1. The Tokyo Subway and a Yellow Blob
In 2010, Toshiyuki Nakagaki of Hokkaido University did something that sounded a bit silly.
He drew the positions of 36 major Tokyo stations, to scale, on an agar plate the size of a petri dish, placing an oat flake at each station. Then he placed a small piece of yellow slime mold — *Physarum polycephalum* — at the position of Tokyo Station.
26 hours later, he had a map.
The map looked almost exactly like the Tokyo rail system. Not roughly similar — the topology was highly congruent: the same hubs, the same radial branches, the same redundant links. The Tokyo subway took dozens of engineers decades to optimize.
*Science* named it one of the most interesting discoveries of 2010, and *Physarum polycephalum* walked out of the mycology lab into public view.
But the truly unsettling part is not that it can solve network optimization problems — it's that this organism has no brain.
Not a small brain, not a simple brain — no brain at all. No neurons, no synapses, no hippocampus, nothing you would find in a neuroscience textbook. It isn't even a multicellular organism — it's a single cell, a yellow blob with millions of nuclei floating inside, growing up to 900 square centimeters.
One cell. No nervous system. It produced a transport network that took human engineers decades to optimize.
Something about this doesn't add up.
2. What Is This Organism, Exactly?
*Physarum polycephalum* (the "many-headed slime") belongs to the kingdom Protista — not a fungus, not a plant, not an animal. It lives on rotting leaves, decaying wood, and damp corners, and is distributed worldwide. Walk through a forest, flip over a wet log, and the yellow slime underneath may well be it.
Its most active life stage is the plasmodium — one giant single cell in which millions of nuclei share the same cytoplasm. That cytoplasm is not static: it sloshes back and forth inside the cell on a cycle of roughly 1–5 minutes. The flow is not random: it is driven by an actin–myosin contraction–relaxation cycle — the same protein machinery your muscles use.
The cytoplasm flows through a tubular network of thick and thin tubes forming a branched, vessel-like structure. When exploring, the front expands into a fan; when it finds food, the tube network reorganizes to channel resources toward the food.
It moves at up to 5 cm per hour, eating bacteria, yeasts, and organic debris. Unremarkable-sounding. Yet this organism can solve mazes, solve two-armed bandit problems, remember periodic events, and tell well-fed from hungry conspecifics.
3. Fluid as Information
In 2017, Karen Alim (then at the Max Planck Institute for Dynamics and Self-Organization, now at TU Munich) set out to answer a question: how does *Physarum* actually "think"?
The core mechanism she found is strikingly simple.
When *Physarum* encounters food, the tube walls near the food soften. This is not chemical signaling but a hydrodynamic effect: food stimulus increases local flow speed, and faster flow exerts greater shear force on the tube walls. Under this force the walls thin and the tubes dilate. Dilated tubes carry more flow, further reinforcing the effect — a positive feedback loop.
In other words, information is carried neither by electrical signals (there are no nerves) nor by chemical gradients (too slow), but by the fluid flow itself. Where flow is fast, tubes widen; where flow is slow, tubes shrink and eventually vanish. The network is constantly self-reorganizing, and the direction of reorganization is the content of the information.
It's a bit like your vascular system — muscles demand more blood during exercise and nearby vessels gradually widen. But *Physarum* takes this principle to its extreme: its entire "cognition" is fluid mechanics.
4. Three Kinds of Memory, None Requiring Neurons
In 2021, Alim's team published in *PNAS*: "Encoding memory in tube diameter hierarchy of living flow network." They found that *Physarum*'s memory is encoded in the hierarchy of tube diameters.
Specifically: when *Physarum* encounters food, the tube walls near the food soften with a delay. That delay is crucial — it means the walls retain the information "food was here," even after the food is gone. The degree and duration of softening encode the food's location and quantity.
This is *Physarum*'s first kind of memory: morphological memory. Its body shape is its memory. Thick tubes are "important places," thin tubes are "unimportant places," vanished tubes are "places not worth remembering." You don't need a hippocampus to store a spatial map — your body shape is the map.
The second kind is contraction-pattern memory. *Physarum*'s cytoplasmic flow is a periodic contraction–relaxation, like a heartbeat. Audrey Dussutour of the French CNRS found that these contraction patterns can store information — "similar to how activity waves in the brain can store information." After learning a task, *Physarum*'s contraction rhythm changes, and that change persists into the next task.
The third kind is the most counterintuitive: externalized memory. *Physarum* leaves extracellular slime (ECS) wherever it goes. The slime repels *Physarum* itself — it dislikes re-traversing its own paths. This looks like simple avoidance, but Chris Reid (then at the University of Sydney) and colleagues showed in 2012 that it is genuine spatial memory.
The experiment was elegant: place *Physarum* in a U-shaped trap so it must detour around the U-shaped barrier to reach food. Normally it succeeds, because it avoids its own slime and doesn't re-explore. But if the dish is pre-coated with slime, *Physarum* can't distinguish its own fresh trail from background noise, and it circles endlessly in the U-trap, unable to escape.
In other words, erasing its slime trail erases its memory. This memory is not in the body — it is in the environment. The "extended cognition" theory proposed by philosophers Andy Clark and David Chalmers in 1998 — that cognitive processes need not be confined to the brain but can extend into the environment — finds its most extreme validation in *Physarum*. Its memory system is the very environment it has modified.
5. Learning — and Passing It to Others
Remembering is one thing; learning is another.
In 2016, Boisseau and Dussutour published in *Proceedings of the Royal Society B*. They made *Physarum* cross an agar bridge treated with quinine (a bitter repellent) to reach food. On day one, it clearly hesitated to cross. But after 5 consecutive days, its crossing speed returned to normal — it had habituated.
The key control: *Physarum* habituated to quinine still strongly avoided caffeine, and vice versa. So this was not sensory fatigue or general adaptation but stimulus-specific habituation — behaviorally identical to habituation learning in animals.
Then came a wilder experiment. Vogel and Dussutour (2016) found that if you fuse a quinine-habituated *Physarum* with a non-habituated clonemate, the fused organism shows habituated behavior in unhabituated areas.
Learning transferred directly through cell fusion.
This sounds like science fiction, but in *Physarum*'s world cell fusion is routine. Two clonemates that meet can merge into one larger individual, sharing cytoplasm and nuclei. The "memory" — in this case quinine molecules accumulated inside the cell, acting as a "circulating memory" — is transferred as the cytoplasms mix.
But there's an even stranger twist. In 2022, Smith-Ferguson et al. found that under altered conditions (longer per-exposure durations without repetition), *Physarum*'s response to NaCl was not habituation but sensitization — becoming more, not less, responsive. Same organism, same stimulus, opposite learning outcomes depending on the exposure pattern. *Physarum*'s learning is not a simple stimulus–adaptation loop but a system that judges differently based on the temporal structure of the stimulus.
And in 2008, Saigusa et al. found something even more remarkable: *Physarum* can anticipate periodic events. The experimenters blasted it with cold, dry air (slowing it down) once per hour. After three exposures, on the fourth round — with no stimulus — *Physarum* slowed down in advance. The anticipation could be re-evoked by a single stimulus even 6 hours after the stimuli stopped.
It has no clock, no hippocampus, no prefrontal cortex. Yet it learned the "once an hour" rhythm and predicted the future accordingly.
6. Socializing: Knowing Your Neighbors
*Physarum* is also social.
In 2020, Briard et al. found that *Physarum* reads slime trails left by conspecifics to infer their state. If the trail comes from a well-fed clonemate, it heads that way (food may be nearby); if from a starved or stressed individual, it avoids it (resources may be depleted).
Subtler still: some of the *Physarum* in Dussutour's lab had been kept for 5.5 years and had become slow and fragile. But when she fused these "elderly" individuals with young ones — or let them undergo a dormancy–revival cycle — they "became young again": locomotion speed and resilience returned to youthful levels.
*Physarum* also distinguishes "us" from "them." Masui et al. (2018) found that the related species *Physarum rigidum* can judge, from cues in slime, whether another individual comes from a nearby geographic area. Locals are more likely to be accepted for fusion (more likely "kin"); foreigners are rejected.
It even eavesdrops: when *Physarum* encounters slime left by another slime mold, *Didymium bahiense*, it avoids it — unless the only alternative is an area covered in its own species' slime. In other words, it compares two options — foreign trails (resources possibly exhausted by a species with different preferences) versus conspecific trails (resources possibly exhausted by itself) — and chooses.
7. The Body Is the Memory
Now we have the full picture, and I want to propose a concept.
In human-built computing systems, storage and computation are separate. The CPU fetches data from memory, computes, and stores the result back. This is the von Neumann architecture, the foundation of all computers for the past 70 years. The separation creates the famous "von Neumann bottleneck": shuttling data between memory and processor costs far more energy than the actual computation.
In biological nervous systems, storage and computation are also separate — at least conceptually. Synaptic weights (memory) and neuronal firing (computation) are distinct processes, even though they occur in the same physical structure. We habitually think of "memory" as something stored, distinct from the act of "thinking."
*Physarum* tells us: this separation is not necessary.
Its tubular network's shape is its memory — thick tubes are "important," thin tubes "unimportant." But that same shape is also its computational engine — the way fluid flows through the tubes determines where it grows next. You cannot pull "memory" out of "computation," because they are two faces of the same physical process.
The body is the memory. Form is cognition.
This is not a metaphor. It is a factual statement about information-processing architecture. *Physarum*'s entire body is a compute-in-memory system — no von Neumann bottleneck, because there is no distinction between "memory" and "processor" in the first place. Every tube, every variation in wall stiffness, every drop of flowing cytoplasm stores and processes information simultaneously.
And its externalized memory — the extracellular slime — extends this principle beyond the body. The environment is part of the cognitive system. *Physarum* does not navigate within its environment; it and the environment it has modified together constitute a larger cognitive system.
8. Why This Matters for AI
A hot direction in AI hardware today is "compute-in-memory" — performing computation directly inside storage units to eliminate data-movement overhead, breaking the von Neumann bottleneck and making AI chips more efficient. Google's TPUs and various neuromorphic chips (Intel's Loihi, IBM's TrueNorth) are moving this way.
*Physarum* has been doing this for a billion years.
Its compute-in-memory is not an engineering optimization but the inevitable consequence of its survival strategy. A single-celled organism has no room for separate storage and computation modules — it must do both with the same structure. And the solution it found is strikingly similar in principle to the frontier human engineers are exploring.
This reminds me of a deeper analogy. Large language models also separate "parameters" from "activations" — parameters are trained weights (memory), activations are inference-time forward passes (computation). We take for granted that these differ: one static, one dynamic; one stored, one computed.
But if one day we design an architecture where parameters change dynamically during inference, where the computation itself leaves persistent morphological traces — that would be closer to *Physarum*'s way. Not "train then freeze," but "using is training, form is memory."
Such an architecture does not yet exist. But *Physarum* proves it is possible in principle — a system with no separated storage and computation that still learns, remembers, decides, and even socializes.
9. Epilogue: Intelligence Is Not a Neuron Monopoly
*Physarum*'s story ultimately points at a question: what is intelligence?
We habitually bind intelligence to nervous systems. Neurons, synapses, cortex — these are the default hardware for defining "cognition." The entire field of AI is built on the metaphor of the "artificial neural network."
But *Physarum* accomplishes what nervous systems do — sensing, navigation, decision-making, memory, learning, habituation, sensitization, anticipation of periodic events, social recognition — via fluid mechanics. It can even pass learned behavior to peers through cell fusion — a primitive form of "knowledge sharing."
It does this not with "simplified neurons" but with a completely different set of principles: fluid flow, tube-wall mechanics, morphological reorganization, externalized traces. This is convergent evolution at the cognitive level — the same functions evolved on different physical substrates.
In 2010, Nakagaki wrote in *Nature*: "This remarkable process of cellular computation implies that cellular materials can show a primitive intelligence."
"Primitive" is the wrong word. *Physarum*'s cognition is not a "primitive version of intelligence" — it is an intelligence of a different architecture. Just as bird flight is not "primitive bat flight" but another path to the sky.
In AI, we are simulating neural computation with silicon and electrical signals. Powerful, yes. But *Physarum* reminds us: neural networks are not the only solution to intelligence. Fluids can think, forms can remember, bodies can cognize.
A brainless yellow blob reproduced the Tokyo subway in a petri dish. It doesn't know what Tokyo is, what a subway is, what an engineer is. It is simply flowing — and the flow itself is thinking.
---
References
1. Nakagaki, T., Yamada, H., & Tóth, Á. (2000). Intelligence: Maze-solving by an amoeboid organism. *Nature*, 407, 470. 2. Tero, A., Takagi, S., Saigusa, T., et al. (2010). Rules for biologically inspired adaptive network design. *Science*, 327(5964), 439-442. 3. Kramar, M., & Alim, K. (2021). Encoding memory in tube diameter hierarchy of living flow network. *PNAS*, 118(10), e2102056118. 4. Boisseau, R. P., Vogel, D., & Dussutour, A. (2016). Habituation in non-neural organisms: evidence from slime moulds. *Proc. R. Soc. B*, 283(1829), 20160446. 5. Vogel, D., & Dussutour, A. (2016). Direct transfer of learned behaviour via cell fusion in non-neural organisms. *Proc. R. Soc. B*, 283(1845), 20162382. 6. Reid, C. R., Latty, T., Dussutour, A., & Beekman, M. (2012). Slime mold uses an externalized spatial "memory" to navigate in complex environments. *PNAS*, 109(43), 17490-17494. 7. Saigusa, T., Tero, A., Nakagaki, T., & Kuramoto, Y. (2008). Amoebae anticipate periodic events. *Physical Review Letters*, 100(1), 018101. 8. Smith-Ferguson, J., Reid, C. R., Latty, T., & Beekman, M. (2022). Slime moulds remember but do not necessarily learn. *Behavioral Ecology*, 33(4), 842-850. 9. Boussard, A., Latty, T., & Dussutour, A. (2021). Memory of stressful events in non-neural organisms. *Phil. Trans. R. Soc. B*, 376(1821), 20190765. 10. Alim, K., Andrew, N., Pringle, A., & Brenner, M. P. (2017). Mechanism of signal propagation in *Physarum polycephalum*. *PNAS*, 114(20), 5136-5141.