What you are looking at
At the top of this page, a form has grown. It was not drawn, nor picked from an image bank, nor generated by an AI. It was computed, here, in your browser, the moment you arrived — and it will be exactly the same on your next visit.
Every page of this site grows its own. Beneath the figure, a line in small capitals gives the rule that produced it and the seed that sowed it. Nothing is hidden: the formula is displayed beneath its result.
Why fractals
It all began with a sense of wonder, long ago: the wonder of fractals, those infinitely rich universes born of formulas that seem, at first glance, cold and inert. That encounter shaped my early years — pure mathematics, computer science, neural networks long before deep learning, and an artificial-life PhD I never ended up writing.
These figures are therefore not decoration. They are the site's argument, in visible form.
For this is exactly the thesis defended here: from iterated computation, something can be born that looks alive. A Barnsley fern is nothing but four affine transformations drawn at random, fourteen thousand times. None of those four rules "knows" the fern. None contains its shape. The fern emerges from their repetition — as, perhaps, intelligence emerges from relations none of which contains it.
If you find it doubtful that a machine could produce something that resembles us, look at the figure at the top of this page. It is made of four lines of arithmetic. It looks like a fern. It is not one, and yet it is not nothing.
It is also a rule I set myself: no generated illustration for the sake of prettiness. On a site about AI, stock imagery or fashionable generative illustration would have been a slightly shameful convenience. The figures are the imagery. They represent nothing — they demonstrate.
Five families, five rules
The form is not decorative: it tells you what kind of content you are reading, before you have even read the title.
The fern — the articles. The Barnsley fern, an iterated function system: four affine transformations, drawn by lot with very unequal probabilities (85% of the time the same one, 1% for the stem), applied fourteen thousand times to a wandering point. No one draws the leaf. It appears.
The phyllotaxis — the concept entries. Vogel's spiral, the one at the heart of the sunflower: point n is placed at the golden angle, 137.508°, at a distance proportional to the square root of n. Five hundred and forty points, and the densest possible order appears without anything having planned it. At the tips, the points turn to pollen: the flowering.
The snowflake — the series. The Koch curve: each segment is replaced by four smaller segments with a peak in the middle, and the process is repeated four times. One rule, applied to its own result — which is what a series of articles does.
The biomorph — the op-eds. An iteration in the complex plane, z → z³ + c, twelve times, with Pickover's criterion: we do not look at whether the point escapes, but at how. Out of it come things that look like organisms. It is the figure reserved for texts in which I take a stand.
The tree — the thinker profiles. A binary ramification: each branch splits in two, each generation measures 0.72 times the previous one, nine times in a row. A lineage, literally — and the tips flower into pollen.
Always the same, for always the same page
These figures look random. They are not. There is no real randomness on this site.
The page's address — its slug — is passed through a hash function (FNV-1a, 32-bit) that turns it into a number. That number seeds a pseudo-random generator (mulberry32): a sequence of numbers that has every appearance of chance, but is entirely determined by its starting point. Same seed, same sequence, same figure. Forever.
That is why the caption displays the seed next to the rule. With those two pieces of information, the figure is reproducible: it is not a pretty accident, it is the result of a computation you could redo.
What varies from one page to another is modest, and specific to each family: the fern curves slightly differently, the phyllotaxis shifts its angle by a few tenths of a degree, the biomorph changes its constant, the tree opens its branches otherwise. The snowflake varies only by its rotation: up to a turn, all the snowflakes on the site are identical. Better to say so than to leave you searching.
A few technical honesties
The figure is painted on a canvas, in your browser, after the page loads. Consequences, which I prefer to state:
- It does not exist in the page's source code. If you share an article on a social network, the preview will not show it. Nor is it visible to a screen reader — hence the textual description attached to each figure.
- It grows on load, in small batches, because watching the form appear is part of the argument. But if your system declares a preference for reduced motion, it is displayed at once, finished. The rule is respected without argument.
- It is recomputed at every display rather than stored. Fourteen thousand points cost a few milliseconds; keeping the image would have cost more than redoing it.
The heart of the matter
The site could have been illustrated otherwise. The choice of these figures says something the articles take thousands of words to establish.
A simple rule, repeated, without intention or overall plan, produces a form we recognise as alive. That does not prove AI thinks. It only makes it harder to maintain that whatever looks organised must necessarily have been willed by someone.
The garden carries the page. The rule carries the structure. Neither hides.