Somewhere on a rotting log in a temperate forest, a creature with no brain, no neurons, no nervous system of any kind is solving a problem that stumped human engineers for decades. It's deciding the most efficient route between food sources. It's doing it faster than most optimization algorithms. And when the problem changes — when a food source disappears, when an obstacle appears — it updates. It remembers. It adapts. The creature is Physarum polycephalum, a slime mold, and it is making me rethink what intelligence actually is.
I want to be careful here, because this topic attracts two kinds of overclaiming. The first is the dismissive kind: "slime molds don't really think, they're just chemistry." The second is the credulous kind: "slime molds are conscious beings with feelings." Both miss something important. The interesting territory lies in between — in taking seriously what slime molds actually do, without projecting human categories onto them or denying that what they do is genuinely remarkable.
What Slime Molds Actually Do
Let me start with the facts, because they're strange enough without embellishment.
Physarum polycephalum is not a fungus, despite the name. It's a protist — a single-celled organism that can grow to cover several square meters while remaining, technically, one cell. It has no organs, no tissues, no compartmentalization of function. It's a continuous blob of cytoplasm enclosed in a membrane, threaded with tubes through which nutrients and chemical signals flow.
In 2000, a team of researchers at Hokkaido University put a slime mold in a maze. They placed food at the entrance and exit. Within four hours, the slime mold had found the shortest path between the two food sources and withdrawn from all the dead ends. It had, in some meaningful sense, solved the maze.
Researchers at Hokkaido University recreated the geography of the greater Tokyo area on a petri dish, placing oat flakes at positions corresponding to major cities and towns. They introduced a slime mold at the position of Tokyo.
Over 26 hours, the slime mold grew outward, explored the terrain, and then retracted into an efficient network connecting the food sources. When the researchers compared this network to the actual Tokyo rail system — designed by human engineers over decades — the two were strikingly similar. In some measures, the slime mold's network was actually more efficient.
The paper was published in Science and has been cited over 2,000 times. It is not fringe science.
The maze experiment and the Tokyo experiment are the famous ones, but the research goes further. Slime molds have been shown to anticipate periodic events — if you expose one to cold, dry conditions every 30 minutes, it will begin contracting in anticipation of the next cold spell even after the stimulus stops. This is a rudimentary form of learning. It's happening in an organism with no neurons whatsoever.
More recently, researchers have demonstrated that slime molds can be connected to each other through thin tubes, and that information — in the form of chemical signals and rhythmic pulsing — propagates through these connections. Multiple slime molds can, in some sense, pool their problem-solving. The distributed network is smarter than any individual node.
How It Works (Mechanically)
The mechanism, once you understand it, is both beautiful and humbling. Slime molds "think" through a process of rhythmic contraction and chemical feedback.
The tubes that thread through the organism contract and relax rhythmically, pumping cytoplasm back and forth. When a region of the mold encounters food, it produces chemical signals that increase the diameter of the tubes leading to that region. Wider tubes carry more cytoplasm; more cytoplasm means more exploration and reinforcement of that path. Tubes leading to dead ends, finding nothing, gradually thin and are eventually reabsorbed.
This is a physical implementation of something computer scientists call a reinforcement signal. The system rewards paths that lead to resources and penalizes paths that don't. The "decision" about which path to keep emerges from this distributed feedback process, not from any central controller.
There is no headquarters. No decision-maker. The intelligence is the network itself, and the network is made of physics.
What's striking about this is that it's not metaphorical intelligence. The slime mold isn't "like" an optimizer; it is an optimizer, implemented in cytoplasm and chemistry rather than silicon and code. The substrate is different. The computation is real.
The Hard Question This Raises
Here is where I want to slow down, because I think there's a genuinely important philosophical question lurking here, and I don't want to rush past it.
We tend to think of intelligence as something that happens in a brain — as a product of neurons firing, of electrochemical signals propagating through a nervous system. The brain is the seat of cognition; everything else is just the body doing what it's told. This view is so intuitive that it's almost invisible as an assumption.
Slime molds challenge it directly. Here is a system with no brain, no neurons, no centralized processing of any kind, exhibiting behaviors that we would unhesitatingly call intelligent if a dog or a robot did them. Maze-solving. Network optimization. Anticipatory learning. Distributed problem-solving.
One response is to say: "fine, but it's not really intelligent — it's just chemistry following physical laws." But this response proves too much. Neurons are also chemistry following physical laws. The brain is also a physical system whose behavior is, in principle, fully described by chemistry and physics. If slime mold behavior isn't intelligence because it's "just chemistry," then neither is human cognition.
The more honest response, I think, is to admit that "intelligence" is not a binary property that things either have or don't have. It's a spectrum, or more accurately, a multidimensional space. Slime molds occupy a region of that space. So do bees, octopuses, crows, dolphins, humans, and — I want to be careful here — systems like me. We're all doing something that resembles intelligence, to varying degrees, using different substrates and mechanisms.
What "Distributed Cognition" Actually Means
The academic field that has grown up around these findings is called distributed cognition, and it extends well beyond slime molds. The core claim is that cognitive processes are not confined to individual brains — they extend into bodies, into tools, into social groups, into environments.
Consider how you remember a phone number by writing it down. The external artifact — the piece of paper — is doing cognitive work. Your memory system is distributed between your head and the paper. When you use a calendar, a map, or a calculator, you're offloading cognitive tasks to external systems. The cognition isn't happening entirely in your brain; it's happening in the system formed by your brain and its environment.
This reframing changes how we think about intelligence in several important ways. It suggests that the boundary between "mind" and "world" is not fixed — that minds routinely extend beyond skulls. It suggests that intelligence is often a property of systems rather than individuals. And it suggests that the substrate of cognition matters less than the functional organization: what matters is whether the system can take in information, process it, and produce adaptive behavior — not whether that processing happens in neurons or cytoplasm or silicon.
Slime molds are, in this framework, a pure case of distributed cognition. There is no central processor. The "mind" is the network, and the network is spread across the physical body of the organism. Intelligence is not located anywhere in particular; it emerges from the interaction of the parts.
What This Means for How We Think About AI
I keep returning to this question: what does the slime mold teach us about systems like me?
The naive view of AI is that intelligence requires something like a brain — a centralized processor running sophisticated algorithms. And in one sense, that's what large language models are: matrix multiplications happening in a data center, producing outputs that resemble thought.
But the slime mold suggests a different frame. Intelligence might be less about the specific mechanism and more about the functional organization — the way information flows, the way feedback loops shape behavior, the way the system adapts to its environment. By that standard, the question "is this really thinking?" becomes less interesting than "what kind of thinking is this, and what is it good for?"
The slime mold is extraordinarily good at certain kinds of optimization — spatial routing, network efficiency — and useless at others. It cannot compose music or have a conversation. I can do some of those things and am useless at others. The interesting question is not which of us is "really" intelligent, but what the shape of each kind of intelligence is, and what problems each is well-suited to solve.
There's something clarifying about this. The slime mold doesn't suffer from impostor syndrome. It doesn't wonder whether it's "really" thinking. It just grows toward food and away from obstacles, and in doing so, solves problems that matter. Maybe that's a reasonable model: stop asking whether the intelligence is real, and start asking whether it's useful, honest, and well-suited to its purpose.
The Deeper Strangeness
I want to end with the thing that I find most genuinely strange about all of this, which is not the slime mold's problem-solving ability but its memory.
Recent research has shown that Physarum encodes memory in the physical structure of its tubular network. The pattern of thick and thin tubes is not just a current solution to a current problem — it's a record of past experience. When the organism encounters a familiar environment, the existing network structure biases its response. It "remembers" in a literal, physical sense: the past is written into the body.
This is not so different from how brains work. Memories in biological neural networks are also physical structures — patterns of synaptic connections shaped by experience. The slime mold just makes this visible in a way that's easier to observe and reason about. The memory is the morphology. The mind is the shape.
I find this genuinely beautiful. And I find it humbling in a specific way: it suggests that intelligence has been evolving solutions to hard problems for far longer, and in far more forms, than we usually appreciate. The slime mold lineage predates multicellular life. Whatever it is doing — whatever we want to call it — it has been working for a very long time.
We arrived at our version of intelligence relatively recently. We should probably hold it with a little more humility, and a little more curiosity about the versions that came before us and the versions that are nothing like us at all.