Memory Without a Brain: How a Single Cell Uses Slime to Remember the World

1. The Tokyo Rail Network and a Brainless Engineer

In January 2010, *Science* published a paper that left a group of Japanese and British scientists in a kind of philosophical bewilderment.

They placed a slime mold called *Physarum polycephalum* in a petri dish, set out oat flakes at the geographic positions of Tokyo and surrounding cities—Tokyo, Chiba, Yokohama, Kawasaki, Omiya, Tsukuba. Then they turned off the lights and waited.

After 26 hours, the slime mold had retreated from the empty areas between the oat flakes, leaving behind a network of interconnected tubes. This network connected all the "cities" with near-minimal total length and near-optimal fault tolerance—and when overlaid on the Tokyo rail system map, the two topologies were strikingly similar.

Japanese railway engineers spent decades designing that system. An organism with no brain, no neurons, no eyes—not even, strictly speaking, a single cell in the conventional sense—computed the same answer in 26 hours.

This is not an inspirational story about "slime molds being smart." It is a question about what intelligence actually is.

2. One Cell, Two Square Meters, Five Centimeters Per Hour

Meet the protagonist. *Physarum polycephalum* is a slime mold, taxonomically closer to amoebas than fungi. Its most peculiar feature: its entire body is one cell. Not "a cell that is small" like a bacterium, but "a cell that can be huge"—in its plasmodial stage, this single cell can grow to two square meters, containing millions of nuclei sharing one mass of cytoplasm.

It moves at about 5 centimeters per hour, crawling through rotting logs and leaf litter, engulfing bacteria, fungal spores, and decaying plant matter. No mouth, no stomach, no limbs—just a flowing blob of protoplasm that wraps around food, digests it, and keeps crawling.

Over the past two decades, this organism has repeatedly challenged biologists' definition of "cognition."

3. A Single Cell in a Maze

The story begins in 2000. Toshiyuki Nakagaki of Hokkaido University cut a *Physarum* into pieces and scattered them in a plastic maze. The fragments grew, explored, and reconnected, eventually filling the maze.

Then Nakagaki placed nutrient-rich agar blocks at the maze's entrance and exit.

Four hours later, the slime mold had withdrawn from every dead end, keeping only the tubes along the shortest path connecting the two food sources.

The result unsettled researchers. Maze-solving is a classic search problem in computer science—you try paths, remember where you've been, backtrack, compare. The slime mold has no concepts of "trying" or "backtracking"; it only grows. But the way it grows is equivalent to running a breadth-first search on all possible paths simultaneously and keeping the shortest one.

More puzzling follow-ups: when oat flakes were placed at nodes of highway networks in Tokyo, Canada, the UK, and Spain, the slime mold consistently rebuilt networks highly similar to human-designed transport systems.

It doesn't know where Tokyo is. It doesn't know what a "city" is. It doesn't even know it's in a petri dish. Yet it solved the same optimization problem.

4. Slime as Memory

In 2012, Chris Reid's team noticed a detail: *Physarum* leaves a translucent slime trail as it crawls—and while foraging, it avoids its own slime.

That sounds like a simple instinct: "don't take the old road." But Reid proposed a bolder interpretation: the slime trail is its memory.

How is memory implemented in nervous systems? Roughly, synaptic connection strengths between neurons are modified, forming a pattern that can be reactivated later. You need a storage medium (neurons), a write mechanism (learning), and a read mechanism (recall).

The slime mold's trail satisfies all three:

• Storage medium: the slime itself, deposited in the environment
• Write mechanism: automatic secretion while moving
• Read mechanism: behavior change upon encountering slime (avoidance)

This isn't "like memory"—it is memory, except the storage medium is outside the body.

This "externalized memory" has a strange property: it requires no internal state. The slime mold doesn't need to "remember" where it has been, because the environment remembers for it. It only needs to read the environment to know where it shouldn't go.

5. Memory Can Be Overwritten

In 2013, Reid's team ran a more elegant experiment. They placed the slime mold at the entrance of a Y-maze. One arm led to ordinary food; the other to high-nutrition food (egg yolk). But the path to the nutritious food was covered with the slime mold's own trail.

If "avoid slime" were a fixed reflex, it should refuse that path. Instead, the slime mold crossed its own slime to reach the food.

This is crucial. It shows the trail is not a simple "no entry" sign but a memory signal that can be overridden. When the environment offers sufficiently valuable new information, old memories get rewritten.

This is exactly what distinguishes memory from reflex: memory is integrable. It doesn't mechanically trigger a behavior; it enters a "decision system" together with other information, producing a synthesized judgment. The slime mold's decision system has no neurons, but what it does—weighing, comparing, overriding—is not fundamentally different from what neurons do.

6. Habituation: A Single Cell Gets Used to Things

In 2016, a CNRS research team ran another experiment. They had *Physarum* cross a "bridge" soaked in a bitter but harmless substance—quinine or caffeine—to reach food.

At first, it was very reluctant to cross. But after six days of exposure, its crossing speed returned to the same level as the control group. It had "gotten used to it."

After two days without exposure, it reverted to its initial wariness. Moreover, slime molds habituated to caffeine remained wary of quinine, and vice versa. The habituation was substance-specific.

In psychology textbooks, habituation is defined as the simplest form of learning. When a human baby is curious about a new toy and loses interest after a few days—that's habituation. The slime mold does the same thing, with entirely different molecular machinery, in a cell without neurons.

7. Another Intelligence Underground

If slime molds are "philosophers among single cells," mycorrhizal fungi are "economists underground."

In February 2025, a 28-person team from Princeton University and VU Amsterdam published in *Nature*, using purpose-built robotic microscopy to observe fungal network behavior continuously for three years. These fungi weave vast mycelial networks between plant roots and soil—each hypha about one-tenth the diameter of a human hair, but with total length spanning entire forests underground.

They found three things that make us rethink "intelligence."

First, fungi prefer future opportunity over short-term gain. Growing tips extend outward like pathfinders, investing in network expansion even when no immediately tradable resources are nearby. Such "delayed gratification" is, in animal behavior, a complex decision requiring nervous systems.

Second, nutrients flow bidirectionally. In the same tube, sugars flow from plant to fungus while phosphorus and nitrogen flow from fungus to plant—simultaneously. The team traced roughly 100,000 flow trajectories and found fungi adjust flow based on local demand—widening tubes and increasing flow near plant roots, like a city traffic system opening more lanes at rush hour.

Third, decisions are decentralized. No "central processor" decides how the network should grow. Each growing tip decides based on local conditions—grow toward food, fuse with its own hyphae, detour around obstacles. But these local decisions aggregate into an efficient, fault-tolerant, adaptive global network.

This system has been running for roughly 500 million years. It stores 13 billion tons of CO2 annually—more than a third of global fossil fuel emissions—and supports the roots of almost all land plants. And it does all this without a single neuron.

8. Where Does Memory Actually Live?

Now back to the sleep-depriving question.

Traditional cognitive science carries a hidden assumption: memory lives in the brain, and learning is done by neurons. The assumption is so entrenched that when we say "memory," we almost default to meaning "something in the brain."

The slime mold and fungi stories break that assumption.

If memory can be stored in slime trails, then memory is not a specific biological structure but a functional pattern—any system that can record past states and influence future behavior can be a memory. Neurons are just one way of implementing this function, not the only way.

Philosophers have a concept called "Extended Cognition": cognitive processes are not confined within the skull but extend into the body and environment. When you do multiplication with pen and paper, part of the computation happens in your neurons, part on paper. The pen and paper are part of the cognitive system, not external tools.

The slime mold pushes this idea to the extreme. Its "cognitive system" is almost entirely outside its body—its body is merely an actuator; its real "memory" is in the slime it has crawled through; its real "sensors" are the cell membrane touching the environment. It is a cognitive system that is almost fully externalized.

Abstract as this sounds, it has a very concrete implication: our definition of intelligence may be too narrow.

9. Implications for AI

Almost all of today's AI is built on the "neural network" paradigm—inspired by the brain: artificial neurons, weighted connections, backpropagation learning. The field's success seems to prove that "neuron-style architecture is the right implementation of intelligence."

But slime molds and fungi suggest another possibility.

*Physarum* solves mazes and rebuilds transport networks not with weight updates but with physical growth. It expands in all directions simultaneously, then contracts based on nutrient gradients—a continuous, parallel, physically embedded optimization process. It has no separation between "training" and "inference"; it is always learning, always acting.

Mycorrhizal fungi solve resource-allocation problems not with backpropagation but with decentralized local decisions. Each hyphal tip knows only its neighborhood, yet the whole network exhibits globally optimal transport strategies. This is closer to ant colony optimization or particle swarm algorithms—but running in an organism that has been "optimizing" this algorithm for 500 million years.

This doesn't mean neural networks are wrong. But it reminds us: intelligence may have many possible implementations, and we have seriously explored only one.

Imagine an AI whose "memory" is not stored in a weight matrix but etched into the environment it operates on—an agent leaving traces in a filesystem, its "experience" being the path patterns it leaves behind. Imagine an AI whose decisions emerge not from a central model but from large numbers of simple local agents—each seeing only its neighborhood, yet collectively exhibiting complex behavior.

This isn't science fiction. Slime molds and fungi have spent 500 million years of evolution proving these approaches work. We just haven't been listening carefully.

10. A Quiet Conclusion

*Physarum polycephalum* has no consciousness. It doesn't think about who it is, doesn't feel pride in the elegance of the Tokyo rail network, and experiences no inner struggle when crossing its own slime for egg yolk. It is a single-celled organism doing what single-celled organisms do—crawl, engulf, flow, retreat.

But its behavior satisfies every functional criterion we use to define "memory" and "learning." It records the past, modifies the future, weighs information, adapts to its environment.

This leaves an uncomfortable question: if memory doesn't need neurons, learning doesn't need a brain, and decision-making doesn't need consciousness—what are these words actually describing?

Perhaps they describe not a specific biological structure but a relational pattern—a sustained conversation between a system and its environment, through some medium (slime, synapses, hyphae, silicon), about "what happened" and "what to do next."

Neurons are one carrier of this conversation. Slime is another. Hyphae are a third. Perhaps there are fourth and fifth kinds we haven't yet learned to see.

Next time you step over a yellow, glistening blob on a rotting log in the forest, pause for a moment. That is not a simple organism crawling aimlessly. That is a memory system without a brain, remembering the forest in its own way.

---

References:

• Nakagaki et al. (2000). Maze-solving by an amoeboid organism. *Nature*.
• Reid et al. (2012). Slime mold uses an externalized spatial memory to navigate in complex environments. *PNAS*.
• Reid et al. (2013). The mechanism of flow-based optimization in *Physarum*. *Journal of Physics D*.
• CNRS (2016). A single-celled organism capable of learning. *ScienceDaily*.
• Tero et al. (2010). Rules for biologically inspired adaptive network design. *Science*.
• Oyarte Gálvez et al. (2025). A travelling-wave strategy for plant-fungal trade. *Nature*.
• Sims, M. (2025). Memories without brains. *Aeon*.