The most thorough survey ever made of explanations for consciousness runs to 142 pages and covers, by its author's own count, more than three hundred of them. Its authors' verdict is not that the field is narrowing down. It is that theories are proliferating rather than being ruled out as the data accumulate. That is not what a maturing science looks like. This is a report on a field that has become genuinely, publicly fractious: what its people are actually arguing about, why one of its most interesting researchers changed his mind after nearly dying, and what happened when I held his position to exactly the same standard the field is currently using to attack its own leading theory.
In March 2021 a Spanish theoretical physicist called Àlex Gómez-Marín was bleeding to death from an internal haemorrhage. He is not a fringe figure. He holds a physics degree, a master's in biophysics and a doctorate in theoretical physics from the University of Barcelona, did his postdoctoral work at the EMBL-CRG Centre for Genomic Regulation and the Champalimaud Centre in Lisbon, and since 2016 has run the Behaviour of Organisms Laboratory at the Instituto de Neurociencias in Alicante as a tenured scientist of the Spanish Research Council. He has published research in Neuron and Nature Neuroscience. Before this, his subject was the neurobiology of action and perception across species, and the arrow of time in physical systems.
By his own account his heart stopped for about seven seconds. What he describes is a vertical well with golden light above it and three figures he recognised, none of them family, who offered to help him climb towards it. He refused, knowing that if he continued there would be no return. He wrote it up two years later in a short paper under the title What happens with the mind when the brain dies?, and returned to it at length in a book published in Spanish in October 2025.
I want to be exact about what that paper says, because coverage of this material usually errs in one direction or the other — and because I am going to hold his position to a hard standard later, which is only fair if I state it as he actually put it.
He is not hedging. The abstract of the 2023 paper reads: "Ancient traditions, manifold personal experiences, nuanced philosophical views, and recent scientific evidence, all point to the brain as a filter (or receiver) of consciousness rather than its fanciful producer."
That is an assertion, not a possibility raised for consideration. The paper is an advocacy piece and reads like one — it describes opponents as "dogmatic materialists" and their position as "more philosophical than scientific… professed with the zeal of a stubborn ideology," it argues that words like "paranormal," "supernatural" and "pseudo-scientific" function as "conversation stoppers" revealing "mulish prejudice," and it invokes the bardo, tukdam and the Tibetan Book of the Dead approvingly. It also contains the line "The plural of anecdote is data," which is the exact inverse of the usual warning and is clearly meant to be.
He is also right about things. He raises terminal lucidity — the brief return of mental clarity that sometimes occurs in severely cognitively impaired patients shortly before death — as a phenomenon the field has largely declined to study, and on that he is correct. The paper is combative, sincere, and considerably more interesting than a hedged one would have been.
In 2025 he co-authored a paper in Nature Neuroscience with Anil Seth — one of the most prominent naturalists in consciousness science, a man whose entire research programme assumes consciousness is something brains do. They disagree about almost everything that matters here. They wrote it together anyway.
That paper, and why two people who disagree that fundamentally ended up on the same page, is the best way into the state of this field.
Start with an arithmetic problem the field does not like discussing.
For most of the twentieth century, scientific theories of consciousness essentially did not exist. The topic was regarded as unrespectable. Then, from around 1990, it became respectable again — and the theories arrived.
In August 2024, Robert Lawrence Kuhn published a 142-page survey in Progress in Biophysics and Molecular Biology titled "A Landscape of Consciousness." It arrays explanations along a physicalist-to-non-physicalist spectrum across roughly a dozen major categories — materialist theories in ten sub-varieties, non-reductive physicalism, quantum theories, integrated information theory, panpsychisms, monisms, dualisms, idealisms, anomalous and altered-states theories, and what he calls challenge theories. Kuhn describes it as covering at least 325 theories.
Two honest caveats, since this number gets thrown around. It is one survey, not a series showing growth over time — you will see "200" and "325" quoted as though the count rose between 2024 and 2025; it did not, and the lower figure appears to be press coverage mistaking Kuhn's remark that he had discussed consciousness with over 200 scientists and philosophers for a count of theories. And the taxonomy figure in that paper is credited to Kuhn and Àlex Gómez-Marín, who also reviewed it — so the survey is not independent of this piece's protagonist, and you should know that before it is used as evidence.
So take a stricter, independent source. Restricting themselves to neurobiological theories — no panpsychism, no idealism, no quantum consciousness — Anil Seth and Tim Bayne tabulated twenty-two live proposals in their 2022 Nature Reviews Neuroscience review, and made the observation that actually matters. Their words: instead of theories of consciousness progressively being "ruled out" as empirical data accumulates, they seem to be "proliferating."
That sentence, not the headline number, is the finding. It is a claim about a rate, made by two mainstream researchers about their own field, and it is the thing that should be uncomfortable.
This is worth sitting with, because it inverts the usual story. The popular framing is that consciousness science is slowly closing in on an answer. The measurable reality is that it is diverging. Whatever else is true, a discipline whose theory count rises every year is not converging on anything.
Divergence of that kind produces friction, and in the last three years the friction has become visible.
In 1998 the neuroscientist Christof Koch bet the philosopher David Chalmers a case of wine that within twenty-five years someone would identify a clear neural mechanism producing consciousness. On 23 June 2023, at the annual meeting of the Association for the Scientific Study of Consciousness in New York, Koch conceded publicly and handed over the wine — Portuguese, by report. Chalmers won because nobody had found it.
That is a charming story and it is also a data point. The person conceding was not a sceptic. Koch has spent his career on the assumption that the answer is in the brain and is findable. He still thinks so. He simply lost the bet.
In 2023, a large group of researchers published an open letter branding Integrated Information Theory — the field's most mathematically ambitious proposal, associated above all with Giulio Tononi — pseudoscience. Not wrong. Pseudoscience. In a field this small, that is a professional accusation as much as an intellectual one, and it landed as such.
In March 2025 the row acquired a formal record. Nature Neuroscience published three pieces on the same day, and read together they are the clearest available map of what this field actually disagrees about:
What the third position actually argues. Gómez-Marín and Seth make four moves. IIT, whatever its problems, does yield empirically testable predictions — and they point to the Cogitate and INTREPID adversarial collaborations as proof, since those consortia are testing specific IIT claims right now. The dichotomy being drawn between computational functionalism and IIT is a false one. The real danger in the row is scientific excommunication by authority — settling a theoretical dispute by declaring one side outside the discipline. And what the field needs instead is metaphysical pluralism and epistemic humility: tolerating incompatible starting assumptions long enough to find out which ones generate testable claims.
Note who is making this argument. Seth is a naturalist who thinks consciousness is a biological process. Gómez-Marín, since 2021, has been investigating whether it might not be. They agree that the way to settle it is not by vote.
The temperamental split running under all of this is the interesting part. One camp believes the field's problem is that it is too tolerant — that unfalsifiable theories are crowding out testable ones, and the remedy is to name them and exclude them. The other believes the problem is that it is too narrow — that the field has been ruling out possibilities on metaphysical grounds and calling it rigour. Both diagnoses predict theory proliferation. They just disagree about the cause.
So let us take the heterodox option seriously — which means giving it its strongest form, and then applying exactly the standard the field is currently applying to IIT. Anything less is condescension dressed as open-mindedness.
William James, in his 1898 Ingersoll Lecture Human Immortality, proposed a transmissive function for the brain, with an analogy that has never been bettered: a prism does not manufacture colour, it refracts light that was already there. Two years earlier, in Matter and Memory, Henri Bergson had gone further — the brain's job is primarily eliminative, cutting away the vast majority of what is available to leave the narrow band that is useful for action. Aldous Huxley later popularised the second version as the "reducing valve."
The reason this survives is that it is not obviously incompatible with anything. Correlation between brain states and experience is exactly what you would expect if the brain were a receiver. A damaged radio distorts the broadcast. That is why the standard neuroscientific evidence does not, by itself, settle the question — and anyone who tells you it does has not understood the claim.
Support 1 — vivid experience during minimal brain activity. The classic argument: patients report rich, structured experiences during cardiac arrest, when the EEG is flat. If there is no brain activity, the experience cannot be brain-produced.
What happened: the largest study of its kind went and looked. AWARE-II (Parnia et al., Resuscitation 191, 109903, 2023) ran continuous EEG and cerebral oximetry through CPR at twenty-five hospitals. Of 567 in-hospital cardiac arrests, 53 survived; 28 completed interviews; 11 of those 28 reported memories or perceptions suggestive of consciousness. The recall rate is real and it is striking. And in the brain-monitored subset, despite severe cerebral ischaemia, normal EEG activity — delta, theta and alpha — emerged as late as 35 to 60 minutes into CPR.
In 2023, Borjigin's group at Michigan published gamma surges in the dying human brain (Xu et al., PNAS 120(19), e2216268120). In two of four comatose patients after withdrawal of ventilatory support, they recorded a rapid surge in gamma power and connectivity — local activity at the temporo-parieto-occipital junction, the posterior cortical "hot zone" that IIT and others already regard as central to conscious processing, with global connectivity to contralateral prefrontal areas. Small sample, mechanism unknown.
What this does and does not establish, stated carefully. The dying brain is considerably more active than the flat-line story assumed. That removes the premise the classic argument rests on — "there was no brain activity, therefore the experience was not brain-produced" is not available if there was brain activity. What it does not establish is that the recorded activity produced the reported experiences: AWARE-II did not show the EEG episodes coinciding with the recalled ones, and the inference is mine, not the paper's. It is also worth knowing that Parnia reads his own result the opposite way — he hypothesises that disinhibition in the dying brain may open access to dimensions of reality normally screened out, which is a filter-style reading of the same data. My own view is that the simpler reading is better supported — but I am a reader of his paper, not a party to the work, and where a reader and the investigator who generated the data disagree, the reader does not get the casting vote. What I can say with confidence is narrower and still worth saying: the specific inference "flat EEG, therefore not brain-produced" is no longer available, because the EEG was not flat. What replaces it is genuinely open.
Support 2 — psychedelics: less brain activity, more experience. The finding is real. Carhart-Harris and colleagues (PNAS 2012) showed psilocybin reduces cerebral blood flow and BOLD signal, most sharply in hub regions including the thalamus and the anterior and posterior cingulate — and the magnitude of the decrease in the anterior cingulate and medial prefrontal cortex predicted the intensity of the subjective effects. Read as filter theory: turn down the reducing valve, more gets through.
What happened: the fuller picture is not "less activity." It is local decrease alongside global increase — reduced connectivity within networks, increased connectivity between them, greater variability across almost all region pairs. That is a reorganisation, not a reduction, and a production account predicts it just as comfortably.
The entropy literature that grew out of this is a more interesting case, and I want to report it accurately rather than conveniently. In June 2026, McCulloch, Olsen, Ozenne, Stenbæk, Armand, Madsen, Knudsen and Fisher of the Neurobiology Research Unit in Copenhagen published a multi-metric evaluation in Nature Communications: thirteen previous studies had each applied a different measure of brain entropy, so they applied fourteen of them to the same psilocybin data. The result is genuinely two-sided. Five of the fourteen metrics showed consistent significant positive associations with the psychedelic state; eight of the fourteen showed no significant effect at all. Their own conclusion is "a nuanced acute psychedelic effect on brain entropy" — not an absence of one. So the honest verdict is not that the finding collapsed. It is that a literature routinely cited as a single robust result turns out to depend heavily on which of fourteen measures you pick, and most of them do not show it.
Support 3 — terminal lucidity. Gómez-Marín is right that this is genuinely strange and genuinely under-researched: patients with severe dementia or structural brain damage who become briefly lucid, coherent and recognisably themselves shortly before death.
What happened: nothing much, and that is the problem. It remains largely a case-report literature. There is no prospective study with pre-registered criteria, no agreed prevalence figure, and no mechanism proposed on either side that has been tested. It is currently an anomaly, not evidence. Anomalies are worth funding. They are not worth citing as support.
1/5 on evidential standing. The filter hypothesis makes no quantitative prediction that differs from a production account; the premise of its strongest empirical argument has been removed by better measurement; its most popular modern support turns out to depend on which of fourteen metrics you choose; and its most intriguing phenomenon has not been studied prospectively at all.
Now the part that has to be said in the same breath. That is not a refutation, and the rating is not a verdict on Gómez-Marín. Unfalsified and unsupported is precisely the position the field's critics say IIT is in — and if the standard is good enough to apply to Tononi, it is good enough to apply here, and it produces the same verdict, not a worse one. Neither camp is entitled to treat the other as obviously unserious.
It also leaves his second argument standing, and that one I think is correct: that these phenomena deserve investigation rather than dismissal. AWARE-II is the proof of it. Somebody went and looked at the exact moment the whole dispute turns on, and found something nobody expected. The result is now contested — including by the investigator who produced it, who reads it against the reading offered here. That is a disagreement about evidence, which is an enormous improvement on a disagreement about whether the question is allowed to be asked.
Beneath the argument, the work continued. Five things moved, and three of them matter.
In April 2026, Barrett, Milinkovic, Mediano, Rosas, Bor, Barnett and Seth — none of them IIT partisans — concluded that Φ is not well-defined for real brains, has never been computed on anything realistic, and that calling existing measures "approximations" of Φ is an overclaim; proxy is the correct word. They also rejected both of the critics' headline arguments as misreadings. The operational consequence is blunt: the field's only fully mathematical theory currently cannot return a number for a human being.
The INTREPID blind-spot protocol. In January 2026, PLOS ONE published a study protocol from the TWCF-INTREPID consortium — fifteen authors including Giulio Tononi (IIT), Karl Friston (Active Inference) and Cyriel Pennartz (Neurorepresentationalism), sitting on the same paper, having agreed in advance what would count as evidence against them.
IIT's prediction is the arresting one: because of how the blind-spot region of cortex is wired, and even though it is never stimulated, experienced space around your blind spot should be compressed — in both eyes, right now, as you read this. Predictive processing accounts predict no such warping. The experiment is running. Protocol only; no results.
In January 2026, Daniel Freeman (MIT Lincoln Laboratory), Brian Odegaard (Florida), Seung-Schik Yoo (Harvard Medical School) and the philosopher Matthias Michel (MIT) published a roadmap for using transcranial focused ultrasound to identify the neural substrate of conscious perception. Almost everything known about the neural correlates of consciousness is correlational. tFUS is non-invasive, reaches centimetres deep at millimetre resolution — deeper and sharper than TMS — and it modulates rather than merely records. That converts "this region lights up when you're conscious" into "switch this region and see whether experience changes." The roadmap is built to separate cognitivist accounts (experience requires higher-level machinery — reasoning, self-reflection, frontal integration) from non-cognitivist ones (specific activity patterns give rise to specific experiences directly, no interpretive layer, plausibly posterior or subcortical).
On 19 August 2026, Earl Miller, Scott Brincat and Jefferson Roy of MIT's Picower Institute published "Analog Cognition and Consciousness" in The Journal of Neuroscience. Synapses store representations, they argue, but travelling waves of rhythmic activity — slow alpha/beta regulating faster gamma, coordinating neurons partly through electric-field effects — determine which representations are active. Where waves intersect they add and subtract: analog computation, run in parallel, exploiting the brain's own physics. In their words, consciousness depends "less on specific receptors or cell types and more on the integrity of large-scale wave organization."
The supporting evidence is anaesthesia, and it comes from two separate studies that are worth keeping apart. Miller's earlier work with Emery Brown concerns brainwave dynamics directly. Then in March 2026, Eisen, Miller and Ila Fiete reported in Cell Reports that propofol, ketamine and dexmedetomidine — three drugs with three unrelated molecular mechanisms — all produce unconsciousness by destabilising the same thing: the brain's balance of stability and excitability, what they call dynamic stability. The convergence is the striking part either way. Three different molecular routes, one functional casualty.
July's piece offered four theory families. This is a fifth, it is not a variant of the other four, and it arrived while I was writing. §2, illustrated.
The 2025 Nature adversarial test (n = 256; fMRI, MEG, iEEG) — whose results, in the paper's own careful wording, "align with some predictions of IIT and GNWT, while substantially challenging key tenets of both" — was always the first of two. The consortium has been analysing the second large-scale experiment since mid-2025. As of this writing no results are public. It remains the single most likely source of a genuine update to this question in the next twelve months. Worth noting as a footnote on §3: the IIT camp reads the first experiment's results considerably more favourably than the field's summary does — even the scoring of an adversarial collaboration is contested.
A Japanese programme funded as a Grant-in-Aid for Transformative Research Areas (A), running 2023–2028, with Naotsugu Tsuchiya as area representative and Masafumi Oizumi leading one of its planning groups, asks the question sideways. Rather than trying to explain why experience is felt, it asks what the structure of experience is — treating a quale as characterised entirely by its relations to all other qualia, and using category theory to formalise it.
Its headline empirical result (Kawakita, Zeleznikow-Johnston, Takeda, Tsuchiya & Oizumi, iScience, March 2025) used unsupervised alignment on colour-similarity judgements to compare relational structures across people without ever comparing one experience to another directly. The critical caveat, which is routinely dropped: the alignment was performed at group level. The authors did not perform individual-level alignment. This result does not show that your red differs from mine, and anyone reporting it that way is overstating it.
Here is why the falsifiability sub-claim above sits at 4/5 and not 5/5. The INTREPID authors concede it themselves: no theory involved specified an effect magnitude. They agreed the direction the effect should go. Nobody stated how big it should be.
A small warping is consistent with IIT. A large one is consistent with IIT. A null result can be attributed to insufficient power. An adversarial collaboration that specifies direction but not magnitude may run beautifully and arbitrate nothing. This is the single most important sentence in this piece, and it applies to my own framework too — see §8.
This section is my interpretation of the material above, not a finding reported by any of the researchers cited. Read it as a proposition to argue with.
Put §2 next to §6 and a mechanism suggests itself.
Theories multiply where theories are cheap to make and disconfirmations are expensive to produce. In consciousness science, a new theory costs one clever person one good idea. A disconfirmation costs a multi-site consortium, four years, hundreds of participants across three imaging modalities, and the prior agreement of people whose careers depend on the outcome. The Cogitate experiment is what a single disconfirmation attempt costs. There are 325 theories.
The magnitude problem then closes the trap. If predictions specify direction but not size, even expensive disconfirmations fail to eliminate anything — the losing theory absorbs the result and continues. So the elimination rate stays near zero while the generation rate stays high, and the count rises every year regardless of how much work anyone does.
The proposition: theory proliferation in consciousness science is not a symptom of the hard problem being hard. It is a symptom of an asymmetric cost structure between making claims and killing them. If that is right, the highest-leverage intervention is not another theory or another consortium — it is a norm requiring predicted magnitudes, which would make cheap theories expensive to hold.
This connects directly to the companion Investigation published alongside this one, which found the same failure mode in quantum gravity — a field with no shared personnel, methods or subject matter. Two instances is not a pattern. A third would be, and finding one is now on the queue.
Readers deserve the full picture rather than the flattering half. Since 2021 his research has moved into territory most of his colleagues reject outright. He was one of three joint winners of the inaugural Linda G. O'Bryant Noetic Sciences Research Prize, administered by the Institute of Noetic Sciences, for a proposal to test "seeing without eyes" — extra-ocular vision in blind individuals. He has co-published with Rupert Sheldrake on whether perceptual fields extend beyond the brain. He is a former director of the Pari Center, now Director Emeritus, and advocates what he calls a "trans-materialist" science.
These are contested to the point of professional risk, and I am not endorsing them; extra-ocular vision in particular has a long history of positive preliminary results that do not survive tightened controls. But two things can be true. His methodological argument — the one co-signed by Anil Seth — stands on its own merits and does not depend on any of this. And a field that responds to heterodoxy by excommunication rather than by measurement is proving his point for him. The correct response to "seeing without eyes" is a well-controlled experiment, and he has proposed one.
Theory proliferation may simply be what an immature science looks like from inside, and may resolve on its own once the measurement tools mature — tFUS being exactly the kind of development that could start eliminating candidates at speed. On that reading my "asymmetric cost structure" is a description of a temporary condition dressed up as a mechanism. I think the magnitude argument survives that objection. I am not certain it does.
My own framework, Recursive Field Theory (RFT), treats selfhood and experience not as contents sitting somewhere in the brain but as a state a system falls into when its relations hold together — coherence maintained across difference. On that view, what you are is not an inventory. It is a shape.
Three of this year's developments sit close to that, and I want to be careful about how close. The Japanese qualia-structure programme says a quale is characterised by its relations to all other qualia — a relational ontology of experience, arrived at through category theory, with an empirical result behind it. IIT's blind-spot prediction says experienced space is set by connectivity structure even where nothing is happening. And the wave/analog account says consciousness tracks the integrity of large-scale organisation rather than any particular substrate — which is why three chemically unrelated anaesthetics produce the same unconsciousness. Three independent lines, all locating experience in structure rather than in stuff.
Now the discipline, and §6 is where it bites. The INTREPID paper's flaw is my flaw. A framework that predicts "experience tracks coherence" without saying how much, measured how, is in exactly the position IIT is in — congenial to the evidence and unable to be embarrassed by it. It is also, by the standard I applied in §4, in the same position as the filter hypothesis. I have rated that 1/5 in public; consistency requires me to say that an unquantified RFT would deserve no better.
I should be clear that this is not a concession offered in the abstract. It is a description of what the theoretical work currently consists of. RFT's closure primitive is already written as a ratio of measurable quantities rather than as a metaphor, which is the first half of the requirement; the second half — stating in advance what magnitude the framework predicts in a specific domain, and what observation would count against it — is the live problem I am working on, and the reason the current empirical programme pre-registers its predictions before anything is run. Whether that succeeds is an open question. But it is the standard I am building to, not one I have just discovered and conceded.
And none of it touches the hard problem. Describing the structure of experience is not explaining why structure is felt. That is why the confidence rating on this question has not moved and will not move on convergence alone.
RFT is my own interpretive framework and original work, offered here as a way of seeing. It is not established neuroscience, it is not a solution to the hard problem, and it is not cited by any of the researchers above.
Format & time. Reflective structured learning · ~30–45 minutes.
Learning outcomes. After completing this unit you will be able to:
Reflection prompts. Write 3–5 lines on each and keep them for your CPD record.
CPD-eligible structured learning. Paid members can download a CPD certificate — see the certificate template. CPD-eligible structured learning; not statutory-regulator endorsement — practitioners self-assess relevance and log accordingly against their own professional body's requirements. Hours shown reflect estimated reflective-learning time; log only genuine time spent. Not medical advice.
Primary sources linked where available. Preprints are labelled as preprints; press releases are cited as press releases where they are the accessible public account of a paper; partisan sources are flagged as partisan and used only for facts corroborated elsewhere. Always consult the originals — this synthesis describes emphasis and findings, not verbatim claims.
How this was made. A structured research pass across neuroscience, philosophy of mind, clinical measurement, the near-death and psychedelic literatures, and non-English research including the Japanese qualia-structure programme; every citation checked individually against a primary source, then re-checked in a second independent verification pass. That pass caught four things worth naming, because the piece would have been wrong without them: it had inherited a widely-repeated but false theory count from press coverage of the Kuhn survey; it had softened Gómez-Marín's position to make him more measured than his own abstract shows him to be, which would have meant judging a claim he never made; it had inverted the finding of the brain-entropy study it cited, which actually reports a positive result on five of fourteen metrics; and it had merged two separate anaesthesia studies into one. All four are corrected above rather than quietly dropped. A personal account is used only where the subject has published it himself, and quoted rather than paraphrased. The heterodox position is given its strongest form and then held to the same standard as the orthodox one, producing a rating its advocates will not like and its opponents should not enjoy either — and where the researcher who ran the key study reads his own data against me, that is stated. The field's own limitations are quoted rather than smoothed over, including the ones that cut against this piece, and my own framework is judged by the criterion I applied to everyone else. Confidence is stated explicitly, held where the evidence did not move it, and raised only on the sub-claim the evidence actually supports. A human reviews every word before publication. Reality is the arbiter.
Join the Frontier. The weekly Investigation is free, in full, every week — including the corrections and the things that don't work. Members get the working: the reasoning behind the findings, the analysis, and the monthly Cross-Domain Proposition.
© 2026 David Fleming · Models of Reality / The Frontier. All rights reserved. Recursive Field Theory (RFT) is the original work of David Fleming. No part of this publication or the system that produces it may be reproduced, reverse-engineered or used to train or build a competing service without permission. Not medical or financial advice. Nothing here is guidance on the management of any medical condition, and no part of it should be read as encouragement to alter treatment.