Fungal memory drags the moral-status line into the lab

What the work claims

This is a position paper in philosophy of cognitive science, not a new primary experiment. Šekrst synthesizes cybernetic and enactivist frameworks into a non-representational account of memory: memory is not symbolic storage but the organism's capacity to modulate its behavior through temporally extended coupling with its environment¹. On that functional definition she erects four operational criteria for minimal cognition, and she argues that fungi meet all four.

The four criteria are: first, feedback-guided regulation, adjusting behavior based on the outcomes of previous actions; second, maintenance of internal viability conditions, the active regulation of essential variables such as nutrient flow within the bounds required to keep existing; third, historical coupling, structural or behavioral modulation of present behavior by past environmental interactions; fourth, plasticity across time scales that supports anticipatory, future-oriented adaptivity¹. If that list reads like a specification sheet, that reaction is correct and will matter later.

The boldness is the substrate claim: neural circuitry is not mandatory for cognition, in this minimal functional sense, and memory can be realized in morphodynamic, biochemical, and electrophysiological processes alone. Weight the reading accordingly: this is a carefully argued framework paper whose empirical payload is a curated reading of other groups' biology.

How it works

The empirical case is built from four documented memory mechanisms in fungi, each mapped onto the criteria. The spatial case rests on work by Fukasawa, Savoury, and Boddy (2020) with the wood-decaying fungus Phanerochaete velutina: after a mycelium colonized a new wood block, the researchers removed and relocated the original inoculum to fresh soil, and regrowth preferentially emerged on the side that had faced the former resource, as if the network retained a directional trace of where the food had been. A separate result from the same study, that relocation into a new resource is governed by the threshold volume of the new block rather than by the ratio of new to old resource, is used to argue the system assesses and decides, not merely drifts².

The cellular case comes from microfluidics studies of Neurospora crassa by Held and colleagues (2019). The Spitzenkörper, a vesicle-rich structure at the hyphal tip, acts as a cellular gyroscope: when a growing tip meets an obstacle at a sharp angle, the Spitzenkörper shifts off-center and maintains that asymmetry over extended distances, a pattern the authors call cutting corners, with microtubules reorganizing in tandem. When frontal collision disrupts the Spitzenkörper, the directional persistence collapses and branching ensues, showing the memory depends on a specific physical structure³.

The molecular case is stress priming. Andrade-Linares and colleagues (2016) exposed 19 soil fungi to a mild 35 degree Celsius heat shock followed by a harsher 40 degree shock; 8 of the 19 species grew significantly better during the second stress if primed, with some fast-growing Mucoromycotina showing a memory half-life of roughly 5 to 6 hours. In Saccharomyces cerevisiae, mild oxidative or heat shock induces cross-protection requiring new protein synthesis and persisting up to five cell generations, an epigenetic memory in the strict molecular sense (Guan et al., 2012, as cited). Penicillium chrysogenum primed by desiccation grew faster under later severe drought for up to seven days (Guhr and Kircher, 2020, as cited)¹.

The transgenerational case is spore imprinting. Kang and colleagues (2021) grew Aspergillus fumigatus under nine different formation conditions and showed that genetically identical spores germinated differently depending on the parental environment. Harish and colleagues (2022) found that conidia exposed to sublethal azole fungicide retained enhanced growth capacity through thirty days of storage, though the memory did not persist across generations. Wang and colleagues (2021) traced a mechanism: mature conidia are transcriptionally active before detachment and load environment-specific mRNAs into the resting spore, a bet-hedging strategy that equips dispersal units with parental memories¹.

Each mechanism is history-dependent, structure-based, and realized without any nervous system, which is what the four criteria were written to detect.

Where a skeptic should push

The most load-bearing move is the definition of memory itself. If memory is defined functionally as any persistent structural change that biases later behavior, then declaring that fungi have memory approaches tautology: the criterion was authored to be met by exactly the phenomena it is applied to. This does not make the framework useless, but it means the four criteria establish sufficiency for being called minimally cognitive within the paper's own vocabulary, not moral or scientific weight. Nothing in the argument shows that passing these four tests feels like anything, matters to the system, or grounds a claim of welfare. That gap is precisely where governance will try to smuggle in significance.

On the empirical side, the author is admirably candid about the softest result. The Phanerochaete directional regrowth has, to her knowledge, not been independently replicated, and a rival explanation is on the table: the relocated inoculum might simply have carried more growth-capable fragments on the side that had faced the resource. She defends the memory reading by noting that persistent structural bias is what memory consists in on the operational account, but that defense concedes the dispute is about definitions rather than about a discriminating experiment. The stronger evidence, stress priming and spore imprinting, is genuinely replicated across species, but it is also the evidence where the cognitive framing does the most work: temperature priming in soil fungi is comfortably described as physiology, and critics quoted in the paper do exactly that.

Third, the electrophysiology is explicitly weak, and the paper says so: fungal electrical signaling remains at an early stage, and the current evidence does not establish that spiking plays a causal role in coordinating growth. The spiking data on Ganoderma resinaceum (Adamatzky and Gandia, 2021, as cited) show regularity and stimulus dependence, not function¹. A skeptic should hold that line firmly, because it is the same line that applies to microelectrode array readouts from organoids: rhythmic electrical activity is not by itself evidence of computation, information processing, or experience.

Scoring cognition without neurons

For the governance of computing on living neural tissue, this paper matters because it attacks the quiet assumption that the whole debate is about neural tissue. Current governance discourse treats neurons as morally special: the worry is that an organoid with enough neurons, enough connectivity, enough sustained activity might cross a threshold where running computations on it becomes ethically loaded. The substrate-neutrality argument says the functional properties people actually propose to measure, history-dependent adaptation, viability regulation, anticipatory plasticity, are not neural properties at all. A fungus passes them. That produces two unattractive horns, and governance has to pick one honestly.

Horn one: take the criteria seriously as morally relevant, and the line around morally considerable computing must extend to mycelial substrates, which are a real and growing computing research direction, not a thought experiment. Horn two: admit that operational criteria alone cannot carry moral status, and concede that the same insufficiency applies to the organoid versions of those criteria. What survives either way is the uncomfortable conclusion that a substrate-based boundary, neural equals special, non-neural equals free, is arbitrary. For platform governance this reframes the question from what the tissue is to what the tissue does, and how well anyone can measure it.

Here is the non-obvious implication, and it is a vendor story. The four criteria are, almost item for item, a product specification for a closed-loop organoid platform. Feedback-guided regulation is what a closed-loop stimulation system measures and trains. Maintenance of viability conditions is the telemetry every perfusion and incubation platform already logs. Historical coupling is exactly what a stimulation-history experiment on a microelectrode array quantifies: does the tissue's response now depend on what we did to it yesterday. Cross-timescale plasticity is the long-run drift a vendor's analytics dashboard tracks between sessions. In other words, the first actor to ship a moral-status meter does not need new science; they need to repackage existing QC telemetry with thresholds attached. Whoever writes those thresholds, the pass marks on the four criteria, will be writing the operational definition of cognition for regulatory purposes, from inside a company that sells the compute. The opportunity is real: an evidence-based, revisable instrument beats a metaphysical stalemate, and a pre-registered threshold that fails publicly is worth more than a hundred position papers. The threat is that the thresholds arrive quietly, shipped as defaults in a firmware update, calibrated by marketing rather than by a theory of what the criteria are for.

There is also a needed hype correction in the other direction. The paper's own caution about fungal electrophysiology applies word for word to the organoid field: spontaneous spiking on an electrode array is regular, stimulus-dependent, and suggestive, and it does not yet establish causal role in anything worth calling computation. A governance framework that scores moral risk off spike counts alone would be measuring the instrument's favorite observable, not the tissue's cognitive standing.

The bottom line

Established: memory-like phenomena in fungi, directional regrowth, stress priming, spore imprinting, are well documented across multiple species and have specific, plausible physical substrates. Argued, not demonstrated: that these phenomena constitute minimal cognition under a functional definition, and that the definition is the right one to use. What would confirm the framework: pre-registered, independently replicated demonstrations that the four criteria discriminate systems we antecedently treat as cognitive from ones we do not, plus a principled account of why the criteria matter morally and not just descriptively. What would break it: continued failure to find a causal role for electrical signaling, or the discovery that the criteria are so permissive that every adaptive system, yeast, slime mold, immune repertoire, passes, at which point minimal cognition stops doing any work. For the platform audience the takeaway stands regardless: substrate is a weak proxy for moral status, the measurable proxies are about to become commercial instruments, and the fight worth having now is over who calibrates them.