This is a superb piece of research, and odd that nobody has pulled this material before.
The Atlas 2 is obviously the successor to the Atlas, which Turing worked on before "The Baby" at Manchester University booted up, at which point he quit and moved to Manchester to work on that.
I seem to recall his early programs for The Baby - and his planned use of the Atlas - was to model "morphogens", his theory of how animals got their strips, spots and other markings. I think it was at the Science Museum in London (worth a trip for computer historians anyway to see Babbage's Analytical Engine, built in xxxx, next to a jar containing Babbage's actual brain), where there was a collection of printouts showing spots like on a Friesian cow (the most common breed of dairy cows in England).
I find this diving back into early uses of computing for creative and scientific purposes fascinating - it's like people were just so excited about being able to do interesting things, they just dived in with the tool that they had.
The poetry of Coetzee's program in this article is interesting - it's better, arguably, than you'd get from an LLM today. It's mildly mysterious, grating, intriguing but uncomfortable. A little like the things I remember of Coetzee's actual later writing (which I've not read much of, TBH).
It appears that Dr Coetzee was working on an important part of Atlas 2..
> "In 1964 Coetzee moved to IBM’s British competitor ICT, collaborating with computer scientists at Cambridge University on Britain’s first supercomputer, the Atlas 2. There Coetzee focused on multiprogramming, working on the Atlas 2's "supervisor," the earliest computer operating system."
Above quote from another article by Rebecca Roach at
That article develops ideas in a literary/cultural direction. I'd like to have a bit more information on the algorithms Coetzee was using.
Christopher Strachey in the early 1950s was using a program with combinatorial word choice to produce 'love poems'. He posted the printouts on notice boards and they were signed M.U.C for Manchester University Computer (Ferranti Mark 1).
That seems to be a simple choice from lists subject-verb-object type thing.
For me, Brian Westley is the greatest code poet of all. His poem was a winner in the 1990 round of the International Obfuscated C Code Competition under "Best layout - poetic exchange between lovers":
Glorious. This is the first time in 40 years of programming that I've encountered the term "waka waka" for the "<>" pair. My google-fu failed me when I tried to look it up today, perhaps due to search pollution generated by the Cameroonian/Shakira song.
EQUIVOCAL HERESIARCH TAMP THE PRESBYOPIA
STYPTIC ILLUSION ABSORB THE EIDOLA
makes me think that it's a sort of mad-libs phrase generator. The leading adjective "PENSILE" appears twice, as does "CATAMITE", and "PARACLETE" occurs three times, as does the noun "EIDOLA", once as subject and twice as object. Hopefully Dr. Roach will be able to present us a full reverse-engineering of the algorithm, but I'm guessing you could probably infer it just from that page of output: f'{random.choice(adjectives)} {random.choice(nouns)} {random.choice(verbs)} THE {random.choice(nouns)}', which produces lines such as STYPTIC PARACLETE TAMP THE PRESBYOPIA.
The raw wind of the smooth horse.
An evil labyrinth of the monstrous, dying waters.
The mere gate of deserted gardens.
The smiling stairway of countless sweet fangs.
The shattered witch of the diseased, ruby monk.
The asphalt house of the smooth unicorn.
A haunted fire of the sorrowful theft.
The stainless raven of countless living gates.
My haunted cloud beneath her glorious father.
The brazen father of their raw villages.
The comments of the program contain a somewhat embarrassing sonnet I wrote with its help.
Cien hijos halagüeños de la emoción infinita y monstruosa.
Los labios caídos de zorros corruptos y dignos.
Un exilio inmaculado dentro de las imprecaciones violáceas y ardientes.
Los dos pies insepultos de mis labios profundos e infinitos.
El oro bello de una libertad corrompida.
El angustia monstruosa de zorros desmoronados.
Aquel campo brillante de su princesa destrozada.
Una joven emoción de un tronco enmascarado.
El viento hermoso de su abandono griego.
Una ley romántica del abandono destrozado.
Now, of course, this sort of thing is much less novel and therefore less entertaining. GPT-5 will be happy to write as many sonnets for you as you want, as long as it doesn't somehow get the idea that they might infringe some kind of copyright or teach you how to do your own electrical work.
I've often wondered if J. M. Coetzee was related to Wikipedia editor Derrick Coetzee with whom I used to collaborate on computer science articles.
This is a superb piece of research, and odd that nobody has pulled this material before.
The Atlas 2 is obviously the successor to the Atlas, which Turing worked on before "The Baby" at Manchester University booted up, at which point he quit and moved to Manchester to work on that.
I seem to recall his early programs for The Baby - and his planned use of the Atlas - was to model "morphogens", his theory of how animals got their strips, spots and other markings. I think it was at the Science Museum in London (worth a trip for computer historians anyway to see Babbage's Analytical Engine, built in xxxx, next to a jar containing Babbage's actual brain), where there was a collection of printouts showing spots like on a Friesian cow (the most common breed of dairy cows in England).
I find this diving back into early uses of computing for creative and scientific purposes fascinating - it's like people were just so excited about being able to do interesting things, they just dived in with the tool that they had.
The poetry of Coetzee's program in this article is interesting - it's better, arguably, than you'd get from an LLM today. It's mildly mysterious, grating, intriguing but uncomfortable. A little like the things I remember of Coetzee's actual later writing (which I've not read much of, TBH).
It appears that Dr Coetzee was working on an important part of Atlas 2..
> "In 1964 Coetzee moved to IBM’s British competitor ICT, collaborating with computer scientists at Cambridge University on Britain’s first supercomputer, the Atlas 2. There Coetzee focused on multiprogramming, working on the Atlas 2's "supervisor," the earliest computer operating system."
Above quote from another article by Rebecca Roach at
https://egomedia.supdigital.org/sections/talking-interfaces/...
That article develops ideas in a literary/cultural direction. I'd like to have a bit more information on the algorithms Coetzee was using.
Christopher Strachey in the early 1950s was using a program with combinatorial word choice to produce 'love poems'. He posted the printouts on notice boards and they were signed M.U.C for Manchester University Computer (Ferranti Mark 1). That seems to be a simple choice from lists subject-verb-object type thing.
Coetzee writes a bit about his early programming career in his autobiographical novel Youth.
https://en.wikipedia.org/wiki/Youth:_Scenes_from_Provincial_...
For me, Brian Westley is the greatest code poet of all. His poem was a winner in the 1990 round of the International Obfuscated C Code Competition under "Best layout - poetic exchange between lovers":
https://raw.githubusercontent.com/ioccc-src/winner/refs/head...
Love is a not a toilet
From: https://www.ioccc.org/1990/westley/index.html
This is pretty astonishing and also very fun to read.
Some previous discussion:
2017 https://news.ycombinator.com/item?id=14776042
Thanks! Macroexpanded:
The computer poetry of J. M. Coetzee’s early programming career - https://news.ycombinator.com/item?id=14776042 - July 2017 (15 comments)
For your consideration: Waka Waka Bang Splat [1]
[1]https://spot.colorado.edu/~sniderc/poetry/wakawaka.htmlGlorious. This is the first time in 40 years of programming that I've encountered the term "waka waka" for the "<>" pair. My google-fu failed me when I tried to look it up today, perhaps due to search pollution generated by the Cameroonian/Shakira song.
The nature of the output displayed
makes me think that it's a sort of mad-libs phrase generator. The leading adjective "PENSILE" appears twice, as does "CATAMITE", and "PARACLETE" occurs three times, as does the noun "EIDOLA", once as subject and twice as object. Hopefully Dr. Roach will be able to present us a full reverse-engineering of the algorithm, but I'm guessing you could probably infer it just from that page of output: f'{random.choice(adjectives)} {random.choice(nouns)} {random.choice(verbs)} THE {random.choice(nouns)}', which produces lines such as STYPTIC PARACLETE TAMP THE PRESBYOPIA.Before LLMs, I found this a very enjoyable activity; my similar program at http://canonical.org/~kragen/sw/dev3/dramaticphrase.py outputs lines such as:
The comments of the program contain a somewhat embarrassing sonnet I wrote with its help.I also wrote a Spanish version at http://canonical.org/~kragen/sw/dev3/frasedramatica.py, which says things like:
Now, of course, this sort of thing is much less novel and therefore less entertaining. GPT-5 will be happy to write as many sonnets for you as you want, as long as it doesn't somehow get the idea that they might infringe some kind of copyright or teach you how to do your own electrical work.I've often wondered if J. M. Coetzee was related to Wikipedia editor Derrick Coetzee with whom I used to collaborate on computer science articles.