Classicery · 005B · 2026
The fifth room, and the last. Two doors open onto it: the hinge of the instruction room, and Volume Three’s warning to keep your hands on the curve. This is where the machine no longer fills a gap or waits for your instruction, but makes the thing itself — the one no hand drew. Perlin’s posture: of course it’s fake, it’s all fake, it’s computer graphics. Generation is synthesis, unapologetic.
Philosophical Transactions of the Royal Society, 1952 · reaction–diffusion
The most important rhyme in the wing, by a man you met at the bottom of the last room. Two chemicals, a few rules, diffusing at different rates — and spots and stripes appear on their own. The pattern computes itself. In 1936 Turing described a machine that does only what its description says; in 1952, a process that makes a form nobody specified. Instruction and generation, one mind, sixteen years apart.
Lindenmayer, 1968 · The Algorithmic Beauty of Plants (w/ Prusinkiewicz, 1990)
A biologist’s grammar for growth. Start with a seed symbol, write a few rewriting rules, apply them over and over — and read the growing string as drawing commands. Out comes a fern, a weed, a tree, branching with the logic of something alive. You do not draw the tree. You write the rule a part grows by, and the tree is what the rule does when you stop holding it back.
Perlin noise, 1983 · “An Image Synthesizer,” SIGGRAPH 1985 · Academy Award 1997
You met Perlin in Volume One as a technique; here he is a manifesto. Challenged that his textures were not real physics, just faked, he agreed and went further: of course it’s fake, it’s all fake, it’s computer graphics. The rule is not a model of reality — the rule is the material. That shrug is the permission the whole generation arc runs on. The question underneath it: when the rule does the making, whose making is it?
“Flocks, Herds and Schools,” 1987 · separation · alignment · cohesion
Three local rules per bird, no choreographer, and the flock emerges. The motion room already runs the field live — go watch it. Here it is the template for everything: author the system, not the outcome, then let it run and accept what comes. The plant grammar, the bred image, the model at the far door are all boids in disguise. The cost folded in: you give up placing any single bird for a richness no direct control could reach.
Conway, 1970 · cellular automata · the purest “what is generation” object
The purest specimen of generation, stripped bare. A grid, and four rules about counting your neighbors — loneliness, survival, crowding, birth. Set down a few cells and shapes appear that hold still, that blink forever, that walk. Nobody designed the thing that walks. The rules were specified completely; the behavior was not, and it is endless. The maker who authors a system is always a little a naturalist of their own rules.
Adaptation in Natural and Artificial Systems, 1975 · variation · selection · iteration
When you can’t write the rule for a good design, let bad ones compete and breed. Holland turned evolution into a method: a population of random attempts, mutated and recombined, scored, the winners breeding the next round, for thousands of generations. What comes out was found, not made. The human specifies only the space of the possible and the measure of good — and the machine gives you exactly what you asked for, not what you meant.
The Blind Watchmaker, 1986 · the human eye as the fitness function
The hardest things to want can’t be put in a formula — so let the definition of good be a person, looking, and pointing. Dawkins grew branching figures from a few “genes,” showed their mutated offspring, and let you pick the one you liked to breed the next litter. Bred by his eye alone, the figures drifted into insects, into creatures he had not designed. You author nothing. You choose. And from choosing, specific form emerges.
“Artificial Evolution for Computer Graphics” 1991 · Genetic Images, Pompidou 1993 · “Evolving Virtual Creatures” 1994
Sims took aesthetic selection out of the demonstration and made it art. In 1991 he bred images from mutating equations; in 1993 the Pompidou public bred them by where they looked; in 1994 he evolved creatures, body and motion both. Keep those apart — images in ’91/’93, creatures in ’94. And here the room’s anxiety is born: when the machine makes and you only choose, are you the artist, or the audience? Carry it to the end.
Rosenkrantz & Louis-Rosenberg, founded 2007 · Kinematics Dress, 2013, in MoMA
Generation stops being research and becomes a living practice. Nervous System writes programs from natural processes — differential growth, branching, plant morphogenesis — and fabricates the output. The Kinematics Dress (2013) was grown as one folded form and printed pre-assembled; it is in MoMA. And the two arcs are one career here: a co-founder came out of the parametric Gehry world the last room mapped. Instruction and generation, end to end, in one studio.
Elite, 1984 · the everyday form of “author the system”
The most-encountered generation in the world is in games, and most players never think of it as the same idea. Write a generator, give it a seed — one number — and the rules unfold a galaxy, a terrain, a world. Elite (1984) stored the seed, not the systems, and regrew each star system identically on arrival. No human placed the mountain, yet it feels designed, because the rules were. The hand is everywhere and on nothing.
Generative adversarial networks, Goodfellow et al., 2014
In the breeder movement the human held one last job — the eye that judged. GANs take that too. A generator makes; a discriminator judges real from fake; they train each other until the faker fools the critic. The fitness function is now automated. Both halves of the breeder’s loop are inside the machine, and the maker has stepped back one more rung — a system that makes what no hand drew and decides for itself whether it’s any good.
Sohl-Dickstein et al. 2015 → Ho, Jain & Abbeel 2020 → public text-to-image 2022
Make an image by first destroying one. Add noise to a picture until it is pure static, then train a machine to undo one step of the destruction — and it can walk the whole process backward, conjuring a form out of static, steered by a sentence. The deepest version of the gap-fill Volume Three began with: there the machine filled between two real frames; here there are no frames, only noise and a sentence, and the gap has widened until it is the entire image.
Transformers · “Attention Is All You Need,” Vaswani et al., 2017 · into language, into image
The two doors turn out to be one. The 2017 transformer learned the shape of language, and the same machinery learned the bridge from language to image. So when you describe what you want and a machine generates it, you are at the top rung of the telling arc and the far end of the making arc at once — the same act, seen from the two rooms it took to understand it. Hopper’s dream and the second Turing’s dream, answered by one sentence in one box.
Classicery · 005B · the wing’s close
Volume Three said keep your hands on the curve; this is where your hands went. They did not leave. Nothing on this ladder is not human — it is human all the way down. What changes, rung by rung, is not how much of the making is yours but how directly you reach the thing, and through how many other hands your wanting travels. Taste is not a consolation prize. It is the work, at a remove. You are still the one who wants it. Wanting it is the making.
Generation is the fifth room in Classicery’s theory wing — From the Literature, Volume Five, and the wing’s close. It is the room two doors open onto: the natural-language hinge of Volume Four, and Volume Three’s warning to keep your hands on the curve. Here the machine does not fill a gap or wait for an instruction. It makes the thing itself.
Fifteen tracks, from Turing’s 1952 morphogenesis — the same man who wrote the universal machine in Volume Four — through grammars that grow plants, flocks no one drew, evolution and aesthetic selection, generation as a shipped design practice, and the machines that learn to make. The arc, in one line: from drawing the thing, to describing it, to breeding it, to asking for it.
The room makes its own backdrop: a live generative field that grows and branches as you cross the volume, and freezes to a single still bloom under reduce motion — legible, and itself a generated artifact. The coda answers Volume Three’s curve, and the whole wing’s question: when the making travels through more hands than you can name, what is still yours.
The wing’s refrain — that works fine in practice, but what about the theory? — is an orphan. Trace it and it leads nowhere clean: it has been pinned on a Chicago economist, an Irish statesman, a South Carolina senator, a Heidelberg professor, and no one in particular, with the earliest strong sighting in a magazine in 1911 that named none of them. It is quoted so freely that no one knows who said it first — which is the joke’s last turn, and why it belongs over this wing. A line that works fine in practice, with no settled theory of where it came from. Its origin remains, fittingly, a question of theory.
Drafted by Claude with Phil Renato, 2026. No invented quotations from any real person, living or dead; living authors are paraphrased and quoted only from the record. Every date and attribution passed a two-pass fact-check. Audio is read by a single narrator; tracks without audio yet show a PENDING state. CART 005B / 027 · THEORY WING · FROM THE LITERATURE.