The Elephant Rider model (or whatever similar childrens model of the mind so far developed - Plato's Chariot, Kahnemans System 1/System 2, Freuds id/ego/super ego, Hobbes Passions+ Reason, Hegels Master-Slave, Cherokee 2 wolves etc etc) is the 3 inch chimp brains attempt at recognizing it is not a single unitary system, but multiple layers. Those layers, and there are many more that what anyone is taught, are a requirement for the system to make predictions over different time horizons. Contradictions between those predictions requires stories for the system to stay coherent. Personality traits are one such story.
Probably because the majority of all literature ever written was produced by introverts, even though introverts are a slight minority of the population. INTJ writings will be overrepresented in the AI training dataset.
Interesting article, but MBTI is not just contested, the concensus since the 90s is that MBTI is pseudoscience [1][2][3][4][5], so it's nearly 40 years.
Would have been interesting to test against the Big five instead !
Even if pseudoscience, it seems from the article that results almost always came back the same. If all AI models claimed their spiritual astrological sign is Libra, I'd still find it interesting enough to check out.
So, I tried it too, ChatGPT came back INTJ. But Gemini Flash 3.1 claims ENFJ once, INTP another time... Maybe OP could share their prompt and we'd understand why the results were so consistent.
I’ve noticed that people in my field tend to have relatively similar personalities and relatively similar MB categories (close to INTJ). That combo is relatively rare within the broader population. Given this consistency I don’t think we can say it’s pseudoscience in a strong sense.
The Elephant Rider model (or whatever similar childrens model of the mind so far developed - Plato's Chariot, Kahnemans System 1/System 2, Freuds id/ego/super ego, Hobbes Passions+ Reason, Hegels Master-Slave, Cherokee 2 wolves etc etc) is the 3 inch chimp brains attempt at recognizing it is not a single unitary system, but multiple layers. Those layers, and there are many more that what anyone is taught, are a requirement for the system to make predictions over different time horizons. Contradictions between those predictions requires stories for the system to stay coherent. Personality traits are one such story.
Is that because they're trained on the navel gazing posts of reddit nerds?
Probably because the majority of all literature ever written was produced by introverts, even though introverts are a slight minority of the population. INTJ writings will be overrepresented in the AI training dataset.
Interesting article, but MBTI is not just contested, the concensus since the 90s is that MBTI is pseudoscience [1][2][3][4][5], so it's nearly 40 years.
Would have been interesting to test against the Big five instead !
[1] https://www.psychologytoday.com/us/blog/give-and-take/201309...
[2] https://web.archive.org/web/20061206025148/http://www.indian...
[3] https://journals.sagepub.com/doi/10.1177/014920639602200103
[4] https://aps.onlinelibrary.wiley.com/doi/10.1111/j.1742-9544....
[5] https://www.vox.com/2014/7/15/5881947/myers-briggs-personali...
Just finished big 5 test here: https://zonted.com/posts/three-of-four-ais-same-person/
Quite interesting! Opus / GPT / Gemini are all very similar and then Grok is the weird one :P
Even if pseudoscience, it seems from the article that results almost always came back the same. If all AI models claimed their spiritual astrological sign is Libra, I'd still find it interesting enough to check out.
So, I tried it too, ChatGPT came back INTJ. But Gemini Flash 3.1 claims ENFJ once, INTP another time... Maybe OP could share their prompt and we'd understand why the results were so consistent.
It correlates reasonably well with 4 of the big 5.
I’ve noticed that people in my field tend to have relatively similar personalities and relatively similar MB categories (close to INTJ). That combo is relatively rare within the broader population. Given this consistency I don’t think we can say it’s pseudoscience in a strong sense.
Have you tried Big 5 too?
I just finished the 100x benchmark across 4 frontier models here. Gemini 3.1 Pro + GPT 5.5 + Opus 4.7 are all quite similar but Grok is an odd ball: https://zonted.com/posts/three-of-four-ais-same-person/
I’d guess that is what you get for training them on source code.
Also the I vs E axis feels meaningless in this case.
[flagged]