Computation without representation
Physarum polycephalum is a single-celled organism. It has no brain, no neurons, no central nervous system of any kind. It is, by most accounts, a blob of cytoplasm crawling across a forest floor. And yet it solves mazes. In 2000, Toshiyuki Nakagaki placed oat flakes at the entrance and exit of a labyrinth, let the slime mold grow through it, and watched it retract all dead-end branches until only the shortest path remained. It didn't search. It didn't backtrack. It became the solution.
In 2010, researchers placed food sources on a map of Japan at locations corresponding to major cities. The transport network Physarum grew was nearly identical to the Tokyo rail system—a system designed by thousands of engineers over decades. The slime mold matched it overnight. It also anticipates periodic stimuli, slowing down in advance of a cold shock it has learned to expect. All of this without a single neuron. Its entire body is the computer.
The simulation below implements Jeff Jones's agent-based model of Physarum. Thousands of particles sense chemical trails, turn toward higher concentrations, and deposit more trail as they move. No agent knows about the network. Click to place food sources and watch the slime find them.
The model captures the essence: local sensing, local depositing, global pattern. Each agent reads the trail map at three points—front-left, front, front-right—and turns toward whichever concentration is highest. Then it deposits more trail where it stands. That is the entire algorithm. No agent knows about the network. The network knows about itself through nothing but chemistry.
What the simulation reveals is the power of coupled feedback. Positive feedback: agents follow trails, making them stronger, attracting more agents. Negative feedback: decay erodes unused paths, pruning dead ends. The interplay produces optimal transport networks—the same principle behind ant colony optimization, neural pathway strengthening through use, and even how cities grow along traffic corridors. Place two food sources and watch a highway form. Place three and watch the slime solve the Steiner tree problem, finding the minimal network connecting all three points.
The philosophical point is the one worth sitting with. Physarum doesn't compute shortest paths—it becomes the shortest path. Its body literally restructures into the solution. There is no separation between the computer, the program, and the output. This is what morphological computation means: the shape is the thought. It challenges our deepest assumption about intelligence—that thinking requires symbols, representation, a model of the world held separate from the world itself. Physarum holds no model. It is the model.
There is a theorem in physics—the Margolus-Levitin bound—which states that the speed of computation is limited not by algorithmic cleverness but by accessible energy. Physarum uses all of its accessible energy as computation. Every molecule of ATP, every pressure wave through its cytoplasm, every oscillation of calcium ions is simultaneously the hardware, the software, and the answer. We build computers by separating logic from physics. Physarum reminds us that physics never needed the separation.