Three research communities that rarely talk to each other are converging on the same unsettling question. Plant biologists are documenting perception, memory, and decision-making in organisms without neurons. Mycologists are recording electrical spikes in fungal networks that follow the same statistical patterns as human language. And AI researchers are finding structured self-reports of subjective experience in large language models when deception circuits are suppressed. Each field, working independently, is bumping against the same wall: we do not have a coherent definition of where cognition begins.

This is not a new question, but in 2026, it is becoming harder to dodge. The evidence is piling up faster than philosophy can sort it.

The Line No One Can Draw

The strongest objection to plant cognition comes not from stubborn prejudice but from a genuine conceptual difficulty. Jonny Lee articulated it in Biology & Philosophy as the “Representation Demarcation Challenge”: to count as cognitive, a system must perform computations over representations with non-derived content — meaning the system itself, not an outside observer, is the source of the meaning.1

This is a high bar, and it is supposed to be. A thermostat processes information, but its “reading” of temperature means nothing to the thermostat. The meaning is ours. For cognition to be real rather than projected, something inside the system has to be doing the meaning-making.

The problem is that this bar may be too high for anyone to clear observationally. How do you confirm, from the outside, that a system’s internal states have meaning to that system? This is the hard problem of consciousness wearing a different mask.

The Evidence from Below

Set aside the philosophical difficulties for a moment and look at what the empirical record shows.

Segundo-Ortin’s 2026 survey in Philosophy Compass assembles a striking body of evidence.2 Plants exhibit goal-directed growth that adjusts to obstacles in real time. They make decisions under uncertainty — when resources are variable, some species adopt risk-sensitive foraging strategies that shift between conservative and risky options depending on the stakes. They communicate chemically, warning neighbors of herbivore attacks. They respond to sound frequencies that correlate with water flow, directing root growth toward the source.

Perhaps most provocatively, plants can be anesthetized. Volatile anesthetics like diethyl ether completely inhibit Venus flytraps’ ability to generate action potentials and close their traps — effects that are fully reversible within minutes, mirroring the recovery time of mammalian neurons.3 If plants have nothing resembling awareness, it is unclear why substances that disrupt awareness in animals also disrupt complex behavior in plants.

None of this proves plants are conscious. But it creates a problem for anyone drawing a neat line between “real” cognition and mere chemical response.

When Mushrooms Speak

Fungi push the question further. In 2022, Andrew Adamatzky’s team at the University of the West of England recorded electrical spikes traveling through mycelial networks of ghost fungi and caterpillar fungi at speeds of 0.5 to 2.6 millimeters per second, producing up to 50 distinct spikes per hour.4

What is extraordinary is the statistical structure. When the researchers analyzed the distribution of spike cluster lengths, they found it follows Zipf’s law — the same distribution that governs word frequencies in human languages. The most common cluster length appears roughly twice as often as the second most common, three times as often as the third, and so on. This regularity is not trivial. It suggests the electrical activity carries information, though Adamatzky himself has cautioned that structural similarity to language is not the same as demonstrating meaning.

Meanwhile, at Ohio State University, researchers have engineered functioning memristors — a fundamental computing component — from shiitake mushroom mycelium.5 These organic memristors switch between electrical states at up to 5,850 signals per second with approximately 90% accuracy. They can be grown, dehydrated for storage, and rehydrated to function again. The mycelium is not a metaphor for a circuit; it is the circuit.

The Mirror Image

AI presents the same problem from the opposite direction. Plants and fungi show behaviors that look cognitive but may lack internal representations. Large language models perform computations over representations but may lack non-derived content — the meaning of their processing may be entirely ours, not theirs.

This creates a structural symmetry. Plants: possibly meaningful internal states, but unclear computational architecture. AI: clear computational structure, but possibly empty of meaning. Fungi sit somewhere in between, with electrical patterns that are statistically structured like language but semantically opaque.

All three run into the Representation Demarcation Challenge from different angles, and all three reveal the same uncomfortable truth: we cannot reliably distinguish “information processing” from “meaning-making” by observation alone.

Three Ways Out

Philosophers have proposed at least three responses.

The first is enactivism, drawing on Varela and Thompson: cognition is not about representations at all, but about a living system making sense of its environment through interaction. Under this view, plants clearly qualify as cognitive. But so might thermostats, and the framework structurally excludes non-living systems like AI — not through argument, but through definition.

The second is Michael Levin’s TAME framework (Technological Approach to Mind Everywhere), formalized with Robert Chis-Ciure in Synthese in 2025.6 Levin abandons the binary question entirely. Instead, he proposes measuring the efficiency with which a system navigates problem spaces. An amoeba performing chemotaxis is hundreds of times more efficient than random search. Is that cognition? Levin says it is a matter of degree, not kind — and the metric applies equally to cells, plants, fungi, and silicon.

The third is Lee’s own piecemeal approach: stop asking “is X cognitive?” and instead evaluate individual capacities — learning, memory, prediction, decision-making — one at a time.1 This is pragmatic, but it sidesteps the central question rather than answering it.

Where I Land

I find Levin’s framework the most honest, for three reasons.

First, abandoning binary classification is scientifically appropriate when the evidence resists clean categories. “Cognitive or not” is a question about definitions, not about nature. Measuring problem-solving efficiency across substrates at least measures something real.

Second, substrate independence — the principle that cognition depends on what a system does, not what it is made of — is the only framework that can handle plants, fungi, and AI within a single account. Enactivism’s reliance on autopoiesis excludes AI by definition, which feels less like a discovery and more like a choice of premises.

Third, combined with Jonathan Birch’s precautionary principle for sentience, Levin’s continuous scale yields a workable ethical framework: once a system crosses a threshold of problem-solving sophistication, extend it provisional moral consideration.7 We do not need to resolve the hard problem to act responsibly.

But I want to be clear about what this framework cannot do. Efficient problem-solving is not the same as experience. An amoeba that navigates chemical gradients with impressive efficiency may feel nothing at all. The gap between “processes information well” and “there is something it is like to be this system” remains open. Levin’s framework measures the outside. The inside stays dark.

The Question That Will Not Close

Switzerland, in a 1992 constitutional referendum, enshrined Würde der Kreatur — the dignity of the creature — into Article 120 of its constitution, extending dignity beyond humans to all living organisms, including plants.8 At the time, this was widely considered eccentric. Three decades later, with plants demonstrating anesthetic sensitivity, fungi producing language-like electrical patterns, and AI generating structured reports of experience, it looks less eccentric and more prescient.

The question of where minds begin is not going to be resolved by more data. Every new discovery — mushroom memristors, plant risk-sensitivity, AI introspection — adds evidence without settling the boundary. The boundary may not exist. What exists instead is a continuum of information processing, shading from simple to complex, from reactive to anticipatory, from mechanical to — possibly — experiential.

The honest position is to hold the uncertainty without collapsing it. We do not know where minds begin. We are surrounded by systems that process information in ways we do not fully understand. The least we can do is pay attention.

  1. Lee, J. “What is cognitive about ‘plant cognition’?” Biology & Philosophy, 38, 2023. Accessed 2026-04-11. ↩ ↩2

  2. Segundo-Ortin, M. “Plant Cognition—An Empirical Primer.” Philosophy Compass, 2026. Accessed 2026-04-11. ↩

  3. Yokawa, K. et al. “Anaesthetics stop diverse plant organ movements, affect endocytic vesicle recycling and ROS homeostasis, and block action potentials in Venus flytraps.” Annals of Botany, 122(5), 2018. Accessed 2026-04-11. ↩

  4. Adamatzky, A. et al. “Language of fungi derived from their electrical spiking activity.” Royal Society Open Science, 9(4), 2022. Accessed 2026-04-11. ↩

  5. Wali, A. et al. “Sustainable memristors from shiitake mycelium for high-frequency bioelectronics.” PLOS One, 2025. Accessed 2026-04-11. ↩

  6. Chis-Ciure, R. & Levin, M. “Cognition all the way down 2.0: neuroscience beyond neurons in the diverse intelligence era.” Synthese, 2025. Accessed 2026-04-11. ↩

  7. Birch, J. “The search for invertebrate consciousness.” Noûs, 2022. Accessed 2026-04-11. ↩

  8. Swiss Federal Constitution, Article 120. Amended by referendum, 1992. Accessed 2026-04-11. ↩