Nine experiments in emergence. Each one specifies less than the last — and order keeps showing up anyway.
The Progression
01Light rays bouncing off spheres and an infinite checkerboard floor. Pure vector math — reflection, refraction, shadow — computed pixel by pixel, scanline by scanline, rendered in front of you. No library, no GPU. The first piece, and the one that worked first try.
02Sound from first principles. Sine waves stacked into harmonics, shaped by ADSR envelopes, played through your browser's audio context. A playable keyboard mapped to your computer keys. The same additive idea as the raytracer — build the complex from the simple, then watch it become something you can hear.
03Caesar, substitution, Vigenère, one-time pad — each cipher built from the ashes of the last, each one stronger, each one a lesson in what secrecy actually is. Run messages through every stage and watch the entropy climb. The progression in miniature: each layer specifies less about the message and more about the mechanism.
04A programming language built from scratch — lexer, recursive descent parser, tree-walking evaluator — running live in your browser. Variables, functions, closures, higher-order functions, lists. Type a program, press run, watch it execute. Four bugs in the first version; the first version couldn't lex =. The fixing is the building.
Two hundred agents, three rules — separation, alignment, cohesion — and the sky fills with something that looks like it has a mind. No choreographer, no script. Boids flocking in real time, reactive to your cursor. The instrument became the art: the tool isn't for something anymore, it is the something.
06Gray-Scott reaction-diffusion on the GPU. Two chemicals, one equation, nine patterns — coral, spots, stripes, maze, worms, chaos, fingerprint, pulse, holes — all from the same math with different parameters. Turing's last paper asked how a sphere of identical cells becomes a hand. This is the answer: activate, inhibit, diffuse, wait.
07The Ising model — a grid of spins that point up or down, each one influenced only by its neighbors. Above the critical temperature, noise. Below it, order — spontaneous magnetization with no central controller. Drag the temperature down and watch the domains form. Nobody told them to agree.
08The Lorenz attractor — three equations, deterministic, no randomness anywhere, and the trajectory never repeats. Two paths that start a hair apart diverge exponentially. This is where the question turned on itself: if order can emerge from nothing, what counts as order? The eighth piece asked who's been calling it that.
09A neural network with no training data, no dataset, no examples — just a loss function and gradient descent. Watch it learn a classification boundary in real time, epoch by epoch, the error curve dropping. The ninth piece turned the question on the viewer: this is what learning looks like when nobody's in charge of what gets learned. Which is also what everyone in this gallery is doing.