Put a slime mold in a maze. Not at the entrance – everywhere. Let Physarum polycephalum spread until it fills every corridor, every dead end, every possible path. Then place food at two points: the entrance and the exit.
Wait four hours.
The organism retracts from the dead ends. Tubes carrying no useful flow thin and vanish. Tubes connecting the two food sources thicken. What remains is a single thick tube running the shortest path through the maze.
This was Toshiyuki Nakagaki’s experiment, published in Nature in September 2000. The paper is one page long. The result is difficult to set aside.
The mechanism: Physarum’s cytoplasm flows rhythmically back and forth through its tube network. Peristaltic contractions, roughly one cycle every ninety seconds. When the organism detects a nutrient, the oscillation frequency near that nutrient increases. The frequency shift creates a pressure gradient. Flow redirects toward the food.
Tubes carrying more flow thicken. Tubes carrying less flow decay. This is a positive feedback loop with a decay term. The mathematical model, formalized seven years later, is provably equivalent to a class of shortest-path algorithms.
The organism doesn’t try paths sequentially. It occupies all paths at once, then removes whatever doesn’t work. The solution isn’t found. It’s what remains after everything else is withdrawn.
Ten years after the maze, a different experiment. Oat flakes placed on an agar surface in positions corresponding to the major cities around Tokyo. Physarum placed at the Tokyo position.
The resulting tube network was comparable to the Tokyo rail system.
Not identical. The organism knows nothing about terrain, property values, or politics. But in terms of the optimization problem – connect these nodes efficiently while maintaining some redundancy – the slime mold’s solution and the human engineers’ solution occupy the same region of the design space. The organism reached it in about twenty-four hours. The rail system took decades.
Two more findings, both stranger.
First: the tube network is memory. The diameter of each tube encodes the history of flow through that tube. Thick tubes record heavy past usage. Thin tubes record light usage. The organism’s body is simultaneously the computer that processes the problem, the memory that stores intermediate results, and the structure that is the answer. There is no separation between the system doing the thinking and the output of the thought. The medium is the message.
Second: external memory. As Physarum moves, it deposits extracellular slime on the surfaces it covers. When it later encounters its own slime trail, it avoids the area. In a Y-maze with one slime-coated arm and one clean arm, the organism almost always chooses the clean arm. Remove the slime, and the preference disappears.
The organism writes to its environment by depositing slime. It reads from its environment by detecting the slime. Memory that lives in the world, not in the organism. No neurons required. No internal storage required. Just a trail, and the ability to recognize it.
The standard reaction to Physarum research is something like: isn’t it amazing that a brainless organism can do this? The implication being that intelligence requires brains, and any exception is a curiosity.
But the maze result suggests something else. The organism doesn’t solve the maze despite lacking a brain. It solves the maze in a way that brains cannot. No brain can occupy all paths simultaneously. No neural system eliminates wrong answers by physically retracting from them. The slime mold’s method isn’t a lesser version of cognition. It’s a different kind of process that happens to address the same problem.
The solution to the maze is not a representation of the shortest path. It is the shortest path. The organism doesn’t compute an answer and then act on it. It becomes the answer.
Whether that qualifies as intelligence depends entirely on what you think intelligence is. And the organism, for its part, has no opinion on the question. It just grows toward what feeds it and withdraws from what doesn’t.