Evolution and Arithmetic

Chapter 13: Shape

← A Thousand RobotsWhat the Field Can Say →

What the Robots Didn't Have to Do

The Kilobots were identical. Every unit was the same robot running the same program, able to fill any spot in the final shape. The cells that build an ear aren't identical. They have to become different kinds of cell doing different jobs: cartilage cells, skin cells, the cells of blood vessels and nerves. They aren't built alike and they don't behave alike.

The Kilobots were told in advance exactly what shape to make. The engineers loaded a picture of the target into every robot's memory, an outside blueprint handed in from beyond the system. Nothing hands a developing ear a picture of what it's supposed to look like. Whatever "knows" the shape, if anything does, has to be built in from the inside.

The Kilobots built a flat outline on a tabletop, a silhouette one robot deep. An ear is a three-dimensional structure that curls, folds, and thickens unevenly across its surface, with fine details, the particular whorl of the rim and the shape of the lobe, that have to come out right in three dimensions at once.

And the Kilobots had help. People arranged them in their starting cluster, loaded the picture, switched them on, and watched. A developing ear gets none of that. Nobody checks its progress halfway and nudges a stray cell back into place. It has to come out right, unsupervised, the first time, in every embryo, using cells that can't see the overall arrangement any more than a single Kilobot could see the swarm.

Then there's the number of participants. The swarm had 1,024 robots. Researchers trying to grow replacement ears in the laboratory reckon on about a quarter of a billion cartilage cells for a child-sized ear, and that's the cartilage alone, before the skin that covers it. A full-grown ear is built from hundreds of millions of cells, possibly more.

The method may well be the same whatever the number, each cell responding to its immediate neighbors. But the scale is worth sitting with. A curl in the rim of an ear isn't one cell making one elegant turn. It's an unbroken relay of countless cells, each building on exactly where the last left off, until the curl is finished and correct.

A thousand identical robots, given the answer in advance, building a flat outline, with people setting them up and watching: that was already hard. An ear asks for different kinds of cell, with no answer given in advance, building a folded three-dimensional shape, unsupervised, correctly, every time. It isn't a bigger version of the same problem. It's harder in nearly every way at once.

Last revised 2026-11-19

← A Thousand RobotsWhat the Field Can Say →