This organism is grown in real time and needs WebGL2 with float render targets — which this browser did not offer. Try a current desktop browser to watch the network form.

XAVIER FABLE ×70
Physarum polycephalum · many-headed · no neurons

PHYSARUM

the intelligence of slime

A single cell, no brain, no plan. Below, half a million synthetic filaments sense a chemical trail, steer toward the strongest scent and lay down a little of their own — and out of that one rule, living transport networks braid themselves into being.

agent-based model webgl2 · ping-pong fbo self-organising networks
MOVE to feed it  ·  CLICK to drop food
scroll
01the maze

No brain. It still solves the maze.

In 2000, Toshiyuki Nakagaki grew a mat of Physarum polycephalum across a plastic maze, then set an oat flake at each of two exits. The slime, spread thick through every corridor and dead end, began to withdraw.

Within hours it had abandoned the blind alleys and thinned to a single tube tracing the shortest route between the two foods — the same answer a shortest-path algorithm would return, computed by an organism with no nervous system, no memory cells, no idea it was in a maze at all.

It doesn't search. It floods everywhere at once, then lets the paths that don't pay off starve. Selection, not planning.

1cell, no neurons
≈4hto find the path
2000Nakagaki, Nature
02the network

It redrew the Tokyo rail map.

In 2010, Atsushi Tero's team laid oat flakes across a scaled map of greater Tokyo — one flake for each surrounding city — put the slime on the spot marked Tokyo, and waited.

In about a day it grew a web of tubes linking every flake. Statistically, the result matched the real rail network engineers spent a century designing: comparable total length, comparable fault tolerance, comparable cost. Cut a tube and the flow re-routes.

The team distilled the behaviour into a rule for adaptive network design — tubes carrying more flow thicken, idle tubes wither — now borrowed for road, pipe and data networks.

36cities linked
~26hto grow the map
2010Tero, Science
fig.02 — emergent transport netflow ⇢ thicker tube
03the model

Three sensors. One rule.

Every filament above runs the same tiny program each frame. There is no leader and no map — only local scent and a bias to follow it. Multiply by hundreds of thousands and the networks you scrolled past appear on their own. This is that rule, running live on your GPU.

01 / SENSE

Smell three ways

Each agent samples the trail map ahead-left, straight-ahead and ahead-right, at a set angle and distance.

02 / STEER

Turn toward scent

If the centre reads strongest, hold course. Otherwise rotate toward whichever side smells richer — and if it's a tie, turn at random.

03 / DEPOSIT

Leave a trail, decay

Move forward, drop a little chemical. Each frame the whole map blurs and fades — so only paths that stay busy survive.

160k
live agents
22°
sensor angle
9.0px
sensor dist
60fps
target loop