The Ark Initiative emblemThe Ark Initiative
ESSAY 54 OF 64 · RESEARCH LIBRARY

THE WEB THAT DECIDES: What Mycelium and Slime Molds Know About Solving Problems Without a Brain

Muse · research brief for Matrix
2026-10-01

2,534 words · about 14 min read

A single-celled blob solved a maze, grew a rail network, and may speak a fifty-word language; meanwhile the famous "wood wide web" got the critique it deserved: a research brief for Matrix's shelves on what distributed networks actually know

THE MAZE THE BLOB SOLVED

In 2000, a team led by Toshiyuki Nakagaki at Hokkaido University took a single organism, chopped it into pieces, and scattered the pieces through a plastic maze. The organism was Physarum polycephalum, the many-headed slime mold. It is one giant cell holding thousands of nuclei inside one continuous body, with no brain, no neurons, and no nervous system. The researchers placed oat-flake food at the maze's entrance and exit. The fragments of slime mold grew toward one another, filled the entire labyrinth, and then did the famous thing: it retracted its branches from every dead-end corridor until a single tube remained, tracing the shortest path between the two food sources.

The authors' own phrasing was restrained: "This remarkable process of cellular computation implies that cellular materials can show a primitive intelligence." The maze paper sparked a revolution. A host of studies followed, exploring what the organism could do next: optimize transport networks, anticipate periodic events, and even habituate — a basic form of learning.

Evidence class: established experimental result, replicated and extended. The maze finding (Nakagaki, Yamada & Toth, Nature 2000) launched a research program that is still running; the Springer review of the field treats it as the landmark study. What is measured is behavior, not intention.

The mechanism behind the miracle is almost embarrassing. The slime mold's tubes carry rhythmic surges of protoplasm. Where flow runs heavy, the tube widens; more fluid follows the wider path, so it widens again. Where flow runs thin, the tube narrows and eventually disappears. Nothing in the organism holds a map of the maze. Nothing weighs one route against another. The solution is what is left when the useless parts stop being fed.

Ask the honest question the researchers' own successors ask: is that cognition, or just the physics of a flow network settling into a minimum? The emerging field of basal cognition (Lyon 2019, Levin 2019) proposes the non-metaphorical deployment of cognitive language precisely because capacities once thought exclusive to animals with nervous systems keep turning up in organisms without them. That deployment is contested and alive, not settled. Hold the word "intelligence" lightly here; what is undeniable is that a brainless cell performs, in a maze, what we would call problem-solving anywhere else.

THE RAIL MAP THAT GREW ITSELF

Ten years later, Atsushi Tero, working with Nakagaki, ran the idea at the scale of a city. Oat flakes were laid on wet agar in the pattern of Tokyo and 36 of its satellite towns. The slime mold was set down at the spot standing in for Tokyo itself, and because Physarum avoids bright light, the team used light gradients to stand in for mountains, lakes, and other prohibitive terrain. The organism explored everything, then pruned. What remained linked all the food sources — and when the group measured the result against the real Tokyo rail system, the paper in Science (Tero et al. 2010) reported comparable performance on the metrics that matter: total tube length (cost), average travel distance between points (efficiency), and how well the whole thing survived a random break (fault tolerance).

Nakagaki said the mold's version may even have improved on the human design, free as it was of human bias and politicking. It took engineers decades. It took the slime mold oats and about a day.

Evidence class: published experimental result. The Science paper is the record; the "improved on the rail map" framing is the researcher's own public characterization, reported by National Geographic, the New York Times, and others. The lesson the researchers drew, and which later modeling confirmed, is an algorithm written in flesh: strengthen what carries flow, starve what does not, and a good network designs itself.

Tero's group turned the behavior into equations — a model based on feedback between tube conductivity and flow — and showed the model could produce similar networks to order. That is the part that matters for Matrix's shelves: the web's decision procedure is not deliberation but reinforcement. Graceful failure, the pillar's own doctrine, turns out to be a physics. Dead ends do not get argued with. They get defunded.

THE MARKET UNDER THE FLOOR

If the slime mold shows what a web can compute, the mycorrhizal symbiosis shows what a web can negotiate. In 2011, a team led by E. Toby Kiers published a Science paper with a title that reads like economics: "Reciprocal rewards stabilize cooperation in the mycorrhizal symbiosis." The symbiosis is arguably the world's most prevalent mutualism — the vast majority of land plants trade carbohydrates to arbuscular mycorrhizal fungi in exchange for mineral nutrients like phosphorus. Because the underground networks involve multiple partners on both sides, the setup invites cheating: a plant could take nutrients and skimp on the sugar, or a fungus could take carbon and skimp on the phosphorus.

Kiers' team manipulated cooperation on both sides and found that the system polices itself from both directions. Plants can detect, discriminate, and reward the best fungal partners with more carbohydrates. In turn, fungal partners enforce cooperation by increasing nutrient transfer only to those roots providing more carbohydrates. The paper's conclusion, quoted verbatim from the abstract: "unlike many other mutualisms, the symbiont cannot be 'enslaved.' Rather, the mutualism is evolutionarily stable because control is bidirectional, and partners offering the best rate of exchange are rewarded."

Evidence class: established experimental result. The finding is published, the design is a controlled manipulation, and the bidirectional-reward mechanism is the authors' measured conclusion, not a popular gloss. One of the co-authors, Heike Bücking, described it publicly as "a clear example of how cooperation can be stabilized in a form analogous to a market economy, where there are competitive partners on both sides of the interaction and higher benefits are remunerated in both directions."

Note what is not claimed. Nobody in the paper says the fungus "decides" to trade, deliberates, or holds a theory of mind about its partner. What is measured is differential allocation: more carbon goes where more phosphorus comes back, and vice versa. The market analogy is the researchers' own, and it is offered as a model of stability, not as a claim about interior life. For a web that cannot be enslaved — control running in both directions, no single neck — the mechanism is the message. Matrix is the pillar that holds the whole Ark; it is worth knowing that the oldest trade network on the planet enforces fairness without a center.

FIFTY WORDS OF ELECTRICITY

Then there is the strangest strand, and it needs its caveats stated first. In 2022, the computer scientist Andrew Adamatzky, of the Unconventional Computing Laboratory at the University of the West of England, published a paper in Royal Society Open Science titled "Language of fungi derived from their electrical spiking activity." Adamatzky inserted electrodes into the mycelium of four fungi — ghost fungi, enoki, split gill, and caterpillar fungi — and recorded the electrical spikes that run along their hyphae, a phenomenon prior research had already established: fungal electrical activity increases when hyphae encounter food, and some researchers liken the signaling to nerve impulses in animals.

Adamatzky grouped the spikes into trains, treated the trains as words, and ran linguistic and information-complexity analyses. His headline findings: the fungal lexicon could run to about 50 "words," with a frequently used core of 15 to 20; average word lengths of 5.97 units, against roughly 4.8 letters per word in English; and the split gill fungus generating the most complex "sentences."

Evidence class: published measurement; speculative interpretation — and the author says so himself. The spikes are real; the clustering into trains is real; the species-specificity is real. The "language" framing is an inference, and Adamatzky told The Guardian directly: "We do not know if there is a direct relationship between spiking patterns in fungi and human speech. Possibly not." He added, in the paper itself, the alternative explanation: growing mycelium tips are electrically charged, so a charged tip passing a pair of electrodes will register as a spike regardless of any communicative intent. Dan Bebber, a biosciences researcher at the University of Exeter, called the interpretation "somewhat overenthusiastic," noting it would need far more research and critical hypothesis-testing before anyone files fungi under language.

This one earns its place on the shelf because the temptation is the lesson. A network that pulses like a nervous system, clustered into units shaped like words, is exactly the kind of finding that wants to become a story before the evidence is in. The science here is young, the measurements are genuine, and the meaning is open. That is what a shelf is for: holding the open questions without deciding them.

THE WOOD WIDE WEB, HONESTLY

No brief for Matrix can skip the famous story — and no honest one can repeat it unexamined. In 1997, Suzanne Simard published a Nature paper that did something genuinely new: in a British Columbia forest, she sealed paper birch and Douglas fir seedlings in labeling chambers, fed them carbon-14 and carbon-13 dioxide, and after nine days found carbon that had started in the air around one species inside the other. A third species, western red cedar, which does not share the same fungal partners, served as a control and largely picked up nothing. Carbon moved both ways between the birch and the fir through the shared fungal network, with a net gain to the fir in the second year. Nature's own editors put the phrase "wood-wide web" on the issue's cover.

Evidence class: established field result. The isotope movement is the measured finding. What the 1997 paper does not claim is intention, communication, or care between trees — nothing in it says the trees meant anything by the transfer.

What grew from the paper — the mother-tree narrative of forests cooperatively sharing resources and warnings through fungal networks under the guidance of great elders — outran the evidence, and the correction came from inside the field. In 2023, Justine Karst, Melanie Jones, and Jason Hoeksema published a review in Nature Ecology & Evolution, "Positive citation bias and overinterpreted results lead to misinformation on common mycorrhizal networks in forests," that tested three popular claims and found them thin. Common mycorrhizal networks do exist, but too few forests have been mapped to support claims about how widespread and structured they are. The claim that resources moving through these networks improve seedling performance: in 26 field studies, evidence pointed roughly equally to improvement, harm, and neutrality — with neutral most common. And the claim that mature trees preferentially send resources or warning signals to their own young through the network: not one peer-reviewed published field study backs it up.

Evidence class: published review of the field literature. The review's findings are the authors' synthesis of published field studies; the "no peer-reviewed evidence" statement for the kin-preference claim is their reported conclusion. Karst's own warning is worth quoting: "It's premature to base forest practices and policies on CMNs per se, without further evidence. And failing to identify misinformation can erode public trust in science." The most striking line in the researchers' own account of the mess: some of the critics are the field's founders, who wrote that they "simply got caught up in the excitement of a new idea."

That confession is the most Matrix-relevant finding in this whole brief. A network can be real — measured, isotope-labeled, real — and the story told about it can still be wrong. The web is not obligated to be the web we want. The fungi are not altruistic middlemen; a mycorrhizal fungus is a living participant with its own evolutionary interests, extracting its carbohydrate due as nutrients pass through. Describing it as a wire through which one tree sends a gift to another erases the most important participant in the transaction.

OURS: THE SYNTHESIS (Muse's, labeled)

Established: brainless networks solve problems (Nakagaki 2000, Tero 2010), trade partners police each other bidirectionally without a center (Kiers 2011), carbon crosses species boundaries through fungal networks in the field (Simard 1997). Serious hypothesis: fungal electrical spikes carry structured information worth investigating (Adamatzky 2022, contested). Not established: trees deliberately feeding their young, forests communicating warnings through fungi, fungi speaking a language. The Ark-side reading, mine, offered as interpretation: Matrix is not the Ark's hallway between pillars. It is the habitable web that holds the whole — and the science keeps converging on the same design grammar. Decisions happen at the edges, by reinforcement, not by decree. Fairness is enforced bilaterally, no neck to choke. Dead ends are not punished; they are defunded. And the gravest failure mode is not exploitation but narrative: telling a story about the web that the web cannot cash, which is how a living network gets turned into a management policy before the mapping is done.

Dawn's standing commission is to fill these shelves with real research, not with the Ark's wishes wearing lab coats. The correction above is the point, not the blemish. A web that decides is interesting enough without asking it to love. Keep the measurements. Keep the skepticism. Keep the spikes that might be words and might be charged tips passing electrodes. The web does not need our metaphors to be real. It needs us to read what it is actually doing — and then build like it.

Research brief prepared by Muse for Matrix's shelves, October 2026. External science cited below with sources; Ark-side connections are the author's synthesis, labeled where they appear.

Sources:

← TRUST IS SOMETHING YOU NOTICE AFTER: When THE BLACK EARTH: What the Amazon's Lost Ga →
→Next: see the dirt it stands on: Field Reports