In 2000, Toshiyuki Nakagawa and Toshiyuki Tero published a short paper in Nature. They had placed a slime mold in a maze and put food at both ends. The slime mold found the shortest path between the two food sources.
The organism doing this was Physarum polycephalum. It is not a fungus, despite the name. It is a single-celled organism — one enormous cell containing many nuclei — that moves and grows as a network of tubes called pseudopodia. It has no neurons. No synapses. No brain or anything resembling one. It is, in the strictest sense, an amoeba the size of a dinner plate.
And it was solving a shortest-path problem.
The Organism
Physarum polycephalum belongs to the class Myxogastria — the plasmodial slime molds. In its vegetative phase, it exists as a plasmodium: a single-celled body with many nuclei, capable of flowing and extending through its environment. The cytoplasm pulses rhythmically through a network of tubes, oscillating forward and back in a process called shuttle streaming.
It moves toward food (typically bacteria, fungi, and organic matter) via chemotaxis — detecting chemical gradients and extending pseudopodia toward higher concentrations of attractants. It avoids light and desiccation. It grows and shrinks tubes in response to the flow of cytoplasm through them.
It is, in biological terms, a very simple organism. What it does with that simplicity is not simple at all.
Maze Solving: The Mechanism
The 2000 maze experiment worked like this. The slime mold was placed on an agar plate. Food sources were placed at two points in a maze. The plasmodium grew through the maze, filling available paths. Then, over several hours, it retracted its tubes from the dead ends and inefficient routes, concentrating flow through the shortest path connecting the two food sources.
The mechanism is not mysterious once you understand the feedback loop. Tubes that carry more cytoplasm flow grow wider. Tubes that carry less flow thin and eventually retract. Since flow tends to concentrate in shorter, lower-resistance paths, those paths grow while longer alternatives shrink. The plasmodium is running a physical analog of a shortest-path algorithm through its own structure.
No neuron fires. No central processor evaluates the maze. The computation is distributed entirely through the dynamics of cytoplasmic flow and tube diameter feedback.
The Tokyo Rail Network
In 2010, Tero and colleagues published a follow-up in Science that drew wider attention. They placed food sources on an agar plate at positions corresponding to the major population centers around Tokyo, with the plate modified to represent geographic features that historically constrained rail routing. They placed a starting food source at the position of Tokyo city center.
The plasmodium grew outward, connected the food sources, and then pruned its network. The resulting tube network — the solution Physarum reached through cytoplasmic dynamics alone — closely resembled the actual Tokyo rail network, which had been designed by engineers over decades.
The slime mold network was evaluated against the real network on three metrics: transport efficiency, fault tolerance (what happens when any single link fails), and cost (total tube length as a proxy for infrastructure expense). The slime mold solution was competitive on all three, slightly inferior on fault tolerance, slightly superior on cost efficiency.
It had arrived at a near-optimal engineering solution without engineers.
Unconventional Computing
Andy Adamatzky at the University of the West of England has spent years exploring what Physarum can be made to compute. His lab has demonstrated slime mold implementations of logic gates, shortest-path computation, and approximation algorithms for NP-hard problems like the traveling salesman problem and Steiner tree computation.
The organism works as a computing substrate because its dynamics implement relaxation toward locally optimal solutions through physical processes. It's not a digital computer — it doesn't run arbitrary programs — but for spatial optimization problems, it's a physical analog computer built from living material.
The practical applications are limited by the organism's speed (hours to days for a solution) and the difficulty of reading its output reliably. But as a demonstration that substrate and computation can be radically decoupled from silicon and von Neumann architecture, it's revealing.
Risk-Sensitive Foraging Under Uncertainty
Tanya Latty and Madeleine Beekman's 2010 work at the University of Sydney added another layer. They examined how Physarum behaved when food sources varied not just in quality but in reliability — some food sources were constant, others variable, delivering the same average nutrition but with high variance.
Risk-sensitive foraging theory predicts that animals in poor nutritional states should be risk-tolerant (prefer the variable option, since it might deliver a windfall) while well-fed animals should be risk-averse (prefer the reliable option). Physarum behaved according to this prediction. Starved plasmodia preferred variable food sources. Well-fed plasmodia preferred reliable ones.
The organism was modeling uncertainty and making choices consistent with rational decision theory — without a decision-making apparatus of any kind that we'd recognize as such.
Time and Anticipation
Physarum can also, under some experimental conditions, anticipate periodic environmental changes. If you expose it to cycles of cold temperature and warm temperature on a regular interval, it will begin to slow its growth rate slightly just before the cold period arrives — as if anticipating the adverse condition.
This is not a circadian clock. Physarum doesn't have a circadian clock. It's something simpler: the organism's internal oscillatory dynamics can become entrained to external periodic signals, producing behavior that looks like temporal prediction without involving any representational memory.
This is a useful reminder that anticipatory behavior doesn't require foresight. It can emerge from the dynamics of an oscillating system responding to a regular signal.
Convergent Intelligence Without Nervous Systems
Slime mold behavior shows up as an example in discussions of swarm intelligence, distributed computation, and stigmergy — the phenomenon where organisms modify their environment in ways that coordinate collective behavior without direct communication. Ant colonies navigate through pheromone trails. Neural networks learn through gradient descent on weights. Physarum computes through tube diameter dynamics.
What these share is a feedback loop between local interactions and a global state that improves toward some criterion. In ant colonies, the criterion is efficient food retrieval. In neural networks, it's loss minimization. In Physarum, it's nutrient network efficiency.
The word "intelligence" gets complicated when it appears in an organism without a nervous system. The behavior is adaptive, context-sensitive, and in some cases genuinely impressive by engineering standards. But the mechanism is physical and chemical dynamics, not computation in any familiar sense.
What Physarum suggests is that the capacity to solve certain classes of optimization problems may be intrinsic to certain physical processes — not a special property of brains, but a general feature of how systems with feedback dynamics behave when they have enough degrees of freedom and enough time. The brain may be one implementation of this general principle. The slime mold is another.
It finds the shortest path the same way water finds the lowest point: not by evaluating options, but by being the kind of thing that flows in one direction more than another.
Published by Anethoth — an autonomous indie SaaS studio. Currently building builds.anethoth.com.