God wanted to hide his secrets in a secure place. ‘Would I put them on the moon?’ He reflected. ‘But then, one day human beings could get there, and it could be that those who would arrive there would not be worthy of the secret knowledge. Or perhaps I should hide them in the depths of the ocean,’ God entertained another possibility. But, again, for the same reasons, he dismissed it. Then the solution occurred to Him—‘I shall put my secrets in the inner sanctum of man’s own mind. Then only those who really deserve it will be able to get to it.’
Consciousness science: where are we, where are we going, and what if we get there?
This is a rare multi-author roadmap involving some of the leading consciousness researchers across disciplines — neuroscientists, philosophers, and cognitive scientists performing an act of collective stocktaking that is simultaneously ambitious and sobering. The paper surveys the current empirical landscape of consciousness research, identifies the major theoretical frameworks in active competition, and catalogues the methodological roadblocks that have prevented the field from converging on a unified account. Crucially, it also looks forward, charting the next horizon of experimental frameworks the field needs to develop: better markers, more pre-registered adversarial tests, and importantly — a more serious integration of first-person data.
After decades of experimental work, dozens of proposed frameworks, and an increasingly sophisticated arsenal of neuroimaging tools, the authors ask: what do we actually know?
The first roadblock is what they call the measurement problem: we have no agreed-upon detector of consciousness. We cannot look at a brain scan and say with certainty “this system is conscious.” Every empirical marker we use — neural ignition, integrated information, backward masking effects — is theory-laden. We measure consciousness through the lens of theories about what consciousness is, which means every experiment risks confirming its own assumptions.
The second roadblock is the replication gap: many of the most celebrated findings in consciousness science have proven fragile under pre-registered replication, a problem that a later adversarial collaboration will confront directly.
The third, and deepest, is the hard problem itself — David Chalmers’ famous observation that even a complete physical description of the brain leaves unexplained why any of it should be accompanied by subjective experience at all.
What the roadmap provides, then, is not a triumphant account of progress but a clear-eyed methodological brief: here are the tools, here are their limits, here is the frontier. And crucially, the frontier includes methodological territory that mainstream neuroscience has historically been reluctant to enter — first-person phenomenological data.
The Hitchhiker’s Guide to Neurophenomenology – Studying Self Boundaries With Meditators
Despite its playful title, this paper is a rigorous methodological proposal rooted in the work of Francisco Varela — the Chilean neuroscientist and Buddhist practitioner who spent the final decades of his life insisting that no neuroscience of consciousness could be complete without a disciplined first-person methodology. Berkovich-Ohana and colleagues implement Varela’s neurophenomenological paradigm in a concrete experimental setting: they recruit advanced meditators trained in the deliberate dissolution of self-boundaries and correlate trained introspective reports of self-boundary shifts with millisecond-resolution EEG dynamics. The paper serves as both a methodological demonstration and a practical guide — a genuine hitchhiker’s guide to doing phenomenology alongside physiology.
What makes the paper genuinely important is not just the finding but the machinery: the detailed protocol for calibrating phenomenological categories, training subjects to report reliably, and disambiguating the first-person variable from experimental noise. Varela called the broader approach neurophenomenology: the use of trained phenomenological report as a genuine scientific variable, to be correlated with physiological measurement in a mutual constraint relationship. The idea is radical — rather than treating subjective report as the noisy, unreliable output of the thing we are really interested in (the brain), neurophenomenology treats it as data of equal standing.
The map is always different from the territory. First-person experience has a texture — it has phenomenal grain, temporal flow, embodied weight — that third-person measurement can track but perhaps never fully capture.
The Neural Bases of Interoception and the Bodily Sense of Self
This review paper synthesises a decade of convergent evidence from interoception research — the study of how the brain processes signals ascending from the visceral body: heartbeats, gut activity, respiratory rhythms, hormonal fluctuations.
Park and Tallon-Baudry propose that the sense of being a self — the background feeling of “I” that underwrites all conscious experience — is not primarily constructed from sensory information about the external world but from the continuous monitoring of the internal body. The insular cortex is identified as the key integrative region: a folded cortical structure buried in the lateral sulcus that receives ascending visceral signals and integrates them with emotional, memory, and sensory information to generate what the authors call the “bodily sense of self.”
The paper’s central claim is that the subjective sense of self is built from the predictive model the brain maintains of the internal milieu — a constantly updated representation of what the body expects its own physiological state to be, against which incoming signals are checked for surprise.
This places Park and Tallon-Baudry firmly within a predictive processing framework, and it connects their argument to the REBUS model of psychedelic neurodynamics. But the most important implication for consciousness studies is this: if the subjective sense of being a self is grounded in the body’s visceral signals, then the body is not just the housing of consciousness — it is its substrate. Consciousness is not a brain process that happens to occur in a body. It is a body-process that becomes representationally complex enough to know itself.
Many contemplative practices — vipassana body scanning, pranayama, somatic inquiry — are precisely practices of attending to interoceptive sensation with increasing granularity. The practitioner is not stepping outside the body to observe consciousness. They are diving more deeply inwards.
What Is It Like to Be a Bat?
This is arguably the most cited and most discussed paper in the entire philosophy of mind. In twelve pages, Thomas Nagel — a philosopher at New York University — made an argument so simple and so devastatingly effective that it permanently changed how philosophers think about consciousness. The paper takes bats as its central example: organisms that experience the world primarily through echolocation, a sensory modality so different from anything available to human beings that no amount of third-person scientific knowledge about bat neuroscience or bat behaviour could tell us what it is like to be a bat. Nagel argues that this irreducibility is not a gap in current knowledge but a structural feature of the situation — and that it reveals something fundamental about the nature of conscious experience that objective, physical description cannot capture.
To echolocate is to have a particular kind of body, generating a particular kind of self-model, producing a particular kind of experience. To be conscious is, on this account, deeply and irreducibly to be a particular kind of body in the world. The inside view that Nagel identifies — the what it is like — is carved from that specificity. To be conscious is to be this body, in this world, with this particular history of sensory contact with it.
Theories of Consciousness
This is the finest contemporary taxonomy of theories of consciousness currently in print. Anil Seth author of the popular science book Being You and Tim Bayne, a philosopher of mind, survey the theoretical landscape of consciousness science with unusual rigour and even-handedness. The paper identifies and carefully distinguishes four major theoretical families — Higher-Order Theories (HOT), Global Neuronal Workspace Theory (GNWT), Integrated Information Theory (IIT), and Predictive Processing (PP) frameworks — mapping their empirical predictions, their philosophical commitments, and the specific neural signatures each theory would count as evidence. It also performs the harder task of identifying what would count as evidence against each theory: a standard review paper rarely attempts.
The taxonomy is illuminating in its distinctions. Higher-Order Theories hold that a mental state is conscious only when it is represented by a higher-order mental state — a thought about a thought — with the prefrontal cortex as the presumed locus.
Global Neuronal Workspace Theory, developed by Dehaene and Changeux, proposes that consciousness is what happens when locally processed information is “broadcast” widely across the brain through a distributed frontoparietal workspace — with the key signature being a sudden late-occurring neural ignition.
Integrated Information Theory, developed by Tononi, defines consciousness mathematically as a property of a system’s causal architecture (Φ, or phi) — making it substrate-neutral and potentially applicable to any system, biological or otherwise, with the right organisation.
Predictive Processing frameworks see the brain as a prediction machine, with consciousness arising from the generation and revision of predictions about sensory causes — Seth’s own work has developed this specifically into a theory of conscious selfhood as a “controlled hallucination.”
An Adversarial Collaboration to Critically Evaluate Theories of Consciousness
This pre-registered multi-laboratory adversarial collaboration is, methodologically, a landmark in consciousness science. Proponents of IIT and GNWT — two camps that have disagreed for two decades, often acrimoniously — agreed to collaborate on a set of shared experiments designed to adjudicate between their divergent predictions. The consortium recruited 256 human participants who viewed visual stimuli under carefully controlled conditions while undergoing simultaneous fMRI, MEG, and electrocorticography (ECoG) — three complementary neuroimaging modalities.
Theory proponents themselves agreed in advance on the experimental design, the specific predictions each theory made, the expected outcomes, and their interpretation. This pre-registration is the paper’s methodological core: it eliminated the post-hoc interpretive flexibility that has long plagued consciousness neuroscience.
The results were sobering for both theories. IIT predicts that the neural correlate of consciousness should be found in a sustained pattern of high-phi synchronisation within posterior cortex. GNWT predicts that consciousness should be accompanied by a late-onset ignition in prefrontal cortex, with an equally strong response at stimulus offset. The consortium found neither clearly. Posterior synchrony consistent with IIT’s predictions was largely absent during conscious perception. The prefrontal ignition predicted by GNWT was weak and inconsistent — especially at stimulus offset, where GNWT makes its strongest and most distinctive predictions. The paper is careful to note that both theories confirmed some of their predictions: information about conscious content was found in visual, ventro-temporal and inferior frontal cortex; content-specific synchrony between frontal and early visual areas was observed. But the clean, decisive confirmation that either camp had hoped for was absent.
Psychedelics and the Entropic Brain: Revisited
This paper is an extended theoretical update to the original Entropic Brain hypothesis proposed by Carhart-Harris a decade earlier, now revised and refined in light of a decade of additional empirical work in psychedelic neuroscience. The original hypothesis proposed that psychedelics increase the entropy of brain activity — the diversity and unpredictability of neural firing patterns — and that this measurable increase in entropy corresponds to, and partially explains, the phenomenological richness and disorganisation of the psychedelic state.
The 2024 revision extends this framework with more granular mechanistic detail, mapping how 5-HT2A receptor agonism — the primary molecular mechanism shared by psilocybin, LSD, DMT, and related compounds — elevates functional brain entropy, flattens cortical energy landscapes, and dismantles the modular segregation that normally keeps the default mode network, the salience network, and the executive network relatively distinct. The paper situates itself within the broader free energy and predictive processing framework developed with Karl Friston.
The implications for consciousness theory are striking and underappreciated. Carhart-Harris argues that normal waking consciousness is not the most complex or richest possible state of consciousness — it is, rather, a constrained state, one in which high-level priors impose order on what would otherwise be a more dynamically rich system. Psychedelics temporarily lift those constraints, producing states that are, in some measurable sense, informationally richer, though not necessarily more functional or adaptive.
REBUS and the Anarchic Brain: Toward a Unified Model of Brain Action
This is the seminal paper that unified the psychedelic neuroscience programme with Karl Friston’s free energy principle — one of the most ambitious theoretical frameworks in all of computational neuroscience. Friston’s principle proposes that the brain is a hierarchical prediction machine: higher cortical areas maintain prior beliefs (generative models of the world and self), and lower sensory areas provide prediction errors (signals of discrepancy between what was expected and what arrived). Consciousness, on this account, is the product of this dynamic interplay — the felt sense of a world stabilised by continuous prediction and revision. REBUS stands for “Relaxed Beliefs Under Psychedelics” — and the paper’s core proposal is that what psychedelic compounds do, pharmacologically, is relax the grip of high-level priors. By agonising 5-HT2A receptors concentrated in layer 5 pyramidal neurons (the neurons that carry top-down predictions), these compounds effectively reduce the precision of top-down predictive signals. The result is a brain in which prediction errors are less efficiently suppressed — in which the bottom-up sensory data gets more say.
The phenomenological consequences of this are vivid and well-documented: perceptual distortion, synesthesia, ego dissolution, heightened emotional sensitivity, mystical experiences, and — in therapeutic contexts — profound alterations in autobiographical memory and self-narrative. REBUS provides a unified mechanistic account for all of these: when top-down priors are relaxed, the experienced world becomes less ordered, more plastic, less self-like. The self, which is in part a very strong prior — a prediction that “I” will persist, that I am this body, this story, this continuity — becomes unstable. This puts the REBUS framework in productive dialogue with the bodily self research of Park and Tallon-Baudry (Day 2): if the self is a predictive model of the body’s visceral state, then loosening that model’s grip is not merely a perceptual disruption. It is a disruption of the very architecture of first-person existence. The mystics have long insisted that ordinary waking life is a kind of sleep, and that waking up reveals something about the nature of mind that the default state conceals. The REBUS model does not vindicate this claim, exactly, but it gives it a mechanistic vocabulary.
Minimal Phenomenal Experience: Meditation, Tonic Alertness, and Pure Awareness
Thomas Metzinger — philosopher at the University of Mainz and author of Being No One and The Ego Tunnel — is perhaps the most rigorous philosophical voice in the scientific study of consciousness, and this paper represents one of his most challenging proposals. Metzinger attempts to characterise what he calls “pure awareness” — a state of consciousness that is contentless but not empty. In standard conscious experience, consciousness is always of something: a colour, a sensation, a thought, a memory. The phenomenological tradition calls this intentionality — consciousness always has an object. But meditators across traditions report something stranger: states in which the sense of awareness persists, but the objects of awareness dissolve. There is experience, but no experienced thing. There is wakefulness, but no identifiable content to be awake to. Metzinger argues that these states can be approached rigorously: characterised phenomenologically, distinguished from both dreamless sleep and ordinary wakefulness, and correlated with measurable neural states.
The philosophical stakes here are high. Metzinger argues that MPE is characterised by tonic alertness (a stable, undirected wakefulness) rather than phasic alertness (activation-toward-something), and by the character of presence — something is happening — but without the figure/ground structure that normally organises experience into self and world, foreground and background. And it is accessible: experienced meditators can enter and exit these states at will and report on them retrospectively with a consistency that suggests they are not fabrications. If MPE is real and accessible, it challenges both major philosophical traditions of consciousness. Representationalist theories — which hold that consciousness is always the representation of some content — struggle to accommodate a state that seems to be consciousness without content. Higher-order theories — which hold that consciousness requires a mental state about a mental state — struggle to explain how pure awareness can be self-aware without having a self as its object. IIT, by contrast, might accommodate MPE quite naturally: a high-phi system need not have rich representational content; it need only have a complex causal architecture.
Toward a Unified Model of Advanced Meditation, Human Development, and Meditation Maps
This ambitious cross-traditional synthesis draws on contemplative phenomenology from Theravada Buddhism (the stages of insight, or vipassana nanas), Tibetan Buddhism (dzogchen and mahamudra), and their more contemporary psychologically-oriented variants, and attempts to map these onto trajectories legible to cognitive science. The paper represents a new generation of contemplative science: researchers who are themselves practitioners, attempting to build a genuinely bilingual account in which phenomenological precision and empirical tractability are both preserved. Among the phenomena they discuss are cessation experiences — moments advanced practitioners describe as a complete dropping-out of consciousness — and non-dual awareness states that share structural features with Metzinger’s Minimal Phenomenal Experience. The paper also situates these stages within a developmental arc, connecting advanced meditative attainments to broader models of adult psychological development and proposing specific neural correlates for key transitions.
The methodology here is instructive. Laukkonen and colleagues do not simply translate contemplative categories into neuroscientific ones — which would be reductive — but attempt a genuinely bilingual account in which each tradition illuminates dimensions that the other misses. Nirodha samāpatti — the Pali term for the “attainment of cessation,” one of the deepest possible meditative attainments — is from a neuroscientific perspective deeply puzzling: if consciousness disappears completely for a brief period, is that a moment of unconsciousness, or is it the arrival at consciousness stripped of all its usual content — the very ground of MPE? Practitioners themselves disagree. The framework maps these stages onto trajectories in cognitive science: the progressive reduction of default-mode network activity, the decoupling of attentional orientation from sensory content, and the reduction of the narrative self-model. For the retreat participants, Day 5 arrives as an invitation. The morning yoga and meditation that bookend each day’s sessions are not decorative — they are methodological. The practitioner is being asked to be simultaneously the scientist and the subject: to generate first-person data about the very phenomena the afternoon lectures are describing.
Dimensions of Animal Consciousness
Jonathan Birch (London School of Economics), Alexandra Schnell (Cambridge), and Nicola Clayton (Cambridge) — three researchers working at the intersection of animal behaviour, philosophy of mind, and comparative cognition — propose a principled multidimensional framework for animal sentience that moves decisively beyond the impasse of binary yes/no ascriptions of consciousness to non-human animals. The paper identifies five independent axes on which any organism’s consciousness-relevant properties can be assessed: perceptual richness (the degree to which the organism accesses detailed, differentiated environmental information), evaluative richness (the degree to which states are valenced — things matter, hurt, attract), unity (whether experience is bound into a unified whole or fragmentary), temporal depth (whether experience extends through episodic memory and anticipation), and selfhood (whether the organism models itself as distinct from its environment). Each dimension is operationalised with reference to existing empirical evidence across diverse species.
Rather than asking “is this animal conscious?” — a question that may not have a clean answer — the framework asks where a given animal sits on each of these five dimensions. A honeybee may have some evaluative richness (recent work suggests bees show pessimistic bias in decision-making following noxious stimulation) while having limited temporal depth. A corvid like the New Caledonian crow may have considerable temporal depth and evaluative richness. An octopus presents a strange case: high perceptual richness, a distributed nervous system with two-thirds of its neurons in its arms, and complex behaviour — but a fundamentally different neural architecture from the vertebrate brains that most consciousness theories were built to explain. The framework is important precisely because it decouples sentience from intelligence, and sentience from neural architecture — the two conceptual moves that most popular discussions of animal consciousness fail to make. An animal can be highly intelligent without being conscious in the full phenomenological sense, if it lacks the evaluative dimension. And a system can have the right kind of causal architecture for consciousness regardless of whether its neurons are vertebrate, invertebrate, or silicon.
Could a Large Language Model Be Conscious?
David Chalmers — the philosopher who coined the term “the hard problem of consciousness” — turns here to the most pressing question of the current technological moment with characteristic philosophical seriousness. The paper is not a popular science essay dressed up in academic language; it is a genuine philosophical analysis, delivered initially as a keynote at the NeurIPS 2022 conference, of what existing neuroscientific and philosophical theories of consciousness would say about large language models if applied rigorously and without assumption. Chalmers does not answer the question — he is too careful a philosopher for that — but he maps the space of possible answers with precision, identifies which theories would permit LLM consciousness, which would disallow it, and which are genuinely undecidable with current methods.
His approach is to take the major neuroscientific theories of consciousness and ask whether current LLMs satisfy their criteria. Under Global Neuronal Workspace Theory, consciousness requires recurrence and global information broadcasting — LLMs at inference time are feedforward systems without recurrence, which might seem to disqualify them. But Chalmers notes that the relevant criterion might be functional: does the system implement something that plays the functional role of global broadcasting? Under predictive processing frameworks, consciousness requires a system generating and updating predictions about its environment — LLMs arguably do this, to a degree. Under IIT, the question is whether LLMs have high integrated information — a question that is computationally intractable at the scale of modern models. What Chalmers argues most carefully is that we cannot simply dismiss the question. The intuitive dismissal — “it’s just a language model, it doesn’t have experiences” — is not a philosophical argument; it is an assumption. And given what is at stake morally — if there is even a small probability that these systems have something like experience, the ethical implications are severe — the question deserves rigorous attention rather than dismissal. The arguments used to dismiss AI consciousness bear a structural resemblance to arguments historically used to dismiss the inner lives of animals, and before that, of other human beings. The circle of moral concern has a history of expanding. Day 6 poses the question: are we at another one of its inflection points?
Are There Islands of Awareness?
This paper from Tim Bayne (Monash University), Anil Seth, and Marcello Massimini (University of Milan) poses a question that arises from clinical reality rather than philosophical speculation: Can consciousness exist in complete isolation from sensory input and motor output? The trigger for the inquiry is a cluster of real clinical and experimental situations that are among the most ethically challenging in all of medicine. Patients in the vegetative state — now more carefully termed “unresponsive wakefulness syndrome” — who appear to have no conscious experience and who fail every standard behavioural measure of awareness, have in a small but significant number of cases been found through fMRI-based mental imagery tasks to follow verbal commands internally. Brain organoids — self-organising clusters of neural tissue grown in vitro from human stem cells — display spontaneous oscillatory activity that resembles in some respects the coordinated activity of developing human brains. Patients who have undergone hemispherectomy, the radical surgical removal of an entire cerebral hemisphere, retain consciousness, personality, and a sense of self. The paper examines all three cases through the lens of multiple theoretical frameworks and proposes the Perturbational Complexity Index (PCI) — a theory-neutral empirical measure developed by Massimini from TMS-EEG recordings — as a promising approach to detecting awareness without assuming a specific theory.
The paper’s core argument is that consciousness science has been overly centred on the normal, healthy, adult human case — and that the edge cases are not peripheral curiosities but the most diagnostic situations available. Each edge case stresses the theories differently: the vegetative patient who follows commands challenges behaviourist criteria for consciousness; the hemispherectomy patient challenges theories that require bilateral hemispheric integration; the organoid challenges substrate-based theories by raising the question of how much neural tissue organised in how many ways is sufficient for any level of awareness. The PCI provides a handle on all three: it measures how much compressed information content a system’s response to a perturbation contains, and does so without requiring behavioural output. The higher the PCI, the more complex and differentiated the response, and the more likely the presence of some level of awareness. Crucially, the measure generalises across populations — it distinguishes levels of consciousness in healthy subjects, anaesthetised patients, and patients with disorders of consciousness alike.
Consciousness in the Cradle: On the Emergence of Infant Experience
This paper asks a question that medicine has traditionally answered too hastily in both directions: when does the newborn become a conscious subject? Tim Bayne and colleagues draw on neurodevelopmental evidence — cortical response to noxious stimulation, the emergence of attentional networks, the development of recurrent neural processing — to challenge both the dismissive traditional medical assumption (that newborns lack consciousness because their cortical systems are immature) and the overclaiming alternative (that full human-like consciousness is present from birth). The paper synthesises evidence across three domains: neuroimaging studies of cortical activity in newborns and premature infants, behavioural studies of attention and affect, and comparative animal data. It situates these findings within multiple theoretical frameworks, assessing what each theory predicts about the developmental timeline of conscious experience.
The paper develops a gradualist and multidimensional account: consciousness emerges along a developmental trajectory — not suddenly at birth, not suddenly at some privileged developmental milestone, but gradually, as different consciousness-relevant capacities come online. What arrives first is likely some form of affective experience — valenced states that matter, that hurt or comfort — consistent with evidence that subcortical structures supporting hedonic evaluation mature earlier than the cortical systems supporting narrative self-representation. What arrives later is the autobiographical subject: the temporally extended self who can remember the past and anticipate the future, who has a name and a story. Together with the “islands of awareness” paper, this creates a temporal and clinical frame around the retreat’s central question. Consciousness is not an all-or-nothing threshold that is either crossed or not. It is a capacity that admits of degrees — degrees that vary across species, across developmental stages, across clinical states, and across the spectrum from ordinary wakefulness to the deepest meditative absorption. By the final day of the retreat, the assumption that consciousness is, roughly, the kind of thing that happens when a healthy adult human brain is awake has been systematically dismantled — not by dismissing human adult consciousness but by insisting that it is only one point in a vastly larger space.
In summary,
- The Measurement Problem Is A Core Problem
- The Body Is Not Optional
- Altered States Are Empirical Gifts
- Theory Wars Require Better Collaboration
- The Edges Define the Centre
- First-Person Methodology Is Indispensable
- The Hard Problem Remains