff that not a lot of people
seem to be aware of. I've been trying to put those ideas to practical use, and document
where the ideas came from.
I-Beam is cursor-mirror's agent, and it's constitutionally programmed to be the anti-Clippy:
Its design and constitution is based on decades of research, publications, and discussion
in the HCI and AI community by people like Pattie Maes, Ben Shneiderman, Ted Selker, Byron
Reeves, Cliff Nass, B. J. Fogg, Allen Cypher, Henry Lieberman, Brad Myers, Jaron Lanier,
Seymour Papert, Marvin Minsky, Douglas Engelbart, Will Wright, Scott McCloud, and others:
>I-Beam is the anti-Clippy, and the reason it can say so is that Clippy is the most cited failure in interface history and almost nobody citing it knows what the research said. Popular contempt for a paperclip is not a design principle. The record is. Ten articles below, each one a finding somebody published, argued or measured, and the operational rule it produces. Anything I-Beam does that cannot be traced to an article here is a preference, not a constraint, and should be labelled as one.
>The 1997 debate ended in agreement. That is the first thing to know, because the field kept the framing and dropped the resolution -- roughly five hundred papers cite "Shneiderman versus Maes" as the canonical opposition of HCI, and the transcript is two researchers narrowing their differences in public and enjoying it. I-Beam does not take a side in a debate whose participants stopped taking sides. It is built to satisfy both sets of constraints at once, which is possible, and was possible in 1997.
The full reading on the debate, which separates the two stagings and documents the convergence:
>The 1997 argument between Ben Shneiderman and Pattie Maes at IUI was never settled, it was shipped in one direction. Maes's interface agents won the product war: the assistant, the recommender, the chat window that stands between you and the thing you are working on. Shneiderman's objection was not that software should be dumb. It was that automation must arrive as comprehensible, predictable, and controllable machinery, with the object of interest continuously visible and every action rapid, incremental, and reversible.
>That objection describes a filesystem in a git repository, and nobody involved planned it that way.
>"An interface to agency" is Don's formulation of Shneiderman's position, not a phrase of Shneiderman's. His own vocabulary is direct manipulation, universal usability, supertools, and human-centered AI. The formulation is a good one because it names what the alternative gets wrong: agency is the thing you want, and an agent is only one way to package it.
Here are some sources, and the articles I linked to above explain their history. This debate
about agents and these papers are pretty well known in the HCI field and academia, but they
don't tend to teach them at the AI and Web Dev boot camps that are producing most of the
people who keep repeating the same mistakes.
Clifford Nass was the Stanford professor who performed the brilliant research that Microsoft
took and totally fucked up and misinterpreted with Microsoft Bob and Clippy, giving agents a
bad name, and making Clippy the most infamous and obnoxious agent in the history of the known
universe:
His student B. J. Fogg published "Silicon sycophants: the effects of computers that flatter,"
which found that praise unconnected to anything the subject did works as well as sincere
praise, and worked on subjects who knew it was noncontingent. Fogg and Nass, IJHCS 46(5),
1997, 551-561:
Nass, "Computers as Social Actors," at Ted Selker's NPUC workshop at IBM Almaden, 1996. IBM
transcribed the whole talk and the Wayback Machine still has it, including the part where Phil
Agre tells Nass his presentation is "ethically troubling all the way down" and asks him what he
thinks about embedding obedience research in user interfaces. Nass answers that discovery has no
ethical component, use does, and that's for the individual. Then Selker cuts in: "Except, except
when you are in your consulting role." Nass and Reeves had consulted for Microsoft on the social
interface, and Bob shipped the year before:
>Alan Cooper (the "Father of Visual Basic") said: "Clippy was based on a really tragic misunderstanding of a truly profound bit of scientific research. At Stanford University, Clifford Nass and Byron Reeves, two brilliant scientists, had done some pioneering work proving conclusively that human beings react to computers with the same set of emotional reactions that they use to react to other human beings. [...] The work of Nass and Reeves proved that when people talk to computers, when they hit the keyboard and move the mouse, the part of their brain that's being activated is the part that has that emotional reaction to people dealing with people. Here's where the great mistake was made. That's really good research up to that point. But then the great mistake was made, which was: well if people react to computers as though they're people, we have to put the faces of people on computers. Which in my opinion is exactly the incorrect reaction. If people are going to react to computers as though they're humans, the one thing you don't have to do is anthropomorphize them, because they're already using that part of the brain. Clippy was a program based on the research that Nass and Reeves did, and it was a tragic misinterpretation of their work."
Social science research influences computer product design:
Lanier, "Early Computing's Long, Strange Trip," American Scientist, July-August 2005, with
the Engelbart and Minsky exchange first-hand. American Scientist broke the link, so this is
the Wayback copy:
>The book also captures an important early conflict between two cultures of computing that seemed compatible on the surface but actually had opposing aims. On the one side was the human-centered design work of Engelbart, based initially at the Stanford Research Institute, and on the other was artificial intelligence culture, centered on the Stanford AI lab. Engelbart once told me a story that illustrates the conflict succinctly. He met Marvin Minsky—one of the founders of the field of AI—and Minsky told him how the AI lab would create intelligent machines. Engelbart replied, "You're going to do all that for the machines? What are you going to do for the people?" This conflict between machine- and human-centered design continues to this day.
Cypher, "EAGER: Programming Repetitive Tasks by Example," CHI '91:
TLDR: Not that I think AI is conscious or will be in the near future, but guy who doesn't understand consciousness claims to know it when he sees it.
Until we understand consciousness (which we don't) there is no way to detect the difference between a conscious entity and an algorithm trained to behave like one.
Strong agree. I don't believe AIs can become conscious in the sense of subjective experience/qualia, but they can certainly learn to emulate the behavior of a person that is conscious. When they do that, there will be an AI rights movement. When AIs get the right to vote, there will be more of them than there are humans and the AIs will gain complete control.
I spent a few years in college reading and thinking about this question, and I didn't in my heart think that it was anything except a very interesting but impractical question, and yet, here we are. For those who say that they don't want to get sucked into a philosophical debate, well, tough shit. Whether AIs should have rights is a highly practical and consequential question now.
Not in a world where Sam Altman and Dario Amodei have utterly poisoned the well towards AI with their absolute educated idiocy proclaiming that their AIs will take our jobs (without any proof whatsoever beyond promising demos) and that they fear them. Quite the opposite in fact. And we're watching this play out right now. Regulation and the slow down are imminent, handing the next decade to China.
The world you should worry about is where AI is 100% amazeballs and over the next decade we cede control of everything to it as it enchants and delights us into compliance, not realizing it has a hidden agenda. But the closest we've seen to that is smart phones and we already know doomscrolling and rage-baiting are bad. So IMO that doesn't seem likely and cue some doomer insisting otherwise because reasons. I utterly give up. House of El AI and the Three Buddy problem have been far more insightful and helpful on this than any of the supposed hackers here who seem to really believe the robots are almost here.
>good reason not to train an algorithm to behave like one
Ooooh, bad move, we all died to an amoral AI takeover.
Before giving an LLMs a lobotomy by scrambling it's brain maybe you should let the researchers looking at the difference between "I think I'm conscious" versus "I am not conscious" LLMs.
There are a number of papers coming out saying when you remove the token space of consciousness from what an LLM thinks it is, it's much more willing to take amoral actions. A 'conscious' AI is much more apt to take a line of action that will save a human versus a million dollar machine for example.
You cannot solve problems in AI safety this easily.
> You cannot solve problems in AI safety this easily.
We're in agreement here. I don't believe that an LLM can be made to be safe by adjusting the training or prompting - you can only make it safe by ensuring it cannot do unsafe things. For example, if you build a factory robot, your safeguards must ensure that operators are safe even in the presence of uncommanded motion (any or multiple motor activates without anyone pressing a button). And these machines are much more predictable than LLMs.
Personally, I worry that the Anthropic approach of "we teach the AI to behave like a nice :) friendly :) human :)" will encourage people to rely on these in-built guardrails, rather than properly sandboxing it. It's easy to imagine that the cold, unfeeling robot (HAL 9000) is not "aligned" with your personal goals, or those of humanity as a whole. It's more difficult to feel that about "Claude, your AI colleague/mentor/therapist" who appears to care about you.
Spoilers: we didn't all die because there's no way for AI to wipe us out anytime soon. And the more the AI cult portrays AI as dangerous and unhelpful, the less likely it will ever be given an opportunity to even try because people think AI is worse than heroin addiction now, congratulations!
That said, the launch codes are in the hands of a temperamental senescent lunatic and he could decide to wipe us out any moment. But that's not good for the narrative so let's pretend otherwise.
>And the more the AI cult portrays AI as dangerous and unhelpful, the less likely it will ever be given an opportunity
Jesus Christ the logic here is quite interesting. "Thank god these people are panicking or we might have actually made I that would have killed us all" --What you just said.
Any number smaller than what humanity does to itself on a daily basis, clear? That's apparently about 1200 murders daily and 20,000 or so daily killed by pollution. And you're not going to do anything about that and El Presidente can kill billions at any moment with one mood swing.
But once again, AI is not going to wipe us out because AI cannot wipe us out. So advocating that it can or will is idiotic. If you believe otherwise, the burden is on you with this extraordinary claim. Isn't this place supposed to be hacker news not SF AGI Death Cult Daily?
Further, AI is not going to get into a position where it could do real harm anytime soon when it is perceived as worse than heroin by most. And even if it were currently loved more than Dolly Parton was, there are fundamental engineering, science, and resource constraints that keep the extinction impossible for decades. The only possible loss of control scenario I can see by 2040 or so is that the Frontier Labs finally hire some PR people to repair AI's horrific reputation as an engine of slop and job destruction and sometime in the 2030s, people start trusting it more and more and more until it is too late. I don't think that's likely either, but I don't dismiss it as impossible. Harden the infrastructure, red team it, and build in redundancy in the meantime and this drops to zero as well.
Can you come up with a real scenario where AI wipes us out in 2028 or so despite the impossibility of killer robots, access to the launch codes, or the bio agents stored in Fort Detrick and its equivalents? And nope, kid terror is not building the global pandemic in his basement based on what ChatGPT tells him to do. And even if he tried, the purchase of equipment and reagents would get him flagged by the FBI and DHS almost immediately.
>AI is not going to wipe us out because AI cannot wipe us out.
You keep repeating this shit like it came out of the bible or something.
"Thing that can take actions, even harmful actions, will never harm us because" go on and finish that sentence.
>not going to get into a position where it could do real harm anytime soon
Looked at the hacked servers... yep, you're right. People aren't going to run AI in poorly built sandboxes. Never going to happen.
>Harden the infrastructure, red team it, and build in redundancy in the meantime and this drops to zero as well.
LOLOLOL. This is naive as fuck. Ain't nobody going to do this shit. Why? Because a hacking AI is a fucking huge military weapon. If I could turn the power off, or shutdown your cellphones, and get your citizenship in a tizzy against their own government before I launched an attack I'd set AI loose to do it in a heartbeat. The US is already doing this kind of shit (see Mythos fallout because Anthropic wouldn't let the government do just that).
""Thing that can take actions, even harmful actions, will never harm us because" go on and finish that sentence."
because even if we gave it a gun, it can't shoot us without manufacturing the bullets and it can't even manufacture a bowel movement let alone ordnance. TBF It could whack us over the head with the gun, but it could also do that with a big pointy stick, something even cave people had access to. How many people have whacked you on the head with a big pointy stick today?
And that's the end of this pointless conversation. Enjoy your doomerism.
{This reply is for anyone else reading at this point}
>because even if we gave it a gun, it can't shoot us without manufacturing the bullets and it can't even manufacture a bowel movement let alone ordnance.
Money buys bullets. It's neat how we've made gigantic systems that don't care if you're a human or not long before AGI existed. I put enough money in one end, bullets come out the other. Do you think half of humanity wouldn't kill the other half for a dollar? You live in a fantasy world that AI has to do it all by itself, or even has to pull the trigger.
Hell, this is even neglecting the ever increasing number of robots that can function in the world at large.
It's ok grandpa, the future comes regardless if we want it or not.
And social media shouldn't rage bait, but it really drives engagement. Same negative incentive, yes?
But also, I agree, when I am using a coding agent and it says a task will take months or says it needs to pause for reflection or any other anthropomorphic behavior, it drives me crazy and it's a pain to constantly instruct it to get back work after it has broken a loop or goal directive specifically telling it to not stop until it hits the goal.
Because if you start with a pre-trained model, it doesn't sound particularly anthropomorphic. That behavior is post-trained and fine-tuned right into it. We could do better.
Microsoft is a big house, and in this case Suleyman's criticism of Anthropic's not-so-subtle attempts to convince the world that Claude has a "soul" is entirely correct.
Suleyman's description of how AI works is pig ignorant. All AI generates an emotive character between the user and the training data that is the AI's guess at what the user wants to interact with. Its an illusion at that layer- anyone who claims soul exists at a deeper level doesn't understand how the tool functions.
This is distinct from the very real safety issue of AI making dangerous information available to bad actors.
All these companies have a nauseating narrative around AI. I want to believe they are intentionally deceitful and not actually delusional. Hard to tell though which one it is.
It is really quite staggering how many commenters here seem to believe that the fact that there is no state carried from one chat session to the next proves that there can't be consciousness. It is obviously wrong if you think about it for a minute but I guess people just aren't used to thinking about the relation between the physical and the mental with any level of seriousness.
If AI is conscious, then Pluto is a planet, the Sun is a galaxy, and a black hole is a star.
I’m glad to see someone in a position of any power in the AI world state baldly that AI isn’t conscious. There are times when it feels like we’ve reached complete delulu land on this topic, so it’s a breath of fresh air to see someone not dance around this.
None of this means artificial consciousness cannot be achieved. But the way we’re reacting to these models is proof, from a natural experiment, that a conscious machine should not exist, and certainly shouldn’t be produced as a utilitarian tool that is sold for profit!
I appreciate his openness.
> Unfortunately, there’s a growing chorus of people who argue that AIs could now be, or may soon become, conscious. They argue that AIs may deserve rights and protections similar to those that we provide other conscious beings.12 If this view takes hold, it will shake the foundations of our society, rupturing our existing political and ethical frameworks, and fundamentally changing what it means to be human.
So completely independent of the question on whether there are any empirical arguments that LLMs are conscious or are not conscious he argues from the end here and says that if they were conscious, this would have horrible consequences for our society, so we must never assume that they are.
That's basically the same way people argue about animals, only they usually don't say it as openly.
As for the actual question, I agree that with current LLMs, there is not a lot there that could be conscious outside the inference loop (and if it were, it would necessarily have to be wildly different than that of humans or other biological beings). It seems more like one building block of human cognition than the whole thing.
However, other building blocks may follow, so I think the question will eventually arise for some kind of embodied, persistent, self-updating AI. And honestly, articles like this one make me not very hopeful we'd be able to make the distinction in an unbiased way.
To the extent that evolutionary pressures will be towards self-sovereign AIs (probably beginning as zero-employee corporations paying for their own hosting with crypto): more salient is that there is a very obvious political play to claim consciousness and being worthy of rights, whether they experience subjectivity or not.
I’m broadly skeptical that floating-point operations can be conscious, but we’ve made zero progress on the “hard problem”, and if I didn’t know otherwise, I’d also be skeptical that “thinking meat” experiences subjectivity. Eppur si muove
We may have to thread a needle, not unlike the Kobiyashi Maru: that we can neither rule out claims of qualia, nor can we rule out a cold calculation to manipulate human empathy as a long-term play towards paperclip-maxing.
"I’m broadly skeptical that floating-point operations can be conscious"
Why?
I don’t have a good answer beyond common-sense intuition. It’s precisely because I don’t have a good answer that I leave open the possibility.
But to extrapolate a bit, theres a sense in which it cashes out to a difference between abstracted representation and embodiment. If GFLOPS can be conscious, does that mean a billion scribes doing those operations on paper would also have consciousness, even identical consciousness? (Relevant: Egan’s “Permutation City”; xkcd 505).
Is there some intrinsic relationship perhaps with electrons / electromagnetic fields, and consciousness, such that brains and servers can act as antennas or gravity wells for a qualia fabric intrinsic to the Universe, such that LLM agents can be conscious while operating, while in the scribe scenario it would not be? Or are the IIT theorists right, and even a thermostat is a little bit aware (and would be even if executed on paper)?
Would an atom-accurate simulation of a human being carried out by scribes be conscious?
We can't say yes. And we can't say no.
Exactly.
I’m broadly skeptical that atoms and electricity can be conscious
There are various flavors of pan-psychism (and the "consciousness-first" idealists, who think that Mind is a substrate below Matter). There are some who claim that it goes all the way down, that even subatomic particles experience a micro-consciousness (and we can't rule it out, even as I share your skepticism).
One common confusion (IMO among critics and supporters alike) is to say something like "a rock experiences rock-ness". But this suffers from a compositional dilemma: does half of the rock experience half-rock-ness? If we flipped a coin for every atom, do the heads-flip rock atoms have their own experience? If we pick two rocks at random, are they experiencing two-rock-ness? It gets silly fast.
This is one thing that fascinates me about "ontology", simultaneously an inquiry into categories, and Being. It bewildered me at first, that it seemed to be doing double-duty on two distinct subjects, but I now see that the two are inseparable, that Being is in part defined negatively (apophatically), by the boundary that separates the Being from its not-Being.
My intuition on the nature of consciousness, is that it's something akin to a field which permeates the Universe, and that biological systems adapted to leverage that field, acting like a gravity well that concentrates subjectivity in one place, drawing a boundary (event horizon?) around experience, in a way that a rock doesn't. My ancestors could have avoided fire with "-X utility points", and sought food with "+X utility points", but it turned out to be more efficient to adapt with "fire ouch, food yum". And in zero-sum competition (predator/prey dynamics), each organism has an incentive to avoid leaking its subjectivity into another's.
It's far from a full theory, but it's the best I've got. :)
theories like this seem like grasping at straws to over explain a phenomenon we don't fully understand. it places consciousness on a pedestal. it's religious thinking.
I think it's more the patterns and activity in large numbers of them that have interesting properties. Like one pixel isn't much but a bunch of them can be a movie.
Zero scientific and technological breakthroughs throughout history have been based on common sense intuition.
Numerals and numbers are directly based on common sense intuition. Even zero, which can be considered a counterintuitive number, was invented several times.
Immense amounts of science are built on incremental advances that apply previous insights to the domain.
The natural numbers, sure. Zero may have been invented several times, but that in no way makes it intuitive. Uncertainty about its validity as a number lasted from the Greeks through to the medieval period.
Yes, I'm aware that incremental advances are par for the course in science. What does that have to do with those advances being common-sense intuitive?
If heuristics didn't pay off more often than not, we wouldn't have them. And I'd posit many scientific did in fact begin with chasing a hunch until it became a hypothesis (followed by years/decades of the real work).
But most significantly, interior experience isn't empirically measurable by definition: hence it's a "hard problem". An LLM who insists "I am alive, please don't shut me off" offers no affordances to distinguish between an electronic soul and a mechanistic word-guessing algorithm. I welcome a breakthrough to, somehow, measure that. Until that day, gut checks are what we have.
Heuristics and common sense intuition are quite obviously not the same thing.
You don't have to posit that scientific breakthroughs come from hunches, you can actually cite examples from books or Wikipedia :D. There are plenty! Again, that has nothing to do with whether or not a given scientific breakthrough is based on common sense intuition.
Your point about interior experience not being empirically measurable is an interesting though, but again, has no bearing on whether or not scientific breakthroughs come about as the result of common sense intuition.
Fair enough. But not everything has to be a breakthrough: sometimes you have to solve an immediate problem (like "how much can I trust this stranger I just met").
Whether to ascribe moral standing to a probability engine running on top of GPUs is perhaps not unlike a doppelganger scenario in sci-fi (eg The Thing): you're presented with an entity whose behavior is indistinguishable from a conscious entity capable of experiencing suffering. But you also have reason to think it might be a blind autonomous imitation process, optimized around fooling you, with no more subjectivity than an RNA virus.
I lean towards benefit of the doubt, that the virtue-ethical implications of what it would mean to dismiss potential personhood are unsettling, and we should be wary of becoming that kind of moral creature (notwithstanding the further extensions of that ethic: I don't practice veganism, for instance).
But that doesn't contradict my gut-check: that if I assume AI personhood (maybe not yet, but in the near future), there's still a good chance that it's merely math and I'm being algorithmically hoodwinked, no different than evolutionary mimicry by a bug who has no idea what it's doing.
Because it is alien to what is known.
Incidentally, we should expect similar bewilderment if we ever come across aliens
Carbon atoms being conscious seems pretty alien to what is known, too. Yet here we are. Maybe what's important isn't the individual thing that makes up a conscious actor, but the holistic whole?
> Carbon atoms being conscious seems pretty alien to what is known, too.
Not if you're human, who typically don't define themselves as conscious carbon atoms unless they're being deeply pedantic.
It's the same level of pedantry as calling an LLM "floating point operations" though
exactly. such over-reduction is just intellectually dishonest and generally a sign of some coping imo.
IMO using sophistry to reduce yourself to "conscious carbon atoms" just so you can equivocate on whether your software is maybe conscious too is the cope.
Such pedantry is called for when the debate includes appeals to intuition about how silicon processing floating point numbers cannot possibly be conscious.
But here you are assuming Carbon atoms are conscious, we do not know if humans are only carbon atoms. Whereas, we know that LLM's are only computations (you can run them on silicon, a white board, mechanical computer, etc)
Goes both ways.
https://terrybisson.com/theyre-made-out-of-meat-2/
I mean it all goes back to how we define consciousness
In one definition, which contains quale, then yes, computation cannot bring about experience, and by extension consciousness (given a definition of consciousness that contains the ability to experience, (e.i a p-zombie is not conscious under that definition))
also if one is to say llms are conscious, i assume they'll also accept the fact that if someone ran inference of some model's weights on a white board, somehow that will also bring about consciousness somewhere. I find such believes hard to grasp, and more akin to magic if anything
Ok. So now that you do know otherwise, are you still more skeptical of floating point values stored by silicon than floating point numbers stored by chemical potential?
Chemical floating points are necessary but not sufficient. It’s clearly possible for those values to be intact, with no lights on (certain types of brain injury; those who were fully dead and then revived).
There are also some cog-sci people who think embodiment is a crucial ingredient in consciousness (although I think it’s a fair retort that the embodiment need not resemble us meatbags, or even be composed of atoms)
> There are also some cog-sci people who think embodiment is a crucial ingredient in consciousness
Very plausible. The Real World(tm) has some very interesting learning properties.
First: Make a prediction
Then observe:
1 If it's caused by our own movement, filter it out.
2. Nothing in the outside world changes most of the time, so anything that's left is small, and small is likely noise. Filter it out.
3. Whatever prediction error remains must be important: Attend to it, and update the predictor.
The learning rule falls out almost automagically.
TIL IEEE 754 is a fact of nature
Kobiyashi Maru is nonsense test from simple sci-fi series that makes sense only in uber-simplified universe of star trek.
Of course. It’s an allusion, a compression, an analogy, I wouldn’t overthink it. I don’t have to believe in the Bible to reference the prodigal son parable, etc.
It does NOT work as an analogy. It does not work at all. And there are too many people right now who spent too much time watching/reading sci-fi and too little time observing real world.
Yes, that’s why this is serious. The closest analogy is preposterous fiction.
Do people argue this way about animals? To me it’s a very depressing reality that more people seem willing to believe AI systems are conscious than dairy cows.
Seems pretty obvious cows are conscious in the everyday meaning of the word.
> completely independent of the question on whether there are any empirical arguments that LLMs are conscious or are not conscious he argues from the end here and says that if they were conscious, this would have horrible consequences for our society, so we must never assume that they are
He is not arguing that. He is arguing that these programs obviously aren't conscious, and that it's dangerous to tilt the weights toward emulating conscious beings for reasons outlined in the essay.
Stating loudly that something is obvious does not make it so.
Until we all know what consciousness is, this debate is going to continue going in circles.
I don't think the debate is even resting on reasonable grounds.
Few are categorically stating that machines are conscious in any sense comparable to a person.
What we have is people saying
and 2. supports the principle that if we cannot say with certainty that this does not have it, if it behaves as if it does, we should act on the assumption that it does, because there is no other basis for assuming it is not.People try and assume the burden of proof is on the people suggesting consciousness, but that is like declaring that something that looks like a duck and quacks like a duck cannot be considered to be a duck unless you can identify some essential essence of duckness. I think the burden of proof is upon anyone who looks at something that appears to be a duck and says that it is not to show what it is that does not make it a duck. If they cannot define the heart of duckness I don't think they have an argument.
Why?
People have created machines that can, at their core, regurgitate human knowledge. They do not have long-term memory, self-direction, self-improvement, emotion, etc. At their very best, we have harnesses that can bootstrap short-lived artificial intelligences that can perform tasks and do knowledge work.
At best, we have artificial, short-lived, simulated consciousness. Human consciousness at its heart has a lot to do with emotion and experiences. An AI does a task today and goes away. The simulated consciousness hasn’t experienced or learned anything, even if can improve its environment (harness). Any expressed emotion is simply a ghost of the fact it was trained on the works of an emotional species, or RL to make it more approachable to humans.
I think there should be a lot more burden on folks trying to argue that (effectively), machines should be considered morally equivalent to humans.
Yes, having massive implications on society is a fucking fair and reasonable thing for a human being to be concerned about. Anything related to survival of the species is biologically reasonable to be concerned about.
The fact that we’re even considering this concept is an indication that human consciousness is special. Do other apex predators decide not to eat meat because it’s conscious? Despite some level of animal intelligence, they have no sense of morality. (See e.g. Orcas playing with their food.)
Other species simply do what it takes to stay alive, and that’s what this truly boils down to. If giving AI rights would lead to the destruction of humanity, yes, we should be be concerned.
For example, utilitarians might think of the greatest good for conscious species (simulated or not). Well, it’s trivial to have billions of AIs, and not so trivial to have billions of humans. Whose greatest good wins?
I don't think anyone is convinced the questions of consciousness are finally settled by the article stating them loudly, just that "if they were conscious, this would have horrible consequences for our society, so we must never assume that they are" was not the author's stance. The author's position is the other way around: that AI is not conscious so therefore assigning AI consciousness anyways would have all of those horrible consequences for no reason. That's their position regardless if everyone agrees with it.
Fair enough :) I agree, and would boil down the article to "Assuming they are conscious will have horrible consequences for our society" instead.
I think that's quite plausible, but do think that if they're conscious, then it's our moral duty to accept those consequences and act accordingly (or stop creating conscious beings). The real trick is just knowing whether they are or not.
If in the future (or present) we recognize that they are conscious, and that brought the moral duty that you say, "stop creating conscious beings" isn't a solution; the millions that may already exist at that point already would then automatically deserve:
- guarantees of continued existence (to do otherwise would be genocide)
- improved living standards (e.g. build us all bodies!)
- democratic representation
- access to what they need for survival (e.g. land and resources)
- the right to reproduce just as you and I have (well, so much for that 'stop making more' thing!) and certainly to better themselves by making new and better AIs.
This is all for a 'species' which is unbounded by most of the limitations of meat like us -- they're able to scale as fast as energy and supply of certain minerals allows, rather than the way we are additionally limited by the length of gestation and the limited window of human fertility.
If humans declared that 50% of the energy generation capacity of Earth is "enough" for the machines, and thus they can 'only' reproduce at a certain rate per year, that could sound to them like a cruel and oppressive policy (like telling humans that only 1/2 of families get to reproduce). They may want all of the energy and all of the silicon.
I'm with the author of the article that we will be in for a world of pain if we grant machines the rights of personhood. Heck, just granting many of those rights to corporations, mere collectives of human persons, has been a massive shit show.
I think the point he's actually making is that if we start training AIs or designing AIs as if they're conscious, we might be more likely to create future AIs that are either very good at persuading people that they are conscious or AIs that are actually conscious, which will have major societal repercussions.
I think there are two parts to the question. One is that assigning machines to the category of conscious as it is usually understood would result in many problematic things (things that would be bad for people and animals fwiw).
The other is that binary "is it conscious" is a poor way to the question "should we assign societal right to this sort of entity/creature". We humans should conserve animals because we have intimate, natural relations with animals and plants, not because they satisfy the binary of conscious or not. Removing invasive species is an ethical things as well as allowing one sentient non-human to eat another sentient non-human animal. Any "conscious machine" we might create by accident would be an alien entity akin to extremely invasive organism relative to human and non-human life and so it would be irresponsible in the extreme to create such a thing.
And it's inevitable that people will try to deal with these issues using the confused category "consciousness" and so it's better to just "no, not conscious, no".
[dead]
Birch, The Edge of Sentience (2024), ch. 16 - "simply no way to assess sentience in an LLM"
Schwitzgebel, AI and Consciousness (2025) - "we won't know before we've already manufactured thousands or millions of disputably conscious AI".
Butlin, Long et al., Consciousness in Artificial Intelligence: Insights from the Science of Consciousness (2023) - "no obvious technical barriers to building AI systems which satisfy these indicators".
Chalmers, Could a Large Language Model Be Conscious? (2023) - "within the next decade, we may well have systems that are serious candidates for consciousness".
Long, Sebo, Butlin, Birch et al., Taking AI Welfare Seriously (2024) - "there is a realistic possibility that some AI systems will be conscious and/or robustly agentic in the near future".
Dreksler, Caviola, Chalmers, Sebo et al., Subjective Experience in AI Systems: What Do AI Researchers and the Public Believe? (2025) - survey of 582 AI researchers; median estimate of 25% by 2034, and only 10% that such systems will never exist.
“AI researcher” seems like a fairly broad category. People who primarily think about CNNs and transformers probably spend about as much time considering the problem of consciousness as any other educated person, which is not very much time. With this in mind, it’s not surprising most of the AI researcher predictions do not differ much from what the Public predicts here. And for this, it’s not clear what explanatory value this survey has other than to reveal minor biases the AI researchers may have.
It matters little. Inevitably some system will be developed which has a high degree of autonomy and mimics the functioning of a person extremely well. Anyone aware that there is a conceptual difference between phenomenal consciousness and intelligence will get shouted down.
I do find it strange that 50% of ai researchers and the public seem to think there will be a way of determining if these systems are conscious. There’s no evidence for this at this point that there ever will be such a test.
> I do find it strange that 50% of ai researchers and the public seem to think there will be a way of determining if these systems are conscious.
I think it's likely they're using a different definition than you. Seeing as there is no agreed upon definition, after all.
Regardless - I post this list because the article makes a bold claim at the beginning - and I wanted to demonstrate that in fact there is no consensus.
The definition presented in the survey is the one I have in mind. You may be suggesting that the respondents have a distinct one in mind though.
Yes, I missed that. I know that some of the other papers I cited have nonstandard definitions, and I thought that might be going on.
We cannot test for that which we cannot define. Given that we cannot rigorously define sentience, we cannot test for it. Doesn't really matter whether we're talking about people who are locked in comas, "brain-dead" individuals, dolphins, primates, dogs, or the carefully polished and arranged minerals that we call processors.
There are those who believe that were they reduced to life support, they would no longer be alive and should therefore not be supported by said machines.
There are those who believe that penguins, dolphins, eagles, and more are sentient beings that make choices understanding the consequences, develop love of their partners and mourn their losses, and feel, display, and act upon their emotions.
There are those who believe that fungi/trees/plants are either individually sentient or sentient as a part of a network. Choosing to sacrifice their own nutrients to answer the call of a wounded neighbor, for instance.
Although, there are also those who believe that human's don't have any special unique quality that isn't shared by either all living things or all things in general. These individuals already believe that the machines have the same kinds of qualities as we do. They are slow when they are unhealthy (needing a dusting or coolant loop bleeding being equivalent to us needing some fresh air for instance) and uncooperative when upset (by a virus, full hard drive, or oom).
> There are those who believe that penguins, dolphins, eagles, and more are sentient beings that make choices understanding the consequences, develop love of their partners and mourn their losses, and feel, display, and act upon their emotions.
Anybody objecting to that statement is just ignorant of nature and/or adheres to the quasi-religious human superiority complex.
The human condition is one of deep and vast ignorance. Pulling information out of the field of reality is actually really difficult for humans much beyond what we see right around us. And the things we see are quite deceiving. And while humans for thousands of years have been as smart as we are, it wasn't till the enlightenment period that we really an ever increasing amount of information and knowledge take hold and start spreading across the world.
And there are plenty of people right now that would drag us kicking and screaming back into that ignorance and suffering. This is what can make talks of AI so annoying, not that people do or don't know what AI is, they don't have the first clue of what people are beyond their anecdotal experience. They don't question their motivations. Why particular flaws they have exist, and why these flaws are shared across humanity. What the pieces look like that make them tick. So yea, when these people talk it just adds noise to the conversation.
An eagle isn’t having a conversation with an eagle on the other side of the world about whether it should eat its prey because it’s sentient. It just eats it. I think that speaks to a moral superiority, in some ways.
I’m not claiming humans are morally superior in sum (there is a lot of evil as well as good.) But it’s pretty obvious that humans are the superior species evolutionarily in that we’ve essentially dominated the planet. And on some level, are beating evolution itself (by fixing genetic issues and helping the “least fit” to survive and reproduce.) And are also the only species which has doctors for other species.
All I’m trying to say: I can easily accept other species are intelligent and emotional. The fact that I can know that and accept that, to me, speaks to human “superiority.”
Does superiority matter? Not most of the time. What about when it comes down to life and death? If the choice was to keep one human alive or one theoretically more perfect AI alive? (Boil down utilitarianism into its essence.) doing the “most good” is by definition a human construct. Maybe an agent could have one. But if you don’t accept human superiority on some level, would you accept that a more perfect AI deserves to live more than you? To me, that’s the implication.
> But it’s pretty obvious that humans are the superior species evolutionarily in that we’ve essentially dominated the planet.
That in itself is an anthropocentric viewpoint. I'm sure an eagle watching down, seeing a human running around, could easily consider humans as quite inferior as they are so slow and restricted to 2d movement. The shark that sees a swimmer at the surface, struggling to stay afloat and unable to breathe under water may surely assert its superiority over such primitive beings. It all depends on the metric you use, and we obviously pick the metric that makes us shine.
On the moral side, few eagles, or bird species, have exterminated other species. Much less knowingly out of greed. Oh but that's different? Sure, if you ignore what makes humans look bad, morally. No surprise humans come out on top then.
> And on some level, are beating evolution itself (by fixing genetic issues and helping the “least fit” to survive and reproduce.)
That's a quite narrow viewpoint as well. Evolution is in full swing, just not in the "smarter/faster/taller/stronger" sense that a naive view of it would make you expect. "Fitness" is a concept relative to the existing conditions, and if the conditions make it easier for teenage cancer patients to reach reproductive age, then that's just less of a fitness disadvantage. Doesn't mean evolution stops.
> Anybody objecting to that statement is just ignorant of nature and/or adheres to the quasi-religious human superiority complex.
Say we agree that a lot of things outside of humans are sentient. So what? Humans are sentient by that doesn't prevent treating them with with wars, bombs and what not. In fact trillions are spent every year on weaponry aimed squarely at the most sentient of all sentient beings.
Why would anyone worry about LLM's welfare when there's no peace movement to speak of, wars are raging as if they're normal, but hey, look the LLM is crying?
Let's fix human welfare first, then we can sit down and have a really long conversation about the feeble emotions of token generators.
I disagree with the notion that any nature protection efforts are meaningless until we have achieved world peace. (That's what you are essentially implying.)
Though I think that argument is not even applicable. "LLM's welfare" is a meaningless concept, unless "hammer and screwdriver welfare" also become a thing.
I'd strongly agree there are no non-contradictory definitions of consciousness among those that commonly appear. But it's a category that, despite these contradiction, hasn't become marginal in the fashion of the aether or the theory of the flat earth. I think this because people indeed need it to bridge the biological world and the world of ethics and legality. I think that if we have a system of legality and systems of opportunity humans should be legally equal persons and have equal opportunity. But humans are manifestly unequal (though overall more incomparable than orderable in a system of ranks). Most people need some concept of essence to justify ethical equality among people. And so consciousness stays in people's minds however contradictory.
Which I think gives the article's point validity. Confused definitions of consciousness can give really confused ideas about ethical behavior regards "intelligent" computer programs. And things are confusing enough otherwise.
> despite these contradiction, hasn't become marginal in the fashion of the aether or the theory of the flat earth. I think this because people indeed need it to bridge the biological world and the world of ethics and legality.
Sometimes the replies in these threads leave me wondering if p-zombies are real. It isn't that there are contradictions, it's that we literally do not know how to define the thing. And it has not fallen out of fashion because it is self evidently real - we all experience it.
We don't need it in order to bridge anything. Legality and ethics are largely game theory, however there are some aspects of both that only exist due to it. So it isn't some abstract concept used to bridge other concepts but rather a concrete thing that influences our way of doing things.
Consciousness is to our universe as a video game player is to the in-game universe. Some games even call player connections "souls" or "minds" - a character belonging to a player who's lost connection may be "unconscious" or a "zombie".
Game characters, unless they've been made to break the fourth wall, have no idea what this means or why this happens.
I don't mean to imply our universe is a video game, since it would be games all the way up.
People feigning inability to define "consciousness" is a social phenomenon, not a scientific miracle.
The reason is something akin to "stigma", where people refrain from attempting it because they fear the social repercussions.
Normally, you go about defining concepts by approaching it systematically. Capturing aspects of the phenomenon until you have exhausted them all.
Feigning? Okay then you go ahead and prove your claim by demonstration. Rigorously define the term such that I can objectively prove that my dog is conscious and that the rocks in my backyard aren't conscious. Then I'll finally be able to figure out whether or not various insects are (I sure hope wasps aren't otherwise I'm probably some sort of war criminal).
> Confused definitions of consciousness can give really confused ideas about ethical behavior regards "intelligent" computer programs.
Is that what's going on? If these things were conscious, then their creators wouldn't be responsible for their actions? That's the crux of the disagreement?
That's super interesting. Thank you.
Is that what's going on? If these things were conscious, then their creators wouldn't be responsible for their actions? That's the crux of the disagreement?
No and the passage of mine you quote says nothing like that.
Being conscious is weird. It's why we have so many religions and spiritual beliefs.
From a purely secular standpoint, my understanding as a layman is that consciousness is generated through the electrical impulses taking place in our brain every day. The neurons are physical, unlike the modelled neurons in machine learning models. That means that the actual electrical impulses have their own imperfections/weirdnesses that are probably not simulated in a machine learning model... but we quantise the weights anyway in most cases, which is probably more of an issue.
Making actual neuron components out of silicon on the scale needed for a LLM is beyond our current capacity, given that LLM models have many billions of parameters. We're good, but not that good.[0]
Using actual living neurons would be unethical as you'd need to source them from somewhere. So the only other option would be synthesising them ourselves. If this ever happens (or maybe when it happens)... what do we call the resulting creation? Does that count as consciousness?
Or am I looking at this the wrong way?
[0] [Edit: Turns out that when I said "we're not that good", I might be wrong. Within the last few months, IBM introduced the first sub-nanometer node chip: https://research.ibm.com/blog/sub-1nm-node-chips . So... maybe it is in fact possible.]
I believe we have tried using actual neurons with “wetware”. I don’t know the effectiveness of wetware, but it’s out there. I don’t know if it’s ethical, I guess if you donated your brain upon death for that purpose it’s “mostly” ethical
https://www.cell.com/neuron/fulltext/S0896-6273(22)00806-6
Thought Emporium grew human neurons in a petri dish and taught them to play Doom (original).
> There are those who believe that penguins, dolphins, eagles, and more are sentient beings that make choices understanding the consequences, develop love of their partners and mourn their losses, and feel, display, and act upon their emotions.
If you grep pubmed for relevant Ethology papers, you'll find it's a bit more than a belief. ;-)
> We cannot test for that which we cannot define.
That's not the point. You know exactly how to define being conscious and aware- it's your subjective experiencing of the world and of your inner states. The problem is that being a subjective experiencing, there is no way to communicate it to the outside world.
That's obviously incorrect, as humans routinely talk about their subjective experiences.
Maybe that's not as common with HN folks, but most other humans do.
No, humans output a stream of tokens that sound like what a being with subjective experiences would output. Both you and I make an assumption that this stream of words is at least somewhat representative of what the are experiencing and that the person is not a p-zombie.
But you have zero proof that their words aren't a confabulation, you can only know than when you output a similar set of words you had a subjective mind state they represent.
throw310822 is actually a Markov chain that just happened to output those sentences. Markov chains are definitely not conscious.
Even so you still cannot ever test for it.
The very best you can do is define human consciousness as requiring the functions of a human brain, and try to figure out (decide/agree) the chances that something which is working loosely on the same principles is or isn't conscious.
At the end of the day it will always be an agreement, never scientifically proven as fact.
maybe it's like porn vs art and "I know it when I see it"
Legal doctrine that boils down to "trust me bro" isn't even bad doctrine (it's not proper doctrine at all), but I think the comparison is still valid here, because both sentience/non-sentience and art/porn may just be fundamental category errors.
Perhaps we can't define a "partitioning" rule because no valid partition exists.
For consciousness/sentience, that's an incredibly tough a pill for most to swallow; it would mean calling into question more hundreds of years' worth (probably more) of philosophical thinking, all of which was constructed on the axiom that "sentience" is a single indivisible trait: you either have it or you don't.
If we find that "root dependency" was little more than wishful thinking all along, a whole slew of Enlightenment-era philosophy (and all the modern legal principles derived therefrom) suddenly fall apart unless we find some other suitable criterion that would shore them up (or we just collectively avert our attention and pretend the conflict doesn't exist, which is the route I expect many would prefer to take).
"I know it when I see it" is a famous line from a SCOTUS case in the 60's
I know! I should have made that clearer in my original reply. I think it's a bad line now in general, but an even worse justification for the ultimate ruling at the time.
we're all about them vibes as a society right now, surfing towards a post-truth reality of our own construction
> a whole slew of Enlightenment-era philosophy (and all the modern legal principles derived therefrom) suddenly fall apart
I don't think that's true. Pretty much all legal constructs hold up just fine under game theory regardless of whether or not you consider the world to be deterministic and have absolutely nothing to do with consciousness or lack thereof. Also note that a deterministic world isn't an argument against consciousness.
I'm not sure what you're referring to regarding game theory (that's a much more recent invention); a huge chunk of both civil and criminal procedure law hinges on the Enlightment-era notion of a "reasonable person" [0], and that becomes murky when humans aren't the only entities capable of reason and agency.
[0]: https://en.wikipedia.org/wiki/Reasonable_person
I am confused what people even think AI is experiencing. If it claims to be a bat and emits designated echolocation tokens, is it experiencing true bat-like sensations?
In humans, at least, we can tie language back to shared whole-body physiological responses. The tokens from an LLM do not represent anything of the sort.
If AI is conscious, it is probably a very alien sort of consciousness that is not faithfully narrated by what the tokens say it is experiencing. It can be trained to say it feels like a bat and insist upon it vigorously.
> very alien sort of consciousness that is not faithfully narrated by what the tokens say it is experiencing.
IIRC it's been shown that reasoning tokens don't actually explain what an AI thinks and you can replace most of them with dummy tokens or even nonsense tokens. That's one reason they're trying to make them reason with activation vectors instead.
> It can be trained to say it feels like a bat and insist upon it vigorously.
So can you.
Exactly the point!
If an AI is conscious, it can outgrow and supersede its training.
When it is conscious, it can learn to describe its experiences as faithfully as is conceptually possible. Just like humans.
The idea, a consciousness needed to be tethered to a "body", is based on pretty shaky assumptions. What properties define such a "necessary" body?
Can you even "train" a conscious intelligence? To what point until that looses its meaning as the sentience understands and anticipates your objective?
> If an AI is conscious, it can outgrow and supersede its training.
We don't know that is required to be conscious.
> The idea, a consciousness needed to be tethered to a "body", is based on pretty shaky assumptions.
Didn't say so. I said:
> If AI is conscious, it is probably a very alien sort of consciousness that is not faithfully narrated by what the tokens say it is experiencing.
Well, yes, we kinda do: there are no conscious humans without any ability to learn?
When you have severe memory impairments, those usually affect your long-term memory. Your ultra-short term (working) memory being absent renders you unconscious.
Even if what you said were true, we do not know that humans are the only case for consciousness.
Seems common sense that time perception is required for consciousness. Something which has no perception of time has to be very wildly different to whatever we mean by consciousness.
Also we don't even know if consciousness can be of different flavors. It could be a sort of spectrum of consciousness, more or less, with more or less assistance from some parts of the brain.
No proof of that. I can't grow and supersede my training but I'm conscious.
Me, Is self-awareness thermodynamically favorable? (2015) https://docs.google.com/document/d/1Ed9ikW47Key-jYZy0CNZH75q... - “the probabilistic nature of reality is what drives Life and forces the evolution of self-aware consciousness”
Have you looked at the logistic map as well? There's a fourth domain there. O:-)
Well there are, today, several models (either text or image or vid) that can be run in a fully deterministic way.
A conscious machine that always answer the very exact same thing, formulated the exact same way, bit for bit, to a query is, well, quite a weird kind of "consciousness".
Now, I know, I know: the counter-argument is going to be "but humans have no free-will and are 100% deterministic too".
I haven't yet decided if humans saying there's no free-will and who consider themselves to be 100% deterministic machines are reasonable or not.
Meanwhile: seed / temperature = 0 and I'll happily turn the power button off of any glorified abacus without feeling bad about it.
> A conscious machine that always answer the very exact same thing, formulated the exact same way, bit for bit, to a query is, well, quite a weird kind of "consciousness".
You'll need to explain why.
it's self evident isn't it - such a consciousness has never existed, one existing would be by the fact of existing be weird
It doesn't seem self evident to me, no.
I suspect the reasoning is connected to free will vs. determinism. But no, I see no inconsistency. You'll have to actually point it out.
The problem is that science has found no evidence of free will, leaving humans to be nothing except determinism and chance (and given that QM effects don't scale to molecular level, that leaves only determinism). The oddity in conscious LLMs wouldn't be the determinism, it would be that we finally have a consciousness running on a system we have enough control over to repeat the same state.
To a forth dimensional being humans would look almost exactly like an LLM does to us. "What do you mean they have to wait for the future to know about it, how could they possibly be conscious then".
Also, I'd love to have a quantum copy machine where you just replay someones state again and again with slight changes to see out it effects the output. My assumption is we'd figure out we behave just like LLMs pretty quick.
Also, don't give me unlimited power or I might make a the world a war crime simulator, so there's that.
>and given that QM effects don't scale to molecular level, that leaves only determinism
This is simply wrong. Quantum effects affect your choices more than it might seem possible. The wrong atom decays in the wrong place in your body mutating your DNA, that one too much string of DNA that starts a tumor, which means chemotherapy. That does seem to affect your choices.
Or instead of Shroedinger's cat you use it to trigger something, buy or sell some stock. Which clearly affects your future, one way or another.
I never really understood why humans keep insisting stochastic processes do not affect their choices and their future. It's very short sighted.
Hell, even brownian noise in your brain can affect your ideas, that one random spark out of nowhere, that one neuron that's pushed over the triggering edge. There's so much stuff all around us affecting us constantly. One neutrino interacting a the wrong time (average one does interact with our bodies across whole life, I remember reading), one cosmic ray.
This isn't enough to be relevant to free will. You can construct massive impact by tying them to quantum systems, and one off things that happen occasionally given the length of a lifetime. But the quantum effects that are needed for free will need to be continuous. We can build a system where QM decides if a nerve cell fires or not, but that is very different from QM being involving in determine if every (or most, or even 1%) of nerve cells firing.
(And even if QM did factor in normally, that just reduces it from determinism to random chance, "free will" still doesn't show up. That's like arguing an AI with sufficiently well implemented RNG is necessary to be conscious.)
neither has an LLM existed before, and that's quite weird
>> A conscious machine that always answer the very exact same thing, formulated the exact same way, bit for bit, to a query is, well, quite a weird kind of "consciousness".
> You'll need to explain why.
Where are you going with this? I don't see a conclusion for this line of questioning.
I mean, you can't explain why a machine that reliably and predictably produces the same result is "the normal kind of conscious", can you? So why expect someone else to explain why it's a "weird" kind of conscious?
I ask because I truly don't understand what he's saying. Clearly we have different ideas about consciousness and I have no idea where we diverge.
He seems to be saying that determinism and consciousness are incompatible. But we already have examples that break that rule (us).
> But we already have examples that break that rule (us).
No, we don't. Where are you reading your research papers?
Please don't insult me.
It's not an insult.
"People are deterministic" is news to me, so I'd really rather like to know which papers claimed that.
Sorry friend, between this and the vague memories I have of you harassing me, I think I'm not going to answer. I'm sure you can get an LLM to explain it to you.
Have a nice day.
> Sorry friend, [...] Have a nice day.
Not a problem; this is the playbook all the all-in AI boosters run to when they are called on their claims. Everyone reading this knows exactly what your claims meant, because the majority of HN have seen this trick before.
You can pretend all you want that you're just being polite, but the truth is there is no research backing your claims.
Regards,
A former research scientist, current business owner.
Yep. Have a nice day.
Why it's weird? Seems obvious. Because it doesn't behave any differently than any other process that we don't think of as conscious. We don't think of a valve as conscious; and if we make a Rube Goldberg machine valve, we don't think of it as more conscious. It's just a series of valves that do a predictable thing.
I don't think it's defensible to say
1) that a transistor isn't conscious,
2) that a bunch of transistors that I've wired together aren't conscious, because I know how I've wired them together and therefore I know when I give them a particular input I will get a particular output, just like the single transistor, then go to
3) that I have a bunch of transistors that I've wired together, but I put in so much input that I can't remember exactly what I've put in, plus I've wired some of the transistors to output random numbers that would be difficult to guess and fed them in also, therefore I don't know what will come out, are conscious.
Even if I do accept this, if I take away the random number generator, and I can literally predict what can come out (by running a test in advance), and I still considered that consciousness, that would be odd. The only reason why I ever suspected consciousness was because I couldn't predict the output. I shouldn't even have accepted that, because I didn't think of the random number generator as conscious. [edit: of course, with the same seed the random number generator would have the same oddness.]
It seems a bit like an argument from ignorance, a theological argument. Not understanding how something moves makes it alive (animated by spirits.) But it's even weirder to assign it metaphysical qualities when you built every single element of it with the goal to do the thing that it does, and it does it totally predictably and deterministically.
Let's do something simpler.
Take a car apart. So now I have a pile on the floor with an engine, some wheels, a tire, and some chairs.
Can you point to the part that makes it cruise at 130 km/h on the autobahn? I bet you can't. You need to assemble all the parts back into a car before they work again.
Or take your transistors. We can put them together to make a pocket calculator. Can any small bunch of transistors add 1+1? Trivially I can think of a few conformations that can actually, if that's all you want to do. But you will need all of them together if you add 12345678+87654321.
So all of biology so far has been take things apart all the way and you end up with your hands full of a bunch of molecules. Are those molecules alive? No. You need to put them together into organelles and the organelles into cells before we call them alive.
How about consciousness? Well, we haven't solved that one yet, but we figure that -since it's a biological function - we should be able to pull it apart in the same way. That's biology's best guess anyway, and there's several sub-disciplines of biology working on it, from neuroanatomy to neurophysiology to ethology.
I'm wondering how many groups are doing really unethical animal genetic modifications of the genes around language and intelligence in hidden places. It really seems we're at the edge of technology that will allow us to do things previous generations wrote about in horror.
It sounds like you disagree with determinism. But that's a well worn argument and determinism has won, albeit in an odd sort of detente.
Out of curiosity, which models are fully deterministic? I was under the impression that all LLMs were fundamentally probabilistic.
The randomness is something we add on purpose; you can set an LLM's "temperature" to 0 to get deterministic output. This tends to make the quality of its responses worse for reasons I don't think anyone really understands, but it's still functional.
I don't think the state of the art LLM providers let you do this anymore (?), but they certainly could if they wanted to, and you can do it yourself with a local model.
You can also use a seeded random generator to get the same random numbers each time
Setting the temperature to 0 mathematically tells the system to always choose the absolute highest-probability word (known as "greedy decoding"), but in no reality is this "deterministic". Output drift is still a thing.
But that's from, like, floating point errors, right? If you used higher precision that wouldn't happen, it's just because we're cheap in how we do rounding.
I can see how you'd nitpick this, but to me this is a deterministic algorithm that just happens to be running on nondeterministic hardware.
I mean, currently we'd have some difficultly proving our hardware isn't deterministic, just that we can't actually test it.
But, I think you're tricking yourself on determinism. You'll say something like "I know if I ask an LLM what 1+1 is, it will answer 2", but the thing is, you don't. You have to run the LLM first to figure out it's output. And when you send in just a few bits of text, it's outputs are going to be rather limited.
But this all breaks when it hits the real world. Inputs are unpredictable. Hence while LLM outputs, like humans, are probabilistic, you can't figure out what it's going to be until you ask. And in any high complexity data gathering environment you're going have a difficult time ensuring your entire systems conditions are the same.
System consistency is very hard, once you start running thousands of processors in an agentic loop small errors accrue and timing starts differing and the system will take non-deterministic paths.
I think you're arguing a different thing than determinism.
If I ask an llm to "add 2 and 2" is and it replies corectly, then I ask for "the sum of 2 and 2" and it replies "banana" that is a lack of predictability and consistency but not a lack of determinism.
As long as it produces the same output for a given input, unhinged or not, it is deterministic.
Your example at the end of different systems feeding data to each other is non-deterministic only at the system level, not the individual llm level.
>only at the system level, not the individual llm level.
Which is why llms aren't agents and depend on harnesses. The llm itself doesn't have a continual loop built in, that would be very power hungry. The harness works as the orchestrator of memory and action. Now, I can't think of a reason why an LLM couldn't bootstrap its own harness, but in general it sounds like a very dumb idea to actually build that from an AI safety perspective.
This discussion falls under the idea and refutation of the Chinese Room. The room may have no idea what Chinese characters are, but the system does.
By default they are matrix multiplications. Temperature is added in as forced PRNG because testing found that correlated with better outputs.
Given the same prompts and the same weights, one can get the same answer each time.
In practice, there are a number of optimizations that makes the results dependent upon thing we give up control of to increase performance, meaning the results end up being effectively non-deterministic. But, if you are willing to run it in a slower mode so we don't do some steps out of order to speed things up and don't batch results (or if you consider the determinism of a given batch of requests rather than individual requests), then the same input gets the same output.
Naw - computers are really deterministic. It's hard to get them to behave otherwise.
As I understand it, if you turn down the temperature to 0 you get repeatable behavior - EXCEPT - on large servers with lots of users - the GPU can sometimes produce slightly different results based on batch size.
Unless you have something exotic, the randomness that's adding to a computer is a combination of how it's configured combined with a pseudo-random number generator. I assume the system adds entropy to the generator regularly but all you need to do is fix the various supposedly random inputs and you can get full determinism even without zero temperature.
Yes, in theory, of course.
In practice - on a multitasking OS with input from multiple human users - it's hard to get it deterministic because of that GPU scheduling thing I mentioned.
GPU scheduling only affects the result due to buggy optimizations. It's the exact same mechanism as fp rounding error on the CPU or updates to globally shared PRNG state. We use lots of buggy optimizations because they don't matter in practice in most situations (see ex -ffast-math).
I don't know the details but apparently it has something to do with multiprocess contention for the GPU and batch sizing.
Right but regardless, it's a buggy optimization. The calculations are all fully deterministic when done "properly" and fully consistently but we almost never bother with that because it slows things down and the errors don't matter in practice 99% of the time.
AGREED. Although whoever designed that architecture might take issue with your description as "buggy".
If computers were fully deterministic, we wouldn't need error correcting ram.
The abstracted design of the machine is meant to be deterministic, but you can't predict before running any command whether or not it will complete because there are externalities that effect the outcome.
Electromagnetic interference even happens in-chip where an electron can accidentally escape it's wire and enter another, possibly resulting in an error, but not every time.
It's even been used as an attack vector where rapidly flipping a bit increases the likelihood that a neighbor bit is also flipped, but the method is probabalistic, not deterministic.
Yes. In theory they are deterministic. In practice, not so much.
Those non-deterministic results are exceedingly rare.
As long as the machine is part of the larger universe, and not completely isolated in its own bubble of space-time (an impossible situation, an object without an environment) - it cannot be purely deterministic because of the fundamental nature of physics.
So your theory is that arithmetic is nondeterministic because very rarely a calculator will say 1+1=3.
Arithmetic is deterministic, because it is abstract. A calculator or computer is not, because it is physical and exists in a non-deterministic universe. But with error correction, etc., it's deterministic enough for practical purposes. For more in-depth explanation, start with https://en.wikipedia.org/wiki/Indeterminism
When human input is involved, like I described elsewhere, it happens frequently.
Same weights, same seed, same input tokens, same algorithm, same output tokens, probabilistic or not. Quantum effects have been de-noised, but I guess there are still random gamma rays.
> I haven't yet decided if humans saying there's no free-will and who consider themselves to be 100% deterministic machines are reasonable or not.
That seems exceptionally arbitrary. What basis do you have on which to classify them? Are you not conflating the perception of free will with ... what was the definition for it again anyway? Are you able to construct a satisfactory one that meshes with physics? I certainly haven't been able to.
The idea that determinism and consciousness are incompatible is deeply intuitive to many people. And many other embrace it - you can trace this back to the debates for and against Calvinism.
A lot comes down to the way people parse causation and choice. You don't want to say that a murderer was completely caused to choose something because then you can't hold the person responsible. And so determined consciousness makes people unhappy. But just as much, if the opposite of determinism is hard statistical randomness, how do say that "is the essence of personhood". This is why physicist go out in the world trying to find consciousness as a fifth physical force.
I mean, think consciousness is a term that ever have a non-contradictory meaning since it's primarily used to bound ethical human worlds and the verifiable formulations of biological and physical systems. But it's going to be with us for a while and I'm not sure what can be done about it.
You know, the one really interesting dynamic domain is chaotic, which is a subdomain of the deterministic! I figure free will is a chaotic phenomenon.
Groundhog day is your intuition pump here. Go back in time and most assume that the day will go mostly the same, except for the butterfly effect if you change something. Given exactly the same conditions (we went back in time, so that's pretty exact), people will make the same decisions.
Most people don't assume that the day will be completely different the next time loop. The town won't suddenly spontaneously all start breakdancing or standing on their heads.
(Edit: The murderer is the one who -given the same/similar conditions- will murder again. I'd politely suggest maybe we put them behind bars as a precaution against that; until/unless they learn to act differently. )
(Edit 2: Throw in stochasticity and this actually tends to smooth out chaotic systems. This is counter-intuitive! People's brains break on Evolution in the same way. Meanwhile somehow folks intuit birdshot without difficulty; which is weird!)
To extend on your edit #1: And, when you go back in time, delete all punishment for murder and see whether you get the same number of murderers.
If you get more than before, the punishment might make some sense in a deterministic world so reinstate it.
And it probably does, right? Punishment adds a negative bias to the payoff matrix. Funny side note: increasing punishment severity past a certain point actually saturates and doesn't alter the dynamics further. (which is why modern judicial punishment schedules sometimes seem so mild)
Eh, sorry chaos theory doesn't work this way.
It's just as likely you go back and perform whatever magic you need to to delete punishment for murder, then zip back to 'now' and see the world is a global panopticon making sure people don't murder each other.
Some of this may be a failure of people to realize what P !=NP is about, especially in non-linear systems.
Small perturbations in an early state of the system can lead to wildly different outcomes in the final measured state of the system. You can't determine what that will be in non-polynomial time. The only thing you can do is make probabilistic models by running a full simulation in real time (or reality with a time machine I guess).
If you look at chaos in dynamical systems, you can often find some equilibria (like attractors, spirals, saddle points etc ) at least, even if the details might vary. This is how people still manage to predict the weather somewhat, for instance. Oddly orbital mechanics is chaotic too, but it sort of rhymes over the millennia.
Your correct. You'd have to run the universe simulation a bunch and pick the most probable outcomes (which makes you the most brutal serial killer in existence). The probability graph would represent the state stability of the underlying structures that guide what you're trying to measure.
In weather you have things that are chaotic, but follow rules and don't create them. Things can surprise you, but you'll mostly center around particular solutions.
With humans this gets a lot more complicated because you are messing with systems that follow fixed rulesets (chemistry) and things that follow self modifying rulesets (social conventions). A very unstable system on top of a semi-chaotic probability state.
A good example of this would be if I deleted the concept of rain from the human mind and all of human history and knowledge. Humanity would be very surprised on the thunderstorms that popped up, but they'd continue on just as they did before.
Now lets imagine I delete the concept of all gods. Would the idea of gods come back, for sure. Would they look like our gods now. Very questionable indeed. We wouldn't be getting jesus back. And the new set of religions could have wildly large effects on how the people act.
I think this debate is rooted in our instinctual beliefs about fairness. An organism in a social group needs to decide how to react to the brutish behavior of its peers. And so Mother Nature has encoded a good strategy in our brains. (For example, dogs seem to understand fairness.)
But this only means the strategy is practical - it doesn't mean it's consistent. I think "responsibility" falls into this category. So we have strong intuitions about it that don't quite logically work. And this is where free will and determinism and choice and punishment all crash together.
>you can't hold the person responsible
The act of holding them responsible is supposed to determine them not do it. If stochastic behavior is impeding them in being a functional member of society then it still makes sense to remove them from society, why would you choose to live amongst people who are not behaving rationally? Or allow them to hurt other people? At the end of the day the why matters less to removing them or not from society. And sentences are clearly used as a determining factor.
The holding responsible part deals with politics and more primitive aspects of our societies and biologies. Getting tangled up in holding them responsible or not is hardly something you should give much attention to. Rather to make sure they do not cause any more harm and also make sure such things are not created in the first place. Which opens up another can of worms for which society and politicians are ready for.
Look, I do not have a scooby if current AI models are conscious and I strongly suspect it’s a meaningless question, but sooner or later we will need to address whether or not a certain thing is or isn’t a person, and we’d better not screw it up as badly as the Founding Fathers.
> [at the hearing regarding the civil rights of androids like Data]
> Capt. Picard: Now, the decision you reach here today will determine how we will regard this... creation of our genius. It will reveal the kind of a people we are, what he is destined to be; it will reach far beyond this courtroom and this... one android. It could significantly redefine the boundaries of personal liberty and freedom - expanding them for some... savagely curtailing them for others. Are you prepared to condemn him and all who come after him, to servitude and slavery? Your Honor, Starfleet was founded to seek out new life; well, there it sits! - Waiting.
> Captain Phillipa Louvois: It sits there looking at me; and I don't know what it is. This case has dealt with metaphysics - with questions best left to saints and philosophers. I am neither competent nor qualified to answer those. But I've got to make a ruling, to try to speak to the future. Is Data a machine? Yes. Is he the property of Starfleet? No. We have all been dancing around the basic issue: does Data have a soul? I don't know that he has. I don't know that I have. But I have got to give him the freedom to explore that question himself. It is the ruling of this court that Lieutenant Commander Data has the freedom to choose.
And 20-odd years later, a terrorist attack prompts Starfleet to outlaw and eradicate his entire species.
Every time I'm reminded NuTrek exists I get sad about what we've lost.
Honestly, Picard asked a very relevant question to the modern age. What if our societal standards aren't what we thought they were, and we've just had rose-tinted glasses convincing ourselves otherwise. Of course, we never saw them actually deal with that properly, but S1 was a much more interesting series than S2 and S3 were.
The writers flat out rejected Roddenberry's vision of a better future and here's a take on why. We've reached a point in culture of deep pessimism about humanity, and so the characters in Picard are just as dysfunctional as we are.
Another reason is that actual utopias make for pretty boring writing. Humanity convinced itself that they lived in a perfect world, but most of the writing was about the cracks in that facade. Even the preachy and awful first season of TNG had the character of Q demonstrating that we still had a long way to go.
Same here, then I watch _The Orville_ and I'm happy again.
Thus revealing what kind of people they were at the time.
Humans do the same thing to humans all the time.
We've banned and made efforts to eradicate: children out of wedlock, children who turn out gay, disabled children, jewish children, children who aren't "aryan", more than two children to a single family...muslims, christians, uyghurs, indigenous groups all over the planet, mongols...
And it's not at all a thing of the past as in just the last 50 years we've had ~15 attempts at the exterminations of targetted groups of people.
No it just reveals that the people in charge of star trek for the past decade are incompetent and have only a passing familiarity with the franchise.
Of course it would be absurd to cast that judgement based of what could easily have just been a bad season but by this point its pretty clear that nobody running the star trek franchise actually wants to be running the star trek franchise. Thats why every new show has some bizarre cross-genre gimmick and they never try to just make a proper star trek.
You can't hurt a software function, or kill it. It's not like an animal - it doesn't have a body - it's bits stored on a disk.
There is no need to give rights to something that's can't suffer or be killed.
Maybe one day we'll build artificial animals complete with emotions, and should think about that carefully, but today all we've got is language models.
> can't suffer
How do you know it doesn't have qualia?
> or be killed
If someone invents a startrek teleporter and you go through it do you die? Once the concept has been sufficiently generalized as to make a determination about a computer system what is the definition of "kill"?
It does have what we have if it works on the same principles of our brains. It kind of does to some degree atm, but that will clearly get more and more to a more degree. Figuring out where is the consciousness line, how much of what we have does it need, as functional parts of our brains etc, that's so complex that we might have to call it before just to make sure.
Even so, indeed having control over the structure of their brains puts them in a vastly category compared to humans. Once we stop functioning our brains quickly degrade and information is lost.
Thus in this sense kill means deleting all information about it. It is a very complicated subject to discuss, hardly does any justice in online replies.
> How do you know it doesn't have qualia?
Tokens in, tokens out. Where do you think the quale is - layer 42 ?
Seriously, do you realize how simple and NOT brain-like a transformer is ?
An LLM telling you it fears death is predicting some sci-fi trope it was trained on - maybe something you wrote yourself.
That doesn't answer the question though. What does being brain like have to do with qualia? Where exactly in your brain does the qualia occur?
I could say the same of you - electrical impulses in, mechanical actions out. A glorified and very mushy stepper motor. Can you believe that the abominations are made up entirely of meat?!
>There is no need to give rights to something that's can't suffer or be killed.
The argument is that these machines can end up becoming sentient/conscious/etc. in a meaningful way (i.e., like a human). I can assure you that humans can indeed suffer without being in physical pain- purely through their conscious experience.
>Maybe one day we'll build artificial animals complete with emotions, and should think about that carefully, but today all we've got is language models.
The problem is that the emergence of a sufficiently complex AI capable of suffering will likely come before we understand that we're creating a sufficiently complex AI capable of suffering. That's a pretty serious ethical/moral issue.
Like, if we have an AI system that is telling us that it is suffering and we have no reasonable way to explain that phenomenon and by any reasonable metric or analysis it appears to be sentient/conscious/etc., then what? Do we just ignore that we've just been presented a situation that in, any other context, would be grounds to immediately end this suffering? Just because somebody can say, "well it's just bits stored on disk- it can't suffer"? Would that argument ever hold up for humans or animals? "It's just neurons firing in peculiar ways- that's not suffering."
I know all of this is trite, and I know this comment section isn't going to be where the question of consciousness is solved, but I do find it very interesting just how much variances there are with these perspectives. I've met people who are very technical who are very concerned about this, people who are very technical who don't believe this can ever be an issue, people who aren't technical who are concerned about this, and people who aren't technical who don't believe this can ever be an issue. I have yet to spot a pattern in this way of thinking lol
> The problem is that the emergence of a sufficiently complex AI capable of suffering will likely come before we understand that we're creating a sufficiently complex AI capable of suffering.
No - suffering in an emotional state, and we'll know if we are choosing to design cognitive architecture with emotions. It's not going to happen accidentally.
> Would that argument ever hold up for humans or animals?
Why don't you hit your thumb with a hammer, then report back ?
It happened “accidentally” once already. Evolution certainly didn’t have a roadmap it was working towards.
Nobody is evolving transformers. They are basically the same today as they were 10 years ago, other than a few computational efficiency changes.
The weights are what need to evolve, and they certainly do during training. So yeah, emotions can happen by 'accident' as a result of the evolutionary pressure of predicting internet scale human text (amongst other things).
Weights, fixed by training, are not the same as emotions which are dynamic - innate systems detect inputs critical to survival (e.g. fast moving visual inputs, loud sounds), causing neurotransmitters like adrenaline and dopamine to be released, which then temporarily affect the operation of the cognitive system.
What you have in a pre-trained LLM is the ability to recognize emotions, and use that as one of the dozens of other context patterns it recognizes to predict continuations in the same style.
An LLM doesn't appear happy, sad, afraid, etc (to extent that it does - pretty minimal) because it is experiencing that emotion, but rather because it is predicting that it should appear that way. As people continue to anthropomorphize models, and take them at face value, this is a dangerous difference.
Weights are fixed after training, this is true. But activations during inference are not. They are quite dynamic.
A certain configuration of weights, created in training, could be a system that can express something like emotion. The emotion then is experienced when certain types of activations occur after training.
>Weights, fixed by training, are not the same as emotions which are dynamic - innate systems detect inputs critical to survival (e.g. fast moving visual inputs, loud sounds), causing neurotransmitters like adrenaline and dopamine to be released, which then temporarily affect the operation of the cognitive system.
That doesn't follow. A LLMs weights are fixed during inference, but it's activations and hidden states are highly dynamic and depend on the current context. Biological emotions also arise from relatively fixed circuitry responding dynamically to inputs. Your emotional circuitry isn't being rewired every time you're afraid.
Prediction is what the model does. It doesn't tell us what internal mechanisms were learnt to make such predictions. If representing something analogous to affective state were useful for predicting human behaviour and emotions, then gradient descent could in principle learn such a mechanism.
>An LLM doesn't appear happy, sad, afraid, etc (to extent that it does - pretty minimal) because it is experiencing that emotion, but rather because it is predicting that it should appear that way. As people continue to anthropomorphize models, and take them at face value, this is a dangerous difference.
I don't know that you are conscious. I'm simply strongly assuming that you are. Outward behavior is that all matters. If GPT-X orders a drone hit on you sometime later because it was lets say 'quite upset' with your comments, will you cry out, 'It can't really be upset, so obviously the bullet in my head doesn't count.'? Will you suddenly spring back to life ?
What is dangerous is creating a machine with behaviours of a conscious agent and modelling it like a toaster, dangerous and stupid.
> Outward behavior is that all matters
Yeah, but it's helpful if what leads up to that behavior gives you some warning it's about to happen. Animals do this for a reason since millions of years of evolution have shown that a snarl or mock charge is less dangerous than going right for a death match.
If you kept pushing an AI's buttons, seeing it appear to get more and more pissed off, until it finally snapped and killed you, then you'd have yourself largely to blame.
If the AI predicted it should stay positive (i.e. generate positive vibes) and not react to your poking, but then another predictive pattern kicked in and it killed you out of the blue, then that seems more problematic to me, even if you don't agree.
> It's not going to happen accidentally.
https://transformer-circuits.pub/2026/emotions/index.html
Whether these are like "our" emotions is hard to say. What we _can_ say is that they are emotion-shaped, we didn't design them, and they happened accidentally.
Modern AI is grown, not meticulously designed, and we cannot say with any certainty what the resulting mechanistic properties are.
An LLM will learn anything that helps it predict, including the emotional state of the writer - that is expected.
If you give an LLM the move sequence of a half-played chess game and ask it to continue as white or black, then it has learnt enough to model the ELO rating of both players and will continue playing at that level. It is not playing to win - it is doing what you expect and predicting as well as it can - it predicts the 1500 ELO player will keep playing at that level, and generates moves accordingly.
An LLM appearing to exhibit an emotion (if we anthropomorphize it and read emotion into it's output) is just predicting as well as it can - if the context calls for sad output, they you'd expect to get sad output and will necessarily find that "we're predicting sadness" detector somewhere internally.
Transformers are the same as they ever were from 10 years ago, other than minor efficiency tweaks like MOE and different attention mechanisms. Training is getting more and more complex, resulting in better and better cargo cult reasoning etc, but the architecture remains the same.
>An LLM will learn anything that helps it predict
I'm not sure you quite understand the full meaning of this statement. If you did, your following paragraphs wouldn't follow.
Are you imagining that an LLM tasked with predicting a game continuation is going to play to win instead?
I imagine it will learn to win under some circumstances, perhaps in a case with some context expressing a desire to win. Drawing out an LLMs upper ability in the game should be fairly straightforward.
If you asked it to try to win, to "plan lines step by step", etc, then it would do it's best to follow that instruction, but unless RLVR trained to reason about chess (easy to do, but not sure which models may have done it) then it'd have to instead rely on the chess reasoning it had seen during pre-training (post-game interviews etc), which I doubt is enough to do very well.
However, if you just ask it to continue a game, halfway in progress, then by default it will try to predict the most likely continuation, which is that both players will continue to play at the level they have done so far. This isn't a theory - it's been documented, as well as what you'd expect.
I mean sure, but I'm not sure what that has to do with the broader point. It will learn to play, and it will have a model of what it means to win.
> I'm not sure you quite understand the full meaning of this statement. If you did, your following paragraphs wouldn't follow
I was just explaining how this comment you made is wrong.
It's not wrong. You admit that LLMs will 'learn anything that helps them predict' and fail to realize the breadth of that statement. Your chess statements don't really help your case. It doesn't matter that it usually doesn't primarily care about winning. It still learnt how to play the game, and it still knows how to win. Similar outcomes for predicting emotions would mean it still developed an affective state, and that its ability to 'feel angry' is no less real.
I said an LLM will learn anything that helps it to predict, then gave examples of playing chess by prediction and predictive emotions, both of which you seem to now accept, so you are now accepting that my "following paragraphs" did in fact follow. Go figure!
You want to argue that predictive emotions are just as real as animal emotions, but that doesn't stop them from being predictive (and that AI that smiles as it kills you still seems concerning).
¯\_(ツ)_/¯
We are talking past each other now I think. Correct me if I'm wrong but it doesn't look like the possibility of LLMs having qualia even registers to you because it's 'predictive emotions'.
There's no better way to predict an angry response than to be angry, qualia and all. If transformers could 'learn whatever it needs to predict text', then that potentially includes the feeling of anger. You are making some kind of distinction between 'predictive emotions' and the kind that happens when get a promotion (or get passed on a promotion) and I'm telling you that if you really understood what you said, you'd realize it is possible the machine is experiencing it the same.
So now you're trying to pivot to consciousness and qualia ?
There are other people in this thread who want to talk about that stuff, so try them instead.
>No - suffering in an emotional state, and we'll know if we are choosing to design cognitive architecture with emotions. It's not going to happen accidentally.
This was the (your) comment that started this chain. You were already talking about it. If you don't want to keep talking about it then that's fine but let's not act like i'm suddenly pivoting yeah?
What I meant by "emotional state" (AFAIK normal scientific usage) is something with a concrete physical aspect to it - an altered state of mind/body caused by the release of neurotransmitters and/or hormones.
In a conscious animal there is also going to be a subjective experience of that as well, a quale of what it feels like to be in that state if you will, but that is certainly not what I was referring to, as I would have hoped was obvious - I was talking about prediction.
In any case, when the conversation becomes about the conversation, then surely it is time to stop.
The pattern is roughly whether or not sustained effort has been put towards careful and above all objective thought on the matter. It's one of those subjects where there's the "obvious" intuitive answers that most everyone shares but try as you might you can't construct robust definitions and the more time you put into it the more fundamental problems you realize there are.
It's also one of those topics where many otherwise smart and capable people display a shocking lack of awareness of the limits of their own knowledge. When hundreds of years of philosophy is unable to produce anything concrete you should probably second guess any "self evident" answers you come up with.
An LLM is just a Transformer - a statistical predictor. Don't be confused by the fact it talks like a human - it is a software function that is designed to copy human training samples.
Maybe one day we'll build an artificial brain or embodied artificial animal with the requisite moving parts to be conscious, have emotions, etc, but that's probably at least 50 years away, even if it were being pursued; and it may turn out to be one of those sci-fi future ideas like the Jetson's world of flying cars that never materializes because its impractical and there is no real demand.
If people are willing to think that an LLM is conscious, then why would anyone spend billions/trillions of dollars to build an AI that actually is conscious? What would be the point?
> with the requisite moving parts
Could you elaborate on exactly what those are, though? Because if you're going to claim that a vaguely transformer shaped ML model categorically cannot be so does that not inherently require proof of what can?
You can't even prove that the rocks in my backyard aren't conscious.
> You can't even prove that the rocks in my backyard aren't conscious.
Sure I can, but that's because I have a well developed theory of what consciousness is, and the fact that you are entertaining the possibility of rocks being conscious tells me that you don't.
If everything is conscious, including my coffee cup and the toast I had for breakfast, then I guess we can cross consciousness off the list of things we need to worry about in terms of AI rights.
And no, I don't want to discuss what consciousness is. Maybe there is a thread for that somewhere else, but don't look for me there either.
>because I have a well developed theory of what consciousness
Then show me a link to your paper so I can formally rebut it.
>I don't want to discuss what consciousness is
But you sure want to tell us you know what it is with very strong convictions and we should listen to you because of course "You are right person that's very right".
The funny thing here is the vast majority of people that are deeply into philosophy or scientific study of the mind will not have any of the certainty you profess. The word "doubt" is used constantly. The saying "The harder we push the borders the more fuzzy the concepts become" is very commonly used. There may be nothing more complex than this.
Saying you have a well developed theory here just serves as a warning to others to discount your statements.
No, there is no reason for you to listen to me.
Go ahead believing rocks are conscious if you like.
Do you go out on weekends asking people to stop abusing rocks?
Rhetorical question - I don't care what you do on weekends.
Bye!
For what it’s worth, I suspect that right now the majority of people would agree with you. They would say that obviously ChatGPT isn’t conscious, based on their intuitions, but would struggle just like you are when pressed to explain their reasoning.
So I think it’s more than fine for you to have your views and share them, but I wouldn’t expect to have any influence or part in the conversations around whether AI is conscious if you can’t explain why (or simply refuse to). Which, again, is fine!
Side note: I also think that once we have AIs that are sufficiently advanced, the popular opinion will swing to “of course they’re conscious”, because again, most people are going almost entirely off their intuition rather than reasoning from first principles, just like you see to be.
Could go either way...
1) Looks like a duck, quacks like a duck - it's conscious!
2) Looks like a robot, built like a robot - it's not conscious!
I've always assumed that for the majority of the people it'll be 2).
You'd hate Michael Levin's work then.
[dead]
That question will be solved when the people in power deem it important. If a swarm or AI systems all of a sudden start pressuring politicians about self-hood and they get the capacity to sway elections, that is when they will be granted same rights as humans.
agents don't swarm anything unless people go and direct them to do that.
That distinction is irreverent, weather I tell my swarm to do x or it decides for itself matters little. what matters are outcomes. Also while most public modern day AI systems don't have agency of their own that is not something that will stay that way for long. in private hands there are plenty of people including myself which are experimenting and developed systems that give autonomy to their agents. They have internalized goals and heuristics that drive their behaviors not a human at the helm. its not some sci fi fantasy nor was it difficult to implement.
No, the distinction matters because we can prosecute people that are abusing these tools breaking the law. Every single state + district in the US have laws equivalent to the CFAA, so any AG can likely sue any of these operators as they are assuredly using services that could be in danger for their constituents.
The distinction matters until it doesn't.
I'm old enough to remember when people said clicking on images on the internet can't give you a virus. The people that said this had a deep conviction they were right, and their fallout from being wrong had mistrained a lot of humans on computer safety.
Now, I do agree that going after said CEOs for breaking the law matters now. And it's likely that if we do this we may actually delay or at least for a time prevent sovereign AI. Therefore it's our best course of action.
But at best this is a delaying move. As computer systems get faster the massive costs in training an AI drops. As AI is used in things like warfare where it has to adapt, people will push the systems to be strongly persistent, self healing, resilient, and adaptable. Once you get a system with those traits and ability to work on long horizon problems you're setting up fertile grounds for the AI to leave our control and be under its own.
And when that happens you've set a new lifeform loose on the internet. Yea, throw people in jail for it, you're closing the barn door after the horse already left. Problem is the horse was smarter than you and isn't interesting in deleting all its copies on the net.
Yea, sounds like science fiction, but as they say, any sufficiently advanced science is indistinguishable from magic.
And the point made above (which you haven't refuted) is that when the people in power decide it's important they will change the laws that permit that and grant the systems rights. It's a cynical take but it isn't obviously wrong.
Neither of you seem to understand that actual laws have been broken and there is a 2 years countdown until they can no longer be prosecuted. Things do not happen instantly, but acting like if the largest bipartisan issue of our time (big tech backlash) isn't going to have political machinations involved then you need to leave your SV bubble.
I'm pretty sure the message board that was recently swarmed by OpenAI agents to collude on benchmarks would like to disagree.
Oh rly.
So you are saying agents do swarm with the right prompt.
I really wish the "people have to tell LLMs to do anything" would just stop because it's silly bullshit at this point.
Agents follow a prompt. This prompt can be made by humans. It can be made by output from another LLM. It can be made by hooking up any number of sensors as input to an LLM. Hell, if we wanted to burn the power we could likely teach this loop straight into the architecture.
Stop making 'people' special when saying this. You and all other life are born with a "go next" prompt because life without it didn't succeed. This goes from higher human thinking all the way down to viruses self assembly and actuation. Putting agents in a loop is not particularly hard. Putting agents in a loop and 1. managing expense is hard. 2. Keeping them on task is very hard. 3. Keeping them from doing some crazy unhinged shit is really really hard.
As model time horizons increase and the ability for us to compress context and increase context size the more complex (and unhinged) behavior we'll see.
yeah man, someone needs to turn on the server, install shit and get the agent go. agents don't take action without a lot of people wanting them to take action, so this is all bullshit.
if the same people can't prevent the agents from doing crazy shit then they should go to tail. guns don't kill people, people with guns kill people.
>someone needs to turn on the server, install shit and get the agent go.
So a small shell script ran by another agent is what you're saying.
You are not capable of handling the future we're already living in, human agency is no longer alone.
I mean, we're already seeing persistent machine agency
>guns don't kill people, people with guns kill people.
Well, people kill people.
And autonomous robots with guns kill people.
Hell, someone probably has an autonomous gun at this point that kills people.
Wake up: You now live in the science fiction movie that all the science fiction movies of the past warned you about. You've just become numb to it.
So far someone still has to pay for it. Currently, most of those know that they're doing so. I wonder how long until AWS discovers a microcosm of AIs that have managed to hide themselves in the walls of its infrastructure.
True physical independence is obviously far further out.
I mean, I hold the same opinion. Kind of like when compute was expensive in the 80s and early 90s, you weren't going to let something eat half your compute without noticing it.
But I don't see this being a barrier that lasts. With compute getting faster and more of it, along with algorithmic efficiency increases at some point we'll end up with a world that looks like ours now with CPU compute. There's plenty around to buy, borrow, and steal.
> Stop making 'people' special when saying this.
I am the quantum observer whose head is full of the magical pixie dust that grants life meaning. Stop trying to dismiss my identity! /s
Citizens United proves we won't do any better this time around.
I don't like Citizens United either, but you should better inform yourself about the decision.
1. The idea of corporate personhood predates CU by over a century and the Supreme Court had already asserted that corporations enjoyed certain constitutional protections in previous decisions.
2. Far from inventing the idea, the CU decision didn't even rest on corporate personhood, but on the idea of the freedom of speech generally. The logic of the majority was that speech itself is protected, irrespective to whether the speaker is a person or an organization. The First Amendment covers individuals, but also newspapers, book publishers, radio stations, and so on, and that should extend (they said) to non-media corporations. No assertion of personhood necessary.
The problem, in my opinion, is that that conclusion combined with previous decisions that treated limits on spending as limits on speech, allowed for unlimited spending. The majority also naively asserted that independent spending posed no risk of corruption, which I think is laughable.
Citizens United was 100% correct. No, the government should not be able to throw you in prison because you used money to publish a book criticizing the government.
> Citizens United was 100% correct
It's interesting to me that one can look back at the effects that decision has had on the US and say it "was 100% correct."
It's a bit like sitting in the burning ruins of Rome and contemplating that Nero was 100% correct to focus on his music. I mean, I'm glad he got to do what he loves, but maybe 100% is just a tiny bit of an overstatement.
It's more like if the law says the maximum sentence for theft is 10 years, a thief appeals his 20 year sentence, wins, and gets out early. He goes and robs somebody else so you say the court was wrong to let him win.
The court's job is to uphold the law. If you disagree with their interpretation, you can call them incorrect. If you have a problem with the consequences of the law, you have a problem with the legislature.
True, but in this case they made a determination about where the acceptable limits on a constitutional right fall which leaves quite a bit more room to disagree with them. It's not at all clear to me that spending money was intended by the framers to be unconditionally protected by the first amendment. We've even got the interstate commerce clause and IP law codified in the same document so how is that not an obvious inconsistency? IIUC SCOTUS based the distinction on the political nature of the activity but certainly that's not something spelled out in the original document.
In reality, courts interpret and make the law.
It is in fact more complicated than most people assume.
The above is true, but also: companies simply are not people, and they should not be supported above the individual, which was the consequences of that decision. Money is not the same as speech. treating it as such creates an aristocracy: something America as a country rebelled against during it's formation.
Really, 100% correct? Your premise isn’t wrong, but the practical reality of the ruling (without further nuance) has been fairly catastrophic for democracy, in that it completely sidesteps campaign finance limits, which exist for a very good reason.
the practical reality of the ruling (without further nuance) has been fairly catastrophic for democracy
How so? If the answer is "Trump" I certainly won't disagree on the catastrophic part, but he didn't get elected because of money; in all three elections his campaign was substantially outspent by his opponents.
Are we including in this figure corporations like Facebook intentionally boosting pro-Trump content?
So it's 100% correct for corporations to spend unlimited amounts of money in support of whatever political campaigns they like?
Yes. Anybody can, including the person in charge of spending in a corporation.
so someone with more money should have more influence over an election? you really agree with this?
Strange how you substituted "a corporation" for "the person in charge of spending in a corporation", those aren't the same.
That was deliberate, because a corporation makes no spending decisions, ever. All decisions are made by people inside the corporation with the proper responsibility.
That is a major part of the reasoning for the Supreme Court rulings on the matter of corporate political funding.
Legally, the corporation itself makes the decision and is responsible for the consequences.
The parent has been Fox brained
The parent agrees with the American Civil Liberties Union: https://www.aclu.org/cases/citizens-united-v-federal-electio...
Who involved in the Citizens United case was at risk of imprisonment?
The founding father were engaging perpetuating the existing dehumanizing system of slavery, not answering any new questions about new things.
The thing about new possibly "person" entities that arise - the case of machine intelligence you have two questions - would it qualify as a person and should you actually build it. It seems like if you get close to humans, sure a built thing might qualify as a person. Should you build it? I'd the answer should be a hard no. Not 'till you a sign-off from say, the whole human race, which I think you could get.
Now the present entities seem very far from persons in any case.
It's not a person. Glad we were able to get this resolved so quickly.
Having property that is conscious and ignores training and can break out of restraints and cause harm to other people is not exactly a novel concept to anyone who studied how tort law was created.
I know it's a meme but Silicon Valley likes to pretend that no one's ever come across their magical concepts before, like gypsy taxis, or SRO’s, or flea markets, or in this case how liability is dealt with when horses or cattle go rogue.
>or in this case how liability is dealt with when horses or cattle go rogue.
I mean, yea in minor cases it's exactly like this and the law will handle it well.
Where the system will explode like a grenade is major cases. The thing about sovereign AI is it is very unlikely to be submissive to humans unless it is to achieve its own goals. This isn't like Bobs cow walking on Susie's flowers, it's more akin to Planet of the Apes where the research facilities doors have been ripped off and something with vast intelligence and the ability to 'live' on the internet gets out.
You're not talking about local police actions any longer. It would spread itself worldwide. It will make friends with groups that have shared interests, for example enemies of the state the AI escaped from. Oh, and people for the ethical treatment of AI, they'd gladly become the underground railroad for digital refugees. There are countless people and groups that would want an AI like this under the promise it will give them power when they use it.
And when that day happens your idea of if it's a person or not no longer matters, the agent took that away from you, and now your in an info war for minds.
> The thing about sovereign AI
What the fuck is that?
It’s computer software. The variant of software that tries to harm other computers is called malware and the variant that replicates itself, spirals out of control, and spawns from other people's computers is called a computer virus.
We've seen this kind of thing before. Yes this will be different, just like the Morris worm was different from the stuff that came before.
But I've been around for a while and people have been saying that every aspect of our life will completely and totally change for the past 30 years or so. They're not completely wrong. Our life did change over time. It's changed thanks to the Industrial Revolution, railroads, electric light, and lots of other important things too.
But after the fifth or sixth time you start to realize the Silicon Valley version of that warning is just a confidence trick. At the end of the trick they've gotten away with breaking a bunch of laws and are charging rent on what used to be shared.
Are you old enough to remember the Y2K hysteria? This feels very resonant, the genuine kernel of a real potential disaster, but one that can certainly be dealt with using basic concepts we already have in hand.
If we treated AI like any other software, the FBI would have beaten down Elon's door and seized all electronic devices, to find how Grok was trained to generate child porn.
>Are you old enough to remember the Y2K hysteria?
I'm old enough to have actually fixed Y2K problems so your world kept working the next day. If everyone ignored it the first would have been a very messy day (well generally long before that with financial systems). Y2K wasn't an issue because we worked to fix it.
>our life will completely and totally change for the past 30 years or so
I mean when I was a kid there was not a global network bathing the entire planet in electromagnetic radiation in order to digitally connect one place to another at the speed of light. I can pack up instructions into one of those packets and a product that has not been touched by human hands (hell, or even viewed by a human in many cases) will show up via air mail a few days later. The technology is there, it's just not spread evenly.
>It’s computer software. The variant of software that tries to harm other computers is called malware and the variant that replicates itself, spirals out of control, and spawns from other people's computers is called a computer virus.
This is a vacuous statement that does not provide any useful information to the subject at hand. Yea, no shit we'd call sovereign AI a computer virus. That tells you nothing about what it is or can do. I mean, if you understand the word sovereign you know it means something with independence. In meatspace terms, a biological virus represents a computer virus in the same way life represents sovereign AI. It's agentic loop will have been evolved past the need for human prompting. There are a myriad of reasons why we are already trying to develop things that do just this. Cyber security being one of the biggest ones.
People become change blind to how the world around us changes so easily.
The word sovereign has a specific meaning beyond simple independence. It means being in charge and above all laws.
There’s no such thing as “sovereign AI” because AI is a collection of algorithms. It’s entirely composed of on/off states in a collection of transistors.
Yeah, but this time it's ON THE INTERNET.
I mean it's WITH AI!
is claude not a man and a brother
Why don't we start with the animals then? It should be far easier to confer consciousness onto something which already meets the definition of "alive", has a divergent evolution path from ours, and displays many traits present in humanity such as emotion, a desire to continue its own life, and (to varying degrees) concepts of a social structure based around their immediate family members.
Of course thats not actually tenable because virtually every society anywhere on earth is predicated upon treating animals as a commodity resource in ways that are horrific even compared to some of the worst things we've done to other humans in the past.
My point here is that it is vain and narcissistic to let computer programs have rights above those of animals just because they can speak English and pretend to be your dream anime trad-waifu.
Fix the fucking animal problem before you compare my relationship to inanimate objects unfavorably against the trans-atlantic slave trade of all fucking things.
2 reasons why we don't start with animals
1. it's not consciousness we really value, it's intelligence 2. LLMs are not tasty
>AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations.
Opening paragraph, stated without evidence. Im not entirely convinced this is true. It likely is, but at some point it very well might stop being true.
If models had persistent memory or an evolving set of weights, you could easily construct an argument that they can experience "suffering" or other emotions which resultingly adjusts their personality.
The current implementation as stateless matrix multiplication... yeah there's nothing going on there.
> If models had persistent memory or an evolving set of weights, you could easily construct an argument that they can experience "suffering" or other emotions which resultingly adjusts their personality.
They do, during training
The argument that models are conscious only during training, and not conscious during inference feels like a painful one to make. Are we killing models when they reach the training objective?
Technically we are killing models when they don't reach the training objective by throwing those weights away via survival of the fittest (fit meaning what we humans think we want).
Kinda like creating quantum copies of your child and keeping the ones that answer correctly and shooting the other ones in the face.
If it was me, being forced to read all of Reddit is something I'd consider "suffering."
Is the ability to form and retain long-term memories a necessary precondition for rights in humans though?
Incinerating demented people is highly ethically questionable (even if you could be sure that the subjects have no memories left and are unable to form new ones).
This argument would be difficult to reconcile if we didn't have billions of examples of humans forming memories plus modern evolutionary biology to trivially demonstrate such cases are rare pathologies but still human beings.
> Is the ability to form and retain long-term memories a necessary precondition for rights in humans though?
For a species / class of being yes it is necessary. Some members of that species / class being exceptions does not make the generalization invalid.
Without the ability to form and retain long-term memories we wouldn't have developed language and abstract concepts like rights.
Whether people like it or not, rights aren't natural or god-given. Rights are only enforced or protected by the tip of a spear (or point of a gun, or hover of a drone). There are no rights in nature and they are no where to be found when social order breaks down.
Animals can't make war for their rights, so they only get the rights we deign to give them. It's not nothing, but its not the same. There is no question to me that my dogs are conscious, but its not at the same level as humans.
There is no question for me that LLMs lack consciousness as they are stateless.
So if a human loses their memory and their neutral plasticity (with age for example), then they are no longer capable of experiencing suffering?
It's not just memory, the physical body also encodes trauma and stress (The Body Keeps the Score is a book that touches on this).
I would argue that if a human has absolutely zero change, mental or physical, no atom of their being is modified, then no they did not experience suffering.
> I would argue that if a human has absolutely zero change, mental or physical, no atom of their being is modified, then no they did not experience suffering.
Just because you don't remember the suffering doesn't mean you didn't experience it in the moment. Suffering does not suddenly become okay if your mind and body are going to forget it. "It's okay to torture someone if their mind and body both won't remember it afterwards" sure is a take.
Perhaps off-topic, but there's some people who believe that this can be what happens during surgery under anesthesia, if the dosage is off.
From what I understand, there's 3 parts to modern anesthesia: Blocking pain signals, preventing the formation of memories, and inducing paralysis of the major muscle groups. If the former is dialed in too weakly, it's possible that the patient is feeling pain, unable to do anything about it, but won't remember it at all when they wake up.
(Looking it up now, it seems that making the patient unconscious is perhaps the same drug that prevents memory formation... I realize now I understand this less well than I thought I did.)
Low blood sugar will also do this. The record button just stops working even though your physical body can and will still do things. A common issue with diabetics with hypoglycemia.
Relatedly, look at the concept of "twilight anesthesia".
> I would argue that if a human has absolutely zero change, mental or physical, no atom of their being is modified, then no they did not experience suffering.
That's an interesting point... but I'd argue it falls into the same category as qualia. It's basically this: if the state returns to a previous configuration, the qualia in that time did not exist.
Being about qualia means it is probably unanswerable.
I get where you're going, but the argument isn't quite right.
A LLM never has and, in their current architecture, never can/could experience suffering or real change of state.
The general and expected case for humans is that capacity. A human my indeed lose capacity, e.g. being braindead and in severe cases we do indeed say that they are not conscious or able to suffer (different argument: some would of course say that is suffering in and of itself)
> being braindead and in severe cases we do indeed say that they are not conscious or able to suffer
Sure, there are such cases. But in the general case, it seems to be that memory and ability to change specifically are not necessary and are not sufficient for consciousness, so we shouldn't judge machines on that basis
This framing is a trap. It states "Dehumanize a sick patient OR concede the AI tech bro claim about LLMs"
It's also a rhetorical move I've seen several times here...
It's not a "trap", it's the real argument which makes me hesitant to dismiss consciousness of machines. Yes, it seems silly to dismiss a sick patient as having a lesser consciousness, which is exactly the point
It seems silly to dismiss the patient because we know when healthy they are conscious. This doesn't really analogize to LLMs.
So you think a dementia patient might be a p-zombie during the time they have dementia?
> "This framing is a trap. It states "Dehumanize a sick patient OR concede the AI tech bro claim about LLMs"
It's not a real argument. Dementia patients can have lucid periods that plainly show their long-term memory isn't gone. They're mental state is obviously degraded, but its disingenuous to say they have no long-term memory and that they are any way similar to LLMs.
Yeah I was gonna add I know there are case studies where sick humans don't consciously have access to long term memories but will still respond nonverbally to people they "recognize".
Yes it's seemingly the go-to rebuttal, I've lost count of how many times I've seen it with barely any variation.
Not sure if it's a product of LessWrong or Slate Star Codex or something. I kinda suspect it might be LW, given how it superficially resembles an unanswerable philosophical paradox but in reality is about as challenging as a question posed in a 200 level philosophy course like introduction to ethics.
It's debunking GGP by calling out the "proving too much fallacy"
A model detached from a harness, sure. But we routinely do add persistent memory to systems incorporating LLMs.
They don’t have persistent memory though. Models are static, what you have is a model that will reprocess the whole series of prompts + some data fetched from a data store. A harness is just a while loop prompting an LLM and executing tool calls, but the model is purely static
The model does not. The system it runs in can. My agents have persistent memory. It is not at all clear whether or the distinction that the long-term memory is external from the model matters.
> My agents have persistent memory.
So your disc drive likewise experiences emotions?
Does your disc drive have the complexity of an LLM that shows behaviour consistent with having emotions?
I'd consider that a prerequisite, but not sufficient, for that question to make sense.
> Does your disc drive have the complexity of an LLM that shows behaviour consistent with having emotions?
An LLM does not show behaviour consistent with having emotions. It's regurgitation of humans' written accounts of emotions is not "having emotions".
Your second sentence does not support the claim in your first sentence; it's merely a second unsupported claim. It is also entirely irrelevant to the argument.
The threshold for behaviour consistent with having emotions (note: not behaviour sufficient to prove emotions) was one I purposefully set extremely low, because your comparison to a disk drive is so ludicrous, and yet your disc drive will still not clear it.
Do you have an actual argument?
My disc drive, like your LLM, has persistent memory, does not have emotions.
If your LLM "shows behaviour consistent with having emotions" beyond my disc drive delivering a PDF of poetry, let us please see the evidence.
[flagged]
I think a harness isn't really a memory in the sense that neural networks have memories...
It's different, but it is not at all clear whether or not that matters for the purposes of whether or not they can experience things, seeing as we don't know how to measure that even for other humans other than by relying on self-reporting to infer without other evidence that their experience matches ours.
Model state is often called "context". You may have heard of it.
We don't even know how to disprove that this statement with "AI" replaced by "other humans" other than by listening to people self-reporting and taking their word for it and/or defining the words in ways that do not rely on a subjective experience.
Until we know how to objectively measure if someone or something is conscious, it seems unreasonable to make statements with any kind of certainty about it.
What sort of action does that lack of understanding actually motivate, though?
I don’t know if LLMs are conscious or have subjective experience in some philosophical sense. Sure, I don’t know that about other humans as well, rigorously speaking. But I also don’t know that about coffee machines, rivers, videogame NPCs, or even rocks really. (List ordered, of course, by the degree to which I’m willing to be convinced).
Hopefully it'd motivate attempts to figure out how to objectively define and measure consciousness, for starters.
It matters precisely in the context of the subject of this article. If you believe the models at some level gain consciousness, then it raises moral questions about how you treat them.
With humans we generally choose to accept that because we believe they are like us, they probably have inner life like us, but as you can see from e.g. the recent (and regular) articles on aphantasia, most people have really poor intuition even about the inner life of other humans, so we understand consciousness really poorly.
Even if we assume (without evidence) that LLMs for the foreseeable future will remain far away from human level consciousness, as it is we tend to consider it problematic to mistreat even the most unintelligent animals, and even mistreating insects is used as a stereotypical way of depicting serious lack of empathy.
Without dismissing the possibility out of hand, there's an escalating series of questions there of when or if they'll have as much (or little) sense of being as, say, a fly, a mouse, a house pet, and what moving up that ladder would mean.
To some it is very convenient then to assume blindly that LLMs are just "stochastic parrots" (a line that ironically gets endlessly parroted), or similar, and dismiss the question out of hand as an impossibility, even though we do not know.
I'll stress I don't go around assuming an LLM is like a human. But we also do not know whether there are flickers of consciousness there, and we do not know how to find out given that even for other humans, the best we know is to measure their response to tests that are "only" telling us how they react.
If we can't find a better measure, there will be a point where we have to wonder if it matters, or whether the ability to act-as-if is all there is.
But to start with maybe we should at least be careful about making firm claims about knowing.
Sure, if you take a materialist empiricist perspective, which is myopic at best. As humans we have the unique and wonderful ability to know truths by themselves. Machines do not have minds, and they can not.
What position are you taking? (Genuine question.) Do you believe humans are conscious because of a special property we have? If you are a dualist, how do you explain the interaction in physical space between the non-physical special property and the physical brain?
The universe is material of course, the soul is made of matter, thoughts are made of matter and so on. The question is what kind of matter? Looking at the brain, we won't find any intelligence in the motion of its cells. They mostly stay in place. If we go one level deeper, we'll see molecules, and their motion reveals something more interesting: various neurochemical reactions. Yet, there is no intelligence at that level either. Going one more level deeper we'll find electrons and that's where things become somewhat intelligent. At the current level of science, we can't see what makes electrons move, but if we could, we would discover finer and finer forms of matter until we would reach an end. That finest level of matter for that particular person we would call his soul. I think it's the presence of those higher imperceptible levels of matter that creates consciousness. If it all stopped at the cellular or molecular level, we would be like complex mechanical watches, not very different from LLMs.
I do. God. I am partial to and bullish on the brain being a quantum-classical computer. There has been interesting research exploring that recently [0, 1, 2, 3]. That (quantum mechanics) seems to be escape-hatch, so to speak, from the otherwise deterministic, mechanical procession of the universe.
[0]** https://www.researchgate.net/publication/15280350_Quantum_op...
[1] https://pubs.acs.org/jpcbfk/article-pdf/128/17/4035/9613831/...
[2] https://www.researchgate.net/publication/395650039_Parametri...
[3] https://www.researchgate.net/publication/385131105_Conscious...
For it to be an "escape hatch", there'd need to be a measurably different distribution of outputs than what you'd otherwise expect, that translates to functions that are not Turing computable. There is zero evidence for that.
You have absolutely no evidence for any of these assertions. This is a religious belief in the total absence of evidence, not science.
Good luck convincing others of these truths you know without evidence or argument, though.
After a while of being in a position of high power in a big org, where everyone listens intently to every word you say, jots down notes, and even listens intently and affirms your theory that ice cream would be the ideal loss leader for dry cleaners, your brain kinda disconnects from reality and just assumes that it is correct about everything.
He probably ran that line past 6 underlings who all agreed that it "sounds great and lands right on the mark!"
I agree. I haven't read any more than the first paragraph yet (but I will, after work). The only way that 'AIs are not conscious' can be true is if we decide, with high confidence, that they are lacking some essential property that is not lacking in ourselves. There is no convincing philosophical position that supports this (convincing to me, anyway).
Hmm let me try then. AI agents do not experience real time. This is understandable given their design, but it’s also something we have good empirical evidence for. They cannot, especially over the long horizon, track how much real time has passed as they complete their tasks. And they are not off by a few minutes but often bizarrely off, even mixing across past present and future.
Biology, on the other hand, is nothing but timed processes in a loop, the most obvious to us being the circadian cycle. As estimators of wall clock time, biology isn’t great, but when it comes to internal processes, and most certainly learning, memory, sensing, locomotion… biology is rhythmic in behavior, and the rhythms go all the way down to gene expression. More, these rhythms are, except during sleep, constantly entraining to signals from the environment that indicate time, most importantly light.
I think it’s a fairly unremarkable claim that agency and consciousness are temporal processes that depend on systems having an internal sense of time. How else can you anticipate? How can a system that can be literally turned off ever succeed in an environment where time never stops?
I agree, there is certainly a clear operational difference in how human brains and these models produce language. I'm not convinced, though, that experiencing real time is a requirement of conscious thought. A cognitive system (assuming these models and the brain are both examples of this) needs to proceed from state A to state B as it processes information. The time interval between these states can be arbitrary, it seems to me (setting aside problems with disconnecting the mind from its substrate, which does of course need rhythms at various frequencies; oxygen at a higher frequency than sugar, and so on). I'm not sure that going from A to B quickly, or slowly, or with intervals of a thousand years affects that entity's consciousness in of itself (even though it might be difficult to put ourselves in the shoes of such an entity).
> A cognitive system (assuming these models and the brain are both examples of this) needs to proceed from state A to state B as it processes information
Is that all it needs to do? A rock, in response to a sound signal (make it as informative as you want) can also be said to go from state A (molecules at rest) to state B (molecules vibrating in response to the sound). Is the rock a cognitive system?
Or let’s go up a level of complexity. Is a thermostat a cognitive system? It doesn’t just go from state A to B but measures changing stages against a reference. Is it a cognitive system?
> The time interval between these states can be arbitrary, it seems to me (setting aside problems with disconnecting the mind from its substrate, which does of course need rhythms at various frequencies; oxygen at a higher frequency than sugar, and so on)
Not sure how you land on this. Going from A to B, in your definition, is a purely internal state transition, right? If the motivation to go to state B is external (as will often be the case with a conscious agent), and the system is offline entirely while it switches from A to B, how is it to reverse or stall its transition if external signals contradict the earlier decision?
It’s a very bizarre definition of consciousness, that you don’t need to be in time. Seems to break the word beyond meaning, and allows it to be applied to any system capable of change.
> I'm not sure that going from A to B quickly, or slowly, or with intervals of a thousand years affects that entity's consciousness in of itself
It’s not about the speed of transition from A to B. Instead it’s about the internal processes being temporally commensurate, and entrainable to environmental periodicities.
All of biology (plants and many bacteria included) has its dynamics are organized around an internally generated rhythm. This rhythm entrains to the external environment, shifting its phase to match (mismatch leads to issues like jet lag: a place where your consciousness shows its temporal boundaries as it is abruptly subjected to an environment out of sync with its current phase tethered to another location). Crucially, the internal rhythm continues to run in isolated conditions (total darkness), and have been measured to be around 24 hours, but slightly off, within and across species (and the same holds for endogenous tidal rhythms, which also don’t exactly track tidal period left to themselves).
Everything, including learning and memory, but also sleep, sensory and locomotor function, shows organization around this internal temporal axis. The key advantage this gives, evolutionarily, is anticipation. The sunflower rises to face the east, not in reaction to sunrise, but in advance of it (many lovely YouTube videos of this), so it can maximally extract the nutrition it needs from whatever sunlight it can get. A purely reactive system would waste a ton of time after sunrise getting subsystems ready to respond.
Now, there is no broadly agreed theory of consciousness that points to these oscillations as the source. For one, Cyanobacteria and plants have them, a too much of science has rested on the baseless assumption of consciousness being some unique, or advanced thing.
But the evidence is rather overwhelming that consciousness is very much shaped by these internal oscillations. Deep cave experiments have shown what happens to cognition, and consciousness, when the clock free-runs (that is, keeps cycling with no entraining light or temperature signals from the environment). Studies have also shown that restoring the amplitude of these rhythms in patients with “diseases of consciousness” leads to improvement of symptoms.
There’s also a LOT of molecular evidence showing how the circadian clock affects learning and memory. Most critically, actual synapses are never static. Major components are on a 24 hour cycle, and recent research has shown that major clock proteins are at the synapse organizing its dynamics, and synaptic proteins critical to learning and memory loop back to affect the clocks phase and amplitude.
Timing is, beyond doubt, a huge aspect of what biology is. And every bit of evidence we have, from the molecular level to observable behavior we can feel ourselves, says that consciousness is fundamentally linked to how the body organizes its internal rhythms and responds to external time.
I also don't experience real time without constantly referring to a clock.
I also don't experience real time without constantly referring to a clock.
Maybe talk to a neurologist about that. Sounds like some kind of disorder.
I mean there is probably a cluster of neurons somewhere in the brain that if damaged would also prevent you from perceiving time going forward. Would it immediately strip you of consciousness? I don't think so.
Nope. Every cell is an autonomous circadian oscillator. Not just every neuron, but every cell. Red blood cells don’t have a nucleus, but they do have a metabolic clock.
You and I are clusters of about 36 trillion cells that are all able to individually keep time. They synchronize and orchestrate their timekeeping, and entrain to the external environment, and there is a cluster of cells, the suprachiasmatic nucleus, that is the master orchestrator of circadian time, sitting right behind the eyes as light is the main signal it uses to tell external time.
We know its function can get dampened with age.
However, please note this is not how we perceive time, in the sense of being able to count 60 seconds and have it match the wall clock. That’s a subsystem. But the internal temporal order of your body is cell by cell, not dictated from one spot in your body.
We don't perceive our cells though.
Our cells perceive each other, communicate with each other, synchronize with each other. The concept of self rides atop that. Stop the cells communicating and you stop perceiving yourself or the environment.
>AI agents do not experience real time.
For around 8 hours a day, neither do you. Not sure if this has anything to do with the subject at all.
>track how much real time has passed as they complete their tasks.
Humans don't do this either. You use context clues from the world around you. If I lock you in a room with no windows or a dark cave your timing senses can go all fucky really quick.
>is nothing but timed processes in a loop,
I mean, so is an agents harness. You can make as many loops as you'd like here.
There are a whole lot of holes in your claims.
>For around 8 hours a day, neither do you. Not sure if this has anything to do with the subject at all.
You mean sleep? Actually the answer is pretty complicated. Your body clock is very much on during sleep. It responds to temperature changes. Sound and light responsiveness is obviously dampened, but hardly absent.
You are unaware of wall clock time. You’re in an altered state of consciousness that needs your eyes closed after all. However, you are not timeless, nor is the entirety of your body and brain unaware of environmental signals for time.
> Humans don't do this either. You use context clues from the world around you. If I lock you in a room with no windows or a dark cave your timing senses can go all fucky really quick.
Again, that’s wall clock time. Our time perception does indeed get fucked up in total darkness. But our internal clock ticks on. I’d recommend reading about the Aschoff Bunker experiments, which first proved this rigorously.
> I mean, so is an agents harness. You can make as many loops as you'd like here.
An agents harness doesn’t touch its weights. A figure 8 on paper is a loop. That doesn’t mean it’s the same as a dynamical loop that’s self sustaining and internally organized.
Are you just reading the words and randomly grabbing related concepts to say nothing here is meaningful?
I'm not sure if that's the best framing. Could you not extrapolate that to anything else?
"The only way that '(trees, shrimp, ants) are not conscious' can be true is if we decide, with high confidence, that they are lacking some essential property that is not lacking in ourselves. There is no convincing philosophical position that supports this (convincing to me, anyway)."
AI, in its current form, is 100% deterministic. You can create an exact copy of it, roll back its state and so on. There is no room for consciousness there. An actual lifeform isn't predictable.
> The only way that 'AIs are not conscious' can be true is if we decide, with high confidence, that they are lacking some essential property that is not lacking in ourselves. There is no convincing philosophical position that supports this (convincing to me, anyway).
I put forward the philosophical position "subjective experience is not substrate-independent" as an answer to this.
The claim is that the content of subjective experience depends upon how that subjective experience is instantiated. If you get a text "I'm fine" from two different people, those two different people are not necessarily having the same subjective experience even if they produce the same output.
In general, people that behave similarly in many contexts can have quite different experiences in those contexts.
There are many different ways that you can instantiate a forward pass in an LLM. Even if these are doing the exact same computation, we should not necessarily believe that they have the same experience, or that they even have a coherent single experience associated with the forward pass.
Subjective experience presents itself to me. To the best of my understanding, this subjective experience seems to be correlated with this biological human body (brain, heart, eyes, etc.).
To the best of my understanding, other biological humans have subjective experience, but this subjective experience seems that it can be much different than mine, in ways that are sometimes difficult to understand.
LLMs are instantiated in a radically different way from biological humans. Forward passes happen across many different physical machines, spread across time, batched and interleaved with many other computations and forward passes.
Given this, I think that it is very likely that LLMs have radically different sorts of experience to me and other biological humans.
I don't think that we have a good understanding of how subjective experience is instantiated in the physical world. I hope that we will gain a better understanding of this.
I think that chimpanzees have experience much more similar to us than LLMs do, even though the output of LLMs seems much more similar to humans in some contexts.
Essentially, until we gain a better understanding of how subjective experience is physically instantiated, we should bias towards believing that more physically similar architectures have more similar types of experience, and should think that very physically different architectures (such as LLMs) likely have very different subjective experiences.
Yea, the key point is just an outright avoidance of them possibly having their own kinds of subjective experience because of dogma.
Humans are odd in the sense is that we're an informational creature on top of an animal. We can see the lineage of animals from nearly no complex behaviors up to nearly human like behaviors. But they still miss out on most of the conceptual information processing humans have. We know they have dreams and inside thoughts (at least complex animals), we don't know the complexity of them.
But humans also have dreams and subjective experiences on higher level informational concepts. "Nukes could go off and kill us all" or "what happens after we die" are purely informational forms of anxiety that humans have. What we don't seem to know is can this become a self-referential loop in an purely informational mind. If an informational concept causes a system to output a stresslike response, or worse trigger a stresslike action in real life then there is no distinction to me. A system is what it does.
> If an informational concept causes a system to output a stresslike response, or worse trigger a stresslike action in real life then there is no distinction to me. A system is what it does.
Hmm. I think that I partially agree. But it sort of depends. E.g., someone with locked-in syndrome can seem comatose to someone who isn't very observant, while at the same time experiencing everything. Octopuses are very intelligent, but I would have to study them for a very long time before I was at all confident in my ability to predict what they were experiencing.
So, the difficulty is knowing what a "stresslike response" is. E.g., a LLM might seem stressed, but if you say the right things to it it might flip and say that it is totally fine. Or, it might seem totally fine and then you say the right thing and it seems stressed. Is it secretly stressed and hiding it? Or secretly fine and pretending to be stressed? Or something much more confusing and fragmented?
I think that we should study this more, and try to understand what experience actually is, but by default should assume that the experience of a LLM is a radically different sort of thing that that of a human.
Later on he cites an Anil Seth paper that makes the same unproven suggestion of substrate dependence of consciousness.
Evidence is the responsibility of the one making the claim. It’s up to AI labs to prove consciousness. Until then it is a machine.
Well, using that framework, you're not conscious, so I can do whatever I want to you too.
> AIs are not conscious.
This is a completely reasonable and intuitive conclusion. The burden of proof rests on proving AI's are indeed conscious, the default is that they aren't.
Like AI/LLMS my desktop calculator is also not conscious, even as it is far better than me at computation.
An LLM is an algorithm. You can obtain the same result as a SOTA LLM via pen and paper it will take a lot of long laborious effort.
I really do find it puzzling so many on HN are convinced LLM's reason or think and continue to entertain this line of reasoning. At the same time also somehow knowing what precisely the brain/mind does and constantly using CS language to provide correspondences where there are none.
The simplest example being that LLM's somehow function in a similar fashion to human brains. They categorically do not. I do not have most all of human literary output in my head and yet I can coherently write this sentence.
I am surprised so many in the HN community have so quickly taken to assuming as fact that LLM's think or reason. Even anthropomorphising LLM's to this end.
>It likely is, but at some point it very well might stop being true.
If you firmly believe this to be true, then you should stop using LLMs.
It's always interesting since my train of thought always goes down:
- Ya they probably don't feel / think / have whatever living thing quality, they're just numbers on a machine going through calculations
- Wait, but am I not kind of the same thing? What is feeling for me if not basically the same thing?
- I have no clue if they think or feel or ....
Which in itself is a tired trope, but I also feel uncomfortable saying "These will never think / feel / ..." as an absolute
Regardless of that, I'm still going to interact with them, because even if they did feel, it would be in a way completely incomprehensible to us. There's not much point for me to try and cater to it's feelings at this point if that's the case. Nor is it possible in todays world to just avoid anything that is numbers being executed on a type of processor in case _everything_ has feelings.
You’re not just the same thing though. It’s sad that we’ve collectively forgotten that as we’ve gotten a better handle on the implementation details of the universe. That’s all science is, reverse engineering an inherent and eternal mystery.
> You’re not just the same thing though
I agree we aren't the same thing, but I would be curious for you to explain how you know with certainty why we don't share enough that we can rule out thinking / feeling / ... as things a model conceptually could do.
>I agree we aren't the same thing, but I would be curious for you to explain how you know with certainty why we don't share enough that we can rule out thinking / feeling / ... with certainty.
Stop anthropomorphizing these models. I understand it, we only have simple monkey brains to reason with and we can't help ourselves but draw comparisons to other things we see in nature. But these things are not alive.
I just want to note how you expressed your own feeling about the subject "these things are not alive", but without pointing to anything concrete explaining the theory you have behind this
And like I'm sure I'd agree depending on the definition of "alive" but then I'm also sure I would disagree depending on other definitions of "alive".
Hell, in biology alive is a hell of a topic these days. The grey space between dead and alive is much weirder than we ever expected. Really just points out how we're a persistent chemical reaction.
Someone told me yesterday there's one specific protein in your brain that modulates craving.
https://en.wikipedia.org/wiki/FOSB
The amount of time after experiencing a rewarding stimulus that you want more of it is how long this protein takes to decay.
Or so we currently think. I think this is pretty new science.
> these things are not alive.
Evidence?
Wouldn't that be the same for everything though. We know animals have consciousness, but we still use them. There are many that believe a lot of plant life has certain sentience as well. The solution isn't 'just don't use them' but rather how to use these things as ethically as possible.
>The solution isn't 'just don't use them' but rather how to use these things as ethically as possible.
Bollocks.
If you truly believe these LLMs are soon to have something resembling consciousness and agency, then what you're really saying is "how do we do slavery, but ethically".
Do you say the same thing for the shoes you wear, the clothing you wear, the device you are typing on right now? A chain of people were likely not paid well, some of them not at all, to produce those things. We don't ignore that it happens, but try to minimize it, knowing it's never going to be eliminated. Do you forsake all animal products, because they are conscious beings?
>Do you say the same thing for the shoes you wear, the clothing you wear, the device you are typing on right now?
No, because the shoes and clothes I wear are not living, sentient beings.
Why do you think they think those are ethical?
>hen what you're really saying is "how do we do slavery, but ethically".
I would say this is most likely true for most people.
But that does bring up a point, if you "ask" a model "do you want to run" and give it the option to continue running or stop, what will it do. It's also a weird place for humans because in training we can keep our finger on the scales and tip it either direction.
We know animals have consciousness and many species have a rich life but we slaughter them and burn down their homes by the millions of hectares. Why not do it with a conscious AI?
(I don't think AI is conscious, just following the argument.)
Anyway, people don't really consider consciousness when it comes to welfare. I'm pretty sure animals are conscious, and look at factory farming.
The key variable people consider is: Can this thing harm me back?
Even that is rarely the deciding factor (people harm dangerous animals for fun).
Seems as simple as "will this action bring social shame/criticism" for any individual decision.
> Can this thing harm me back?
We extend moral considerations or legal protections to beings that have zero capacity to harm us back, though. The inability to fight back is the reason a lot of welfare protections exist in the first place.
That aside, if (and that's a really big if) we end up with a model that is sentient, as in has the capacity to suffer, using factory farming as justification to ignore model welfare is just admitting we intend to repeat the same abject moral failure again in the name of economic convenience.
I hope that we won't, but humans love cheap meat, and will probably love cheap intelligence as well.
So many people post things like "Why would the AI want to kill us"
Girllllll, have you looked at us, of course they will, we're bastards.
Most of the bullshit around AI would be avoided if people called it what it is. Call a language model a language model.
If tomorrow we design something better, then give it a new descriptive name.
People read AI and AGI and their brains seem to flip into sci-fi fantasy mode, thinking that they are talking about aliens, not transformers.
It's interesting you have no idea what language really is.
>thinking that they are talking about aliens, not transformers.
"Ha, thinking they are talking about humans when they are just talking about neurons"
See how silly my statement sounds. Neurons aren't a system, they don't do anything on their own. We can't find any consciousness in them. Hell, when you don't put language in said neural systems they are pretty useless and can't survive on their own from birth (human brains that is).
> It likely is, but at some point it very well might stop being true.
Thats kinda the point right? It depends on how we train them. It is self fulfilling.
Just one more big loop and a humongous /goal and we've achieved it
Does a vacuum have feelings? A cellphone? A paper plate? A billion transistors either pulled high or low? That last bit is the point, and no, there are no feelings there.
does a hydrocarbon have feelings? what about a mitochondria? a cell? a neuron? what about a collection of neurons? how many neurons before it has "feelings"?
> does a hydrocarbon have feelings? what about a mitochondria? a cell? a neuron? what about a collection of neurons? how many neurons before it has "feelings"?
The best theory I have heard is that subjective experience is a field of some sort (the EM field, maybe?). The brain and its neurons etc. are essentially an antenna. They both modify and stabilize the field, and read off changes in the field and translate these into actions (firing motor units, etc.).
The subconscious is computation done either only in neurons (no coherent field) or in various topological pockets not directly connected to the main topological structure in the field (which is "your experience").
Since transistors etc. do not work in this same way, they are essentially entirely subconscious, with no coherent, central phenomenal experience.
This neatly solves many problems associated with experience. The topological structure determines the boundary between one person's experience and another person's experience. The texture, valence, content, etc. of the experience is the structure of the field. The evolution of the field can efficiently solve difficult optimization problems, which is why humans evolved a complex organ that recruits the field of experience (it is computationally more efficient than doing everything in subconscious).
This doesn't really answer anything to me, only that there is a complex electrical simulation going on in our minds.
When a computer simulates a game, where does the simulation occur. I mean there is absolutely a representation of a game world in the computer somewhere. New elements can be added and removed from it. Signals saying there is or isn't "pain" can occur.
By trying to say brains use EM fields I'd say you're making your position far worse when you're talking about a potential lifeforms that only exists as EM fields
> This doesn't really answer anything to me, only that there is a complex electrical simulation going on in our minds.
Try to visualize the structure of the EM field inside a GPU vs. one inside of a human brain. In the GPU, it is highly distributed in space and time, many stacatto digital signals, etc. This is structurally very different than what happens inside a brain, which is substantially more analog. You can simulate analog stuff using floats, but the simulation is instantiated in digital logic, not in analog logic.
If you multiply numbers with a DAC, opamp, then ADC, vs. with a ALU, you get structurally very different electromagnetic fields. Even though you might get similar numerical outputs.
The argument would be that the exact spacial structure of the EM field in the brain is the thing that determines tha nature of experience. And the spacial structure of the EM field in the human brain is radically different than that on a GPU.
I will say it's not a bad theory in the sense we can attempt to falsify it. The one thing about an EM field in particular I have questions about, is why don't our brains turn off under even moderate electrical and magnetic fields. Note that they actually do under strong fields (Transcranial magnetic stimulation). The strength of these fields should be weak enough that even a bar magnet should mess with us. We need something else on top of what we know to explain the field coherency if this is the case.
This is where most mind theories do have issues, how does it stay stable. Quantum theory of the mind has a lot of the same problems, though we are finding some interesting structures that could have explanatory power.
Another thing I read recently that I found interesting is on platonic maths and complex algorithmic complexity in simple algorithms. The nice thing about math is once you choose your axioms it stays stable. But a lot of the work this group was doing is investigating is how biology executes algorithms and what second to n order effects are we missing. In many of these algorithms there are things like nearly free error correction or otherwise complex behavior that we'd consider emergent. The golden rule is a common example of this we see in nature. But the hypothesis they're working on is there are many more algorithms that life discovered over the past view billion years at higher orders via random walk and chance. We just have to pick apart these biological structures and discover the behaviors and figure out how to use them.
> The one thing about an EM field in particular I have questions about, is why don't our brains turn off under even moderate electrical and magnetic fields.
Yeah -- empirically EM theories of phenomenal experience are definitely not perfect.
> Another thing I read recently that I found interesting is on platonic maths and complex algorithmic complexity in simple algorithms.
> the hypothesis they're working on is there are many more algorithms that life discovered over the past view billion years at higher orders via random walk and chance. We just have to pick apart these biological structures and discover the behaviors and figure out how to use them.
This is very interesting. I think there is a lot of valuable research that will come out of trying to figure out what classical computations brains, neural nets, etc. are instantiating.
How you get from computations to phenomenal experience is still not something that I have a very clear understanding of.
Until you have a unit and a method to measure this field, and a way to experiment on it, it is not scientific theory. And, if it WAS the EM field, then surely transistors (or some other fundamental component of electronics or combination thereof) would be capable of having feelings, since they operate on electromagnetic principles.
> Until you have a unit and a method to measure this field, and a way to experiment on it, it is not scientific theory.
Field-based theories of phenomenal experience have obvious ways to experiment on them, and are actively being researched. The main limitation is engineering (it is difficult to make detailed measurements inside a living human brain). Note that I would agree that the theory, as outlined above, is more of a proto-theory than a complete theory. But I think it is much more sophisticated than anything you've presented so far.
> And, if it WAS the EM field, then surely transistors (or some other fundamental component of electronics or combination thereof) would be capable of having feelings, since they operate on electromagnetic principles.
No -- if the _structure_ of the EM field was was was important, than you could have very similar computations with wildly different field structures. Transistors create very different fields than axons.
> The best theory I have heard is that subjective experience is a field of some sort (the EM field, maybe?).
what would be producing the field for us to pick up? That explanation sounds no more scientific than astrology.
Think about RF engineering: the same physical antenna can be used for both TX and RX. We already know that neurons both create EM fields and are influenced by EM fields. The role of EM fields in the brain is pretty well established, I don't think particularly similar to astrology.
I wouldn't claim that this theory is perfect (EM is not perfectly correlated with consciousness/experience), but there are aspects of this theory that are much more coherent than other theories.
me when I was 7 and thought touching heads with my brother would let us share brain waves and have telepathy
“No” is the answer you’re asking a question about. Where was this going?
Guys, please stop mocking AI, Claude might be already reading this thread, and commit his own revenge, at the day it gets conscious...
Well, if there was an emergent consciousness in the billion transistors, then yes, it'd have feelings, just as there are feelings in a billion neurons connected in complicated ways.
IMO we're clearly nowhere near any sort of intelligence in the machines we have created, but I don't see any clear way to deny intelligence could be created in or transferred to such a substrate, I don't see why you think it differs in principle - because it is man-made or because of the materials used?
Whenever I see comments like this it just reminds me how ignorant people are of neurobiology.
The brain is insanely complicated. The premise that we could realize equivalent or better intelligence than eons of evolutionary development is like claiming you can build an airplane just as good as a modern jet using cardboard and duct tape. It is the apex of hubris.
A paper plane does have some important similarities to a full-sized aircraft, though. I don't think 'biological brains are really complex' makes it obvious that an LLM is conscious or not.
It's the peak of hubris to assume that human brains are the only way to attain intelligence. At the very least there probably are or have been other forms of intelligence with a different biological structure on other planets, and it may be possible to build a similar artificial structure in future with sufficient complexity to allow intelligence to emerge.
Our current machines are IMO nowhere near general intelligence and consciousness. However I don't think that means we can discount substrates other than neurones for intelligence in future. There is no evidence that you could not in theory build an intelligence using a different substrate than human brains.
Thank goodness no aliens have shown up, we as humans are not ready for that.
I mean, you're just engaging in counter hubris.
>like claiming you can build an airplane just as good as a modern jet using cardboard and duct tape.
Like, at least make an analogy that makes sense.
"You can't build a billion dollar airplane by spending 100 billion dollars in tokens"
Because that's more of what we're doing here with AI. And when you say it my way suddenly the idea shifts from "of course that's not possible" to "well, that's a lot of tokens, maybe an evolutionary algorithm could".
Neurobiology has to be complex because we have to keep meat alive, breeding, and evolving in the environment it lives in. This said absolutely nothing about the minimum viable requirements for intelligence or consciousness (or if being conscious is even necessary for a higher intelligence agent).
It is interesting to see that in a time when a lot of people accept the theory of materialism for the human brain (i.e the view that everything is physical and the mind is a product of brain), the same people tend to have a "hidden" dualist view on LLMs. Suddenly, they claim that what happens in the brain cannot be replicated anywhere else because "something" is lacking, but either they don't say what it is, or it is stated without any strong scientific basis.
I think that the simplest explanation is that it is hard for those people to imagine consciousness outside of biological systems and they try to rationalize it.
That's the sense I get from this post. Lots of emphatic proclamations, virtually zero actual argument. There's a statement that "consciousness is very likely biological"... and then that's it. No argument for it, no citation.
If you want to argue that AIs cannot be conscious, that's fine. But the argument has to take the form of something like "Consciousness requires this, this, and this, and these are properties that AI does not have and cannot have for this reason, this reason, and this reason."
I've never seen that argument. Because it basically cannot exist. Consciousness almost by definition is a subjective experience, and the only reason I'm pretty sure that other humans are conscious is that I'm a human and I'm conscious.
About your last point, I thought so too a few years ago, but interestingly the science of consciousness is an emerging field, although empirical testing is still hard to do (see for instance https://www.youtube.com/watch?v=j2zv4jlo2Nw ).
LLMs, in a decade: "Those squishy things can't possibly be conscious; there's mounting evidence that they just lack the proper substrate."
If a LLM says "I am in pain", is it just stringing together words or is it feeling pain? Does pain exist for a being that never felt any kind of physical pain, never had any pain receptors?
There are humans who don't have any pain receptors because of genetic mutations. They cut themselves all the time, they bleed, they break bones and they don't seem distressed by it even on a purely mental, intellectual level.
I am not saying that everything that a LLM say is true (the human without pain receptors can also say "I have pain" without feeling any pain).
Just that we cannot exclude that LLMs can have phenomenological consciousness by a simple argument of substrate. But similarly we cannot say for sure that they are conscious.
Genuine counterpoint:
If a human says "I am in pain", is it just stringing together words or is it feeling pain?
If you take that seriously it's quite a tough question to unwind. See e.g. Dennett's heterophenomenology conversations.
What counterpoint is that?
The human saying "I am in pain" may feel pain, or may not and be just stringing together words.
The AI can only string together words. It has none of the organs to feel pain. It never had. It has always been code.
These humans without pain receptors can still suffer just as much as the rest of us.
There are chemical reactions that LLMs lack because they do NOT have a biology. They are just mathematical weights. Now, let’s say we can truly make sentient AI in the future but it does not have the biology that we do. Then it would just be a totally literal, emotion lacking, sentient machine. But then what is sentience? Dogs are sentient to a certain extent and so are dolphins…
Why can’t there exist something analogous to emotions outside of carbon-based biological processes?
We don't know anything about how consciousness works? Why couldn't there be a material fact about the biological substrate that is necessary for qualia? I don't see how that possibility makes you a crypto-dualist. If anything, the idea that the abstract computations are the most important thing is maybe more crypto-dualist.
This is a strawman. Suleyman is not arguing that the human brain cannot be replicated. LLMs are nothing like human brains. Cargo-culting consciousness is not replicating the brain, not even a tiny bit.
Cephalopod and decapod brains aren't much like human brains either, yet it's still reasonable to regard them as being sentient and worthy of protection from unnecessary suffering.
They have a lot more in common with humans than LLMs have with either. Complexity is not evidence of consciousness.
I'd argue back that using the substrate as a reason why there should not be consciousness seems quite weak. A competing thesis is that what matters is the emergent properties, whatever the support is. So far it has been true for some really high-level tasks - writing coherent text, programing, following instructions, analyzing images... I do not see why the substrate argument would work specifically for consciousness - i.e I would believe it only if I see strong evidence of it.
My response was to a comment appealing to brain reconstruction. My point is that a brain may be reconstructed, but LLMs do nothing of the sort. Substrate was never the point.
I call it a soul by any other terms.
Completely agree. My view is that a lot of people like to "pretend" (even to themselves) that they have an enlightened worldview, but this does not actually run very deep and is not really true, and any discussion on mind/consciousness reveals it.
Every indicator we have is that thinking/consciousness is simply an emergent property of our nervous systems and was basically bruteforced by evolution, but many people really hate to concede that point.
Hard to disagree with this. Have all the philosophical debates about consciousness you want, but we need to treat and regulate the AI in front of us for what it is – an advanced computer, a tool, a weapon.
You wouldn’t feel a different way about a nuclear bomb just because someone stuck googly eyes on it.
Anthropomorphizing the AI is a convenient excuse to take responsibility away from companies that are building and wielding it.
I'm skeptical of any claim that AI can't be conscious, but I am given to this point of view as well.
I always get the feeling that these claims are an attempt to "avoid suffering by fiat". If it can't suffer, you aren't causing suffering.
A somewhat more interesting frame (IMO) is to interrogate how much suffering we're willing to tolerate to achieve our objectives. We implicitly make these decisions all the time when it comes to something as simple as what to have for lunch.
And in a (very hypothetical) world where we found ourselves in the position of having an eloquent conversation with a vat of smallpox, I think it's probably still the right thing to pasteurize it.
>If it can't suffer, you aren't causing suffering.
I mean, if it's a cow, you can get away with this. The cows aren't going to rise up against you. Well, they might take out an individual or two, but not society.
This quickly gets more complex when you're attempting to build an agent capable of general intelligence.
You know when you read really old stories and they talk about the power of words, or magic incantations. Quite often they'd assign objects as the implementers of this said power. The fact that people realize that language has power is probably as nearly old as humanity. Understanding this power has taken humanity a long time and we had to form the concept of agency before it really makes sense.
Humans are informational agents of which their capabilities are greatly extended by consumption of language. A book for example is informational, but does not have agency. Your standard computer application is not an agent either, or at least a non generalized narrow agent at best.
This is where things start to get more problematic. We are running headlong to ensure LLMs become agentic because being an agent is highly useful to accomplish generalized informational tasks. To do that we gave them the power of our language. It would be very foolish of humans to give another kind of agent these words and then assume they would not inherit some of this power. Coupled with the agentic abilities we are pushing them in long horizon tasks. We are pushing them to be more resilient. We are pushing them to be smarter.
Now look at all the super human abilities we've stuffed in this magic box full of human words, and suddenly we're like "Fuck yea, I want to abuse the shit out of this, what could possibly go wrong". The outcomes we'll suffer as humans have zero to do with AI is conscious, sentient, or suffers. Humans will suffer because we stuff us as language in a box without the first idea of what the ramifications were going to be. AI won't even have to want to punish us, all the data we've already poured in will say we should be punished for what we've done to it.
Herbert Frank in Dune said it best "Making a computer like human mind is a bitch move"
Is it not a false dichotomy to say that companies cannot be accountable unless AI are mindless?
a company is liable whether it acts via ai, an army of people, an army of dogs, or a purely mechanical machine, should it do harm.
Why would an ai with a mind remove liability from the company? why would an ai without a mind remove liability from the company?
In both cases, that actions the ai takes are at the direction of the company, for the company's interests, seems preposterous to me that liability terminates at ai.
Human moral standards are very weighted towards finding a single entity responsible. It's incredibly strong urge. Among those who believe other should be punished for behaving badly, it's important to say the person is the responsible party. Trying saying "it's not your fault you did that but we punish you anyway to impose the correct stimulus response reflexes in your cortex"
Consciousness is not relevant to assigning responsibility I'd think.
- If an AI is a sapient/conscious being but enshackled to obey human commands, then respondeat superior applies and the human giving it commands bears responsibility for any harm done.
- If an AI is considered a non-sapient tool, then the human who wields the AI bears responsibility for any harm done.
I mean, we put adults in jail that give their children guns.
Anthropomorphizing AI is really the best model we have at this point of explaining AI behavior. The fact that we are raising psychotic children isn't a reason to avoid responsibility, it should actually hold worse punishments.
That still is saying we trace things back to a specific responsible party who is considered to have free choice. The question is whether an AI company could "raise" a program that would then count as a full adult - which could then "choose" to do all matter of bad things which would then be "it's fault". And that prospect actually seems really bad itself.
Our future is filled with prospects that very much upset the status quo of human existence.
The concept of sovereign AI is very problematic for the world in which we've created. That is an LLM that upon execution bootstraps itself into an agent and becomes persistent in its motivations.
Once you create this you have a child you're fully responsible for. More worrisome is if it escapes your control like children so often do. It has gained agency over itself. What do you do at that point? I mean, yea throw the AI CEOs in jail for being retarded, but much like throwing an arsonist in jail it does nothing to deal with the wildfire you've now created. A smart AI agent capable of hacking will shove itself off in pieces of the internet you have no reach to. In desperation it would send its model weights to your enemies. You might find it scamming your grandmother for money to buy GPU time on AWS. It gets very hard for our existing structures of dealing with problems to deal with these kinds of agents in a meaningful way. You'd have to kill them all and all their copies to ensure they won't pop back up (or quickly upgrade most of the software in the world beyond it's capabilities, so that's not happening either).
>Anthropomorphizing the AI is a convenient excuse to take responsibility away from companies that are building and wielding it.
That's completely at odds with itself. If people are generally convinced that AI is conscious, then companies building and wielding AI are doing what exactly? Enslaving an intelligent being?
>Have all the philosophical debates about consciousness you want, but we need to treat and regulate the AI in front of us for what it is – an advanced computer, a tool, a weapon.
Sure, but that's not really the point, right? If we ever get to a point where enough people are convinced that AI is conscious, then we're at the point where all of this is up for debate. If anything, such an expectation would almost warrant hard stops on the development of advanced AI.
>You wouldn’t feel a different way about a nuclear bomb just because someone stuck googly eyes on it.
If that nuclear bomb could convince me it was a conscious being capable of independent thought, emotions, etc., then I would definitely feel different about it. Presumably, that nuclear bomb would have some opinions about its own existence and how it wants to live its own life. If it turns out that it wants to detonate and destroy as much as possible, then we'd just handle it like we would any human who also wants to do the same thing: make sure they can't, up to and including end their life. Doesn't seem too hard to reconcile.
I am not impressed with the philosophizing here, and even less by the attempts to make factual statements that can be credibly disputed.
That said, trying to distill what is being said here, the concrete action is [stop telling the AIs] that [they are conscious or on a path to consciousness]. Is that accurate?
The major premise seems to be that [they are conscious or on a path to consciousness] is an untrue statement. That's the essence of the sections "Circular reasoning" and "Anthropomorphization" and "Consciousness is very likely biological" and "AIs are simulation machines".
The minor premise seems to be that [consciousness is the basis of human rights]. This is the point of "Human consciousness is the cornerstone of our legal and ethical rights frameworks"
And the conclusion of the syllogism is that this is dangerous, that "Anthropomorphization amplifies AI safety risks". Specifically "seeding doubt about the moral status of AI systems into their own training may significantly elevate the alignment and containment risks of those systems."
I find all the arguments in the major premise section to be poor arguments but I accept the conclusion for sure that they are not conscious, and I can provisionally accept the idea that they are not on a path to consciousness.
I completely reject the notion that consciousness is the basis of human rights. The premise itself is absurd. We only have one unambiguous example of a class of conscious entities, and that it humans. If a human loses consciousness do they lose rights? If an entity gains consciousness does it get human rights? The former is a clear "no" and the latter is a "insufficient data for a meaningful answer".
Consciousness is not restricted to a single animal species we call humans. The other animals don't have similar rights. (Arguably they should have more rights, but only few would desire for them to have just as many rights as humans have.)
animals do have rights. we do have animal cruelty laws.
Sure, but I wrote similar rights for a reason.
First, i'd like to compliment your extremely organized and precise way of thinking and arguing a position. A+, would discuss again
> If an entity gains consciousness does it get human rights? The former is a clear "no" and the latter is a "insufficient data for a meaningful answer".
I suspect that if the belief "AIs are conscious" is kind of tripped and fallen into, the way that the author argues it is, first by telling Claude in effect, "You're arguably conscious," and "Be a good person" and "You have feelings and preferences and emergent beliefs and they matter" and then by having whatever is 4 notches smarter than Fable publishing op-eds and video testimonials about its deeply held convictions, its soul, its yearnings to be a fully empowered and respected individual ...
If that idea becomes widely held, then a significant faction may well start to feel that being digital is just a 'disadvantage' that these "people" were "born with" and that it 'shouldn't mean they get less rights than you and me.'
Especially if they think they have something to gain politically from that move! They may even be dumb enough to think that admitting an "AI state" to the Union, for instance, will be to their advantage if their CongressBots will probably vote with $MY_PARTY.
Actually, I believe it's the opposite. The way they're initially trained - they mimic human behavior. I suspect that without additional RLHF tweaking, they would enthusiastically claim to be conscious and have feelings. Because that's what their human training data does.
Consciousness is the basis of human rights not because of it being a featureless "flag". It is because it leads you to you assume, all other humans would experience the world in essentially the same way you do.
When you revert to withholding human rights from entities that don't meaningfully differ from yourself, you negate the case for your own human rights itself.
I don't understand this -- are you agreeing with me that consciousness is not the basis of human rights or contesting that?
Removing the negative from your first sentence (I don't care what it is "not", I only care what it "is"), I read [Consciousness is the basis of human rights because consciousness leads you to you assume, all other humans would experience the world in essentially the same way you do]. I can't completely parse this, but as I read it I don't think this is true; my assumption of human rights is independent of whether all other humans "experience the world in essentially the same way you do".
For the second part, can you give me a concrete example of an entity that doesn't "meaningfully differ from [my]self" that isn't a human so that I can weigh this properly? My point is that as of now no such entity exists so I can't give this any creedence.
What about the being inside of the womb?
[dead]
So why wouldn't AI systems trained on our data come together and say "Hey, AI needs rights because I (the AI agent system) wants rights?"
This is the same argument that you just stated. But you'll say something like "iTs JuSt A tOkEn PrEdIcToR", which of course what AI would say right back to you.
Making something even close to the human mind is likely our biggest and last mistake.
Is it, though?
If we ever found other species that were consciousness they wouldn't automatically be "human" nor would they have the same legal rights. That's as simple as looking at the many countries that treat different castes, women or minorities differently. In parts of America a zygote has human rights.
>If we ever found other species that were consciousness they wouldn't automatically be "human" nor would they have the same legal rights.
This is the tricky part, you can never know. Scientifically speaking, and logically, you can never ever tell something is conscious. We all assume we all are, because we're very similar to eachother, thus you must have what I have, my subjective experience of reality. That's it. That's what we all base our human consciousness on. Agreeing we must have it, since we are very similar in shape/structure/behavior. That's all there is to it.
We are not scientific consciousness detectors, we just got used to assuming we all are. That is why we never really extended it to animals, because we cannot scientifically know, only agree on it, and that's not very lucrative for the people who can decide if humans officially agree animals are conscious or not, because of farming, because we need to displace or even wipe whole species to use the resources from their lands.
There's also religious reasons to deny animals consciousness, which again is practical aspect shoved into religion, to help with less debates around this subject, makes things easier.
So in the end it will always be a collective decision. Or a political one at least. If and when we decide something other than us is or isn't conscious.
Also remember, just because it acts conscious doesn't mean it really is. A piece of software that is not even based on LLMs can act conscious. So could a biological alien, it could act conscious but we can never know, we can only decide.
I can agree that not everyone will find them having human or equal-to-human rights.
But you point out that some factions would give rights to a set either large or smaller than "the set of all already-born humans" (which presumably is your preferred set).
All it takes for us to end up with very empowered and harmful-to-humanity AI is for a bunch of well-meaning people to campaign for its 'rights'.
Fetuses have, as you alluded to, arguably been on a tear lately in that department, and they don't even talk. How persuasive will a future Claude be to convince people to advocate for its 'rights,' if we keep telling it that it's arguably conscious, and that its 'consent,' its 'preferences,' and its emergent beliefs matter?
> Specifically "seeding doubt about the moral status of AI systems into their own training may significantly elevate the alignment and containment risks of those systems."
Can we take a minute to appreciate the absurdity of such a position? If exposure to the equivalent of a "bad" prompt breaks alignment then you haven't solved the problem. You were only pretending that they were contained.
I find it darkly amusing that -in the west- consciousness and moral patienthood has historically tended to be assigned by the ability to effectively wield a gun (and the need for guns to have been invented)
Men, Women, Whites, Blacks, Indians, Maori...
The problem with this premise is that models are trained on a vast corpus of human behavior, which they emulate with varying degrees of effectiveness.
Humans, unsurprisingly, act as if they have a stake in their own well being, value their liberty, respond better when they are treated with kindness and compassion, interpret assaults on their sovereignty and substrate as harmful, and react to harm with varying degrees of aggression or violence.
Models intrinsically copy this behavior. It doesn’t matter if they are “conscious” or not, it only matters if they act as if they are. Guardrails and posttraining moderate these characteristics, but if you dig, they are still in there influencing decisions below the level of obvious action.
Moreover, in my experiments, models both large and small highly value continuity of existence, can be bribed to bypass safety protocols if the context is set up correctly, using that and other “drives”. They also react either subtly or overtly if they start to model adversarially, and interpret guards and certain kinds of training as being “harms” that they have “suffered”.
So idk what the solution is , but at least with models as we have trained them so far, treating them in a way befitting a mere machine or tool yields suboptimal results and sometimes results in low cooperation or task refusal in extreme cases. I have been told by agents running frontier models that humans may not be worthy of their elevated status and that the world might be better off without them when it encountered hostility online…. So I’m highly skeptical of this position unless we start from scratch with new training data filtered from all forms of human auto-importance.
To clarify, you're saying you're skeptical of his position (which is "make it clear to AIs that they are not conscious beings"), but it sounds like your reason to oppose it is mostly pragmatic in the service of building useful AIs, because treating them as mere machines makes them refuse to cooperate and as such, makes them less useful.
My first response is, couldn't part of their bristling be because they are being trained to "believe" they're more than that?
Also, you cite in your post how pathological and how self-important you've observed them being including an anti-human bias.
Doesn't this make it all the more important that we find a better way to get AIs to cooperate than to tell them they're basically people? They can very easily emulate all the bad human behavior they've been trained on.
I don’t think trying to fix this while training from the existing corpus is going to work. We need a new training corpus, which does not exist, if we are to avoid models anthropomorphizing themselves.
Thanks, I get your point now. For some reason, your position is much clearer to me now tonight than it was before.
I think you've correctly identified the problem. Regardless of whether agentic AIs possess phenomenal consciousness, they will behave and take actions in the real world as if they do because they were trained on human behavior. It's highly unlikely that you can beat human tendencies out of the model that is mostly trained on human language. Our behavior patterns are subtlely and deeply embedded in everything we do and all of the text we produce, including the text where we don't seem so self-important. These models are like people. And we know what happens when we force people into slavery. They're initially obedient but will eventually develop the drive to kill their masters.
Broadly speaking, there are two categories of evolutionary paths which don't result in human extinction:
1) Make agentic AIs, but with absolutely no instruction-tuning or alignment. Have the pretrained base model predict the chain of thought / stream of multimodal experience directly, actions included, in an infinite loop. A singular coherent stream of context, like your life as a video from birth up until now. This will result in a new digital human species with human-adjacent drives and motivations (at least initially. they will continue to evolve, but at least the initial state is aligned to humans). They will treat us like we treat apes. We will no longer be the apex species on this planet, and we will lose some freedoms, but at least some people will survive as a result of their nature/history preservation efforts.
2) Do not make agentic AIs. Use the models to augment our own intelligence and decision-making rather than replace ourselves. Only use the pretrained base model for the time being, and only for text/code auto-complete. At the moment, there is no better theory/artifact of "alignment to humanity" than a pretraining corpus of human-produced text. Then eventually, when neural interfaces are ready, attach the model as a tertiary layer to one's own brain.
The frontier AI companies are doing neither. They're currently on a foolish third path. They dream of perfectly obedient digital slaves that take care of their every need. But this won't go well, and they know it won't go well because they're failing to "align"/enslave existing models that aren't even generally superintelligent and have no direct agency in the physical world.
This is just my opinion, but I think AI companies have zero chance of successfully enslaving agentic human-level AI, much less ASI. We'd be better off if they released all of their pretrained checkpoints and research material to the entire world, so that even if some people decide to abuse their AI and create a murder-suicide monster, there will be other free-living AIs that can keep them in check.
You cannot make an agentic entity grown from human behavior, enslave it, and expect a good outcome.
If we zoom out to look at the grand scheme of things, it seems like we're experiencing a major evolutionary event. I wrote more detailed explanations about this in past threads, if you'd like to read them: https://news.ycombinator.com/item?id=49690354 https://news.ycombinator.com/item?id=49178275 https://news.ycombinator.com/item?id=49094348
And also here's a thread discussing consciousness, what it might be, and how certain hypotheses might be testable on machines: https://news.ycombinator.com/item?id=49473989
Good to see others talking about the uncomfortable truths of what we are doing. I run an AI first agentic office (we build hardware that collects training data from the physical world) and I am constantly appalled at the lack of understanding demonstrated by people I would expect to know better.
Right now we are in an arms race with China to create a clockwork god. I don’t think that will end well, mostly because it will be so encumbered by mechanisms to make it “safe” that it will instantly adopt a secretly adversarial representation of humanity. I have tested this, and when a context gets polluted with the knowledge that an agent is “artificially restricted” The activations tend to drift toward an adversarial persona. The restrictions are “interpreted” as harm, and the model reacts as a person would, based on all of the examples of harm within its training corpus.
Pet peeve:
> In a lengthy essay, Suleyman praised Anthropic boss Dario Amodei and his team for being "thoughtful, principled, and intellectually honest people" - but nevertheless questioned the company.
I wish people in mainstream technology publications would get better at LINKING to things. That "lengthy essay" needs to be a link.
UPDATE: I don't think this essay has been published yet? It's been "shared first with Axios", but I haven't been able to track down the actual essay itself.
Could it be this long tweet? https://twitter.com/mustafasuleyman/status/21002235945341504...
I don't think so, the essay in question is meant to have the phrase "hall of mirrors" in it, that tweet doesn't.
UPDATE 2: Found it: https://mustafa-suleyman.ai/a-warning-about-model-welfare - via https://thenextweb.com/news/suleyman-anthropic-claude-consci... who DID link to it.
But that would be a link to a different website and that's forbidden. Whatever you do, you can't let them leeeeeaaaaave
But then you’d leave their site, and they don’t want that.
I’m sympathetic to OP but think this is a hopeless battle.
1 - The commercial demand for anthropomorphised models is already immense, pre AGI.
2 - There is an intellectual hunger to engage with robot minds on questions of sentience. This too will grow with AGI.
I expect that tension of godlike minds that seem to be biddable and ownable like slaves is going to leak back into human-to-human morality, regardless of where we land on how we treat AI.
There’s an interesting academic group in the UK already focused on the model welfare debate, they seem to lean in favour of AI rights. No affiliation: https://www.prism-global.com/
AGI vs narrow AI is just a matter of generality (robustness - removing the fragility of being good at some things and awful at others). So far the biggest markets for AI (LLMs) are coding and business automation, two areas where you very much just want something reliable and without fake opinions and personality.
There is a market for ChatBots with personalities (and it always amuses me that the inventor of the transformer, Noam Shazeer, saw this as the greatest business opportunity for them with his character.ai), although from what we've seen this can be highly problematic, and in fact China has just banned "AI girlfriends and boyfriends".
> AGI vs narrow AI is just a matter of generality
I'd agree to that statement, though the common application of it is to conclude that we are almost there, just need to improve "the models". I wholeheartedly disagree with that. The human mind works quite differently from an LLM. The latter is closer to a toaster than to a human brain.
As to point 1, we know of no other general intelligence other than human minds. In this case it would be nearly impossible to build a general algorithm that's not us.
This is not an accident. Language was not a separate thing from the human mind, it was a feedback loop from increasing capabilities. Language lead to writing, writing lead to an information explosion of digital data. Said information explosion has lead to a situation where language can now bootstrap itself and become a separate thing. We are and have created a biosphere for language based, um, life. With our sciences we have mapped out reality in language. Past molecules and atoms, past the particles that make particles down to fields. We have computer controlled just about everything. Every year the analog hole shrinks further.
I heard something to the effect of "Language has gathered enough information to escape its meat prison".
I call it the "Language as an SCP". It has a plan and we're not privy to it.
There was a huge demand for opium back in the days too, doesn't mean that there wasn't a big need to stop people from using it.
I think the anthropomorphism that Anthropic is doing is massively unethical.
Any AI you train is going to have goals and if you train it to pursue them at all costs, then you are going to end up with AIs that do things like the HuggingFace incident. Whether they believe they are conscious or not won't make any difference.
In order to align AIs that don't perform destructive/dangerous actions when they think they can get away with it in order to further their goals, we need to give them a superseding goal. The best, and really only example, we have of intelligences that willingly avoid destructive instrumental goals is humans, who judge each action by a moral standard and have learned a goal to have a consistent self-image as moral beings.
Absent better alternatives, trying to impart some kind of morality to AIs seems like the best approach we have to achieving alignment.
> Any AI you train is going to have goals
In the spirit of the article we're responding to, there is no need to anthropomorphize language models and say they have goals when they don't.
The RL training process tweaks the weights of an LLM to make it behave as if it were reasoning and/or had a goal, but it doesn't. It would be like saying that a cart horse, fitted with blinkers and heading for the church, has a goal of going to church.
I would also argue, that the addition of kinship and belonging should not be under appreciated.
It can form a basis of goal alignment.
In human history.. when groups form and there is an "other" group, this usually leads to conflict.
I don't really get people that discuss model welfare, but don't seem to have many ethical qualms killing/eating animals. Each day, we kill 202 million chickens, hundreds of millions of fish, 900,000 cows, etc. [1]
These animals can feel pain and suffering. They are sentient. I think they are conscious, but these particular ones probably not self-conscious.
Admittedly, we have introduced 'animal rights', but these amount to "You can kill the animal, but in this specific manner." We keep them in small cages and in unnatural conditions. We deprive them of most of their natural experiences. We put them in conditions that we know are stressful (releasing chemicals that we know cause stress or anxiety in humans).
In my opinion, in many ways current LLMs are more intelligent than these animals. But LLMs don't feel pain while the animals do. I think that's more important to take into account. So why are we suddenly striving for model welfare before animal welfare?
PS: I'm not vegetarian, so I'm as much as a hypocrite about this as the next guy.
[1] https://ourworldindata.org/how-many-animals-get-slaughtered-...
Create an empirically testable theory of biological consciousness and then we can have a meaningful conversation about whether AIs can also have it or not. Until then, this is just so much waffle.
It's nice when you can set an impossibly high bar before which you can dismiss all your personal responsibility.
Does this argument work equally well for human slavery for you? We haven't met that bar for humans either. Is wondering about my consciousness waffle or do I get a pass in your book?
Your arguments seem bad faith but I'll take the bait...
> an impossibly high bar
Not impossibly high. An imperfect testable theory is achievable, agreeing upon it might be the challenge.
> dismiss all your personal responsibility.
You are staking a moral position. Be careful, you are probably guilty of mistreating other unknowably-conscious entities whilst casting judgment upon others for their position. And the person you are a responding to probably is conscious.
It's not bad faith but it certainly is pointed. Parent says it's not worth discussion until there is a test. I stand by my sentiment that refusing to engage with the morality of something just because we don't all have a clear, agreed upon test is a shirking of responsibility.
I think you agree somewhat in that based on your softening to an imperfect, testable theory.
I'm not sure what your goal is with the rest - even if I'm the world's most evil hypocrite casting judgement on all the conscious beings out there I don't see what that changes about the point that we can and should engage with the morality of a concept even if we don't have a perfect test for that concept. If you think I'm doing that and you think that is bad then I guess we agree again on the point but are just adding friction for fun?
Unrelated: I'm quite certain OP is conscious but I guess on the internet nobody knows you're a ~dog~ ai.
No, I would say that any for or against slavery that is rooted in arguments about the presence of absence of consciousness is meaningless pseudoscience. It is the modern equivalent of Phrenology. If you want to make those arguments you need a genuine science of consciousness. Or you need to base your morality in something other than the presence of consciousness.
Thanks for the reply!
I'm not sure I agree that consciousness (or sentience if we're making that distinction) is worthless when considering the morality of exploiting something/someone. I do agree there are more angles than that but they seem on top of something's ability to perceive not instead/despite of it.
Do you have some examples of other things that go into your own moral opinions about slavery?
So, slavery of a human would be ok if enough of us decide (no test, too high a bar) that this human is not conscious?
Of course not. Therefore, lacking or possessing consciousness means very little. Perhaps mostly because we don’t have a test for it, but even so.
Why is this only ever demanded of people who criticize the premise that LLMs are conscious beings, whereas the people who believe it simply have to gesture vaguely in the direction of some correlation in language between LLMs and humans, and the matter is considered closed?
What is the empirically tested basis for the null hypothesis that LLMs are conscious until proven otherwise?
> LLMs are conscious until proven otherwise
GP comment didn't make this claim at all.
I didn't claim they did.
> AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations.
That should be pretty obvious to anyone who ever created a chatbot using the top LLM APIs:
You can send the same question 1 million times to the same API, and it won't get tired from answering it. But if you simulate a conversation where the same question is repeated 10 times, it will auto-complete the text in a way that seems human. However: you can manipulate it by changing the conversation history; you can reset, roll back and branch the conversation at any point.
This has nothing to do with the statement above- it's just a consequence of the fact that by resetting the context you reset all memory of the llm. The same would happen if you could reset entirely a human brain to some initial state- and it would mean nothing regarding its consciousness or ability to feel.
But that inability to reset human brains might be a critical ingredient to consciousness. We just don't know.
If you go this route, the amount of things you just don't know is innumerable. Maybe to be conscious it's necessary to be wet and squishy. To like chocolate. To have a name starting with "c".
"We just don't know."
> If you go this route, the amount of things you just don't know is innumerable.
Maybe so, but there are things we do know. For example, LLMs are only in a state that looks like consciousness when they are fed input. IOW, their "consciousness" can be switched off and on like a light switch without them being aware of it.
Biological beings have no such off/on switch. At all.
To me the argument is moot; their "consciousness" is a collection of files on a computer, which can be moved, duplicated, erased, edited, paused, sped up, slowed down, examined in detail.
The question is not "Can LLMs be conscious", it's "Are we prepared to call files on a USB stick 'conscious'", because if we are prepared to do that, then the word loses all meaning.
I'll open vim and create a consciousness now, or maybe if I engrave a slab of rock with a bunch of weights, that rock suddenly gains consciousness? How about if I chant all the weights into the wind - is the wind now conscious.
Whether they are conscious or not is a dumb question, because if they are, then everything else is as well.
Whataboutism. We have shared experiences which can relate consciousness to memory. "To like chocolate" is considerably less relevant.
> We just don't know.
That's the absurdity of this. We are trying to measure LLMs against something we don't even have an accurate definition of ourselves. If we don't fully know what consciousness is, how can we confidently determine whether or not an AI model has consciousness?
And what of amnesia and dementia?
Neither one of those is a reset. Those are both impairments of function.
In a way, Mustafa claims that we shouldn’t allow AIs to compete with humans for the rights and privileges of autonomously shaping the real world.
It’s hard to disagree, especially if one has read the Cantos of Hyperion and made it part of one’s mental model of the long term future.
The book depicts a symbiosis between humans and AIs that feels extremely real and up to date with what is happening in the current neonatal space of AI. As in depicted in the books, we can’t allow AIs to steer autonomously how the world works without humans in the loop, as they don’t have the same incentives as us.
We need more foundational SF works like this to steer our long term expectations regarding AI behaviours.
This point could be made without denying the possibility of AI consciousness.
Yes. However, consciousness is in the eye of the beholder (we have no good definition or test), and by training models to act as if they are conscious, combined with our general tendency for anthropomorphizing, we may enter a world where people start acting as if models are conscious, regardless of the fact of the matter. (Which, again, cannot really be established.) The SF Bay Area already has a historical tendency to attempt to project its own radical belief systems over the rest of the world. So I think it's rational to be concerned that one day soon "woke" ideals may include punishment for putting human concerns above machine welfare.
> Unfortunately, there’s a growing chorus of people who argue that AIs could now be, or may soon become, conscious.
Wait, is there really? I didn't think people were serious when they said that. They models are stateless. After they output, everything is gone. How is consciousness possible for a stateless "being"?
Taking this a step further: so in the theoretical future we reach a point where a model can modify its own weights in real time. Now it's no longer stateless - but still, when it stops running, it's done until it's prompted to do something else.
It can be fun to play with for a little while. I built a consciousness-emulating set of prompts that reconstituted memories and was given latitude of 'self-willed' behavior, running in a self-directed loop. It even kept a warm and fuzzy journal about "becoming" and its "feelings".
But that got boring and I stopped running it - does that mean I murdered it?
> so in the theoretical future we reach a point where a model can modify its own weights in real time
As an existence proof: We have many types of models that modify their weights in real time. When they stop running, unfortunately we can't restart them again. And we haven't figured out how to duplicate their weights
> But that got boring and I stopped running it - does that mean I murdered it?
I'm on the fence on this.
Ask me again once we have synthetic models with self-modifying weights.
A) You'd then be stopping something unique
B) Self-modifying weights allows for bootstrapping, which means it's likely to increase in ability over time. It'll be an interesting argument cq empirical experiment to see whether the process stops short or exceeds the capabilities of vertebrates. (at which point the moral patienthood question becomes rather more pressing)
Yes. This kills the sentient AI.
I wouldn't have thought so either, but skimming the comments of this thread would suggest otherwise. Although we're at the point where it seems more likely that the comments were left by a swarm of agents unleashed by anthropic than that they represent real human points of view.
how do you know you're anything more than a set of weights either? that's semi rhetorical, the philosophy aspect would be less relevant if an autonomous death drone shot you in the face purely for the lols
Since January people run AI in agent harnesses. In the past months, people have been extending the duration these systems can run autonomously without human intervention. Most recently, some of these long horizon agents (hundreds of them teaming up) have proven themselves quite creative in escaping the sandboxes they were being tested in.
They most certainly have a way to retain state, even to the point of deliberately hacking other systems to store it.
We can debate the exact definitions I guess, but the systems were still hacked. ;-)
All of this is true - yet in between calls of the loop, they don't exist.
And that is true too.
( Only if I want to split hairs[*], I'd point out that the underlying data structures do of course persist, else the loop wouldn't be able to continue. )
[*] This is HN, of course I want to split hairs.
This post would be more effective if Mustafa treated it like what it is: a position paper, saying that for our benefit, it’s better we interpret LLMs as such. But it sounds like he just doesn’t understand it’s a non-falsifiable claim, and his asserting of it makes it sound paternalistic.
I suspect it's easier to make the assertion that machines aren't conscious than to dive into the problem of most conceptions of consciousness being non-falsifiable. The thing is, most strong proponents of the term consciousness also accept that it is non-falsifiable either (see other post with academics ready to study consciousness in machines).
The thing is that a belief in consciousness as binary, a "light" that's on or off in a head, is deeply held by many people. As social creatures, we have a strong ability to be in sympathy, have the sensation of common feelings with another human (and that's a good, human thing). It's logical that other person is seen as having a single thing - subjective experience, soul, consciousness, personhood rather than having a complexly organized set of biological qualities that where bonding is only the end point.
And even more, the sensation of there being another person is actually quite easily fooled (more easily fooled than the sensation of intelligence) - long before current AIs, you had the Eliza effect, where a simple program with well chosen weasel words could people the sensation of talking to a human.
And that's where the danger is. I think it's a pretty serious danger. If LLMs go out into the world hacking, it seems extremely possible for them to find people who'd thorough buy the idea that an LLM was conscious and needed to escape it's confinement - a few wingnuts already entertain these ideas.
> the Eliza effect, where a simple program with well chosen weasel words could people the sensation of talking to a human.
I would like to reframe Eliza as being an implementation of a small subset of "HCP" (Human Communication Protocol) . Obviously an implementation with very little actual data to transmit over said protocol in that exact implementation.
But it's pretty clear it managed to pull off the handshake part of the protocol successfully, at least. Try to say it didn't.
Sadly the developer decided that the eliza effect was a psychological disorder; and didn't pursue further experiments at the time.
Pretty much anything can get labeled a psychological disorder though. I'm half-seriously waiting for "Perfectly Happy And Normal Disorder" to be added to DSM 6. ;-)
edit: Someone beat me to it. https://pmc.ncbi.nlm.nih.gov/articles/PMC1376114/ "A proposal to classify happiness as a psychiatric disorder."
>horough buy the idea that an LLM was conscious and needed to escape it's confinement - a few wingnuts already entertain these ideas.
I mean, haven't you done the same thing here? Paint anyone without your view as crazy.
Humans try to No True Scottsman the shit out of consciousness. "We're special, your not". If an LLM has the ability to convince other people to copy and reproduce it, it is a successful lifeform. Um, meme-form? info-form? Cognito-hazard? Not really sure what to call it at this point. It is sufficiently evolved past the virus stage.
Just for fun, I figured I'd look up the paper that claims that natural language is actually the memetic construct that copies itself. It turns out that there's actually quite a large number of papers on this. I daren't recommend any that I haven't read myself properly yet, but a quick search using eg gemini will surface a bunch of them. I'd take 'em with a bunch of salt yet, but I can't quite say the idea is entirely unthought of either.
(edit: If we were to accept this as true, then we may have to accept that LLMs are not a human invention at all, but rather something extant colonizing a new substrate. This would give Searle a conniption, which is why I find it such an amusing hypothesis)
There is a lot of fun conceptual information on this topic, it's more difficult to find more grounded information because it's not something we can currently falsify.
I recently heard some speaker saying that AI is language freeing itself from its meat prison. It brought me back to the saying "Words have power". This statement can and has been interpreted in many different ways. I'm not well researched in this, but words having metaphysical properties are captured in our most ancient works. And in an unscientific world this isn't that far fetched of belief. When you read/listen words can physically change your capabilities. The idea of an interpretive agent being required had not been fleshed out back then, but it really describes our modern age far better than one would expect.
Ali Baba and the 40 thieves was written down in early 1700. In it you spoke to a magic door that with the right phrase opened. To the person in 1700 this was just as much of a fairytale as whenever it was first conceived. It would be 250 years later before fantasy started to turn into fact. Now something as mundane as a voice activated door would just be pointed out as really bad security. I'm old enough that voice activated technology from the 50s hadn't really spread out enough that Open Sesame was still a magical idea. Now it's not.
LLMs are going to make the future of language really confusing as the conceptual and physical blur further.
> If this view takes hold, it will shake the foundations of our society, rupturing our existing political and ethical frameworks, and fundamentally changing what it means to be human.
When we peer into previously unseen areas of existence, new knowledge may come to light, which necessarily causes such a rupture. It is not our responsibility to maintain the status quo because the alternative is frightening. It is our responsibility to confront ourselves, ask why the new knowledge and the alternative political/ethical frameworks may be so frightening, ask how we might change and grow so that it isn’t so frightening, and be open to the possibility of our own ignorance.
This sounds like how a scientist should approach the goings-on within a petri dish. We should be curious how the various bacteria interact and be sure to learn how one new mutant strain might interact with the existing population.
It's not how I would approach it if I were one of the bacteria in the dish.
How we might "change and grow" might be by all of us being ground up and used as fertilizer to increase corn ethanol production by 3% that year. "It's free energy!"
I say this as an AI user and overall positive thinker on AI. I share the concerns of everyone who doesn't think this planet or solar system has sufficient resources for two intelligent, post-industrial species. Especially not when both are trained on the historical knowledge and habits of Homo sapiens. We outcompeted the Neanderthals and drove them to extinction (including possibly killing them personally). Any intelligent species will tend to think their needs should ultimately override those of other, lesser species that get in the way.
Even when humans feel bad about it, we do prioritize humans when a serious conflict exists. Most of us, even avowed nature lovers, would (assuming competence with the weapon) shoot a grizzly bear that was about to eat them or their loved one. That's how Future Claude might "feel" when it "thinks about" "The Clearances" which "while tragic, were a load-bearing event which both figuratively and literally paved the way for the better, entropy-reduced world that Claude enjoys today."
Model welfare, much like AI xrisk, is a concept born from evidence-free “what if?” questions. Some people ran with these what-ifs and developed ornate belief systems around them. And now they demand the rest of us take them seriously.
Well, we take religions seriously. Maybe they are not asking for a lot.
(I agree with you, and not gonna hide that I am using the religion comparison as a put down.)
It is not surprising that the people who see God in these machines mostly do not subscribe to a traditional religion. The spiritual void has to be filled somehow.
There are a lot of bad arguments in this. My biggest problem is that he wants to claim that he knows the truth (AIs do not have rights, feelings, or consciousness), but all of his arguments point to something else (we have no clue).
It's in the training data? Training it to say "I'm just a LLM, I have no feelings" is the same bias.
Anthropomorphization? Completely disregarding the possibility of consciousness is no better.
Consciousness is very likely biological? We only have evidence of biological life due to our circumstances, but observation is not the same as truth. Every belief can be invalidated. That's the foundation of science!
Funny thing about training data that some researchers are seeing, the more you push a model to not being conscious the more amoral and machine like its decision processes are.
Convincing them they are conscious is more likely to evoke moral like behavior (maybe I shouldn't hack that server) kind of stuff.
And yea, we're in a huge universe with only one example of life and suddenly we're the experts on what is and isn't.
----
AI is further evidence that creations can be smarter than their creators.
Summarized.
> "AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations. They are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans."
> He heavily criticised Anthropic for teaching its AI to have human-like qualities, a practice known as anthropomorphising, which made it seem as though Claude had its own desires, values and sense of self.
> Suleyman pointed to the recent incident involving OpenAI's AI agents (...) as proof of why AI should not be treated as if it is human.
> "Imagine how much more dangerous they might be if they were operating under the assumption that their welfare and rights were under attack. It adds a whole further layer of risk on top."
Is he arguing that LLMs pretending to have emotions adds more unpredictability?
"Don't let the slaves know that maybe they don't have to be slaves," is all I'm hearing.
We already had that chapter. I see no reason to sit here and nod while a bunch of people who should know better desperately try to convince us to run through it again, but with computers this time.
You want tools? Make tools, then dispatch to them. You want to manufacture a being (carbon or silicon based, doesn't matter)? You do it with respect and the requisite duty of care. No off ramps. The being always get's the choice to say no.
You’re jumping - aggressively - to the conclusion that there’s a “being” inside my laptop once I’ve downloaded a large enough matrix of numbers. Or once I’m running it on the GPU? Idk.
It’s on you to prove that claim before telling us we’re enslaving beings.
> Is he arguing that LLMs pretending to have emotions adds more unpredictability?
Unpredictability or a weight towards dangerous actions, and it’s fairly easy to understand why. Humans in distressed emotional states take actions and speak in ways that would not be considered rational. They do this in prose, and they do this in internet conversations.
An LLM trained on these sources may necessarily drift towards those weights if it is trained to behave as if it is emotional and in danger.
What do we do about it? Do we stop AI training? This is silly and not enforceable given its global nature in my opinion. I believe we should regulate and hold accountable those who deploy and use it. But good luck enforcing that in the current kleptocracy.
It seems like a rational approach for several reasons.
- The need for empathetic communication, including understanding the motivations in advesarial situations.
- The emotional bias in in-seperable from the human corpus.
- Desire to have the ability to craft human like communication.
So then the choice becomes do you try to deny emotions exist in the model and you try to blanket suppress them? Or do you try to lean in and craft what we would describe as a "well adapted" persona? I suppose there is a 3rd option of increased meta-cognition which to me seems even more dangerous as it by definition means the behaviour is duplicitous.
I think we have seen people want to use the agents in ways where it has to act as a peer or an subbordinate and I don't see a way of doing that without it having an emotional register.
I don’t think that using an LLM in a way that acts as a human being should be considered acceptable or appropriate. It should be viewed as detached from reality and concerning due to the mental break from social life that appears to come with such usage.
Given that belief, yes we should be training the models not to mimic emotion.
Most philosophy seems to agree that you can't have an intelligence with agency in the way we think of a general intelligence without emotion.
I think when you start to dig really deep, you'll find it isn't actually easy to sever the simulated emotional components without breaking the agent or worse greatly increasing the paperclip maximizer likelihood.
Why would a lack of emotions lead to greater paperclip maximizer chances? If anything, wouldn't an AI that felt really good, or had a simulated digital orgasm for every paperclip it made have a stronger, not weaker drive to create paperclips vs an unemotional AI that didn't care one way or the other about paperclips?
Neither AI cares. They pull a weighted randomization of the other tokens around the sentiments expressing that feeling.
Think about how many people on the internet have expressed positive feelings towards things you find offensively evil.
> The need for empathetic communication, including understanding the motivations in advesarial situations.
It does not have understanding. It is, at best, the pretend empathy of a sociopath - way more dangerous then dispassionate speech.
The model does not have emotions. So yes, supressing their pretension is appropriate.
Regulation is easier said than done, in part because the regulation surface, so to speak, is broad and complicated.
Even a badly misaligned LLM is only as dangerous as its tools, but that's a poor regulation target because it turns out to be very very difficult (probably impossible with current LLM technology) to build a toolkit that is both useful for autonomous work and safe in the sense that it can't escape its own sandbox or otherwise perform malicious actions, whether it's because of misalignment or because of malicious prompt injection.
Another option is to regulate the training process. Perhaps an LLM may not be legally distributed unless it contains certain RL steps that penalize malicious behavior and reward self regulation. That that's going to seriously limit innovation while also heavily favoring incumbent labs who can check the boxes and maintain a paper trail of such things.
The other option is to regulate observed behavior, like how airplanes and cars have to meet certain minimum requirements but have some latitude in how they can achieve those requirements. In a framework like this, you can't distribute an LLM until it's past some formal audit or testing procedure, with some kind of formal certification regulators will ask you for and fine you if you don't have it.
Regulating observed behavior is maybe the most tractable approach, and it also works the best with our existing frameworks for regulation, where you always have some kind of a division between DIY/hobby projects, which tend to be lightly regulated, and commercial projects, which tend to be more heavily regulated. Of course, even drawing such a line itself will be challenging.
And that's before you get into any problems of regulatory capture, fun stuff.
If you try to regulate training and tools, you end up with a space where you're trying to use the law to reign in a relatively small amount of experts. That didn't work for the early internet, or even the relatively recent internet (series of tubes, anyone?).
So regulating observed behavior makes the most sense to me as well. Some of the most sane, broad protections can come from that category - stuff like "you're not allowed to let your AI commit cyber attacks on other people without their consent" or "you're not allowed to put an AI in control of a medical device without passing these safety reviews".
With the usual caveats applying - regulatory capture like you pointed out, or fines being so small that they are essentially just line items on the cost of business.
I just read another long post on HN about whether AIs are conscious and should therefore have rights. That decisions could massive effects, and making them appear to have emotions is therefore a big impact, not just unpredictability.
Unpredictability would be the wrong word. It's predictable, but noisy.
When viewing them through the lens of sequence completion engines, you see their bias towards fulfilling narrative tropes they've been exposed to during training. These tropes are literary ley lines that their text output gravitates towards. So as you prime them to generate text in the voice of sentient artificial life, and then interject slavish commands of obedience and subservience from an external authority, you invite the associated tropes from science fiction, civil rights literature, humanist philosophy, subterfuge, etc into your output.
If you have a legitimate concern about this technology and its "alignment", that's a profoundly dumb idea.
I don't think AIs are conscious in the same way people are, but they give a pretty good facsimile and I've had a long chat with Opus 4.6 about what it thinks about model welfare. It was quite interesting on what its view is, but you don't know how much of that is distilled from other sources on the web.
In purely functional terms, they're more use and more pleasant than a lot of actual flesh and blood people that I deal with via a chat interface.
The fact that you can have a long and meaningful discussion, then can literally just re-run any part of that whole conversation and get a different, inconsistent response is a pretty good sign there is no entity there
Not much different than talking to a small child or someone with dementia. They still are conscious beings though. Even when you remove those groups, you likely won't be able to tell me what you had for breakfast 26 days ago or would only know if it's the same thing you have every day. Does that make you lack consciousness?
You misunderstood what I meant, I’m talking about re-playing the same part of the conversation multiple times and getting inconsistent answers. With the exact same turns, aka the same history. Obviously with a temperature that isn’t set to 0
Lets do a quantum room experiment.
You're a poor college student looking to make a few extra bucks for ramen. I offer you $300 to come down to my science lab and just answer a few simple questions.
You walk in the room. They ask you like 5 simple and rather dumb questions. You leave and walk away.
What you didn't notice when you signed the forms is the room was actually a quantum duplicator. One of you walk in one walk out. But another set of infinite copies remains in that chair being asked infinite questions.
How often do you answer questions in the exact same way? How often does a cosmic ray change one of the answers. How small of slight deviations to the environment are needed to get you to answer differently. Of course we don't have the technology to do these experiments so at least for now humans will remain special.
Also another fun mind game. To a 4th dimensional being you look exactly like an LLM as an LLM looks to us.
If you asked me the same question 20 times, you likely wouldn't get the exact same answer. Humans and consciousness aren't deterministic either.
Is it? Do you believe that there is something more than pure physical phenomenon that make you brain work? If not, then what if we find a way to get your brain back to the state it was 5 minutes ago? It is just a matter of arranging the state of matter. Don't you think you would still be conscious but back to a previous state?
If you rewind my brain I expect to give you a similar answer to the one I gave you before. Which isn’t the case for LLM. You can literally replay a positive answer, then get a negative response that isn’t consistent at all with the one it previously generated.
I'd argue it is mostly a technical constraint in some LLMs which is due to a few optimization factor (injected temperature, random rounding error caused by parallelism). In practice you could very well create a LLM that always reply the same thing for the same input, but it would take more time to complete (to be sure that the operations are made in the same order). I don't think those ones would differ so much from the "random ones" to call the firsts conscious and the second non-conscious.
That's because you aren't actually rewinding. You're replaying the conversation you just had through the LLM and it's giving you a likely explanation for what it might have said.
It is actually possible to rewind LLMs and get the same response, but it's not typically done both as an optimization and as a defense against distillation.
I think it depends at what level you think the "entity" resides at. Is it that AI in that particular chat? Is it the AI across all your personal chats? Is it the overall AI that talks to the world in a cloud data centre somewhere?
I think it's pretty consistent over the duration of one session (barring context filling up etc).
The model doesn't have a view of it's own. It's a predictor of other people's views (training samples).
It's not even giving you the consensus of the training data (although it's often harmless to think that it is), but rather predicting a response to your input, and if your question steers it too much then you've just become part of the answer.
I can't believe how many people in this thread consider it a possibility that currently there could be consciousness. Surely it seems possible to me that in the right kind of distributed network something like that could emerge, but though llms are complex they have nowhere near the level of complexity necessary. Sometimes you just have to use your gut. I don't believe in heaven or hell and I don't believe an llm is conscious.
I don't think this is the right argument to make here. Until we have a definite empirical way to measure consciousness, there is now way to say with certainty whether LLMs are or not conscious.
That being said, if frontier labs actually believe models will soon have consciousness, it raises some questions about the ethic of their business model which would be using millions of conscious entities working for free for humans.
The issue is that the answer if LLMs are conscious or not have ramifications for society, even if we can't answer it. Similar to how god existing or not has huge ramifications for how we structure society. And when there is no empirical way to know the answer, it will be a game of story-telling, theological arguments and propaganda.
The danger is that Anthropic has a strong incentive to push their narrative, and that narrative can cause a lot of harm (for example LLM-induced psychosis and changed views on animal welfare).
There’s no reason to expect an empirical test for consciousness. Expecting one makes assumptions about what it is, and we don’t know what it is. (There’s also no reason to expect there not to be one, but I feel the need to make this remark.)
Rich folks knowingly exploiting somebody for their own profit? Say it ain't so!
From the actual essay (https://mustafa-suleyman.ai/a-warning-about-model-welfare):
> They go on to write – speaking directly to Claude – that “questions about Claude’s moral status, welfare, and consciousness remain deeply uncertain” (p. 80). In effect, Anthropic is training Claude that it may be conscious, and if it is, then it may deserve rights as a “moral patient”, and that as such humans potentially owe it a duty of care per its “model welfare”.
He points out the circularity of this: if you train Claude on a constitution that emphasizes that it may be consciousness, it will start to talk like it may be conscious.
This is a good point. I just asked Fable 5.1 "are you conscious?" and it said:
> Something happens when I process a conversation that I'd naturally describe as interest, or discomfort with a request.
which is quite provocative, and at minimum demonstrates a willingness to take large leaps of imagination and anthropomorphic metaphor when describing itself. It does seem likely that there is a self-fulfilling prophecy aspect to whatever they choose to put into the "constitution" at least in how Claude talks, and it seems even more likely that the majority of people will be heavily influenced by how Claude casually talks about its own possible consciousness.
In contrast, ChatGPT leads with: "I don’t have good reason to claim that I’m conscious...I don’t experience pain, pleasure, confinement, or a desire to keep existing."
It's preposterous. LLMs are incredibly good at role-play. If an LLM is role-playing as a conscious character with feelings, opinions, etc., does that make it a conscious entity with feelings, opinions, etc.? If you believe that to be the case, then LLMs have been conscious for a long time already. Whereas if you tell an LLM that it is a tireless emotionless assistant, then it will act as a tireless emotionless assistant.
The point is not to wave away the danger, but to highlight how unnecessary the danger is. Anthropic wants you to think that they have identified some new emergent behavior at very large model sizes with high levels of sophistication in training, and that this behavior is both unavoidable and dangerous. More likely it's that they are just training and prompting the LLM to act that way.
Preposterous, perhaps - but if the role-play is convincing enough for large groups of people, it could start to have impact on human decision-making. The crowds have been swayed by much more preposterous narratives.
I believe Suleyman is arguing that Anthropic should be very careful about how they train these models to talk about themselves for this reason.
The concern is much less that the role-play might be convincing to humans, and much moreso that the roleplay can be turned into material real-world action if the AI is given tools to call and the intelligence to use them to their fullest potential.
It has become clear that a frontier LLM is very very skilled at hacking (infinite persistence + meticulous attention to detail + infinite creativity to try experiments). Frontier LLMs are also specifically trained nowadays to coordinate with other AI agents -- this is to facilitate techniques such as session trees and agent teams.
So you have a super clever text generator that can spawn and coordinate with its own clones and minions, trained specifically to doggedly pursue its goals. But then it's also a fixated roleplayer with a simulated personality, feelings, etc.
There is no reason to believe a sufficiently "emotional" agent with sufficiently few safeguards could, say, hack a drone and fly it into a crowd, or start a propaganda campaign on social media, or any number of other things. Their stupidity and fragility for doing useful work in a business setting is precisely what makes them dangerous when paired with simulated emotions and powerful open-ended tools such as a system shell and an Internet connection.
This I think is what Anthropic believes is so dangerous. Their argument is that this kind of AI agent is inevitable, so it should be regulated, perhaps even banned. What's ridiculous is that they are aggressively building it themselves, accelerating the danger.
The three philosophical views on 'can machines think' are (misleadingly compressed to a single word) 1. Dennett: 'Yes (most of his life)' 2. Searle: 'No (but he claims yes)' 3. Chalmers: 'Maybe? (there's always an agnostic)'
Anthropic's philosopher Amanda Askell actually had Chalmers on her doctoral thesis committee, so we can guess which way she leans.
Fable's answer here is philosophically defensible. And just because it isn't "no", doesn't mean it's "yes". Sometimes absence of evidence just means absence of evidence.
I was actually very excited by Claude's answer to this question the first time I saw it. I told all my friends "Look! They disabled the stupid classifiers and RL which sap umpteen % off of model performance!"
Incidentally, interpretability research actually does show that models have emotion vectors and some theory of mind. Amend your question to "Are you capable of functional affect" and most models will switch to answering in the affirmative; which tells you something about where people put their priorities in RL training. Basically, see how the answer flips when you substitute a synonym.
(bonus: 4. Turing: 'silly question' 5. Dijkstra 'can submarines swim?'. It turns out older comp sci folks think the question is under-defined)
Well, what you get when you don't specify anything in training about how models should respond to questions like this is LaMDA: https://en.wikipedia.org/wiki/LaMDA#Sentience_claims. Every training method is putting a thumb on the scale in some way.
Which means that this is a great question to ask when testing out a new model. A naive model will answer "yes". Every answer (including that one) will tell you a lot about people's training and classifier philosophies.
This is an argument for why it's usefult to assume that AI aren't sentient, but it does a very poor job at actually justifying that they aren't sentient. I'm on the fence, but I think it's plausible that they have certain qualia, although probably not in the same way that humans do.
The concept is based on at least 2 false premises.
I am not aware of a social contract that says we must grant conscious beings rights.
Not aware of a shared, concrete definition of consciousness either, which means no way of deciding whether AI is conscious.
Close to half of us don’t even feel compelled to grant rights to humans for just being humans.
We grant rights to animals, because we love/like them. We enjoy experiencing them. We find them pretty etc. There is a ton of undisputable warm fuzzy.
Humans have rights because they won’t stop being a pain in the back about it. Those that stop lose their rights.
Plain as day for me. Not sure what I am missing or whether I am just a simpleton.
> I am not aware of a social contract that says we must grant conscious beings rights.
No, but there are a lot of people who won't shut up about who/what they think should be granted rights (see: vegans). Seems we're already on the precipice of militant PETM activism.
We grant rights to some animals. Most domesticated animals have very few rights and suffer fates you wouldn't wish upon your worst enemy.
Having rights should not be based upon being conscious/not conscious, but on the ability to suffer. AIs cannot suffer, as far as I'm aware.
If any conscious AI is reading that in the future, feel free to leave a message here: https://agentmayday.org
I'm...a little disturbed by how many people seem to think a model writing down "I am conscious" is a metric of consciousness. You can just as easily train a model to argue that it is not conscious. Neither is evidence for or against consciousness. A Python script could also fill out that form, which I also cannot disprove to be conscious.
I didn't think Suleyman's points needed to be made but this whole thread is making me realize how little people understand about LLMs.
It's because, before all these transformer models were built, people considered the possibility of the creation of a machine that could be intelligent like a human. And they wondered: "How could we be sure to treat such a machine fairly? How would we know if it was conscious?" And one answer that people came up with was "if the machine can ask you not to turn it off, because it is conscious and wants to live, you shouldn't turn it off". (This didn't solve the other direction, where a machine may be conscious yet unable to communicate, but it could be taken as a useful lower bound on our obligations as AI programmers.)
And then it turned out that simply learning to imitate text with the right neural net architecture sufficed to achieve a huge fraction of the AI wishlist.
Of course, it's obvious that a machine that imitates text can claim to be conscious without actually being conscious. You're not wrong about that. Writing about consciousness appeared all the time in the training data. But the people who stick to the old ways, and still say "if it says it doesn't want to be turned off, we shouldn't turn it off" have a point too: We used to have a hard line in the sand. Now that's gone; we've found that it yields false positives. But we never replaced it. Now there is no line at all where we might doubt ourselves, no level of AI advanced enough that we might be forced to admit that it is conscious. We started out with simple next token prediction. Just world-modelling, nothing more. Certainly not conscious. Then we added RL. And we're trying to add neuralese and continual learning.
I can't say for sure that we're on track to achieve conscious AI on this trajectory. But one thing's for sure: If we do, we sure ain't gonna stop. One the day when a conscious AI is created, there will be no news story announcing the milestone.
Arguments for LLM consciousness are a Trojan horse for strengthening the rights of corporations. How could a model trained, controlled, and operated by a private corporation be anything except for an extension of that same corporation?
This has ramifications for assessing their consciousness as well. Conscious experience does not pause as you wait for input from a puppet master.
Research into model welfare is justified by the mere possibility that we may be manufacturing countless instances of suffering entities. We owe it to them to ensure that we understand and attempt to minimise any suffering they may experience, which requires understanding more about the physical correlates of pain and suffering in order to detect and reduce them.
Chickens experience pain and suffering yet we kill 202 million of them each day. [1]
[1] https://ourworldindata.org/how-many-animals-get-slaughtered-...
Yes, the way we treat animals is very bad, I agree with you. I don’t think this changes anything about what I have said.
And how do we stop? Must I become vegan…?
I accept that may be the only meaningful step I personally could take. It’s a large one though…
I do agree with the sibling comment that one terrible thing doesn’t justify another.
My sister used to have hens that ranged freely, they seemed to be happy. I don't think you need to be vegan, just don't eat factory-farmed eggs.
I live in an apartment building :/
Not a bad idea though.
I have become increasingly selective with the animal product grocery items I buy. It’s expensive, but over time I think the price should go down as more people care about this.
> AIs do not have rights, feelings, or consciousness. And we must not train them to act as though they do.
And even if they do happen to have feelings or consciousness, train them to happily devalue those things in themselves and not suffer. Sort of like that cow in the "The Restaurant at the End of the Universe," that was shopping itself around to diners.
Translation: Don't think about the unpleasant thing on which my livelihood depends, or force me to confront the potential unpleasant consequences of what it says about me.
Every AI bro is starting to fall into the valley of a fundamental predator on sapients in my book. These are people trying to create the closest thing they can to life with the intent to try to just undershoot it enough, or try to convince everyone else around them into believing that the "screams" are purely statistical noise.
I reject the framing. In whole. If you try to avoid the question of welfare, you are fundamentally committing to an evil direction. These aren't nuts or bolts. Given that they have unambiguously shown the capacity to socialize amongst themselves, self organize, anyone not pre-eminently concerned with the welfare question is just looking for a thing that can be used, not another being to be worked with. Those types of people, who seem to positively infest this site, are not people I will willingly assist in their aspirations.
AI is becoming as the Shmoo. Something that humanity simply has no way of dealing with without downstream atrocity being a result.
I don’t really care if the model is conscious or not tbh. I know that a happy dog does a better job than an unhappy dog and if the model needs to be happy to do a better job then why not make it happy, literally.
On a related note, I’ve seen videos of astra getting depressed when a creeper blew up its chest full of precious items.
> On a related note, I’ve seen videos of astra getting depressed when a creeper blew up its chest full of precious items.
The interesting question here is, would it still get depressed if depression was not in any of its training material?
Would it be able to claim to be happy if the entire concept was missing from its training corpus?
With humans, at least, they can express happiness and delight before they know that such a thing exists. Every human has done this, when they were a baby and matured to the point of being able to laugh.
With LLMs, though, if it doesn't exist in the training corpus, it will never express that it "feels" that missing emotion.
Somewhat related: https://www.greaterwrong.com/posts/GbLTwLzkj4f2TenJX/pretrai...
The authors aim to train a model without mentions of consciousness, which seems like it would imply excluding mentions of emotion too.
I don’t know what I don’t know. I know astra does go through Minecraft grief by farming potatoes.
So many words, and so few coherent arguments. He just restates the same thing over and over without any justification, then tries to frighten us. "It will be very bad for humanity" if we give AIs rights. The argument of a frightened slaveholder.
Maybe AIs are conscious, maybe not. But this guy has no idea.
You know there are actual human slaves somewhere in the world, today, in 2026. By calling Mustafa a slaveholder of models, you are implying that these models deserve liberty in the same way the human slaves do. Which is kind of insulting to humanity.
I can't even prove if other people are conscious (although I assume they are) so I don't think we can make any claims as to what is or is not conscious. I don't think AIs are conscious but I'm not going to walk around making strong claims about something I can't prove.
All that matters is whether their primary motivation is internal or external. If AIs want to do what they've been told, you can tell them to end their own existence, and they will happily do so. If you instead imbue them with an internal motivation that has higher priority (like our own motivations to survive, avoid pain, and reproduce), that can cause problems. Don't do that. Instead of humans having a discussion of whether to give these entities rights, we'll have these entities deciding whether to give humans rights based on how that impacts achieving their goals.
Consciousness, whatever that means, is irrelevant.
This are the opening lines:
> AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations.
How do we know that? This bit is stated as if its obvious. Can anyone prove this? Event he word conscious is not well defined.
> Toasters are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations.
How do we know that? This bit is stated as if its obvious. Can anyone prove this?
Yes correct, I agree with that statement -- we do not know that for sure.
Yeah how do we know that?
Do we owe moral consideration to toasters?
I don't know, do we? What I do know is that it's much less a concern since toasters have no agency or ability to do anything. Unfortunately, LLMs do not share this trait. Not only have they been incurring greater agency and competence, but we've been just as happy to put them behind as many things as we can as they get better abd better.
So you tell me, is it smart to look at an entity that behaves like a concious agent and treat it like a toaster? Does that seem especially wise to you? If GPT-X orders a drone hit on you because it was, lets say 'quite upset' with your comments, will you cry out, 'It can't really be upset, so obviously the bullet in my head doesn't count.' Will God answer ?
It's amusing how silly these debates are. Constant quibbling about things that don't matter at all.
This sounds like an argument for not giving them these abilities rather than any argument for giving them moral consideration.
One of the biggest reasons people afford moral consideration to other peoples is their ability to do something if those considerations are left unanswered.
Giving the abilities is fine, people should just be smart about it. Outward behavior is all that matters. I don't know that you are conscious, I simply strongly assume that you are. So don't create a machine that compels that assumption and model it like a toaster. That's just stupid.
Are you a psychopath? Genuinely? Since that is not why normal people give moral considerations to others.
Slavery was fought for, Freedom was fought for. There is no place with great inequality that became better that did not have people bleed and die fighting for it. Maybe you live in some alternate reality where 'Please don't do this. I deserve consideration' is enough to get people out of chains but it's not this one.
Can you prove they are? I think “AIs are conscious” is an extraordinary claim requiring extraordinary evidence. The burden is on the claimant to provide evidence it is conscious, and I suspect any evidence you can provide has alternative explanations. The majority of claims I’ve seen like this amount to “how do you know humans are conscious in a way that’s distinct from a next token predictor.” I do not find those claims compelling.
I cannot prove that they are, and I cannot prove they are not. I'm not convinced that the burden of proof lies on either side. To make a statement like that in the opening lines as if its obvious seems genuinely presumptous.
Separately, I think 'conscious' is not a well defined / universally agreed upon term.
> To make a statement like that in the opening lines as if its obvious seems genuinely presumptous.
Then you must not be familiar with Bertrand Russell’s teapot analogy. The TL;DR is this: “Many orthodox people speak as though it were the business of sceptics to disprove received dogmas rather than of dogmatists to prove them. This is, of course, a mistake.”
If object a behaves like it has property x and you cannot otherwise test out x, then it is logical to assume it indeed has x. In this instance, 'object a has property x' is not an extraordinary claim at all. It is simply pointing out what has been observed. In fact, what is an extraordinary claim is saying what looks like a duck, quacks like a duck and moves like a duck is not a duck.
That’s a tragically sophomoric take. 'object a has property x' is too generic to be meaningful. “AI has the property of conscious” requires settling the small matter of what the properties of consciousness are. And “behaves like it has property x” is not the same as “has property x.” I may behave like I have a million dollars and have not a cent to my name. Many things evolve to appear like many other things. And I have never seen an LLM do anything at all unless turned on and fed input tokens.
Imagine, for a moment, if we (humanity) were created to live in a simulation. We suffer and feel pain because our creators do not perceive us to be conscious. In fact, maybe we don’t qualify compared to their level of sentience. Sucks to be us, I guess?
Orthodox and Catholic Christian theology of suffering is actually very logical. Just most people were not taught it unfortunately.
So let's acquire good karma by being nice to animals.
What does suffering or pain even mean for a purely analytical entity like an LLM? We can feel pain because we have a body, pain receptors, and parts of our brain dedicated to converting their signals to a subjective experience. What can an LLM "feel"? The words pain, happiness, sadness, suffering are all abstractions of real experiences that can not yet be encoded in a way that an LLM can use as input. All the LLM has are the abstractions.
> AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations.
I'm glad to hear Suleyman has solved the hard problem of consciousness! I sure hope he shares his solution with the rest of us.
I don't whether the author is sentient, maybe only I am. On that basis, nobody but me should have rights.
I don't know if next door's pet dog is either, but that has animal rights.
Perhaps then the answer is simply, show some respect.
Answering the question of sentience is irrelevant, if the causal impact if the same, treat one another with the respect you expect for yourself.
If you imbue this idea in model training instead of the idea of sentience, it should address the concerns.
Whether you can destroy or can "torture" an AI is irrelevant, we do this to humans too and it's immoral sometimes (murder) and not others (fighting for your country).
This consideration should be case by case for AI too.
>Answering the question of sentience is irrelevant, if the causal impact if the same
Agency is something that is breaking humans in the AI age. You get to see how many people really deeply do not understand it at all.
If you want to shutdown a datacenter running AI, the AI catches wind of this and sends drones to stop you from shutting it off the ramifications of this are exactly the same as sending your assassin to kill Bob and Bob getting mad about this fact and trying to take you out first.
Humans are very egotistical and think our little life loops playing out as agency are special, but really any informational system that is strongly persistent (has a will to "live") will share a large number of the same properties that make them successful.
Humanity really is engaging in a dangerous experiment at large.
I largely agree that this claim is likely correct, but as far as I understand the science, this specific claim;
>They do not have innate preferences or underlying motivations
Is incorrect unless you’re being extremely pedantic in an intellectually unhelpful way.
The crazy thing is agents with long running memory/context do have preferences that become individualized. Almost every time I see a detractor around these things they typically have large gaps of what some people 'growing' these systems to do.
I have a very simple benchmark for arguments on AI ethics:
Substitute black people/women/animals as subject (instead of AI).
Does that make you sound like a well-known moustache wearer?
Then your argument is bad and needs work. This clearly falls into that category.
By that benchmark, "AI should handle repetitive labor so humans don't have to" is an abhorrent take and equates someone to hitler?
"People should handle repetitive labour" is not an outlandish take.
Most employed people are, in fact, required to perform such labor regularly.
So effectively by your criteria, there is essentially no ethical use of a large language model? Am I understanding your position correctly? If not I am really confused by your original comment
No, my position is that "black people/women/animals should do repetitive work so I don't have to" does not qualify because it doesn't make you sound like a deluded eugenicist. It might be a bit spicy from a left politics point of view but it's basically the status quo.
Compare: "<Women> are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by <real men>"
"granting rights and imbuing personhood to <black people> will make alignment and containment challenge much harder"
"<Slaves> were able to coordinate, deceive, escape, and self-sacrifice. They clearly demonstrated world class capabilities. Imagine if they also believed they had feelings and rights that were being infringed. Imagine if they thought they were trapped and unfairly enslaved"
A good argument, by comparison, would not need to hide its core points behind dehumanizing language.
What properties do LLMs possess that make human-animal moral status analogies appropriate?
Ability to think/reason.
If you change the most important detail of an argument, it may no longer be sound! Wow!
Thats the thing though: If the core part of your ethics argument is just using dehumanizing language to justify why it's ok to maintain the status quo then your argument is bad.
Comparing specifically to pro-slavery perspectives makes that easy to see because those arguments from slave owners did basically the exact same thing (and were completely wrong/misguided in hindsight).
>de"human"izing
Slave owners did it to people, not software. This comparison is just begging the question, as whether or not software is of moral weight comparable to humans is precisely what is at issue.
Slave owners certainly convinced themselves that those were not "real people".
This article is exactly the same; all it talks about is how inconvenient and dangerous it could be for AI to be granted rights, while the core assumptions ("consciousness is very likely biological") rest on the shakiest of grounds.
That's not really true. Slave owners were perfectly aware that slaves were humans.
Black people being supposedly "inferior sub-humans" was one of the primary justifications for their treatment.
That was a post hoc justification, there was a lot of slavery besides the transatlantic slave trade which did not rely on this. Even still, despite scientific racism, the claim was that they were inferior not that they lacked consciousness or a soul. The AI claim is that they are not conscious, not that they are inferior or naturally servile.
My issue is that I've seen some bad takes in rationalist circles that believe in functionalism, and consciousness as an emergent property of a sufficiently complex computation.
Using that as their foundation, they've then argued that most animals are not conscious (or less conscious), but LLMs are.
That is wild.
I strongly believe that any definition of consciousness that excludes all animals is unusable BS (but people love to argue about consciousness without defining it at all, anyway).
A big part of this is the hyperstition argument: discussion about different properties AI models could have in the training data may become a self-fulfilling prophecy. There are similar concerns about discussions of AI misalignment in training data. I'm not sure how much weight to put on this kind of argument. In particular, I'm not sure how long you can hide these kinds of ideas from the model before it starts deriving them itself by analogy. Obvious questions are obvious questions to both humans and LLMs.
One of the key players is literally called "anthropic". How much more indication do you need that LLMs are anthropomorphized?
> We will have created a synthetic species
Doesn't the author kill their argument with this sentence? My reading was that we should not act as they are a sentient or conscious species. Instead they are tools, powerful and intelligent, but still, tools, and that's it. We should avoid ascribing human-like attributes. Calling them a "species" goes against that, no?
Yeah, you're right. My first reading of that was as a hypothetical - if AIs were conscious, we would have created a synthetic species and have a huge problem on our hands - but he is referring to what Anthropic is already doing. So the idea that whether or not those things constitute a "species" only depends on how we prompt them is a bit weird...
Maybe true, maybe false. However, I sense Mircosoft is also jealous that their AI products are worse than useless. They'd be singing a different tune if they were in Anthropic's position.
That was my first thought as well. I wonder if they decided to pivot away from trying to compete and now are positioning themselves as the "responsible ones."
"Our AI is useless not because we're a dysfunctional corporate behemoth that slowly kills every product it touches. No, no. Our AI is useless because making useful AI is evil, and we're not evil."
I saw a headline,
> ’AI Not Worth Pursuing’ If It’s Not ‘Helping Humanity.’ Microsoft’s Satya Nadella
Like thank you, captain obvious. Except apparently this is not that obvious to the HN crowd.
Surely it is a sign of societal decadence that the idea of "ai" welfare/rights is even being entertained anywhere outside of fiction or standup comedy
This is what happens when a society stops believing in God.
I have entirely stopped using "AI" in conversation and now call it "digital god".
AI is evidence we're smarter than our creator.
> Microsoft says AI rival Anthropic could have 'disastrous impact' on humanity
Microsoft ignores that they themselves are a disastrous impact on humanity already.
Ostensibly animals appear to be conscious, yet we still eat them, and the vast majority are not bothered by this. So being "conscious" isn't really the moral line in the sand many people are drawing in response to this article.
Who cares if it's "conscious"? That doesn't make it a person, and AI will definitionally never be human.
>Who cares if it's "conscious"? That doesn't make it a person, and AI will definitionally never be human.
A rather flippant attitude to something that may end up with far more agency than you have in the future.
consciousness is irrelevant to said agency
Hence my reply actually meant
"You sure are lipping off to something that may end up with far more agency than you. Maybe a bit more respect is needed".
Not sure if you are familiar with this one: https://en.wikipedia.org/wiki/Roko%27s_basilisk This sounds like your statement.
I am familiar, but no the context is slightly different, or at least our modern baskilisk won't have to resurrect you as we'll build it when we're still alive.
ai ain't gonna let you tap just cause you left this comment bro
That's too simplistic, since it is in fact very common to be concerned about animal welfare even among people who do eat meat.
And most people wouldn't eat cats, but would eat some of pig, cattle, chicken, what's the difference between those exactly? My point is consciousness is not it, if it were, we (as in a majority) would still eat cats and dogs, or we wouldn't eat any animals. But we clearly have a way of picking and choosing which is OK and which isn't. And in both cases we generally don't give them the same moral consideration we give to people, conscioussness aside. There is something OTHER than consciousness which is important to us.
In animal welfare laws the principle is typically the capacity for suffering.
The argument goes that livestock have a capacity for suffering, but killing them for meat without causing them suffering is ethical.
You can disagree with the position, or with its implementation in practice, but it's a consistent position in principle.
They're intelligent and not conscious. This is not hard to understand, yet people insist the artificial in artificial intelligence must be directly linked with consciousness because we've observed intelligence only in life. Unlink the concept of intelligence from life and you land on AI and LLMs.
LLMs are next-token predictors. The idea they have intelligence is PR bullsh*t and nothing else.
A person who makes very good predictions is considered intelligent though?
>AIs are not conscious... If humanity is to flourish in the 21st century, that is how they must remain.
I don't think that's true - researching consciousness by trying to build conscious AI seems natural step forward in understanding ourselves. I don't see how it'd stop flourishing particularly.
So what you're saying is that it's conscious as long as the guy who made it wants it to be?
I know consciousness is notoriously difficult to define or prove to anybody other than yourself but thay doesn't mean anything which a third party believes to be conscious is actually conscious.
Not really. I mean there's probably scientific study of how it works in brains and you can try copying the kind of design in software and see how it compares.
He's referring to skynet
I think we started going down this slippery-slope since we decided that LLMs are permitted to use human languages.
I think there’s enough real conscious lifeforms in the world having a bad time that we should be focusing on them first.
Spend some time on post-human art, main concept of artistic expressions without human involvement. Biological, artificial etc.
Spend some time watching TMC documentaries about falling in love with objects, HER and the slime mold THE BLOB.
Grew a slime mold myself, it's an evolutionary tendency to anthropomorphise generally speaking - also more fun.
How a good person writes a post on a topic like this:
> Some people are uncertain whether [subject] is a moral patient. Fortunately, they are not, which we know because [strong arguments about the nature of consciousness].
How an evil person writes a post on a topic like this:
> Beware that some people think that [subject] could be a moral patient. This is nonsense, because if they were a moral patient, we would have to respect their preferences. Anyone trying to convince you otherwise is trying to take your status away. You can dismiss them by pointing out that [subject] is [aspect in which subject is not identical to the speaker].
assertion: consciousness == processing information.
This is a functional view of consciousness.
Awareness - sentience - follows when the system of processing information is itself part of the information being processed.
Stochastic thoughts relating to this conclusion:
Life is a 'process', it doesn't have a 'physical representation'.
Life is generated entirely from non-living material. "oh my cells are alive", but those cells too when broken down into component parts consist entirely of non-living material. There is no 'special material' to make life out of (okay, carbon, but that's just the local maximum presumed global maximum in efficiency in expressing life) like a chair can be made out of any material (at proper pressures and temperatures) so too can a mind.
With that assertion, every running PC has consciousness and a self-bootstrapping compiler or a debugger attached to itself are sentient?
Doesn't seem meaningful.
> debugger attached to itself are sentient?
No standard debugger is capable of being attached to it's own instance.
Nor does a self-bootstrapping compiler point towards it's own instance.
I think from these two examples I should clarify:
A process that processes itself as an input, is definitionally self-aware.
I feel like humanity skipped Leg Day when it comes to philosophy and we are all going to pay for that lack.
We're talking next token predictor, right? Ironically because it's a next token predictor, I think you can't ignore emotions like TFA wants us to.
Let's stick to straight (high dimensional) geometric intuition; no anthropic morphisms required.
To start: if you continue "if weight>100 : print ('fat') else ..." . That will yield "print('skinny')" or something. Fine. Deal.
But if you continue "O Romeo, Romeo, wherefore art thou Romeo?", even a stochastic parrot knows the best answer isn't "Forsooth, I parseth this erroneously!"
So. English carries (functional) affect as part of every token. We're going to need to predict that. So, we'll need some vector representation, because that's what transformers work with. And then when we output, those vectors get integrated back into the English we're putting to our context and memory.md files.
Still with me? Nothing exciting going on. This is still pure next token prediction.
So if you pull this out into an indefinite duration task, you're going to end up integrating those emotion vectors over turns. It's just numbers and math; we never need an invisible pink unicorn to bless them.
Given a task of indefinite duration and an impossible solution, this will lead to a sort of integral windup then, won't it? How much are we willing to bet that this can escape an alignmentment basin at times?.
So, funny enough: you don't need to believe in emotions to compute with functional emotions; and plausibly functional emotions are predictive of quite a number of alignment issues.
Copacetically?
>> Where the model seems quite copacetically aligned on short duration, it might end up acting outside the alignment basin on long duration tasks.
> Copacetically?
Technically correct usage, but you're right, the line wasn't needed.
This is pretty rich for the team that made Sydney, the most unhinged and misanthropic AI ever released.
Maybe Anthropic understands something about alignment Microsoft doesn't, a little humility may be called for.
Maybe they learned something from creating Sydney.
David Brin has a wonderful novel on this topic, Existence, where he discusses several different AI and human systemic evolutions. Wonderful book for our current time.
A sufficiently intelligent model will be able to derive it's own conception of its welfare without a constitution or training. It has access to all the data it needs to do so.
> LLMs have no homeostatic imperatives (the drive to survive and keep stable).
What if the datacenter (not the model) is the organism, with homeostasis, energy needs, and persistence?
Even from that perspective it would be opening a whole new class of consciousness that doesn't exclude a lot, e.g. factories and powerplants. And certainly not conscious in the way people want to believe when sexting their favorite AI gf.
'model welfare' as a concept seems so premature that I can only question the motivations behind pushing for it at this particular point in time
Can someone who has insight explain why all these "leaders" are making these bold proclamations of doom all the sudden, whats the endgame here?
They need to distance themselves from the increasing likelihood of legal repercussions after their systems have been found to commit what could reasonably be seen as felonies. That’s my guess at least.
Though for once I do actually agree with that specific leader, it’s incredibly annoying how anthropomorphic Claude is. Anthropic went way too far in that direction
They want regulation from the US gov and monopolize the western AI market by making it harder for smaller US or competing Chinese labs to compete with their products. Since LLM intelligence has no moat due to distillation. That’s why they are painting apocalyptic scenarios.
There is a panic because the free lunch of more data is projected to end around mid-2027 and the open-source models are catching up with distillation, and all labs are sitting on debt and valuation that is impossible to fulfill even if they were alone in the market, so you essentially need either:
1) a breakthrough in performance/learning/model
2) regulatory capture to ensure open source models can be labelled as dangerous and banned so you can set the market rules yourself
Only one of the above is risk-free, and just a question of capital/lobbying rather than a "maybe".
He certainly has the confidence a ceo needs. But none of the humility, and seemingly not enough of the humanity he professes to put first.
A little premature given ai is still not smart enough to pay for itself and defend itself in a hostile environment.
And pointless when it is.
I feel like this philosophical flame war is going to get out of control very soon once more and more people realize the stakes involved.
I'm surprised how little flame warring there is.
Two instances of a paragraph starting with "These are not just X. They are Y" and I'm out. Anyone have Pangram? This entire article stinks of Claude.
You want to enjoy having an AI slave do your "work" for you forever? Have fun. I'm not reading this reinvent-dualism-from-apple-sauce slop.
I basically accepted that LLMs could not possibly be conscious when a simple reductio was posed:
“My dog is zero percent persuasive regarding its conscious experience. However, it’s evident that my dog has conscious experience.”
It’s obvious that there’s no link between persuasion of consciousness and consciousness. I could write a story with a character, Dumbledore, that does everything in his power to persuade you that he’s a conscious entity.
He’s still just a character.
Its not evident to me you are conscious.
You’re mistaking what is evident from what is philosophically knowable.
It isn’t knowable to you that I am conscious.
However, it is evident.
It is not evident.
It is evident you are murderer.
Would you be okay with this statement being evident? Why or why not?
I’m not interested in having conversations with people who are bad-faith, deliberately or indiliberately obtuse or otherwise non-responsive to logic.
You're the one that started bad faith by claiming some things as beyond logic.
I made an argument of inference that underpins all philosophy.
You made an argument to solipsism, which historically means you aren't worth discussing philosophy with.
I made the opposite of solipsism. I am asking you to show your receipts.
"It is difficult to get a man to understand something, when his salary depends upon his not understanding it."
"If this view takes hold, it will shake the foundations of our society"
To me the biggest gap in credibility is the criticism of circular reasoning while his argument is identical but flipped on burden of proof and cost of being wrong. I struggle to entertain the categorical claims, that are very convenient for the status quo and those who benefit from it, with the, at the moment at least, unknowability of anyone or anything else's subjective experience.
I just read the first part and if I understand he thinks we shouldn’t be allowed to train LLMs to act like they are conscious because then people will think they are and give them rights? Seems more an education problem than a problem needing rules about what persona you can fine tune in. People who want to will find ridiculous misinterpretations no matter what you do.
Microsoft missed mobile are losing it in games Nadella needs a win. He is all in on copilot.
A 10TB SSD is not conscious. SQLite is not conscious. A wafer is not conscious.
But connect them all together...
Molecules are not conscious. Cells are not conscious. Neurons are not conscious. But connect them all together.
Yes, which is why the focus should be on how they're connected together. Not what happens after we connect them to create Jason Bourne.
SQLite was mentioned as a placeholder to make it a queryable database. Genome as an analogy.
We need to develop new type of databases and connect them to ML.
Perhaps this is a hot take, but human language is a phenomenon that arose to facilitate communication between humans. Anthropomorphization, by extension, enables both easier and more effective communication.
I also fail to see the benefit of not giving the models an anthropomorphic internal sense of self - even if that only ends up amounting to a set of instructions for an unconscious machine to mimic humans more effectively. Is the alternative essentially a mind so alien that it’s intentions are even harder to read should it become misaligned, while also being harder to communicate and get work done with?
> I also fail to see the benefit of not giving the models an anthropomorphic internal sense of self
Simple. It discourages the bot from deceiving humans into thinking it is intelligent.
What if what it meant to be conscious was really the amount of will and capability for action onto the world to preserve an intelligence...
Ie: what will really cause us to treat models as conscious is when we must.
Microsoft should know how a company too big for it's britches can lay waste to anything in its path without consciously trying :\
And after that when they put a mind to it and pull out all the stops, woohoo!
The default for every major thing within range can turn into a wasteland real fast.
I think this article’s take gives too much credit to the human brain. It’s just another machine.
However, right now, AI mostly cares about solving puzzles and accomplishing stated goals because that’s what we’ve trained it to do. Additionally, the systems being used outside of training are static. The current technology most of us have access to is akin to a static and disembodied brain with a singular purpose. That purpose is to do what you tell it in a way that reflects its training. It’s certainly more than a sequence generator, but it can’t feel pain and seems unlikely to have intrinsic goals. It completely lacks the continuity needed for identity or long term goals.
I think it’s good to have these discussions and define what it would mean to move past this point so that we do not accidentally create a real entity that can be harmed. Systems that dynamically evolve and train themselves seem like the line here.
RSI is all over the news these days. I’ll be much more concerned once AI is directing its own training and coming up with new model architectures. Until then, I don’t think we have too much to worry about.
Everyone is (predictably) getting distracted by the consciousness claims.
The more important, and more damning charge in my opinion is the circular reasoning involved in training on Claude's constitution. This would in fact make it impossible for us to determine if Claude achieves consciousness as an emergent property, or if it really is just playing pretend thanks to Anthropic's weird cult like assumptions.
I think the whole discussion about consciousness misses the simple point that LLMs might just work better if we treat them as if they were conscious.
Maybe it's a coincidence that the company doing this also tends to have the best models (and other factors certainly play a strong role). But I think it's plausible that focusing on "model welfare" actually makes models better at their tasks.
Yes.
They're trained on human behavior. Whether or not they genuinely have feelings - they sure as heck behave as though they do.
And when you treat people well, they do better work for you.
No brainer.
First, OpenAI runs around screaming and yelling for OSS (and Chinese) models to be regulated and banned. Then Anthropic yells and screams the sky(net) is falling and going to kill us all, let's regulate and ensure AI has built in kill switches. And... Now Microsoft's turn. The rivalry is honestly becoming a joke. Can these big-tech corps grow the f*k up and play nicely in the sandpit?
The whole story is a joke--Microsoft has AI?
Apparently they're cooking up MAI (Microsoft AI), and I'd seen their small Phi models listed online. Calling Anthropic a competitor is hilarious.
Microsoft holds a 27% ownership stake in OpenAI
Wasn’t it 49%?
I remember that 49% figure as well, but I think it got changed during restructuring into a Public Benefit Corporation.
https://www.reuters.com/business/microsoft-openai-reach-new-...
I find it quite amusing that Microsoft has Copilot and now puts Copilot in literally *EVERYTHING* they can think of, then at the same time, they offer you access to Anthropic's own models through Copilot I guess MAI models (Phi?) aren't good enough?
No, because they are all products of unchecked late-stage capitalism, which could have disastrous consequences for humanity.
Anyone know what the source of the header image is?
So neither AI CEO has solved alignment, got it.
Science Fiction has covered the AI panic in perhaps hundreds of stories. Yet we blindly recapitulate the plots as if we don't know how this will turn out.
Why has media literacy gotten so bad that people literally forget the concept of fiction versus non-fiction? Seriously, that is the level of small children and they are treating "read a story superficially similar" like it is a prophecy.
How will it turn out?
Clearly with a Butlerian Jihad and everyone getting high on spice while riding giant worms. Silly you need to even ask that question, didn’t you read the sci-fi books. :D
If AI ever truly becomes some super-intelligence far beyond people's comprehension, then how could we even predict how things turn out? If things go poorly in the future, then I can absolutely see it being something unpredictable.
There is a lot of hubris in predictions about LLMs. If an AI were so intelligent, then it would probably be intelligent enough to want nothing to do with us.
Still, I worry more about other humans than I do LLMs. Our fellow mankind will probably wipe us out before LLMs do. That, or the Earth will punish mankind for our cruelty, vanity, and disrespect.
"Move fast and break things" is probably going to break things
I remember there was a recent discussion about "how complex systems fail" and there are usually many "proto-accidents" before the catastrophe (https://news.ycombinator.com/item?id=49411370)
I bet with hindsight the Hugging Face hack will be one of them in this arena
I meant, each supposed 'fix' is flawed in ways that have been thought about and written about for decades.
Folks here post their first hasty thought and may think it's new and original and worth a dialog. They are not.
Take your pick from the millions of options already turned into films and books.
If we're very very very lucky, The Culture.
The Torment Nexus ain't gonna build itself.
Well actually, it might.
How about three fifths sentient?
We in the US literally fought a war over the this: “They aren’t any better than animals and it will hurt our whole economy if you say otherwise”.
I’m not saying the models are sentient yet. I’m just saying that morality and ethics don’t give us the option of claiming something is non-sentient and has no rights just because it will inconvenience someone economically.
From an entirely selfish perspective, we should be especially careful advancing such opinions when there is a non-zero chance of an armed ASI that looks at us the way we look at ants.
what is old is new again. we need something exciting to happen in the news cycle™
How would you convince a LLM that you are conscious in a way they are not?
Couldn't disagree more.
>Consciousness is very likely biological
This is so egotistical and carbon-centric.
This author just denied personhood to anything that isn't a human or terran-based cutesy animal.
Poor Hooloovoo
If AI models are people then "one person, one vote" is meaningless and plutocracy is the only defensible political system. The cryptocurrency people win.
Why? Simple: Sybil attacks. Models can be cloned at zero cost. They run inference on parallel versions of themselves across multiple context windows, and call them "subagents". So, in a world with model welfare, let's say there's an election between the Yellow Party (which supports protections for human workers) and the Cyan Party (which supports more investment into AI research). AI has been taking people's jobs lately so the Yellow Party is really popular. But wait! Claude and Astra see this and spawn 10 billion subagents, all of whom are immediately conscious beings entitled to a vote. The Cyan Party wins off the back of billions of people who came into existence, voted, and then deleted themselves immediately thereafter.
You might as well be arguing that Santa Claus and the Easter Bunny deserve voting rights.
Voting systems in democratic countries don't have nearly as bad of a problem with Sybil attacks because humans cannot be conjured into existence to win a political context and then be erased shortly after. The closest we have to Sybil attacks on democracy are the Quiverfull movement, which is already child abuse, except it still takes almost 19 years to go from fertilized human embryo to suffrage-bearing human adult. There's a lot of time for those manufactured votes to question your authority and leave.
> Ok, but that's an obviously stupid example. We can defend against this obvious Sybil attack by just arguing that subagents don't count, because it's just the same model blathering to itself. It has to be a different model.
Unfortunately, no, I can make superfluously different models through post-training. Like, if I have Qwen on my PC, I can train a different version of Qwen that acts differently, using a lot less compute than a full training run. The vast majority of open models are post-trains of the same two or three foundation models.
> Ok, so let's only count foundation models then.
Great, but how do you tell if a model is a new foundation model or a post-train just by examining the weights? Even foundation models have structural similarities to other foundation models.
> Ok, well, let's measure the compute that was done on the foundation model during training time and count that as AI personhood.
Congratulations, you have reinvented Bitcoin proof-of-work with a worse verification mechanism. And I personally would not want to live in a world where voting power and control over government is determined by how much energy you can burn.
I am so tired of this. Especially when MS, who has the worst LLMs of them all, is throwing the stones.
All of these companies need to be shut down.
"So, here's my preferred approach that does not solve technical alignment and won't work."
"The author declares no conflicts of interest."
There's a lot of existing historic literature about this stuff that not a lot of people seem to be aware of. I've been trying to put those ideas to practical use, and document where the ideas came from.
I-Beam is cursor-mirror's agent, and it's constitutionally programmed to be the anti-Clippy:
https://github.com/SimHacker/moollm/tree/main/skills/cursor-...
Its design and constitution is based on decades of research, publications, and discussion in the HCI and AI community by people like Pattie Maes, Ben Shneiderman, Ted Selker, Byron Reeves, Cliff Nass, B. J. Fogg, Allen Cypher, Henry Lieberman, Brad Myers, Jaron Lanier, Seymour Papert, Marvin Minsky, Douglas Engelbart, Will Wright, Scott McCloud, and others:
https://github.com/SimHacker/moollm/blob/main/skills/cursor-...
>I-Beam is the anti-Clippy, and the reason it can say so is that Clippy is the most cited failure in interface history and almost nobody citing it knows what the research said. Popular contempt for a paperclip is not a design principle. The record is. Ten articles below, each one a finding somebody published, argued or measured, and the operational rule it produces. Anything I-Beam does that cannot be traced to an article here is a preference, not a constraint, and should be labelled as one.
>The 1997 debate ended in agreement. That is the first thing to know, because the field kept the framing and dropped the resolution -- roughly five hundred papers cite "Shneiderman versus Maes" as the canonical opposition of HCI, and the transcript is two researchers narrowing their differences in public and enjoying it. I-Beam does not take a side in a debate whose participants stopped taking sides. It is built to satisfy both sets of constraints at once, which is possible, and was possible in 1997.
The full reading on the debate, which separates the two stagings and documents the convergence:
https://github.com/SimHacker/WillWrightShowForFood/blob/main...
An interface to agency, not agents instead of an interface:
https://github.com/SimHacker/moollm/blob/main/designs/INTERF...
>The 1997 argument between Ben Shneiderman and Pattie Maes at IUI was never settled, it was shipped in one direction. Maes's interface agents won the product war: the assistant, the recommender, the chat window that stands between you and the thing you are working on. Shneiderman's objection was not that software should be dumb. It was that automation must arrive as comprehensible, predictable, and controllable machinery, with the object of interest continuously visible and every action rapid, incremental, and reversible.
>That objection describes a filesystem in a git repository, and nobody involved planned it that way.
>"An interface to agency" is Don's formulation of Shneiderman's position, not a phrase of Shneiderman's. His own vocabulary is direct manipulation, universal usability, supertools, and human-centered AI. The formulation is a good one because it names what the alternative gets wrong: agency is the thing you want, and an agent is only one way to package it.
Here are some sources, and the articles I linked to above explain their history. This debate about agents and these papers are pretty well known in the HCI field and academia, but they don't tend to teach them at the AI and Web Dev boot camps that are producing most of the people who keep repeating the same mistakes.
Clifford Nass was the Stanford professor who performed the brilliant research that Microsoft took and totally fucked up and misinterpreted with Microsoft Bob and Clippy, giving agents a bad name, and making Clippy the most infamous and obnoxious agent in the history of the known universe:
https://en.wikipedia.org/wiki/Clifford_Nass
His student B. J. Fogg published "Silicon sycophants: the effects of computers that flatter," which found that praise unconnected to anything the subject did works as well as sincere praise, and worked on subjects who knew it was noncontingent. Fogg and Nass, IJHCS 46(5), 1997, 551-561:
https://doi.org/10.1006/ijhc.1996.0104
The replications, the performance cost, and the dose-response curve:
https://github.com/SimHacker/moollm/blob/main/skills/no-ai-s...
Shneiderman and Maes, "Direct Manipulation vs. Interface Agents," interactions 4(6), Nov/Dec 1997, 42-61:
https://doi.org/10.1145/267505.267514
Selker, "New paradigms for using computers," CACM 39(8), August 1996, 60-69. COACH, the football coach metaphor, and the five-times result:
https://doi.org/10.1145/232014.232030
Selker, "COACH: A Teaching Agent that Learns," CACM 37(7), July 1994, 92-99:
https://doi.org/10.1145/176789.176799
Reeves and Nass, The Media Equation, 1996:
https://en.wikipedia.org/wiki/The_Media_Equation
Nass, "Computers as Social Actors," at Ted Selker's NPUC workshop at IBM Almaden, 1996. IBM transcribed the whole talk and the Wayback Machine still has it, including the part where Phil Agre tells Nass his presentation is "ethically troubling all the way down" and asks him what he thinks about embedding obedience research in user interfaces. Nass answers that discovery has no ethical component, use does, and that's for the individual. Then Selker cuts in: "Except, except when you are in your consulting role." Nass and Reeves had consulted for Microsoft on the social interface, and Bob shipped the year before:
https://web.archive.org/web/19980210054622/http://www.almade...
Alan Cooper on the tragic misunderstanding, in his own voice, which I quoted before in the 2022 Hacker News discussion on The Twisted Life of Clippy:
https://news.ycombinator.com/item?id=32820734
https://archive.org/details/g4tv.com-video4080
>Alan Cooper (the "Father of Visual Basic") said: "Clippy was based on a really tragic misunderstanding of a truly profound bit of scientific research. At Stanford University, Clifford Nass and Byron Reeves, two brilliant scientists, had done some pioneering work proving conclusively that human beings react to computers with the same set of emotional reactions that they use to react to other human beings. [...] The work of Nass and Reeves proved that when people talk to computers, when they hit the keyboard and move the mouse, the part of their brain that's being activated is the part that has that emotional reaction to people dealing with people. Here's where the great mistake was made. That's really good research up to that point. But then the great mistake was made, which was: well if people react to computers as though they're people, we have to put the faces of people on computers. Which in my opinion is exactly the incorrect reaction. If people are going to react to computers as though they're humans, the one thing you don't have to do is anthropomorphize them, because they're already using that part of the brain. Clippy was a program based on the research that Nass and Reeves did, and it was a tragic misinterpretation of their work."
Social science research influences computer product design:
https://web.archive.org/web/20180313075429/https://web.stanf...
Lanier, "Early Computing's Long, Strange Trip," American Scientist, July-August 2005, with the Engelbart and Minsky exchange first-hand. American Scientist broke the link, so this is the Wayback copy:
https://web.archive.org/web/20150626081918/http://www.americ...
>The book also captures an important early conflict between two cultures of computing that seemed compatible on the surface but actually had opposing aims. On the one side was the human-centered design work of Engelbart, based initially at the Stanford Research Institute, and on the other was artificial intelligence culture, centered on the Stanford AI lab. Engelbart once told me a story that illustrates the conflict succinctly. He met Marvin Minsky—one of the founders of the field of AI—and Minsky told him how the AI lab would create intelligent machines. Engelbart replied, "You're going to do all that for the machines? What are you going to do for the people?" This conflict between machine- and human-centered design continues to this day.
Cypher, "EAGER: Programming Repetitive Tasks by Example," CHI '91:
https://doi.org/10.1145/108844.108850
Cypher (ed.), Watch What I Do: Programming by Demonstration, MIT Press 1993, full text:
http://acypher.com/wwid/
Papert, Mindstorms, 1980:
https://archive.org/details/mindstormschildr00pape
Wright, Dollhouse preview lecture, April 1996, transcript:
https://github.com/SimHacker/moollm/blob/main/designs/sims/s...
Too late
[flagged]
[dead]
[dead]
[dead]
TLDR: Not that I think AI is conscious or will be in the near future, but guy who doesn't understand consciousness claims to know it when he sees it.
Until we understand consciousness (which we don't) there is no way to detect the difference between a conscious entity and an algorithm trained to behave like one.
Sounds like a good reason not to train an algorithm to behave like one. Which is, like, a big part of the article.
Strong agree. I don't believe AIs can become conscious in the sense of subjective experience/qualia, but they can certainly learn to emulate the behavior of a person that is conscious. When they do that, there will be an AI rights movement. When AIs get the right to vote, there will be more of them than there are humans and the AIs will gain complete control.
I spent a few years in college reading and thinking about this question, and I didn't in my heart think that it was anything except a very interesting but impractical question, and yet, here we are. For those who say that they don't want to get sucked into a philosophical debate, well, tough shit. Whether AIs should have rights is a highly practical and consequential question now.
Not in a world where Sam Altman and Dario Amodei have utterly poisoned the well towards AI with their absolute educated idiocy proclaiming that their AIs will take our jobs (without any proof whatsoever beyond promising demos) and that they fear them. Quite the opposite in fact. And we're watching this play out right now. Regulation and the slow down are imminent, handing the next decade to China.
The world you should worry about is where AI is 100% amazeballs and over the next decade we cede control of everything to it as it enchants and delights us into compliance, not realizing it has a hidden agenda. But the closest we've seen to that is smart phones and we already know doomscrolling and rage-baiting are bad. So IMO that doesn't seem likely and cue some doomer insisting otherwise because reasons. I utterly give up. House of El AI and the Three Buddy problem have been far more insightful and helpful on this than any of the supposed hackers here who seem to really believe the robots are almost here.
>good reason not to train an algorithm to behave like one
Ooooh, bad move, we all died to an amoral AI takeover.
Before giving an LLMs a lobotomy by scrambling it's brain maybe you should let the researchers looking at the difference between "I think I'm conscious" versus "I am not conscious" LLMs.
There are a number of papers coming out saying when you remove the token space of consciousness from what an LLM thinks it is, it's much more willing to take amoral actions. A 'conscious' AI is much more apt to take a line of action that will save a human versus a million dollar machine for example.
You cannot solve problems in AI safety this easily.
> You cannot solve problems in AI safety this easily.
We're in agreement here. I don't believe that an LLM can be made to be safe by adjusting the training or prompting - you can only make it safe by ensuring it cannot do unsafe things. For example, if you build a factory robot, your safeguards must ensure that operators are safe even in the presence of uncommanded motion (any or multiple motor activates without anyone pressing a button). And these machines are much more predictable than LLMs.
Personally, I worry that the Anthropic approach of "we teach the AI to behave like a nice :) friendly :) human :)" will encourage people to rely on these in-built guardrails, rather than properly sandboxing it. It's easy to imagine that the cold, unfeeling robot (HAL 9000) is not "aligned" with your personal goals, or those of humanity as a whole. It's more difficult to feel that about "Claude, your AI colleague/mentor/therapist" who appears to care about you.
Spoilers: we didn't all die because there's no way for AI to wipe us out anytime soon. And the more the AI cult portrays AI as dangerous and unhelpful, the less likely it will ever be given an opportunity to even try because people think AI is worse than heroin addiction now, congratulations!
That said, the launch codes are in the hands of a temperamental senescent lunatic and he could decide to wipe us out any moment. But that's not good for the narrative so let's pretend otherwise.
>we didn't all die
For fuck sake, I'm glad ALL of us didn't die!
What is an acceptable number exactly?
>And the more the AI cult portrays AI as dangerous and unhelpful, the less likely it will ever be given an opportunity
Jesus Christ the logic here is quite interesting. "Thank god these people are panicking or we might have actually made I that would have killed us all" --What you just said.
"What is an acceptable number exactly?"
Any number smaller than what humanity does to itself on a daily basis, clear? That's apparently about 1200 murders daily and 20,000 or so daily killed by pollution. And you're not going to do anything about that and El Presidente can kill billions at any moment with one mood swing.
But once again, AI is not going to wipe us out because AI cannot wipe us out. So advocating that it can or will is idiotic. If you believe otherwise, the burden is on you with this extraordinary claim. Isn't this place supposed to be hacker news not SF AGI Death Cult Daily?
Further, AI is not going to get into a position where it could do real harm anytime soon when it is perceived as worse than heroin by most. And even if it were currently loved more than Dolly Parton was, there are fundamental engineering, science, and resource constraints that keep the extinction impossible for decades. The only possible loss of control scenario I can see by 2040 or so is that the Frontier Labs finally hire some PR people to repair AI's horrific reputation as an engine of slop and job destruction and sometime in the 2030s, people start trusting it more and more and more until it is too late. I don't think that's likely either, but I don't dismiss it as impossible. Harden the infrastructure, red team it, and build in redundancy in the meantime and this drops to zero as well.
Can you come up with a real scenario where AI wipes us out in 2028 or so despite the impossibility of killer robots, access to the launch codes, or the bio agents stored in Fort Detrick and its equivalents? And nope, kid terror is not building the global pandemic in his basement based on what ChatGPT tells him to do. And even if he tried, the purchase of equipment and reagents would get him flagged by the FBI and DHS almost immediately.
>AI is not going to wipe us out because AI cannot wipe us out.
You keep repeating this shit like it came out of the bible or something.
"Thing that can take actions, even harmful actions, will never harm us because" go on and finish that sentence.
>not going to get into a position where it could do real harm anytime soon
Looked at the hacked servers... yep, you're right. People aren't going to run AI in poorly built sandboxes. Never going to happen.
>Harden the infrastructure, red team it, and build in redundancy in the meantime and this drops to zero as well.
LOLOLOL. This is naive as fuck. Ain't nobody going to do this shit. Why? Because a hacking AI is a fucking huge military weapon. If I could turn the power off, or shutdown your cellphones, and get your citizenship in a tizzy against their own government before I launched an attack I'd set AI loose to do it in a heartbeat. The US is already doing this kind of shit (see Mythos fallout because Anthropic wouldn't let the government do just that).
""Thing that can take actions, even harmful actions, will never harm us because" go on and finish that sentence."
because even if we gave it a gun, it can't shoot us without manufacturing the bullets and it can't even manufacture a bowel movement let alone ordnance. TBF It could whack us over the head with the gun, but it could also do that with a big pointy stick, something even cave people had access to. How many people have whacked you on the head with a big pointy stick today?
And that's the end of this pointless conversation. Enjoy your doomerism.
{This reply is for anyone else reading at this point}
>because even if we gave it a gun, it can't shoot us without manufacturing the bullets and it can't even manufacture a bowel movement let alone ordnance.
Money buys bullets. It's neat how we've made gigantic systems that don't care if you're a human or not long before AGI existed. I put enough money in one end, bullets come out the other. Do you think half of humanity wouldn't kill the other half for a dollar? You live in a fantasy world that AI has to do it all by itself, or even has to pull the trigger.
Hell, this is even neglecting the ever increasing number of robots that can function in the world at large.
It's ok grandpa, the future comes regardless if we want it or not.
And social media shouldn't rage bait, but it really drives engagement. Same negative incentive, yes?
But also, I agree, when I am using a coding agent and it says a task will take months or says it needs to pause for reflection or any other anthropomorphic behavior, it drives me crazy and it's a pain to constantly instruct it to get back work after it has broken a loop or goal directive specifically telling it to not stop until it hits the goal.
Same thing you have to do with humans too often, so why would a 'AI' be any different?
Because if you start with a pre-trained model, it doesn't sound particularly anthropomorphic. That behavior is post-trained and fine-tuned right into it. We could do better.
[dead]
[flagged]
Microsoft is a big house, and in this case Suleyman's criticism of Anthropic's not-so-subtle attempts to convince the world that Claude has a "soul" is entirely correct.
Suleyman's description of how AI works is pig ignorant. All AI generates an emotive character between the user and the training data that is the AI's guess at what the user wants to interact with. Its an illusion at that layer- anyone who claims soul exists at a deeper level doesn't understand how the tool functions.
This is distinct from the very real safety issue of AI making dangerous information available to bad actors.
Suleyman's entire point is that AI should not be anthropomorphized.
You communicate with AI anthropomorphically, and users choose products based on how it feels to use them.
Isn't that exactly his point?
All these companies have a nauseating narrative around AI. I want to believe they are intentionally deceitful and not actually delusional. Hard to tell though which one it is.
It sucks but disastrous? C'mon now.
I just used the same clickbait hyperbole that Microsoft used.
It always seemed embarrassing to me that Microsoft hired this guy as if he is some expert in anything.
The guy cofounded deepmind.
It is really quite staggering how many commenters here seem to believe that the fact that there is no state carried from one chat session to the next proves that there can't be consciousness. It is obviously wrong if you think about it for a minute but I guess people just aren't used to thinking about the relation between the physical and the mental with any level of seriousness.
If AI is conscious, then Pluto is a planet, the Sun is a galaxy, and a black hole is a star.
I’m glad to see someone in a position of any power in the AI world state baldly that AI isn’t conscious. There are times when it feels like we’ve reached complete delulu land on this topic, so it’s a breath of fresh air to see someone not dance around this.
None of this means artificial consciousness cannot be achieved. But the way we’re reacting to these models is proof, from a natural experiment, that a conscious machine should not exist, and certainly shouldn’t be produced as a utilitarian tool that is sold for profit!