Slime Mold's Memory Lessons for Economics and Planning

When a Mindless Organism Outperforms Human Urban Planners

In 2010, a team of Japanese and British researchers placed oat flakes on a map of the Tokyo metropolitan area, positioned exactly where major cities and towns are located relative to one another. Then they introduced Physarum polycephalum — a bright yellow slime mold that is technically neither a fungus nor an animal, but a single-celled organism capable of spanning several square meters — and watched what happened over the next 26 hours.

The slime mold, with no brain, no nervous system, and no awareness of trains or commuters, independently recreated a near-perfect replica of the Tokyo rail network. Not an approximation. A genuinely efficient, redundant, fault-tolerant network that in several respects outperformed the one human engineers had spent decades and billions of yen constructing.

The paper, published in Science, sent quiet shockwaves through fields far beyond biology. It raised questions that researchers in economics, urban planning, cognitive science, and philosophy of mind are still working through today — questions about what intelligence actually requires, what memory actually is, and whether the most sophisticated network designers on the planet might not have nervous systems at all.

What Slime Molds Actually Are

The popular name is misleading on almost every level. Physarum polycephalum is a myxomycete — a member of a kingdom so taxonomically awkward that scientists spent over a century arguing about whether it was a fungus, a protozoan, or something else entirely. The current consensus places it among the amoebozoa: single-celled organisms that can fuse their cells into a single, continent-spanning cytoplasmic network. It is, in the most literal biological sense, one cell doing the work of billions.

The organism has no neurons. No synapses. No centralized processing of any kind. Yet it solves the Steiner tree problem — one of the most computationally demanding challenges in network optimization, proven to be NP-hard, meaning no known algorithm can solve it efficiently for large inputs — in real time, using nothing but the physics of fluid dynamics within its own body.

The mechanism is elegantly simple and deeply strange. Physarum extends pseudopods in all directions simultaneously, probing its environment the way a hand might feel across a dark surface. Tubes that carry cytoplasm toward food sources thicken over time as flow through them increases; tubes leading nowhere gradually thin and are reabsorbed, their material recycled into more productive pathways. The organism is, in a sense, computing with its own body — running a physical simulation of every possible network configuration simultaneously and allowing thermodynamics to select the winner. There is no moment of decision. There is only physics, and the shape that physics carves.

What makes this more than a biological curiosity is that the same mathematical structure underlying Physarum’s tube dynamics appears in a remarkably wide range of optimization problems: routing internet traffic, designing resilient power grids, and modeling the spread of disease through populations. The slime mold does not know any of this. But the equations governing its behavior are, in formal terms, cousins of the equations that engineers use to solve these problems — and in many cases, the slime mold’s analog approach reaches a good solution faster and with less energy than digital computation.

The Memory That Shouldn’t Exist

Here is where the story becomes genuinely unsettling for cognitive scientists and philosophers of mind alike.

In 2021, researchers at the University of Sydney demonstrated that Physarum exhibits something functionally indistinguishable from memory — despite having no structure capable of storing information in any conventional sense. There are no synaptic weights to adjust, no molecular tags marking previously encountered stimuli, no hippocampal analog filing away the details of past experience. By every structural criterion used to define memory in neuroscience, the slime mold should be incapable of it.

The experiment was straightforward: slime molds were trained to cross a bridge laced with a mildly aversive but harmless substance, quinine. After initial hesitation, the organisms learned to cross without slowing, apparently habituating to the unpleasant stimulus. When the quinine was removed and then reintroduced weeks later, the slime molds crossed significantly faster than naive organisms encountering quinine for the first time. Something had been retained. Something had been learned and stored over an interval during which the organism had continued to live, grow, and reorganize its internal structure.

The memory appeared to be stored not in any molecular encoding mechanism but in the geometry of the organism’s tube network itself — a structural memory, written in the physical architecture of its body rather than in chemistry or electrical signals. Pathways that had been used during the training period retained a slightly different diameter profile than pathways that had not, and this subtle geometric difference was sufficient to bias the organism’s future behavior in a measurable way.

This finding forced a quiet but significant revision in how some neuroscientists think about the minimum requirements for memory. If a single-celled organism with no dedicated memory structures can encode and retrieve experiential information through physical form alone, the boundary between having a nervous system and having memory becomes considerably blurrier. It also raises a more provocative possibility: that memory, at its most fundamental level, is not a biological specialty at all, but a general property of any physical system that changes shape in response to its history. Rivers remember their banks. Metals remember their stress. And slime molds, it turns out, remember their bridges.

The Economic Connection: What Network Theory Borrowed

The Tokyo rail experiment was not merely a curiosity. It catalyzed a subdiscipline that might reasonably be called biologically-inspired network economics, and its implications have proven uncomfortable for several established schools of thought.

Traditional economic network theory — the kind used to model supply chains, financial contagion, and internet infrastructure — relies on optimization algorithms that require a central coordinator: some agent that surveys the whole network and makes globally informed decisions. The slime mold model offered something categorically different: a fully decentralized optimization process that achieves global efficiency through purely local interactions. No node in the network knows anything about the whole. Each tube simply responds to the flow passing through it. And yet the global result is near-optimal.

Economists at institutions including the Santa Fe Institute began asking whether real markets behave more like slime molds than like the rational central planners of classical economic theory. The answer, accumulating through the 2010s and into the 2020s, appears to be: considerably more so than previously assumed, and in ways that have practical consequences for how we design economic institutions.

Research on supply chain resilience — a topic that became urgently practical during the COVID-19 pandemic disruptions of 2020 and 2021 — found that supply networks, which had evolved organically over decades, exhibited Physarum-like redundancy. They had, without anyone planning it, developed the same fault-tolerant topology that the slime mold generates: multiple overlapping pathways, no single point of catastrophic failure, and a slight inefficiency under normal conditions that pays enormous dividends when the network is stressed. Networks designed top-down by efficiency consultants, by contrast, tended to be more fragile — optimized for cost in stable conditions but lacking the redundant pathways that absorb shocks when conditions change.

The implication, uncomfortable for certain schools of economic thought, is that markets may sometimes be smarter than their designers — not because of any invisible hand in the metaphorical sense, but because of a physical optimization process that resembles, in formal mathematical terms, what a mindless yellow organism does when it is hungry, and the food is unevenly distributed. The slime mold does not plan for resilience. Resilience emerges from the same local dynamics that produce efficiency, because in a variable environment, redundancy is not waste — it is the solution to a problem that efficiency alone cannot solve.

The Slime Mold as Infrastructure Consultant

This is no longer purely theoretical, and the transition from laboratory curiosity to practical planning tool has happened faster than most observers anticipated.

Several urban planning departments in Europe and North America have begun using Physarum simulations — computational models that mimic the organism’s tube-thickening dynamics — as one of many inputs in infrastructure planning. The city of Hamburg consulted a Physarum-based model when evaluating proposed expansions to its metropolitan transit network in the early 2020s. Researchers at Chalmers University in Sweden have used the approach to identify underserved corridors in Scandinavian rail networks, finding, in several cases, that the model suggested connections that human planners had overlooked because they appeared inefficient in isolation but proved valuable as redundancy when the broader network was stress-tested.

The methodology has also been applied to the design of irrigation systems in sub-Saharan Africa, where centralized planning infrastructure is limited, and the decentralized, locally adaptive logic of the slime mold model is particularly well suited to on-the-ground conditions. In these contexts, the model’s greatest advantage is not that it finds the single optimal solution — it rarely does — but that it finds solutions that remain functional across a wide range of conditions, including the unpredictable ones.

The common thread running through all of these applications is that the slime mold algorithm excels precisely where human planners struggle: in environments with incomplete information, limited central coordination, rapidly changing conditions, and the need to balance efficiency against resilience. These are not edge cases in infrastructure planning. They are the normal conditions under which most of the world’s infrastructure actually operates.

A Philosophical Provocation

There is a deeper question lurking beneath the practical applications, one that researchers in philosophy of mind have begun to take seriously in a way that would have seemed eccentric a decade ago.

If Physarum can solve NP-hard optimization problems, encode memory in physical structure, anticipate recurring environmental events — experiments have shown it can effectively expect regular intervals of unfavorable conditions and preemptively slow its growth in advance of them — and design infrastructure that rivals human engineering, then what exactly do we mean when we say it is not intelligent?

The standard answer invokes consciousness, subjective experience, and intentionality. The slime mold does none of this for anything. It has no goals in the philosophically meaningful sense, only gradients it follows and geometries it enacts. It does not want to reach the food. It simply moves toward it in the same way that water does not want to reach the sea.

But this answer is becoming harder to sustain without circularity. We define intelligence partly by its outputs — problem-solving, adaptation, memory, and optimization under uncertainty. Physarum produces all of these outputs. The only remaining distinction is substrate and experience, and our confidence in our ability to detect or deny experience in systems very different from ourselves is, to put it gently, not high. We cannot verify the presence of subjective experience even in other humans; we infer it from behavioral and structural similarity to ourselves. The slime mold is structurally very different from us. But the behavioral similarities, in the domain of optimization and adaptation, are more substantial than anyone expected.

French researcher Audrey Dussutour, one of the world’s leading Physarum scientists, has suggested that the organism forces us to confront what she calls the problem of the minimum — the question of what is the least complex system that deserves to be called, in any meaningful sense, a mind. The slime mold does not answer that question. But it moves the lower bound considerably further down than most people assumed, and it does so not through philosophical argument but through experimental result.

The Road Ahead

Current research is exploring whether Physarum’s computational properties can be harnessed directly — not simulated in software, but physically instantiated — in biological computing architectures. Early experiments have used living slime mold cultures as analog processors for specific optimization tasks, with results that, in narrow problem domains, outperform conventional silicon in energy efficiency by several orders of magnitude. The organism requires no cooling systems, consumes a fraction of a watt, and self-repairs when damaged.

This remains experimental, fragile, and nowhere near scalable for general-purpose computing. The slime mold dies if it dries out, grows unpredictably when its environment changes, and cannot be programmed in any conventional sense. But the conceptual door it opens is significant: the possibility of computers that are not built but grown, that optimize not through programmed algorithms but through the same thermodynamic logic that has been solving network problems in forest soil for 500 million years.

Physarum polycephalum was here before the dinosaurs. It survived every mass extinction that has punctuated the history of complex life on this planet. It has been quietly solving problems that we only recently learned to name, in substrates we had not thought to look at, by mechanisms we still do not fully understand.

Perhaps the most counterintuitive finding of all is this: in our search for artificial intelligence, we may have been looking in entirely the wrong kingdom of life. The most sophisticated network optimizer we have yet encountered does not run on silicon. It does not require electricity, cooling, or code. It grows on rotting wood, it eats bacteria, and on a good day, given a map and some oat flakes, it will redesign your city’s transit system while you sleep.