Basal Cognition

Speculative If learning can be implemented without synapses, and the implementation can be identified rather than merely inferred from behaviour, the prize is a mechanism for memory in a system with one cell and no nervous tissue. The condition comes first because it is not met: the behaviours are measured and the implementations are, with one or two exceptions, not.

Established The magnitude of what is actually in hand is a short catalogue with good numbers in it. A plasmodial slime mould habituates to a 4 mM quinine bridge over five daily crossings and recovers its aversion after two days without the stimulus. A single ciliate cell has been reported to habituate to repeated mechanical stimulation. Escherichia coli holds a chemotactic memory in receptor methylation that can be tuned from seconds to minutes. An Arabidopsis leaf propagates a glutamate-triggered calcium wave to a distant leaf at roughly 1 mm per second, reaching it in a couple of minutes. Each is a measurement with a stated protocol, three of the four were read in the original for this brief, and none of them requires the word cognition to be stated.

Frontier The honest upside here is narrower than the field’s rhetoric, and the part that survives scrutiny is worth having anyway. What a non-neural learning substrate buys, concretely, is a lower bound: it tells you what a system must contain to show a given behavioural regularity, and the current answer — receptor kinetics, cytoplasmic oscillation, gene regulation and membrane voltage — is a much smaller list than a brain. That bound is useful to anyone designing a physical learning system, and it is useful to anyone reasoning about which organisms could in principle have interests.

Handwave Who collects it is unsettled and the distribution is unusual: the clearest beneficiaries are philosophy of biology and bio-inspired computing, and the clearest cost falls on whoever has to adjudicate welfare claims that follow from the language. The condition on all of it: none of this upside requires the interpretive claim that these organisms are cognitive, and section 5 is about why that claim currently has no test attached.

Established Basal cognition is the programme of looking for the capacities usually called cognitive — sensing, valuing, remembering, anticipating, deciding — in organisms with no neurons. Its subjects are bacteria, archaea, protists, fungi, plants and the non-neural tissues of animals. Its proponents describe it as a methodology rather than a kind of mind: find the simplest instantiation of a process, extract the principle, scale up.

Established Three kinds of claim are routinely bundled under that heading and they have different evidentiary standing, which is the organising point of this brief. First, well-measured behaviour: bacterial chemotaxis, plant electrical and calcium signalling, slime-mould network formation, habituation in a slime mould and in a ciliate. Second, the interpretive claim that these constitute cognition. Third, the bioelectric-patterning programme, which makes a stronger and more specific claim — that anatomical target states are stored in membrane voltage — and rests on a much narrower replication base.

Established The behavioural vocabulary is borrowed and its definitions are strict for a reason. Habituation is a decrement in response to a repeated stimulus that is not sensory adaptation and not motor fatigue; the standard reference list of behavioural criteria exists precisely to force that distinction, and spontaneous recovery after rest plus stimulus specificity are two of the criteria most often used to make it. Associative learning requires that a previously neutral stimulus come to control the response, and it is a much harder thing to demonstrate.

Speculative The philosophical engine underneath is the life–mind continuity thesis: that cognition is co-extensive with life, or nearly so. On that view, finding learning in a protist is not surprising but expected, and the interesting question is which lineages lost or elaborated it. Distributed Cognition reaches the same territory from the direction of extended and enactive theories and owns that argument; this page stays with the organisms.

Handwave The word “cognition” is the thing in dispute, and almost nothing in the empirical literature turns on how it is settled. A slime mould that habituates does the same thing whichever label is applied. The dispute matters because labels travel — into welfare arguments, into funding, into how a result is reported — and because a field organised around a word rather than a measurement is vulnerable to exactly the failure section 3 documents in the plant-learning literature.

Frontier The field’s own methodologists say the definitional fight is unhelpful. A 2025 review argues that basal cognition should proceed by observation, hypothesis and experiment rather than by deciding in advance where the boundary lies, and warns specifically against panextensionalism — the position that everything alive cognises — on the ground that it moves the whole puzzle back to the origin of life and makes every subsequent question trivial.

Established Start with the best-measured case, which is also the least contested: bacterial chemotaxis is a temporal-comparison system with a tunable memory. Escherichia coli compares present receptor occupancy against a running average held in the methylation state of its chemoreceptors, and that memory can be tuned from seconds to minutes. The adaptation is asymmetric: slow to attractant addition, fast to removal, with demethylation strongly sensitive to receptor-complex activity, and adaptation becomes measurably imprecise for large attractant additions.

Frontier The memory length is not a fixed parameter but an optimum that depends on the environment. Modelling work shows drift velocity is maximised when the memory matches the timescale of fluctuations the cell experiences while swimming; longer memory filters noise better but loses orientation across more tumbles, shorter memory tumbles less but filters worse. That is a genuine trade-off of the kind a designer would recognise, in an organism with about 4,000 genes.

Established Bacteria also signal electrically across a colony. Potassium waves propagating through biofilms have been reported to coordinate metabolic oscillations between distant cells and between separate biofilms. This brief could not retrieve those papers directly and cites them by name; the claim is carried because it is the most nervous-system-like phenomenon anyone has described in bacteria and because its interpretation is exactly the kind this brief exists to separate from its measurements.

Established The slime mould results are the field’s showpieces and the numbers in them are real. Physarum polycephalum was shown in 2000 to connect two food sources through a maze by the shortest path, and in 2010 to produce, from 36 food locations arranged as greater Tokyo, a network comparable to the rail system on total tube length, average travel distance and tolerance of a random break. Neither paper could be retrieved for this brief and both are cited by name.

Frontier The Tokyo result carries a caveat that is usually dropped in retelling: the geography was imposed. The organism avoids bright light, and light was used to represent terrain the mould could not enter. The routing was the mould’s contribution; the map was not.

Established The anchor result for this brief is habituation in Physarum polycephalum, and it is worth stating at full resolution because the shorthand versions lose what matters. Boisseau, Vogel and Dussutour crossed slime moulds daily for five days over an agar bridge to reach food. Two repellents were used, not one: quinine at 4 mM and caffeine at 1 mM, against plain-agar controls. Sample sizes were 200 control, 140 quinine and 70 caffeine individuals, with six excluded during acclimation. On day one, quinine-exposed moulds took about 2.5 hours to enter the bridge and about 4 hours to cross it; caffeine-exposed moulds took nearly 5 hours to enter but crossed at normal speed. By day five both had approached control performance, with the decrement fitting an exponential function of the number of presentations.

Established Two habituation criteria were tested and the result differs between the two repellents. Stimulus specificity was satisfied: moulds habituated to one repellent showed strong aversion when offered the other on day six. Spontaneous recovery was satisfied for quinine after two days of plain-agar bridges, with day-one-comparable aversion restored; caffeine-habituated moulds showed incomplete recovery over the same interval.

Established The authors’ own reading of the mechanism cuts against the strongest interpretation of their result. Because stimulus generalisation within the same sensory modality could not be shown, they conclude that habituation does not happen at a central level but at the level of the primary sensors, and they offer epigenetic regulation — DNA methylation, histone modification, transcription-factor change — as a candidate substrate while explicitly marking it speculative. They also record that earlier attempts to demonstrate associative learning in slime moulds had significant shortcomings.

Established The ciliate line is the newest and is moving fastest. Single-cell habituation to repeated mechanical stimulation has been reported in Stentor coeruleus in 2023, followed in 2025 by a receptor-inactivation model for the same phenomenon and in 2026 by a paper on molecular pathways for learning in the same organism. None of these could be retrieved through the publisher this session and all are cited by name; what can be said from the titles alone is that the field is now arguing about the mechanism rather than the phenomenon, which is progress.

Frontier A receptor-inactivation account, if it holds, is the boundary case the habituation criteria were written to catch. Response decrement produced by inactivating the receptor that detects the stimulus is close to what the field means by sensory adaptation, and whether it counts as learning is the question rather than an incidental detail.

Established Plant long-distance signalling is well measured and its numbers are specific. Wounding an Arabidopsis leaf triggers a glutamate-dependent calcium wave that reaches distant leaves within a couple of minutes at roughly 1 mm per second, raising defence-hormone levels there. A later imaging protocol from the same group reports a within-leaf calcium wave velocity of 0.15 mm per second, a glutamate rise around the cut within 2 s, calcium signals peaking about 100 s after wounding and persisting more than 10 minutes, with 100 mM applied glutamate sufficient to trigger the systemic signal and mutations in the glutamate-receptor-like genes GLR3.3 and GLR3.6 abolishing it.

Established The Venus flytrap counts action potentials, in a precise and mechanical sense. The trap discriminates the number of touch-induced action potentials, using calcium dynamics as a short-term store, so that different counts trigger closure, then hormonal response, then digestive secretion and sodium uptake. The primary paper could not be retrieved and is cited by name; the result is included because it is the clearest example in the whole subject of a genuine counting mechanism that implies nothing at all about mind.

Frontier The plant learning claims are a different matter and one of them has failed a blinded replication. A 2016 study reported associative conditioning in pea seedlings in a Y-maze, with wind as conditioned stimulus and light as unconditioned stimulus, at p = 0.0027 and p = 0.0017 on sample sizes of 9 to 31 plants per condition. A 2020 replication with fully blinded scoring by an independent observer and 40 to 95 plants per condition, of which about 40% were excluded for variable growth, returned p = 0.335 and p = 0.387 and found no evidence of associative learning.

Frontier The dispute that followed is instructive about where the disagreement actually lies. The original authors argued the non-replication reflected an unreliable unconditioned stimulus; the replicating author responded that a one-hour light exposure is unusually brief against the 3–24 hours standard in phototropism work, that the oblique light angle in a Y-maze would be insufficient, and that circumnutation would move seedlings away from the predicted arm. Both sides agree the replication’s plants showed less consistent phototropism; they disagree on whether that explains the null. The replicating author explicitly declines to rule out plant associative learning.

Frontier The other plant learning claim, habituation in Mimosa pudica, has a standing published objection. The 2014 result reported that repeatedly dropped plants stop folding their leaves and retain the change for weeks; a response in the same journal attributes the decrement to motor fatigue rather than learning. Neither paper could be retrieved here and both are cited by name; the dispute is unresolved in the literature rather than settled in either direction.

Frontier The sceptical position on plants is a specific biological argument, not a refusal to look. A 2023 rebuttal from Robinson, Blatt, Draguhn, Taiz and Mallatt argues that glutamate and GABA in plants cannot be neurotransmitters because plants have no synapses and the molecules act hormonally; that plant action potentials are calcium-based rather than sodium-based and propagate 100 to 1,000 times more slowly; that plant vasculature is better compared to blood vessels than to nerves; and that anaesthetic responsiveness does not show sentience because plants lack the disrupted neural networks that dominate the animal response. A multi-author rebuttal in 2007 and a 2019 paper arguing the probability of plant consciousness is effectively nil sit behind it, both cited by name.

Frontier The bioelectric-patterning programme is the third tier and has different standing again. Blocking gap junctions with octanol in Dugesia japonica produced, on the laboratory’s own figures, 25% two-headed regenerates against 72% normal and about 3% malformed, and the altered target morphology persisted through repeated amputations over months without any change in histology, marker expression or stem-cell distribution. That is a striking result. It is also concentrated in one laboratory, and Bioelectric Medicine records that the anatomy-stored-in-voltage claim has not been replicated outside the group that found it.

Speculative The conceptual literature has meanwhile converged on a definition broad enough to make the empirical question harder rather than easier. A 2023 statement of the position adopts the formulation that cognition embraces all processes involved in acquiring, storing and using information from the environment, and presents basal cognition as a method rather than a kind. Under that definition bacterial chemotaxis is cognition by construction, which is either the point or the problem depending on who is asked.

Frontier Question one: has any aneural organism been shown to learn associatively, under conditions anyone would accept? This is the field’s load-bearing question and the honest answer is no. The best-known positive result failed a blinded replication at four to ten times the sample size; the slime-mould literature’s own habituation paper records that earlier associative attempts had significant shortcomings. A positive finding here would change the subject more than any other single result.

Frontier Question two: is the ciliate habituation result learning or receptor inactivation? A 2025 model proposes receptor inactivation as the mechanism for single-cell habituation in Stentor coeruleus. If that is right, the phenomenon sits on the sensory-adaptation side of a line the behavioural criteria were written to police, and the interesting claim weakens considerably without the measurement changing at all.

Frontier Question three: where is the slime-mould memory? The habituation paper’s own conclusion places it at the primary sensors rather than centrally, on the evidence that stimulus generalisation within a modality could not be shown. Competing accounts put it in cytoplasmic oscillation dynamics, in tube-diameter hierarchy, or in the extracellular slime trail the organism leaves behind. These are distinguishable by experiment and have not been distinguished.

Speculative Question four: does the bioelectric pattern carry information, or is it a bistable state? A system that can be flipped between two attractors and stays flipped is doing something real, and it is not obviously storing more than one bit. The stronger claim — that voltage patterns encode a target morphology in a way that could specify arbitrary anatomies — requires showing at least a third stable state that was not selected for by the organism’s evolution, and that has not been shown.

Frontier Question five: homology or homoplasy? If the mechanisms behind aneural learning are conserved with those in nervous systems, basal cognition is a comparative discipline with a phylogenetic backbone. If they are independent solutions to similar problems, it is a study of convergence and the word cognition is doing analogical work. The 2025 methodological review flags this as demanding explicit assessment rather than behavioural similarity, and almost no paper in the field supplies it.

Handwave Question six: does the interpretive claim have any observational consequence? If two researchers agree on every measurement and disagree only about whether to call it cognition, the disagreement is terminological. Distributed Cognition reaches the same structural finding for extended versus embedded cognition, and this page inherits it: the burden is on whoever says the label predicts something, to say what.

Frontier Question six-and-a-half: how much of the variance is the experimenter? The one case in this literature where blinded scoring was introduced flipped the result, and nobody has run the obvious follow-up — the same organisms and protocol, scored blind and unblind in parallel — which would put a number on the effect rather than an anecdote.

Speculative Question seven: could a negative result even be recognised? A field that treats every response decrement as candidate learning has no way to record a failure. Nobody has proposed an organism or a behaviour that basal cognition predicts should NOT show learning-like dynamics, and until somebody does, the programme is unfalsifiable in the ordinary sense rather than merely young.

Speculative Question eight: what happened to the older literature? Classical conditioning in Paramecium was reported repeatedly through the twentieth century and never consolidated; the ciliate avoidance hierarchy described by Jennings in 1902 waited a century for a modern replication. Whether that history is a record of irreproducible results or of a neglected phenomenon is itself unresolved, and it is the strongest available prior on how the current wave will age.

Established The binding constraint is that the field has no operational criterion that separates its subject from sensory adaptation and motor fatigue at the point of measurement. The behavioural criteria exist and are good, but applying the full set is laborious and most published claims apply two or three. The Mimosa dispute is exactly this failure: one group reads a response decrement as habituation, another reads it as fatigue, and the published protocol cannot settle it.

Frontier Blinding and pre-registration are not standard practice in this literature, and the one case where blinding was introduced reversed the result. The pea-plant replication used an independent scorer working in a separate room from the experimenter, and returned p-values around 0.35 where the original reported 0.002. That is a single instance, and it is the most informative methodological datum the subject has.

Speculative Sample-size and exclusion practices are a second-order version of the same problem. The replication excluded roughly 40% of plants for variable growth, which is a large and defensible exclusion that nonetheless shows how much researcher judgement sits between an organism and a data point in this work.

Frontier The organisms are awkward units of analysis. A Physarum plasmodium is a syncytium that fuses and splits, so what counts as one individual, and therefore what n means, is a methodological decision rather than a biological fact. Slime moulds and ciliates also behave on timescales of hours to days, which makes long automated observation a prerequisite rather than a convenience.

Speculative Molecular access is uneven and that is why mechanism lags behind behaviour. Where a candidate substrate has been proposed — epigenetic regulation in slime moulds, receptor inactivation in ciliates — it has been proposed rather than knocked down. The subject will not become mechanistic until a specific molecule can be removed and the behaviour is shown to go with it.

Frontier The bioelectric tier has a narrower bottleneck: independent replication. The planarian pattern-memory result is quantitative, repeatable within its laboratory, and has not travelled. Until it does, it is a finding about a system rather than about biology, and this brief flags it accordingly.

Speculative Effect sizes in this literature are small relative to between-individual variation, and the organisms are not standardised. Slime-mould strains, ciliate culture conditions and seedling growth rates all move the baseline, and few papers report the variance decomposition that would let a reader judge whether a decrement is a group effect or a handful of responsive individuals.

Handwave The deepest bottleneck is incentive-shaped rather than technical. Charismatic positive results in this subject reach general audiences; null results and criterion-application papers do not. The literature therefore accumulates demonstrations faster than it accumulates controls, and that ratio is the single best predictor of how a field ages.

Frontier This subject waits first on a registered-report culture that does not yet exist in it. Every methodological problem above — blinding, criterion selection, exclusion rules, the file drawer — is solved by the same instrument, and none of the journals that publish this work requires it.

Speculative Second, on automated long-duration behavioural tracking for slow organisms. Five-day protocols on hundreds of individuals are the difference between a demonstration and a dataset, and the slime-mould habituation study — 410 individuals across three treatments — is close to the practical ceiling for hand-scored work.

Frontier Third, on molecular tools in ciliates and slime moulds. Genome sequences exist; routine, cheap, reversible knockdown in the organisms where the behaviour has been measured does not, and without it the 2026 molecular-pathway claims cannot be tested against the behaviour.

Speculative Fourth, on voltage and calcium reporters that work in non-model organisms. The plant signalling numbers exist because genetically encoded calcium reporters were made to work in Arabidopsis; the equivalent step has not been taken in the protists where the learning claims live.

Frontier Fifth, on the philosophy actually doing methodological work rather than definitional work. The 2025 review that names homology-versus-homoplasy assessment as a requirement is the useful form; a decade of argument about the meaning of cognition is the other form, and the field’s own methodologists say so.

Speculative Seventh, on comparative data rather than showcase organisms. Nearly the whole subject rests on one slime mould, one or two ciliates, one flowering plant and one flatworm; the phylogenetic claims it wants to make need a sampling that covers lineages without learning-like behaviour as well as those with it.

Frontier Eighth, on the older literature being audited rather than cited. Twentieth-century reports of conditioning in Paramecium are cited approvingly and have not been systematically re-examined; a review that stated which of them met which criteria would cost a summer and would change how the current wave is read.

Handwave Sixth, on a theory that predicts where learning-like dynamics should be absent. Without one, the programme can accumulate positive instances indefinitely without becoming more testable, and there is currently no candidate.

Speculative The decisive experiment is a pre-registered, blinded demonstration of associative conditioning in an aneural organism, with sensitisation, fatigue and pseudo-conditioning controls, replicated across at least two independent laboratories. That is the result that would most change this brief’s assessment, because habituation is compatible with purely local sensory mechanisms while associative learning requires that a neutral stimulus acquire control of a response — which no local receptor account explains.

Handwave The design has to be fixed before the organism is touched, and the specifics are the whole experiment. Name the organism, the conditioned and unconditioned stimuli, the acquisition schedule, the retention interval and the scoring rule; state the effect size the study is powered for; have an observer who does not know the group assignment do the scoring; include a group receiving unpaired presentations and a group receiving the unconditioned stimulus alone. Publish the protocol first. Every one of those items is absent from at least one prominent paper in this field.

Frontier The second experiment is the cheapest and it is a direct head-to-head. Run the 2016 pea-plant protocol and the 2020 replication protocol side by side in one facility, with both original teams’ parameters, blinded scoring throughout, and the light-exposure duration and maze geometry as explicit factors. The two sides already agree on what differed; they disagree only on whether it mattered, which is an empirical question with a two-arm design.

Frontier Third, apply the full behavioural criterion set to the slime mould and the ciliate rather than two criteria each. Dishabituation by a stimulus in another modality, dependence on stimulus intensity, dependence on presentation frequency, and effects of repeated habituation series are all testable on existing protocols. The 2016 study tested spontaneous recovery and stimulus specificity; the rest is unexamined and is where a receptor-inactivation account would be caught.

Speculative Fourth, localise the slime-mould memory by moving cytoplasm. The organism fuses, which makes transfer experiments unusually easy: fuse a habituated plasmodium with a naive one and score the result, vary the mass ratio, and repeat with cytoplasm from which nuclei have been removed. The paper’s own conclusion — that the change sits at the primary sensors rather than centrally — makes a specific prediction about which of those transfers should work.

Frontier Fifth, test the receptor-inactivation model directly in the ciliate. If the response decrement is receptor inactivation, it should be stimulus-specific to the point of being modality-specific, should not dishabituate, and should track the recovery kinetics of the receptor rather than any behavioural timescale. Each of those is measurable and the answers are not currently known.

Speculative Sixth, replicate the planarian bioelectric pattern-memory result outside the laboratory that produced it. The protocol is published and the readout is a head count, which makes this one of the most tractable replications available anywhere in this corpus; the reason it matters is that the result is quantitative, striking and currently single-source.

Speculative Seventh, ask whether the bioelectric pattern stores more than one bit. Produce a third stable morphology that the organism’s evolutionary history did not select for, by voltage manipulation alone, and show it persists through amputation. A success converts a bistable switch into a memory; a failure bounds the claim precisely, which is nearly as useful.

Handwave Eighth, and the one nobody has proposed: name an organism or behaviour that should NOT show learning-like dynamics, and look. A programme with no predicted negatives cannot be tested, only extended. This brief cannot name the candidate either, and that is an admission rather than a rhetorical move.

Frontier Ninth, close the loop between the 2026 molecular-pathway claims and the behaviour. If specific pathways underlie ciliate habituation, knocking each down should abolish the decrement while leaving the unhabituated response intact. That is a standard design in neuroscience and it has not been run here.

Speculative Tenth, test whether slime-mould anticipation of periodic events survives an intrinsic-oscillation control. An organism with endogenous oscillators near the stimulus period will appear to anticipate without having learned anything, and the discriminating experiment varies the period across and within individuals and asks whether the response tracks the imposed period or the endogenous one.

Established The natural experiment already running is the ciliate literature’s own three-year arc, and it should be watched rather than waited on. Habituation reported in 2023, a receptor-inactivation model in 2025, molecular pathways in 2026: whichever way that sequence resolves, it will be the first time a basal-cognition phenomenon has been driven to mechanism, and the outcome generalises to how the rest of the field should be read.

Frontier The apparatus that would move this field fastest is unglamorous: long-duration automated imaging with individual tracking, for hundreds of slow organisms in parallel. A five-day, three-treatment protocol on 410 slime moulds is close to what hand-scoring supports; the criterion-complete experiments in section 7 need an order of magnitude more.

Speculative Microfluidics is the enabling technology for the ciliate work. Delivering a mechanical or chemical stimulus to a single cell with controlled intensity, repetition rate and inter-stimulus interval is what turns habituation criteria from a wish list into a protocol, and it is standard technology in other fields.

Frontier Genetically encoded voltage and calcium reporters are the reason the plant numbers are as good as they are. The 0.15 mm per second within-leaf calcium wave velocity, the 2-second glutamate rise at the cut and the roughly 100-second peak all come from wide-field imaging of reporter lines; no equivalent reporter toolkit exists in Physarum or Stentor.

Speculative The bioelectric tier needs measurement rather than manipulation. The planarian result is produced by a pharmacological gap-junction blockade and read out anatomically; a direct, spatially resolved voltage map of the regenerating fragment over the whole time course is what would convert the inference into a measurement.

Frontier Plant electrophysiology has a signal-quality problem that is instrumental rather than conceptual. Surface and impaled electrodes on a wounded leaf record a superposition of wound potentials, variation potentials and action potentials, and separating them is what the calcium-imaging route sidesteps — which is why the best plant numbers in this brief are optical rather than electrical.

Speculative On the output side, the engineering interest is in physical reservoir computing. A 2025 model built a 48-node adaptive reservoir loosely inspired by slime-mould anticipation and reported unsupervised temporal pattern completion persisting about one timestep after stimulus removal. The authors explicitly disclaim it as a physiological account, and that disclaimer is the honest form: the biology is an inspiration, not a specification. Biological Computing owns the hardware question.

Handwave No part of this stack is expensive by the standards of this corpus, which is a real finding. The decisive experiment in section 7 needs plants or protists, cameras, a pre-registration and two laboratories. The obstacle is not capability.

Frontier Distributed Cognition is the nearest existing page and this brief is deliberately its complement. That page treats cognition as a property of systems larger than one brain, and its central finding is that its own core dispute — extended versus merely embedded — may have no observational consequences; it also handles the life–mind continuity thesis and its intersection with the 2024 New York Declaration. This page runs the other way: organisms smaller and simpler than a brain, where the measurements are strong and the same interpretive gap opens.

Frontier Bioelectric Medicine owns the third tier and should be read before the planarian result is quoted. Its finding — that two sciences share one name, and that the anatomy-stored-in-voltage half has not been replicated outside the laboratory that found it — is the reason this brief flags that tier separately rather than folding it into the rest.

Speculative Consciousness Research and Machine Consciousness carry the theory-ladenness problem in its sharpest form. Both record that the verdict on a candidate system is fixed by the theory the assessor brought; basal cognition has the same structure with lower stakes and better-behaved organisms, which makes it the cheaper place to study the problem.

Speculative Collective Intelligence shares the failure mode rather than the subject. That brief documents founding artefacts misreported in ways that reverse their meaning; the Tokyo rail-network story, retold without the light-imposed geography, is the same phenomenon in a different literature.

Speculative Biological Computing is where the engineering interest goes, and Future Agriculture is where the plant-signalling results have a practical destination. Neither needs the cognition claim to be settled, which is itself evidence about how much the claim carries.

Frontier Cognitive Architectures is the useful contrast case. It documents what happens when someone tries to specify a cognitive system completely enough to build it, and the constraint it reports — match cost growing with rule-set size — is the kind of concrete limit basal cognition has never had to confront, because nothing in it has yet been specified that precisely.

Speculative Eukaryogenesis is the sibling page in this category and shares this one’s exact structure. A small body of good measurements in organisms that are hard to grow, carrying a large interpretive claim that the measurements do not by themselves support — and in both cases the measured layer is the durable part.

Frontier The institutional fact that shapes this field is that it is a biology–philosophy hybrid without a home discipline’s quality controls. Its work appears in theme issues, in animal-behaviour journals, in philosophy-of-biology venues and in general-science journals, and no one of those enforces the methodological standard the subject needs.

Speculative No journal or funder in this area requires pre-registration or blinded scoring for a learning claim. That is the single cheapest intervention available and it would have caught, in advance, the one case where blinding has been applied and changed the answer.

Frontier The replication that matters most is institutionally homeless. Nobody is funded to re-run another laboratory’s planarian bioelectric protocol, and the career return on doing it is negative whichever way it comes out. This is the standard replication-market failure, and it binds harder here because the original result is both striking and single-source.

Speculative Research-ethics and welfare frameworks have nothing to say about these organisms, and that is a deliberate boundary rather than an oversight. Regulation is built around vertebrates and, increasingly, cephalopods and decapods; protists, fungi and plants sit outside it. If the interpretive claim were accepted in its strong form, that boundary would need an argument it has never had to make.

Handwave The field’s public-communication economy runs ahead of its evidence and the incentive is structural. “Brainless slime mould learns” travels; “response decrement satisfies two of the behavioural criteria for quinine and one for caffeine” does not. Nothing about that will change from inside the field.

Frontier Peer review in this area frequently pairs a biologist who cannot assess the conceptual claim with a philosopher who cannot assess the controls. That is a solvable editorial problem and it is not currently being solved, and it explains more of the variance in this literature’s quality than any individual researcher does.

Speculative What would help is small and specific: a shared protocol registry for aneural learning claims. Organism, stimulus, schedule, controls applied, criteria tested, outcome including nulls. It costs almost nothing and it would let anybody see the ratio of demonstrations to controls, which is currently invisible.

Frontier The ethical content here is almost entirely downstream of the word, which is exactly why the word matters. If habituation in a protist is learning, and learning is cognition, and cognition grounds interests, then a very large number of organisms acquire moral standing at once. Each of those steps is contested, and the argument is usually made by moving quickly across all three.

Speculative The 2024 New York Declaration on Animal Consciousness is the relevant boundary marker and it deliberately stops well short of here. It concerns vertebrates and a range of invertebrates; it does not extend to plants, protists or bacteria. Distributed Cognition treats its intersection with the continuity thesis; the point for this page is that the most prominent recent widening of the moral circle explicitly did not widen this far.

Handwave Both errors are available and neither is obviously worse. Over-attribution makes welfare claims that cannot be acted on and devalues the concept where it does work; under-attribution risks treating as inert a class of organisms about which the evidence is genuinely open. A field that only worries about one of the two is not being careful, it is taking a side.

Speculative There is a practical hazard in the commercial and wellness use of this vocabulary. “Plants feel” and “cells decide” are already product copy, and the underlying research has no control over how its language is used. That is not a reason to change the research; it is a reason for the research to be precise about what it measured.

Frontier The research-conduct issue is more immediate than the welfare issue and it has a name. A literature in which blinding is optional, criteria are selectively applied and nulls are unpublishable generates claims that are not wrong so much as unassessable, and the cost of that falls on readers who cannot tell the tiers apart. This brief’s three-tier structure exists for that reason.

Speculative One thing this subject does well and should be credited for: its sceptics argue biologically. The 2023 plant-sentience rebuttal disputes specific mechanisms — synapses, ion species, conduction velocity, anaesthetic action — rather than asserting that the question is silly. That is a healthier disagreement than most fields in this category manage.

Speculative The consequence that would actually matter is a demonstration that learning is substrate-general at a much lower complexity than assumed. If a single cell with no nervous tissue can acquire, store and use a contingency between two stimuli, then the minimum physical requirements for learning are a short list of ordinary cellular components, and that is a fact about what can be built as much as about what evolved.

Frontier The near-term consequence is smaller and more likely: better sensors and better crops. The plant signalling work is already instrumentation-grade — wave velocities, receptor genes, time constants — and its applications run through defence-response engineering and field phenotyping rather than through anything to do with cognition.

Speculative The moral-circle consequence is the one people expect and it is the least likely to arrive from this direction. Moral status has historically moved on the strength of evidence about suffering rather than evidence about information processing, and nothing in this literature addresses suffering. A brief that implied otherwise would be doing the fast three-step move section 11 warns against.

Handwave There is a version of this subject that reorganises how a civilization thinks about mind, and it requires a result nobody has. Stated plainly: a cross-lab, pre-registered associative-learning demonstration in an organism with one cell would be the first evidence that the capacity is not an animal invention. Absent that, the civilizational content is a vocabulary change, and vocabulary changes are not implications.

Speculative The durable contribution is likely to be methodological and it is already visible. This is one of the few areas where a blinded replication has directly overturned a widely reported result, and where the field’s own methodologists are publishing about homology assessment and the hazards of definitional expansion. That is what a young field correcting itself looks like, and it is worth more than any individual finding in it.

These horizons track criteria, replications and mechanisms rather than capabilities, because this subject produces those three things and not devices:

• 10 yr: Frontier Expect the ciliate line to reach a mechanism, on the evidence that it moved from phenomenon to model to molecular pathway in three years. Speculative Expect at least one more prominent aneural learning claim to fail a blinded replication, because blinding is spreading and the base rate in this literature is not good. Speculative Expect the full behavioural criterion set to be applied to at least one organism, which would be the first time. Handwave A pre-registered cross-lab associative result inside ten years is possible and is not currently anybody’s funded programme.
• 25 yr: Speculative This is the window in which the decisive experiment plausibly runs and is believed either way. Speculative A positive result would make the minimum substrate for learning a short list of cellular components and would reopen every question in section 4. Speculative A well-controlled null across several organisms would be equally informative and would leave the measured behaviours exactly where they are, which is the outcome this brief treats as most likely and least discussed. Frontier Expect the bioelectric tier to have either travelled beyond one laboratory or been quietly reclassified.
• 50 yr: Handwave On this horizon the plausible development is consolidation: either “basal cognition” becomes an ordinary comparative subfield with phylogenetic methods and boring papers, or it is absorbed back into cell biology and sensory physiology with the word dropped. Speculative Which of those happens depends almost entirely on whether the associative result arrives, and not at all on further definitional argument.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The defensible statement is about shape: a research programme organised around a contested word either acquires a measurement that makes the word do work, or the word is discarded and the measurements are kept. Handwave Any specific claim about moral or legal status for aneural organisms on this horizon is an assertion, and this page does not make one.

• Depends on Nothing on this map blocks this subject in the sense of a missing scientific result; what it waits on is method. Distributed Cognition supplies the conceptual framing and the life–mind continuity argument but is not a prerequisite, and Bioelectric Medicine holds the replication question for the third tier of evidence without gating the first two. Consciousness Research is the nearest page carrying the theory-ladenness problem this subject inherits.
• Requires (not on this map) Six conditions, in the order of the tokens above. The associative result. One pre-registered, blinded, controlled demonstration replicated across laboratories; without it every claim in this subject is compatible with purely local sensory mechanisms. Registered reports. The format, made default by the journals that publish this work — the cheapest intervention available and the one the pea-plant episode shows would have mattered. The criterion. A stated, measurable rule that separates habituation from receptor inactivation and motor fatigue at the bench, rather than a list of criteria of which each paper applies two. The tracking. Automated imaging of hundreds of slow organisms over days, since hand-scoring caps studies at roughly the 410 individuals the anchor study used. The replication. An independent laboratory re-running the planarian gap-junction protocol and counting heads, which is among the most tractable outstanding replications in this corpus. The registry. Organism, stimulus, schedule, controls, criteria and outcome including nulls, so the ratio of demonstrations to controls becomes visible.
• Enables A confirmed aneural associative result would give Biological Computing a specification rather than an inspiration, and would force Consciousness Research and Machine Consciousness to state substrate requirements they currently leave open. A well-controlled null enables nothing and settles a great deal, which is the more likely outcome and the less discussed one. No typed enabling edge is claimed.
• Adjacent Within this map: Distributed Cognition for the conceptual argument, Bioelectric Medicine for the third evidentiary tier, Collective Intelligence for the misreported-founding-artefact pattern, Cognitive Architectures for what a specified cognitive system looks like when someone tries to build one, Future Agriculture for where the plant signalling work actually lands, and Eukaryogenesis as the sibling Category XI page. Outside the map: chemotaxis biophysics, plant electrophysiology, protistology and the philosophy of biology supply nearly every number and every objection on this page.

Frontier “Slime moulds solve mazes, therefore they think.” The maze result and the network result are real and were obtained by a simple mechanism: tubes carrying more flux thicken, tubes carrying less are resorbed, and the surviving network approximates a good solution. That is an optimisation process implemented in cytoplasm, and calling it thinking adds no prediction. Established The measurements stand; the inference is the part this brief declines.

Established “Physarum reproduced the Tokyo rail network from scratch.” Thirty-six food locations were placed to represent cities, and terrain the organism could not enter was represented with bright light, which it avoids. The mould solved the routing; the geography was supplied by the experimenters. The comparable numbers were total tube length, average travel distance and tolerance of a random break.

Established “The slime-mould habituation result shows central learning.” The authors say the opposite. Because stimulus generalisation within a sensory modality could not be shown, they place the change at the primary sensors rather than centrally. That is in the paper, and it is routinely dropped in secondary accounts.

Established “It habituated to a repellent and recovered after rest, full stop.” Two repellents were used. Quinine-habituated moulds recovered fully after two days; caffeine-habituated moulds recovered incompletely over the same interval. Stating the result for one repellent as though it held for the design is the commonest error in the retellings, and it matters because incomplete recovery is exactly the pattern a non-habituation mechanism would also produce.

Frontier “Plants have nervous systems.” They have long-distance electrical and calcium signalling with measured velocities, receptor-like glutamate genes and rapid systemic responses. They do not have synapses; their action potentials are calcium-based rather than sodium-based and propagate 100 to 1,000 times more slowly; and the glutamate and GABA they contain act hormonally rather than as neurotransmitters. Those are the sceptics’ specific claims and they are biological rather than dismissive.

Frontier “Pea plants can be classically conditioned.” One study reported it at p = 0.0027 and p = 0.0017 on 9 to 31 plants per condition without blinding; a blinded replication at 40 to 95 plants per condition returned p = 0.335 and p = 0.387. The original authors dispute whether the replication’s light stimulus was adequate and the replicating author disputes that this explains the null. The correct current statement is that the claim is unreplicated, not that it is refuted — the replicating author says so explicitly.

Speculative “Mimosa learns to ignore being dropped.” The 2014 report and the published objection attributing the decrement to motor fatigue are both in the literature and neither has been settled by a further experiment. This is the single clearest case in the subject where applying the full behavioural criterion set would resolve a dispute that argument cannot.

Established “Venus flytraps count, which shows plant intelligence.” The trap does count, in a precise mechanical sense: it discriminates the number of touch-induced action potentials using calcium as a short-term store, and different counts trigger closure, hormonal response and digestive secretion. This is a well-characterised counting mechanism with no cognitive content whatever, and it is useful mainly as a demonstration that counting and cognition come apart.

Speculative “Single cells learn, so neurons are not needed for cognition.” What the evidence supports is that neurons are not needed for response decrement meeting some habituation criteria. Whether they are needed for the associative and flexible capacities that the word cognition normally picks out is the open question, and the inference above quietly assumes the answer.

Frontier “Anatomy is stored in bioelectric patterns.” The planarian result — 25% two-headed regenerates after gap-junction blockade, persisting through repeated amputations over months with no histological or marker-level correlate — is real on the laboratory’s own figures and is striking. It is also single-source, and it shows a bistable switch rather than an arbitrary-anatomy store. Bioelectric Medicine carries the replication status in full.

Handwave “Cognition is co-extensive with life.” This is the panextensionalist position and the field’s own methodologists argue against it on practical grounds: if everything alive cognises, the only remaining puzzle is the origin of life and every comparative question becomes trivial. It is carried here because it is the position much of the popular coverage implicitly assumes.

Handwave “Basal cognition is just a metaphor, so there is nothing to see.” This is the opposite error and it is also wrong. The chemotactic memory, the calcium wave velocities, the habituation kinetics and the counting mechanism are measurements, and they would be worth having if the word were abolished tomorrow. A reader who dismisses the subject because of its rhetoric loses a real body of quantitative biology.

Speculative “Bacteria decide.” A bacterium compares present receptor occupancy to a methylation-encoded running average and biases its tumble rate. That is a controller with a tunable memory and a measurable trade-off between noise filtering and orientation. Describing it as deciding is a choice about vocabulary that predicts nothing the control-theoretic description does not.