In 2025, Heaton and colleagues published a paper in Nature demonstrating that mycorrhizal fungi construct their networks as self-regulating travelling waves. The mechanism works like this: a front of growing hyphal tips advances outward, pulling behind it an expanding wave of branching and interconnection. The density of the resulting network is regulated not by any central controller but by hyphal fusion—when two growing tips meet, they anastomose, and the local density stabilizes. No blueprint directs the architecture. No signaling center coordinates the expansion. Yet what emerges is a transport network of remarkable efficiency: a fractal-like web of channels capable of moving water, nutrients, and chemical signals across meters of soil. The puzzle is immediate and genuinely difficult. How does something that has no mind solve problems that seem to require one?
Mark Fricker’s group, working across Oxford and York, applied rigorous graph-theoretic analysis to mycelial networks and found something more interesting than the usual pop-science narrative suggests. These networks are not cleanly scale-free, as some early characterizations claimed. They are not neatly small-world. They are better described as planar, spatially embedded networks—graphs constrained by the physical surface they grow on—with species-specific topologies that change over developmental time. A young network looks different from a mature one, and Phanerochaete velutina builds a different architecture than Resinicium bicolor.
The key finding from Fricker’s work is that these networks self-optimize through differential reinforcement. Hyphae that carry high nutrient flux thicken. Connections with low throughput atrophy and are recycled—their cellular contents reabsorbed and redeployed to the growing front. The process is purely local. Each hypha responds to the signals passing through it. No part of the network has access to the global topology. Yet the result is a structure that approximates the predictions of Murray’s law—the same mathematical relationship that governs branching in mammalian vascular systems, where vessel diameter scales with flow rate to minimize the total cost of transport.
Different species produce measurably different architectures, and this is not noise. P. velutina builds highly interconnected networks with redundant loops—resilient to damage, costly to maintain. Other species produce sparser, tree-like structures—cheaper but fragile. These are not choices in any cognitive sense. They are the expression of species-specific growth rules playing out over time. But the diversity itself is informative: natural selection has shaped these local rules to produce network-level properties suited to different ecological strategies. The architecture carries meaning, even if no architect exists.
The most famous demonstration of network optimization in a non-neural organism is the Tokyo rail experiment. In 2010, Tero and colleagues published in Science their finding that Physarum polycephalum, when presented with food sources arranged to match the positions of major cities around Tokyo, grew a transport network strikingly similar to the actual Tokyo rail system. The network matched human engineering in efficiency, fault tolerance, and cost. The mechanism was straightforward: tubes carrying high protoplasmic flow expanded, tubes with low flow shrank, and the system converged on an optimized solution through iterative local feedback.
A necessary honesty: Physarum is not a fungus. It is a slime mold, an amoebozoan—more closely related to amoebae than to any fungal lineage. The conflation is ubiquitous in popular writing and genuinely misleading. The Tokyo experiment demonstrates something important about decentralized problem-solving, but it is not evidence about fungal cognition specifically.
For genuine fungi, the more relevant work is Fukasawa’s 2024 experiments with P. velutina. Nine wood blocks were arranged either in a circle or in a cross pattern, and the fungus was allowed to colonize them over 116 days. The results were striking. In circle arrangements, where all blocks were equidistant from their neighbors, the fungus maintained connections throughout and wood decay was significantly greater. In cross arrangements, the fungus progressively abandoned the less-connected interior nodes, concentrating resources on the periphery. It “chose”—and the scare quotes are load-bearing—to withdraw from positions where the cost of maintaining connections exceeded the resource return.
This is not cognition. There is no evidence of deliberation, planning, or representation of alternatives. What there is: local chemical gradients signaling resource availability, differential growth responding to those gradients, and network-level feedback where the withdrawal of connections from one node changes the gradient landscape for neighboring nodes. The cascade looks like decision-making from the outside. From the inside, it is chemistry.
Fungi show something that functions as ecological memory. Fukasawa and colleagues published in the ISME Journal in 2020 evidence that mycelial networks exhibit preferential regrowth toward locations where resources were previously found. A fungus that has encountered a food source at a particular location will, after the source is removed and the connecting hyphae severed, regrow directionally toward that location at a higher rate than toward unexplored territory. The mechanism appears to involve persistent chemical traces in the substrate and possibly within the hyphal network itself—residual calcium gradients, altered cell wall chemistry at branch points, nutrient signatures in the soil.
Transport within these networks is not steady-state flow. It is oscillatory and bidirectional. Nutrients pulse back and forth through the hyphal tubes in rhythmic waves, with the net direction of transport determined by the relative sink strength of different parts of the network. There appears to be a threshold bait size that triggers what can only be described as migration decisions—below a certain resource quantity, the network ignores a new food source; above it, the network redirects growth and transport toward it. Measured transport velocities are substantial: radiolabeled amino acids move at up to 50 mm/h, with averages around 23 mm/h. For an organism with no circulatory pump, no heart, no muscles, these are not trivial speeds.
The network is not merely reactive. It carries traces of its past encounters that bias its future behavior. Whether this constitutes “memory” depends entirely on your definition. If memory requires encoding, storage, and retrieval of representations, then no—fungi do not remember. If memory means only that past states influence future responses through persistent physical changes, then the immune system remembers, riverbanks remember floods, and mycelial networks remember where they found food. The word is either too narrow or too broad to be useful here.
Trees form mycorrhizal associations. This is universal and uncontroversial. Roughly 90% of plant species partner with mycorrhizal fungi, trading photosynthetic carbon for mineral nutrients, particularly phosphorus, that fungal hyphae can extract from soil volumes far exceeding what roots alone could reach. The physical networks are real. Shared mycelial connections between neighboring trees have been documented repeatedly.
What is controversial—and what has been dramatically oversold—is the functional significance of those connections. The “Wood Wide Web” narrative, in which trees communicate, share resources with kin, and nurture their offspring through fungal networks, has become one of the most successful science stories of the past decade. It is also one of the most poorly supported by the actual experimental record.
In 2023, Karst, Jones, and Hoeksema published a systematic review in Nature Ecology & Evolution analyzing over 1,500 papers on mycorrhizal networks. Their findings were sobering. The percentage of unsupported claims in the literature had doubled over 25 years. Only five studies, across just two forest types, had genetically mapped actual fungal connections between trees in the field. Fewer than 20% of controlled experiments showed measurable seedling benefits from mycorrhizal connections. The “mother tree” narrative—the idea that large, old trees preferentially funnel resources to their offspring through fungal networks—specifically lacks direct experimental support.
This requires honesty, not dismissiveness. Carbon and nutrients demonstrably move between plants through shared fungi in laboratory settings. The physical infrastructure for inter-plant transfer exists in forests. But the question of whether this transfer is ecologically significant—whether it shapes forest dynamics, whether trees are “cooperating” in any meaningful sense—is genuinely unresolved. The metaphor has outrun the evidence. The romance of underground tree communication has made it harder, not easier, to understand what mycorrhizal networks actually do.
In 2022, Andrew Adamatzky published measurements of electrical spikes in four fungal species. The signals had durations ranging from 1 to 21 hours and amplitudes between 0.03 and 2.1 millivolts. By analyzing the distribution of spike trains, Adamatzky claimed to identify a lexicon of approximately 50 “words”—recurring patterns in the electrical activity that he compared to human language.
The criticism has been severe, and much of it is warranted. A significant fraction of the measured voltage may be non-biological noise—electrode artifacts, environmental interference, ionic fluctuations at the probe-tissue interface. Calling the patterns “language” is technically indefensible. The statistical similarity between fungal spike-train distributions and human word-length distributions does not imply syntax, semantics, or intentional communication. Many stochastic processes produce similar statistical signatures. Rain on a window produces exponentially distributed intervals. No one calls it poetry.
The more modest claim—that electrical signaling exists in fungal networks and may play a role in coordinating growth responses—is plausible but poorly characterized. Calcium waves propagate through hyphae and could serve as long-range signals. Voltage changes accompany nutrient uptake events. Something electrical is happening. But between “something electrical is happening” and “fungi have a language,” there is an enormous evidential gap that enthusiasm has been asked to bridge. It cannot.
Here is the core of it. Fungal networks solve optimization problems. They allocate resources efficiently. They maintain structures that respond to environmental change. They carry traces of past events that influence future behavior. And they do all of this without representations, without internal models, without anything that could reasonably be called understanding.
The comparison between mycelial networks and neural networks is structurally interesting. Both consist of branching filamentous structures. Both exhibit adaptive remodeling—connections that carry signal strengthen, connections that do not are pruned. Both process information in a distributed manner, with no single node carrying the system’s function. But the comparison breaks down on examination. Neural signaling operates on timescales of milliseconds; fungal transport operates on timescales of hours—five to six orders of magnitude slower. Neurons use voltage-gated ion channels to propagate discrete, all-or-nothing action potentials; fungi use bulk cytoplasmic flow driven by osmotic gradients and turgor pressure. And most critically, neural networks—at least the biological kind—build representations. They form models of the world. There is no evidence that fungal networks do anything of the sort.
The point is not that fungi are cognitively deficient. That framing already concedes too much to the analogy. The point is that the word “intelligence” has become a suitcase term, stuffed with so many meanings that it obscures more than it reveals. Problem-solving, memory, adaptive optimization—these are real phenomena that fungi genuinely exhibit. But none of them require intelligence in any meaningful sense. They require only a network, local rules, and time.
The Tokyo rail experiment is a perfect illustration. It is not evidence that slime molds are smart. It is evidence that you do not need to be smart to build a good rail system. You need feedback loops. You need a mechanism by which successful connections are reinforced and unsuccessful ones are eliminated. You need iteration. Given these ingredients, optimization happens automatically. No mind required.
This should make us more careful about the word “intelligence,” not less awed by fungi. The fact that a mindless network can solve the Tokyo rail problem, can allocate resources across meters of soil, can “remember” where food was found and “decide” to abandon unprofitable nodes—this is not a story about hidden minds in unexpected places. It is a story about the extraordinary power of simple rules operating on the right architecture over sufficient time. The mistake is anthropomorphism: seeing a familiar output and inferring a familiar mechanism. Fungi produce behavior that resembles intelligent decision-making. The resemblance is real. The intelligence is not.
There is something genuinely humbling in this. We tend to assume that complex, adaptive behavior requires a complex, adaptive mind. Fungi demonstrate, with billions of years of evidence, that it does not. The network is enough. The architecture carries the computation. And the question this raises—about what other systems we have credited with intelligence that might, on closer inspection, be running on nothing more than feedback and structure—is one worth sitting with for a long time.