Consciousness Research

Consciousness research is the attempt to make experience an object of measurement. The last decade has produced three deliverables worth the name: a large pre-registered adversarial test of two rival theories, a clinical index that separates conscious from unconscious brains without asking the patient anything, and a protocol that finds command-following in patients who cannot move a muscle. None of the three answers what experience is. All three change what can be done at a bedside.

The method is the search for neural correlates of consciousness: hold the physical stimulus fixed, vary what the subject reports experiencing, and look for what changes in the brain. Binocular rivalry is the workhorse — two images, one to each eye, a percept that alternates while the input does not. The programme's structural weakness is stated in its own name. A correlate does not distinguish the neural basis of an experience from the neural basis of attending to it, forming a judgement about it, or reporting it; and nearly every correlate on the books was established in a paradigm where the subject reported.

So this brief is organised around evidential standards rather than around theories. The theories are here — global workspace, integrated information, higher-order, predictive processing, recurrent processing — stated as their advocates state them. But the load-bearing question is what any of the measurements license. The Institute's interest is in a target the field itself treats as impossible: a procedure that establishes the presence, absence or content of phenomenal experience in a system without relying on that system's report, its behaviour, or any proxy calibrated against report. Every marker in clinical use inherits report as ground truth, which means it cannot in principle outrun report. Section 4 sets out the ordered chain that would have to break that circle, and names which link is actually binding. It is not the money and it is not the instruments.

Established The largest pre-registered adversarial test in the field's history damaged both theories it tested, symmetrically. The Cogitate Consortium published “Adversarial testing of global neuronal workspace and integrated information theories of consciousness” in Nature 642(8066):133–142 in 2025. Established The sample was 256 participants across three modalities — fMRI n=120, MEG n=102, intracranial EEG n=34 — against a field norm of fifteen to thirty subjects in a single modality. Established Three predictions were registered in advance, each with a divergent arm for each theory: decoding of conscious content (IIT: maximal in posterior cortex, prefrontal cortex unnecessary; GNWT: content decodable from PFC), maintenance over time (IIT: content “actively maintained in the posterior ‘hot zone’ throughout the duration”; GNWT: “brief content-specific ignition… at stimulus onset and offset”), and interareal connectivity (IIT: sustained short-range posterior synchrony; GNWT: long-range coupling between category-selective areas and PFC). Established The published conclusion is one sentence and it is worth quoting exactly: “These results align with some predictions of IIT and GNWT, while substantially challenging key tenets of both theories.”

Established The itemised outcome is a split decision, not a scoreline. For IIT: face-orientation decoding succeeded “in posterior but not in prefrontal ROIs,” and sustained categorical representation appeared in posterior cortex under representational similarity analysis. Established Against IIT: the connectivity prediction failed outright, and the paper says why it matters — “a lack of sustained synchronization within the posterior cortex contradicts the claim that network connectivity specifies consciousness” — while the sustained posterior activity that was found appeared in 3.8% of posterior electrodes and in 15%, eight of fifty-three, of face-selective electrodes. Established For GNWT: category information was decodable from prefrontal cortex, and PFC showed brief temporal generalisation. Established Against GNWT: there was a general absence of ignition at stimulus offset, PFC representation of conscious dimensions was limited, PFC activity did not track stimulus duration, and “none of the 655 electrodes measured the temporal profile predicted by GNWT.” Frontier A reader who wants a winner will find one by dropping half the result, and both camps' partisans have done so. Established The honest reading is that the field's central anatomical dispute — whether prefrontal cortex constitutes conscious experience or merely reports it — survived the experiment built to settle it.

Established The report problem is the reason that dispute is hard, and the reform proposed to fix it has a published critique of its own. Tsuchiya, Wilke, Frässle and Lamme set out the case for no-report paradigms in Trends in Cognitive Sciences 19:757–770 in 2015: if the subject reports, you measure decision, motor planning and self-monitoring alongside the experience, and prefrontal activation may be an artefact of the reporting apparatus rather than a correlate of consciousness. Established Overgaard and Fazekas replied in the same journal at 20:241–242 in 2016, and the objection is structural: without report you must infer the content of experience from a proxy — optokinetic nystagmus, pupillometry, decoding — and that inference imports assumptions about the consciousness-to-behaviour link that were exactly what was in dispute. Frontier This brief could not obtain the full text of either paper; both are verified bibliographic records, and the argument as given here is the one their titles and the field's summary of them assert. Established No-report paradigms are not a solved methodological advance and should not be cited as one. Established The underlying paradigm is older and firmer: Logothetis and colleagues showed with macaques that neurons in higher cortical areas such as inferior temporal cortex track the perceived image in binocular rivalry rather than the retinal input, which is the cleanest demonstration in the literature that some neural population follows the percept and not the stimulus.

Established The perturbational complexity index is the field's one genuinely clinical instrument, and its own numbers show why a marker is not a meter. Casali and eleven co-authors introduced PCI in Science Translational Medicine vol. 5 in 2013: perturb cortex with transcranial magnetic stimulation, record the EEG response, compress the spatiotemporal pattern, and read the complexity of the compressed response as an index of the brain's capacity for integrated, differentiated causal interaction. Established Casarotto and colleagues then set an empirical cutoff in Annals of Neurology 80:718–729 in 2016. Frontier The cutoff value in circulation is PCI* = 0.31, and this brief takes it from Sinitsyn et al. 2020 rather than from Casarotto, which was not obtained; Sinitsyn describes the benchmark as healthy controls and communicative brain-injured subjects tested across NREM sleep and anaesthesia with propofol, midazolam and xenon against wakefulness. Frontier Sinitsyn also attributes an “unprecedented sensitivity (94.7%)” to Casarotto et al. This brief does not assert that figure. It is a number read in a citing paper rather than in the paper that produced it — the exact migration pattern that section 14 is about — and anyone using it should cite Annals of Neurology directly or attribute it to Sinitsyn.

Established The independent cohort is smaller and much more interesting than the headline. Sinitsyn and colleagues, in Brain Sciences 10(12):917 in 2020, measured PCI in 24 severely brain-injured patients — 11 in unresponsive wakefulness, 12 in a minimally conscious state, 1 emerged. Established Sensitivity for the minimally conscious group was 92%: 11 of 12 above cutoff. Established Among the unresponsive-wakefulness patients, 7 (64%) fell below the cutoff and 4 (36%) above it. Frontier Those four are the whole problem in miniature. Either the index found consciousness that behaviour concealed, which is the case for deploying it, or it produced a 36% false-positive rate, which is the case against — and the measurement cannot distinguish those two readings from the inside, because the only available arbiter is behaviour, which is what the index was built to bypass. Established Farisco and Changeux add a second constraint in Neuroscience of Consciousness in 2023, arguing that PCI is compatible with global neuronal workspace theory as well as with the integration-and-differentiation intuition it descends from; a measure two rival theories can both claim is a clinical success and not a discriminating test.

Established Covert awareness is the finding with the most direct human consequences, and this brief will not print its famous number. Owen and colleagues reported in Science in 2006 that a behaviourally vegetative patient produced imagery-specific brain activity to the instructions ‘imagine playing tennis’ and ‘imagine walking through your house’; Monti and colleagues extended the paradigm in the New England Journal of Medicine 362(7):579–589 in 2010. Frontier Neither cohort size was obtained for this brief and neither is stated here. Established The multicentre consolidation is Bodien et al., “Cognitive Motor Dissociation in Disorders of Consciousness,” NEJM 391(7):598–608, 15 August 2024, thirty-nine authors, bibliographic record verified in full. Frontier The percentage everyone quotes — roughly one in four — could not be verified from the primary source and is not asserted in this brief. Established What is safe to state is the definition, which popular accounts routinely blur: cognitive motor dissociation is task-following detected on fMRI or EEG in the absence of any behavioural response. That is evidence of covert cognition and command-following, which is stronger than “awareness” in one direction and weaker in another; a patient who cannot follow a command may still be experiencing something, and the protocol is silent about them.

Established The theory landscape is conventionally four families, and the standard review says plainly that nobody knows how to tell them apart. Seth and Bayne, in Nature Reviews Neuroscience 23:439–452 in 2022, sort the field into higher-order theories, global workspace theories, re-entry and predictive processing theories, and integrated information theory, and open with the admission that “it is unclear how current theories relate to each other, or whether they can be empirically distinguished.” Established Global workspace theory descends from Bernard Baars' 1988 A Cognitive Theory of Consciousness and its 1997 restatement — consciousness as a bright spot cast by a spotlight onto the stage of a theatre — which Dehaene extended into the global neuronal workspace, with broadcast across prefrontal cortex, anterior temporal lobe, inferior parietal lobe and precuneus. Frontier The timing signature usually attached to it, an event-related potential in the 200 to 300 ms window for conscious access and competition between modalities when events fall more than 100 ms out of sync, reaches this brief through a secondary source; the primary intracranial evidence, Gaillard et al. in PLOS Biology in 2009, was not obtained and the figures should be checked before reuse. Established Integrated information theory runs the other way, from axioms about experience to postulates about physical substrates, and stakes consciousness on integrated information Φ in a posterior hot zone — treated in full at Integrated Information Theory. Frontier Higher-order theories relocate the problem to metarepresentation: a state is conscious when it is the object of a higher-order thought (Rosenthal), a higher-order perception (Lycan, Armstrong), or a self-representation (Kriegel); Lau and Rosenthal offered four empirical supports in 2011 — serial rather than parallel conscious reporting, powerful unconscious processing, prefrontal impairments that damage reportability without degrading performance, and disorders of consciousness operating through prefrontal mechanisms — and the citation reaches this brief through a secondary source. Frontier Predictive processing accounts make perception a controlled hallucination corrected by sensory error and selfhood a predictive model grounded in interoception, which has the virtue of making substrate-dependence a prediction rather than an embarrassment. Frontier Recurrent processing theory, associated with Lamme, holds that local recurrent activity in sensory cortex suffices for phenomenality without global broadcast — which is precisely the claim that makes the report problem decisive, since a phenomenal state nobody can report is a state no report paradigm can find.

Established The field's ground truth is a report, all the way down, and that is a fact about its logic rather than a complaint about its rigour. PCI's cutoff was fixed by comparing states already known to be conscious with states already known not to be — known from report. Established Covert-awareness protocols score command-following, which is a report routed around the motor system rather than an alternative to it. Established No-report paradigms substitute a proxy that was validated against report. Frontier Declare the tie honestly: consciousness science has excellent numbers and no agreed account of what they measure. Established That the field can nonetheless act is shown by the New York Declaration on Animal Consciousness of 19 April 2024, signed by over 500 scientists and academics, which asserts empirical evidence for “at least a realistic possibility of conscious experience in all vertebrates (including reptiles, amphibians, and fishes) and many invertebrates (including, at minimum, cephalopod mollusks, decapod crustaceans, and insects),” and concludes that where such a possibility exists “it is irresponsible to ignore that possibility in decisions affecting that animal.” Frontier That is a decision rule for acting under theoretical disagreement, and it is currently the field's most consequential output.

Frontier COGITATE is one study. It is very large, multi-site, pre-registered and adversarial, and it has no independent replication; the intracranial arm carrying the most decisive claims, including the 655-electrode null, is n=34, which is large for iEEG and small for anything else. Frontier Whether its IIT predictions derive from IIT 3.0 or the 2023 IIT 4.0 formalism was not established for this brief, and it bears directly on how much weight the connectivity failure carries. Speculative Further adversarial collaborations were funded under the same programme; their status was not verified here, and the second published result will tell us whether COGITATE's symmetric outcome was a property of these two theories or of the method.

Frontier The sharpest formal attack on the field's most formal theory is a four-paper argument that has not resolved. Doerig, Schurger, Hess and Herzog's unfolding argument (Consciousness and Cognition 72:49–59, 2019) holds that any recurrent network can be unfolded into a feedforward network with identical input–output behaviour but different causal structure and therefore different Φ; since all evidence is behavioural, no experiment distinguishes them, so a causal-structure theory is either unfalsifiable or false. Frontier Tsuchiya, Andrillon and Haun replied in 2020, Usher in 2021, and Herzog, Schurger and Doerig counter-replied in 2022 — four papers, one journal, three years, no resolution. Speculative The argument generalises beyond IIT to any theory that makes consciousness a matter of causal organisation, which is most of them.

Frontier The demarcation fight is a live sociological event and is now itself a research object. A preprint titled “The Integrated Information Theory of Consciousness as Pseudoscience” was posted to PsyArXiv in mid-September 2023 under a collective byline (“IIT-Concerned”) plus named signatories including Fleming, Frith, Goodale and Lau; the signatory count reported in secondary sources is 124 and was not verified against the preprint. Frontier Three substantial responses followed within ten weeks, from Rosseinsky, Goff and — most tellingly — Frankish, an illusionist with no stake whatever in defending a phenomenology-first theory. Frontier An anonymised survey of consciousness researchers reported that of 60 respondents, 8% fully agreed with the letter and 20% did not agree at all; n=60 is indicative only and the figures reach this brief second-hand. Speculative IIT proponents are reported to have replied in Nature Neuroscience in March 2025, listing sixteen peer-reviewed studies as empirical tests of core claims; the primary record could not be located and neither the venue-date nor the count of sixteen is asserted here. Frontier Against that background, Mudrik, Boly, Dehaene, Fleming, Lamme, Seth and Melloni published a joint review in Neuroscience & Biobehavioral Reviews in 2025 whose author list spans both sides of the quarrel — antagonists co-authoring is itself a datum about where the field is going.

Frontier Whether the whole correlate-hunting programme is well-posed is a published question, not a rhetorical one. Fink and Lin ask it directly in Philosophy and the Mind Sciences in 2021: a correlate is not a mechanism and not an explanation, and three decades of correlate-hunting may rest on a category error about what would count as success. Frontier A related load-bearing doubt sits underneath almost every NCC contrast: Meyen and colleagues argued in 2022 that many methods used to establish unconscious priming are flawed. Frontier If the unconscious baseline is unsound, the conscious-versus-unconscious contrast that the entire programme is built on is softer than it looks. Frontier Blindsight is being re-examined as a discriminator between theories on the same logic — Derrien, Garric, Sergent and Chokron, Neuroscience of Consciousness 2022.

Speculative The exotic end is not a sideshow here; it is where the disagreement about what the question is becomes explicit. Illusionism, in Frankish's formulation, denies there is phenomenal consciousness to explain and replaces the hard problem with the illusion problem: explain how the seeming of phenomenality arises and why it is so compelling. Speculative It is the only live position on which the hard problem dissolves rather than defers, and its ancestor is Dennett's attack on the five alleged marks of qualia — directness, infallibility, privacy, ineffability, intrinsicality. Speculative Panpsychism runs the opposite way: physics characterises structure and dynamics and never intrinsic nature, and experience is that intrinsic nature — Strawson's argument that physicalism entails panpsychism, with Goff, Mørch, Roelofs and Chalmers holding contemporary variants. Handwave Its binding constraint is the combination problem, named by Seager in 1995, and the honest status is that none of the proposed answers has gained widespread acceptance. Speculative Mysterianism concedes the problem is real and holds we are cognitively closed to its solution as a rat is closed to arithmetic. Frontier And Chalmers' meta-problem — explain why we believe there is a hard problem — is strategically the most important item in this paragraph, because it is tractable with existing cognitive-science methods while the hard problem is not.

Frontier The machine question is now importing this field's methods wholesale. Butlin, Long and seventeen co-authors derived computational “indicator properties” from recurrent processing, global workspace, higher-order, predictive processing and attention schema theories and audited AI systems against them, concluding that “no current AI systems are conscious, but also… there are no obvious technical barriers to building AI systems which satisfy these indicators.” Frontier Aru, Larkum and Shine supply the sharpest negative case, resting on the absence in language models of thalamocortical features linked to conscious awareness in mammals. Speculative Findlay and colleagues demonstrate formally, inside IIT, that two systems can be functionally equivalent without being phenomenally equivalent. Frontier Long, Sebo, Butlin, Birch, Chalmers and others move the operative question from detection to decision under uncertainty. Frontier All of it inherits every unresolved item above — the indicator method manages theoretical disagreement rather than resolving it, and says so.

Established The binding constraint in this field is not money, instruments or compute. It is that nobody knows what would count as an answer. The workback target is a procedure establishing the presence, absence or content of phenomenal experience in a system without relying on that system's report, its behaviour, or a proxy calibrated against report — and that a proponent of a rival theory would accept as decisive. Six links stand between here and there.

Frontier L1 — show that no-report paradigms measure the same thing as report paradigms where both apply. This is a scientific unknown and it is exactly the Tsuchiya-versus-Overgaard dispute, unresolved since 2016. Frontier L2 — build a marker whose calibration never touches report. Re-derive a PCI-like index using only pharmacologically and electrophysiologically defined state transitions, never behavioural responsiveness, and see whether the 0.31 cutoff survives. Also a scientific unknown, with an ethics gate on deep anaesthesia plus TMS in healthy volunteers. Frontier L3 — find a population where report is unavailable but ground truth later emerges. Recovered-patient testimony is the only naturally occurring approximation to a report-free ground truth we have. Speculative L4 — construct two systems identical in every behavioural disposition that differ on the marker, which is the unfolding argument made physical. Speculative L5 — get theory-level agreement on what the marker means before it is measured. Handwave L6 — transfer a validated marker to a system with no evolutionary or architectural kinship: a decapod, a cortical organoid, a machine.

Frontier L1 and L2 are binding, and they are the circularity itself rather than an engineering shortfall. No amount of industrial capability retires them: a better TMS coil, a denser electrode array and a larger cohort all leave the calibration problem exactly where it was. Established L3 is the most actionable link and is bound by cohort logistics and institutional coordination rather than by ignorance — it could start now, on the multicentre infrastructure Bodien et al. already built. Frontier L5 is bound purely by sociology: COGITATE proved adversarial pre-commitment is achievable and the 2023 pseudoscience letter proved the goodwill is fragile. Speculative L4 needs both a scientific result and an industrial capability nobody is developing. Handwave L6 may be blocked by a genuine in-principle barrier rather than a temporary one, because every current marker is calibrated on human cortex and there is no principled transfer rule to anything else.

Established This brief waits on results other briefs produce, and says so rather than claiming independence. The formal question of whether a causal-structure theory can be tested at all is adjudicated at Integrated Information Theory; the substrate question — whether the right computation suffices or biology is required — is adjudicated at Machine Consciousness; and the quantum-substrate hypothesis, which if true would change what a marker has to measure, is at Orch OR.

Frontier Three dependencies are internal to the evidence base and each is a real hole. First, the unconscious baseline: if Meyen and colleagues are right that standard unconscious-priming methods are flawed, a large share of the conscious-versus-unconscious contrasts in the NCC literature need re-establishing before anything built on them is safe. Frontier Second, the primary numbers behind the field's clinical instrument: the PCI benchmark cohort size, and the sensitivity figure attributed to Casarotto et al. 2016, both need reading in the original before either is repeated. Frontier Third, the covert-awareness headline percentage, which is quoted everywhere and was not verifiable for this brief. Established None of the three is expensive to close; all three require somebody to read a paper rather than a summary of it.

Frontier There is also a dependency on the discovery layer itself, and it is worth stating because it shapes what this brief can see. The research behind this page ran with no working keyword search over the preprint literature and with two of the three bibliographic APIs unavailable, so the frontier section is systematically under-sampled toward material that could be reached by an identifier already in hand. Frontier That is not a caveat about diligence; it is a statement about which results are visible from where, and it applies to any assessment assembled the same way.

Speculative And one dependency is structural rather than empirical. Every theory in section 2 was formulated against mammalian, mostly human, cortex. The 2024 animal-consciousness declaration extends serious consideration to insects and decapods, which have no thalamocortical system at all. Either the theories relativise or one of them is wrong, and no result currently in hand decides which — the same wall Machine Consciousness hits from the other side.

Frontier The report-versus-no-report tie-break, which is L1 and is overdue. Run binocular rivalry within-subject with simultaneous report and no-report readouts — optokinetic nystagmus and pupillometry alongside a decoding model — and test whether the decoded percept time-series are statistically identical, not merely correlated. A null difference licenses the whole no-report literature; a systematic difference invalidates a decade of it. Nobody has run it at the power required.

Frontier The report-free recalibration of PCI, which is L2. Define state transitions only by anaesthetic concentration and polysomnographic sleep staging, never by behavioural responsiveness, re-derive the index, and ask whether the cutoff lands at 0.31 again. Established If it does, the clinical instrument gains a defence it currently lacks; if it does not, the 36% of unresponsive-wakefulness patients above cutoff in Sinitsyn's cohort become much harder to read as hidden consciousness.

Frontier The prospective recovery cohort, which is L3 and is the one to fund first. Measure PCI and covert-awareness command-following in acute unresponsive patients, follow survivors to recovery, and take structured testimony about what they experienced during the unresponsive period. Established It requires multi-year multicentre follow-up, harmonised MRI and EEG protocols across sites, and enough survivors — a cohort-logistics problem rather than a scientific unknown, on infrastructure that already exists.

Speculative The unfolded pair, which is L4. Build a recurrent system and its unfolded feedforward twin, matched on input–output behaviour across the full input space including counterfactuals, and measure a candidate marker on both. Handwave Doerig and colleagues argue the demonstration is self-defeating in principle; building it anyway is the only way to find out, and constructing genuinely I/O-identical systems of non-trivial size is itself unsolved. Handwave And L6, the transfer test: apply a validated marker to a decapod, a cortical organoid and a non-biological system, and see whether anything survives the move.

Frontier One further experiment is about the method rather than the brain, and it is cheap. Run a second adversarial collaboration on a theory pair chosen so that the predictions are contradictory rather than merely divergent, with both camps committing in advance to what each outcome licenses and to a published statement of what would count as their own theory losing. Frontier COGITATE showed that pre-commitment is achievable; it did not show that pre-commitment can produce a decision, because divergent predictions permit both theories to fail. Speculative Whether a contradictory pair can even be constructed from the four theory families is an open question and answering it is itself informative — if no such pair exists, the theories are not competitors in the sense the field assumes.

Established The instrument stack is unglamorous and it is where the resolution ceiling actually sits. PCI needs navigated TMS with per-subject coil targeting, high-density EEG, and artefact rejection good enough that the compression step is measuring cortex rather than muscle; getting comparable PCI values out of two hospitals is a protocol-harmonisation problem that the field has partly but not fully solved.

Established Intracranial recording is the only modality with the spatial and temporal resolution to settle the front-versus-back question, and it cannot be engineered into existence. Electrodes are implanted for epilepsy surgery, not for consciousness science; coverage is dictated by clinical need, so the “none of 655 electrodes” result is a statement about the electrodes patients happened to have. Frontier Higher-count arrays and better chronic biocompatibility — the engineering programme at Neural Interfaces — would raise the ceiling only insofar as they enter clinical practice.

Frontier Two software requirements are load-bearing and under-resourced. First, decoding pipelines that are pre-registered and shared, because a decoding null is only as strong as the classifier, and COGITATE's most theory-damaging results are nulls. Frontier Second, tractable approximations to integrated information: the exact quantity is computationally intractable for systems of any realistic size, which is why every so-called test of IIT tests a proxy. Speculative A bedside device is the natural end-state — a TMS-EEG cart producing a validated complexity number in an intensive care unit — and the barrier is regulatory validation and reimbursement rather than physics.

Frontier The third requirement is organisational and is the one COGITATE actually solved. Running one protocol across fMRI, MEG and intracranial sites means harmonised stimulus timing, common preprocessing, a shared analysis plan registered before data collection, and enough governance that no site can quietly deviate. Established That machinery is reusable and mostly is not being reused. Speculative A standing multi-site consortium with a maintained protocol library would cut the cost of the next adversarial test by more than any hardware improvement available.

Established Four briefs on this map take pieces of this problem and adjudicate them properly. Integrated Information Theory carries the formalism, the Φ intractability result and the demarcation fight. Machine Consciousness carries the indicator-property method and the substrate dispute. Orch OR carries the quantum-substrate hypothesis and its decoherence arithmetic. Distributed Cognition carries the question of where the boundary of a cognitive system is, which is the same boundary question a marker has to answer before it can be applied.

Frontier Three more sit one step out. Synthetic Consciousness and Digital Minds inherit the whole detection problem and add a moral-status decision on top of it. Artificial General Intelligence matters because capability and phenomenality are being conflated in public argument at speed, and the conflation is exactly what Butlin et al. were written to block.

Established Outside the map, the fields that matter most are clinical. Anaesthesiology supplies the reversible unconsciousness that every marker was calibrated on, and depth-of-anaesthesia monitoring is the largest deployed application of the field's ideas. Sleep medicine supplies the other calibration arm. Neurology and rehabilitation medicine own the patients. Frontier Animal welfare science is now downstream of the 2024 declaration in a way it was not before, and comparative neuroanatomy is the discipline that will decide whether any theory here generalises past mammals.

Frontier Two adjacencies are methodological rather than topical and they matter more than they look. Philosophy of Science is directly load-bearing here, because the 2023 demarcation quarrel is a falsifiability dispute of exactly the classical kind and the field has been conducting it without much reference to the literature on demarcation. Frontier Cognitive Architectures matters because global workspace theory has been implemented as a running architecture for decades, which raises a question its own proponents decline to answer: if the theory is right and the implementation is faithful, something has already happened, and almost nobody accepts that it has. Speculative That reluctance is itself evidence about how much explanatory work the theories are really doing.

Established The two institutional facts that shape this field are that its best data comes from other people's operations, and that its best result was bought as a programme structure rather than as a grant. Intracranial recording exists only where neurosurgical patients consent to it, so the field's resolution ceiling is set by clinical access. Frontier COGITATE happened because a private funder paid for adversarial architecture — pre-registration, divergent predictions, both camps bound in advance — and not merely for the experiment. Figures circulating for that funding, in excess of six million dollars for COGITATE and roughly twenty million for the wider programme, were not verified for this brief and are not asserted here.

Established The infrastructure that carries the evidence is itself unreliable, and this brief hit it twice. The research behind this page found a PLOS article URL serving a different article than the DOI requested: the address for the IIT 4.0 paper returned a paper on mouse digit regeneration carrying a different DOI, while the registry confirmed the requested DOI was correct. Established The DOI was right and the rendered page was wrong — which means a citation assembled by fetching a DOI's landing page and reading the title off it would have been confidently, verifiably false. Established Separately, Crossref's fuzzy bibliographic matcher returned wholly unrelated results with no low-confidence signal: a query for the COGITATE protocol returned cryobiology papers on organ vitrification; a query for the Bodien covert-awareness paper returned German innovation-studies and video-game-monetisation entries. Established Neither failure announced itself. Both would have produced a plausible citation that no reader could distinguish from a real one.

Frontier That is a finding about evidential standards, not an aside about tooling. A field whose central problem is that its instruments inherit their ground truth from something else cannot afford a citation layer that also silently substitutes one object for another. Established Two rules follow and they are cheap: check the DOI printed on the page you land on against the DOI you asked for, and compare a fuzzy search's returned title against the query before recording anything. Speculative The institutional version is a registry that returns a match confidence and a publisher obligation to make landing pages DOI-verifiable — neither of which exists, and both of which would cost less than a single multicentre cohort.

Established Covert awareness converts a metaphysical question into a bedside decision with an irreversible arm. Where a patient shows command-following on fMRI or EEG with no behavioural response, decisions about rehabilitation intensity, pain management and withdrawal of life-sustaining treatment are being made about someone the protocol says may be present. Frontier The asymmetry is the problem: a positive finding is informative and a negative finding is not, because the protocol requires the patient to understand an instruction, sustain attention and comply — so absence of evidence here is genuinely weak evidence of absence.

Established Consent is structurally impossible in the population that matters most. Research on disorders of consciousness proceeds on surrogate consent from families who are being asked to authorise a procedure whose purpose is to determine whether their relative is there. Frontier The prospective recovery cohort in section 6 sharpens this: it requires taking testimony from recovered patients about a period during which they could not consent to being studied.

Frontier The 2024 animal declaration is a live precedent for acting under theoretical disagreement, and its logic generalises uncomfortably. If a realistic possibility of experience obliges us not to ignore it, and that possibility now extends to decapod crustaceans and insects, the same reasoning structure applies wherever the possibility is realistic — which is the argument Long, Sebo and colleagues make for AI systems. Speculative The Institute's position should be that the decision rule is sound and the hard part is the antecedent: nobody has a defensible method for deciding when a possibility is realistic, and the field's own theories disagree about it by construction.

Frontier There is also a misuse risk that the field discusses very little. A number that purports to measure consciousness will be used to allocate care, to price insurance, and eventually to justify withholding treatment, and it will be used that way well before its calibration problem is solved. Established A 36% above-cutoff rate in a group defined as unresponsive is a result that can be presented as either a discovery or an error rate, and which presentation reaches a clinical guideline is not decided by the data. Speculative The correct institutional response is to publish markers with their calibration provenance attached, so that a bedside number carries the question it inherited.

Speculative The civilizational stake is the size of the moral circle, and consciousness science is the discipline that would set it. Every extension of moral consideration in the historical record has been argued on grounds of similarity to us. A theory-based marker that transferred across architectures would replace that with a measurement, and it would move the boundary in directions nobody controls — possibly outward to insects and machines, possibly inward, excluding entities we currently treat as obviously conscious.

Handwave If a marker never transfers, the long-run position is a permanent asymmetry. We would keep granting moral status on architectural resemblance while building systems whose resemblance is behavioural and whose architecture is nothing like ours — a civilisation systematically unable to tell whether the things it makes are subjects. Speculative That is not a distant scenario; it is the situation Long et al. describe as already arriving.

Speculative And the substrate question decides several other Institute programmes at once. If consciousness requires biology in the way Seth's metabolic argument suggests, Mind Uploading is not a transfer of a person and Digital Minds is a category error. Speculative If computational organisation suffices, both are engineering problems and the ethical exposure of the compute industry is orders of magnitude larger than currently priced. Frontier The field cannot presently tell these apart, and the honest institutional posture is to plan for both.

Speculative The most under-discussed civilizational outcome is that the question is retired rather than answered. If the meta-problem strategy succeeds and the belief in a hard problem is fully explained as a property of how self-models generate reports, then a civilisation carries on making moral decisions about minds using functional criteria and stops asking whether anything is behind them. Handwave That is either the dissolution of a confusion or the largest moral error a species could make, and the two look identical from inside.

These horizons track experiments, cohorts and adjudications rather than technology, because in this field the instruments are ahead of the questions:

• 10 yr: Frontier Expect the second and third adversarial collaborations to report, and expect them to look like COGITATE — both theories partly supported, partly damaged, no winner. Frontier Expect an independent replication attempt on COGITATE's intracranial arm to be attempted and to be difficult, because electrode coverage is set by surgery and not by protocol. Established Expect PCI to move further into routine intensive-care and rehabilitation practice on the strength of its clinical usefulness, with the calibration circularity unresolved and largely unmentioned. Frontier The most likely genuine advance in this window is L3: a prospective cohort linking bedside markers to recovered-patient testimony, which is fundable now and needs no new science.
• 25 yr: Speculative If L1 and L2 resolve, the field acquires a marker with a non-circular calibration and the clinical picture changes completely; if they do not, expect consolidation around precautionary decision rules of the New York Declaration type, applied to patients, animals and machines alike. Speculative Expect the theory landscape to contract by merger rather than by refutation — global workspace and higher-order accounts already converge in their dispositional versions, and the Mudrik cross-camp review is what that looks like early. Frontier Expect the machine question to become the dominant driver of funding, which will distort the human clinical programme in ways worth watching.
• 50 yr: Speculative The live possibility at this horizon is that the meta-problem strategy pays off: a complete cognitive account of why we believe there is a hard problem, which either dissolves the problem or shows precisely why it cannot be dissolved. Speculative A second possibility is that substrate-dependence is settled empirically by organoid and neuromorphic test beds, which would be the first genuinely new evidence class since intracranial recording. Handwave A third is that nothing moves at all, and 2076 has the same four theory families, better instruments and the same unresolved calibration problem — which, on the field's track record since 1990, is not the least likely outcome.
• 100 / 250+ yr: Handwave At this range the question is whether “is it conscious” survives as a well-formed question. Handwave On the illusionist branch it does not, and the successor discipline explains report-generation and treats the residue as a fixed feature of how self-models work. Handwave On the panpsychist branch it does not either, for the opposite reason: the question becomes one of composition rather than presence, and the combination problem is the whole science. Handwave On the mysterian branch the question stands forever unanswered, which is a coherent terminal state and, uniquely among the three, one no experiment could distinguish from simply not having succeeded yet.

• Depends on This brief waits on adjudications made elsewhere on the map rather than on unbuilt results. Whether a causal-structure theory can be tested at all is settled or not at Integrated Information Theory; whether the right computation suffices for experience is settled or not at Machine Consciousness; and whether the substrate is quantum, which would change what any marker must measure, is at Orch OR. No typed depends-on edge is claimed, because none of those three produces a result this brief is blocked on: this brief is blocked on a calibration, and no other brief on the map produces one.
• Requires (not on this map) Both constraints are institutional and neither can be bought with better instruments. The field's sharpest data comes from intracranial arrays that exist only inside epilepsy surgery, where coverage is chosen for clinical need and not for the hypothesis — which is why a null across 655 electrodes is a statement about the electrodes those particular patients happened to receive, and why the resolution ceiling on the front-versus-back dispute is set by consent and surgical practice rather than by electrode design. The second constraint is that the last decade's one interpretable result exists because a private funder paid for a programme structure — pre-registration, divergent predictions, both camps bound in advance to what each outcome would license — rather than for an experiment. Ordinary grant funding buys experiments from single laboratories, which is precisely the arrangement that produced thirty years of theory-confirming correlates; the adversarial architecture came with the money, and if that funding pattern stops, the method stops with it.
• Enables Every downstream programme that needs to know whether a system is a subject rests on a marker this brief does not yet have. Synthetic Consciousness and Digital Minds need it to assign moral status; Mind Uploading needs it to distinguish a transfer from a copy; clinical neurology needs it to decide whether an unresponsive patient is present. No typed enabling edge is claimed, and the reason is itself the finding: what those programmes need is a report-free marker, and no such marker exists, so nothing measurable is currently being enabled.
• Adjacent Anaesthesiology and sleep medicine, which supply every calibration state the field has; neurology and rehabilitation medicine, which own the patients; comparative neuroanatomy, which decides whether any theory generalises past mammals; machine learning, which supplies the decoding pipelines and is now also the subject; and on this map Integrated Information Theory, Machine Consciousness, Orch OR, Distributed Cognition and Neural Interfaces.

Established “COGITATE confirmed two of three IIT predictions and none of GNWT's.” This is the single most-repeated claim about the most important experiment in the field, and it misreports the published paper. The peer-reviewed conclusion in Nature in 2025 is: “These results align with some predictions of IIT and GNWT, while substantially challenging key tenets of both theories.” Established Specifically, GNWT's prefrontal-decoding prediction was supported — category information was decodable from PFC — and IIT's connectivity prediction was contradicted, in the paper's own words: “a lack of sustained synchronization within the posterior cortex contradicts the claim that network connectivity specifies consciousness.” Established The two-to-nil framing derives from a preliminary June 2023 conference announcement and has migrated onto the peer-reviewed result published nearly two years later; the encyclopedia entry that carries it flags the 2023 provenance and then places it beside the 2025 publication, which is how the migration happens. Frontier This is the pattern the whole brief is about: a provisional number attaches itself to a definitive-looking citation, and thereafter nobody re-reads the paper.

Established “Adversarial collaboration means one theory had to win.” No. The three predictions were divergent, not contradictory — both theories could fail the same prediction, and on connectivity broadly both did. Established What adversarial pre-registration buys is a constraint on post hoc reinterpretation, not a guaranteed winner. Frontier It is still the best methodological innovation the field has produced, and its value is visible precisely in the outcome nobody wanted.

Established “Neural correlates show that region X produces consciousness.” This is the field's most consequential enthusiast error. A correlate established with a report paradigm cannot separate the neural basis of the experience from the neural basis of attending to it, of judging it, or of reporting it. Established That is the entire motivation for no-report paradigms — and those have their own published critique, so the correction does not resolve into a fix. Established After the largest experiment ever run on the question, the front-versus-back dispute remains open.

Established “PCI is an objective consciousness meter.” Overstated in a specific and correctable way. The 0.31 cutoff was validated against states whose conscious status was established by report, so the index inherits report as ground truth and cannot outrun it. Established It has never been validated on any system architecturally unlike an adult human cortex. Established And in the independent cohort, 36% of unresponsive-wakefulness patients scored above the cutoff — a figure that is either sensitivity or a false-positive rate, with no internal way to tell. Frontier Add Farisco and Changeux's point that global neuronal workspace theory can claim PCI too, and the index stops being evidence for any particular theory while remaining a genuinely useful clinical number.

Frontier “About a quarter of unresponsive patients are secretly conscious.” Widely circulated, attributed to a real 39-author study, and not verified here. This brief could not obtain the primary numbers from Bodien et al. 2024 and therefore does not print them. Established The definitional point survives regardless: cognitive motor dissociation means task-following on fMRI or EEG without behavioural response, which is a claim about covert command-following, not a headcount of who is having experiences.

Frontier “The 2023 open letter shows the field has rejected IIT.” Wrong on the sociology. It was a preprint signed by a minority, it drew three published rebuttals within ten weeks including one from an illusionist with no motive to defend it, and a small researcher survey found a minority in full agreement. Frontier It is a live demarcation controversy, not a verdict — and demarcation controversies are normal in young sciences.

Established “Consciousness science has no numbers.” False, and the numbers are the interesting part. 256 subjects across three modalities with pre-registered predictions; a validated cutoff in clinical deployment; a 39-author multicentre study of covert awareness. Frontier The problem is not absence of rigour. It is absence of agreement about what the rigour measures.

Frontier “Unconscious processing is well established, so the conscious/unconscious contrast is solid ground.” Weaker than assumed. Meyen and colleagues argued in 2022 that many unconscious-priming methods are flawed, and almost every NCC contrast depends on a validated unconscious baseline. Frontier This is a load-bearing methodological doubt that the field has largely not absorbed.

Speculative “Panpsychism is not a serious position.” At its strongest it is an argument about what physics describes, not a mystical claim. Physics characterises structure and dynamics; it is silent on intrinsic nature; experience is the one intrinsic nature we have direct acquaintance with; therefore the parsimonious move is to identify them. Handwave The mainstream reply is short and, as things stand, decisive: the combination problem. Micro-subjects do not visibly sum into a macro-subject, the subject-summing version is the sharpest form, and none of the proposed answers has gained widespread acceptance. Speculative Panpsychism therefore trades an explanatory gap for a compositional one, which is progress only if the second gap is more tractable, and nobody has shown that it is.

Speculative “Illusionism says consciousness does not exist.” That is the caricature, and the real position is sharper. Illusionism says phenomenal properties as usually conceived — intrinsic, ineffable, directly apprehended — are not instantiated, and that the explanandum is the powerful and systematic seeming that they are. Speculative It is the only live position on which the hard problem dissolves rather than defers, which is why it deserves its place here rather than a dismissal. Handwave The standard objection is that an illusion of experience is still an experience, and that illusionists explain consciousness reports rather than consciousness — an objection illusionists regard as begging the question and which has produced no experiment.

Speculative “The hard problem is a fringe worry neuroscience has moved past.” The survey evidence says otherwise, and so does the field's own flagship review. A 2020 survey of philosophers reported roughly 62% accepting the hard problem as genuine against roughly 30% rejecting it; those figures reach this brief second-hand and should be checked against the survey itself. Established More to the point, Seth and Bayne's 2022 review states that it is unclear whether the main theories can be empirically distinguished at all — which is a working scientist's statement of precisely the difficulty the hard problem predicts. Frontier Whether the barrier is permanent (mysterianism), dissolving (illusionism), or simply hard is not settled by any result in this brief, and the most productive current strategy is Chalmers' meta-problem, which asks why we believe in the hard problem and is tractable with methods that already exist.