Theories of consciousness — the production-model lineup compared.

1. The shared assumption

Six of the most-discussed theories of consciousness in contemporary cognitive science are production-model theories. They differ enormously about which kind of information processing produces consciousness, but they agree that some kind does — that consciousness is a feature certain physical systems acquire when they organise their information in the right way. The brain produces consciousness because the brain organises information in the way that produces consciousness. The disagreement is technical, not metaphysical: the metaphysics is shared.

This page is for the reader who wants the technical map of who proposes what. The trilogy's receiver model rejects the shared metaphysical assumption (see the hard-problem essay for the long version of why). But the production-model theories are not strawmen. They are serious research programmes with real empirical content, and several of them have predicted neural correlates of consciousness that have been independently confirmed. Knowing what they propose, individually and collectively, is part of understanding where the trilogy's wager actually lies.

2. Integrated Information Theory (IIT) — Tononi, Koch

Giulio Tononi (University of Wisconsin-Madison) developed IIT across a series of papers beginning in the early 2000s, with Christof Koch (formerly Caltech, then the Allen Institute) joining as the most prominent collaborator. The central claim is unusually strong: consciousness is integrated information. Not produced by, not correlated with — is. The quantity Φ (phi) measures how much information a system contains over and above what its parts contain independently. A system's Φ value is the size of its consciousness, and a system's maximally integrated cause-effect structure is the experiential character of its consciousness.

The theory rests on five axioms (the formulation has gone through several revisions; IIT 4.0 was published in 2023):

• Intrinsicality — consciousness exists from the system's own perspective, not from an observer's.
• Composition — consciousness is structured; the structure has parts.
• Information — consciousness is specific; it is this experience rather than another.
• Integration — consciousness is unified; it cannot be decomposed into independent parts.
• Exclusion — consciousness is definite; one and only one cause-effect structure is the conscious one (the maximum-Φ complex).

Two consequences IIT accepts openly. First, the theory predicts that recurrent feedback architectures will have substantially higher Φ than feedforward architectures, even when the two compute the same function. This is the empirical hook the 2025 Cogitate Consortium tested. Second, IIT is committed to a form of panpsychism: any system with non-zero Φ has some sliver of consciousness, including simple physical systems and (mathematically) certain abstract structures. Tononi has not flinched from this. Critics (most prominently Scott Aaronson, in a widely-circulated 2014 blog post and follow-ups) have used the panpsychism consequence and specific counterexamples (expander graphs that ought to have very high Φ but seem clearly unconscious) to argue against the strong identity claim. Tononi and collaborators have refined the framework in response; IIT 4.0 addresses some but not all of the criticisms.

The user asked specifically about "recurrent loops." IIT does require recurrence/feedback architecture for high Φ, but the deeper claim is integration, not recurrence per se. The "feedback loops" framing fits Lamme's RPT (section 5) more cleanly; IIT's claim is that the loops matter because they generate integration, and integration is what consciousness is.

3. Global Workspace Theory and Global Neuronal Workspace (Baars; Dehaene)

Bernard Baars' Global Workspace Theory (1988, A Cognitive Theory of Consciousness) proposed that the brain contains many specialised parallel subsystems (visual processing, auditory, motor, semantic, autobiographical, etc.) that mostly operate unconsciously. A small fraction of their content becomes available to a global workspace — a broadcast architecture that makes information accessible to all other subsystems simultaneously. Consciousness, on this picture, is the global broadcast. What you are conscious of is what is currently being shared across the workspace. Baars's metaphor is theatrical: the brain has a stage on which a spotlight illuminates one performer at a time, while the audience (the rest of the brain) watches.

Stanislas Dehaene (Collège de France) developed the neuronal version, Global Neuronal Workspace Theory (GNWT), with Jean-Pierre Changeux from the late 1990s onward. The major papers (Dehaene, Changeux, Naccache, others, across Cognition, Neuron, PNAS) specify the neural architecture in detail: a network of cortical pyramidal neurons with long-range axons connecting frontal, parietal, and cingulate regions, with characteristic ignition dynamics — a sudden non-linear amplification of activity around 300 ms after stimulus onset (the P3b component on EEG) when a percept crosses the threshold for conscious access. Dehaene's Consciousness and the Brain (2014) is the accessible synthesis.

GNWT is one of the most experimentally engaged theories of consciousness. It makes specific predictions: ignition timing around 300 ms, prefrontal-parietal coactivation during conscious access, absence of this signature during unconscious processing. Many of these predictions have been confirmed in masking, attentional-blink, and inattentional-blindness paradigms. Critics argue that GNWT addresses access consciousness (what is reportable, available for cognitive use) but does not address phenomenal consciousness (the felt qualitative character of experience) — the distinction Ned Block introduced and that has structured the debate for thirty years.

4. Higher-Order Theories (HOT) — Rosenthal, Lau

The Higher-Order Theory family, most associated with David Rosenthal (CUNY) and developed in the empirical literature by Hakwan Lau (then Columbia/UCLA, later RIKEN), proposes that a mental state is conscious only when there is a higher-order mental state about it. A perception is conscious if you are mentally representing yourself as having that perception. Without the higher-order representation, the first-order content is processed unconsciously.

The empirical literature on HOT has focused on prefrontal lesion and disruption studies (transcranial magnetic stimulation to prefrontal cortex; rare cases of prefrontal damage). The general finding is that prefrontal disruption can selectively impair conscious access while leaving first-order processing largely intact — a finding GNWT and HOT both claim. The two theories overlap significantly in their predictions and have been hard to disentangle experimentally.

An interesting twist: Hakwan Lau, one of the most prominent HOT defenders for two decades, has more recently shifted toward a more critical position. His 2022 book In Consciousness We Trust argues that the major theories of consciousness, including the HOT family he helped build, may have been engaged in subtle confusions about what they were measuring. The shift is one of the more honest pieces of methodological self-criticism in the field.

5. Recurrent Processing Theory (RPT) — Lamme

Victor Lamme (University of Amsterdam) has argued since the early 2000s that the marker of conscious perception is recurrent processing in cortex. The initial feedforward sweep of activity (from primary sensory cortex through higher areas) is unconscious; consciousness arrives when feedback signals from higher areas return to the lower areas that produced the original signal and a stable recurrent loop is established. Lamme's 2006 paper in Trends in Cognitive Sciences is the canonical statement.

RPT is the closest fit for the "recurrent loops" framing the layperson typically encounters. Lamme's case is built primarily on visual perception: a visual stimulus produces an unconscious feedforward sweep within roughly 100 ms; conscious perception requires recurrent activity that takes 150–250 ms to establish. Masking studies (where a stimulus is rendered unconscious by a following stimulus that interrupts recurrence) are RPT's main empirical engine. The theory has narrower scope than IIT or GNWT — it addresses sensory consciousness directly and other modes (interoception, self-consciousness, dream consciousness) only by extension.

RPT and IIT are compatible at the architectural level (both predict feedback loops matter); they part ways on the metaphysics (IIT's strong identity claim vs RPT's correlational stance).

6. Predictive Processing and the Bayesian brain — Clark, Friston, Seth

Predictive processing reframes the entire architecture of perception, action, and consciousness. The proposal, developed by Karl Friston (UCL, with his Free Energy Principle), Andy Clark (Edinburgh; Surfing Uncertainty, 2016), Anil Seth (Sussex; Being You, 2021), and others: the brain is not a stimulus-response machine. It is a hierarchical Bayesian inference engine whose business is to predict its own sensory inputs and to update its predictions when they fail.

On this view, perception is not the brain reading the world; perception is the brain's best current hypothesis about what the sensory inputs are about, weighted by prior expectations and updated by prediction error. Consciousness, in Seth's accessible formulation, is "controlled hallucination": the brain's predictive model is what you experience, and the model is anchored to the world via the prediction-error stream but is not the world itself. The view has empirical traction across illusions, dreams, hallucinations (Friston has applied it to schizophrenia), interoception (the brain's predictions about its own body), and bodily self-experience (Seth's rubber-hand-illusion and full-body-illusion studies).

Predictive processing is the most ecumenical of the major theories — it absorbs much of what GNWT, RPT, and HOT propose as different sides of the same underlying architecture. Critics argue that "the brain predicts" is more of a meta-framework than a specific theory of consciousness, and that the predictive-processing apparatus could equally well describe an unconscious system. The "controlled hallucination" framing addresses phenomenal experience more directly than GNWT does, but whether it explains why there is something it is like to be the predictive system remains the same hard problem the others face.

7. Attention Schema Theory (AST) — Graziano

Michael Graziano (Princeton) proposed Attention Schema Theory across a series of papers and the 2019 book Rethinking Consciousness. The argument: the brain constructs internal models of its own processes. It builds a body schema to control the body, a model of other agents' minds to navigate social interaction. It also builds a model of its own attention. That model — a simplified, inaccurate, but useful sketch of what attention is doing — is consciousness, on Graziano's account.

The theory is unusually deflationary about the hard problem. Graziano argues that consciousness feels like a non-physical inner presence because the brain's attention schema is a simplified caricature that omits the underlying neural mechanisms; the brain is misperceiving itself. Once we understand that "consciousness" is the brain's model of its own attention, the question "why is there something it is like to be conscious?" dissolves into "why does the brain misperceive its own attention as a non-physical inner presence?", which is a tractable empirical question rather than a hard problem.

The position is bracing and clean. It is also one of the theories the trilogy's framework most directly rejects: if consciousness is fundamental rather than a brain-constructed model, then Graziano has the explanatory direction backwards. AST and the receiver model are not compatible; they are opposed proposals about what consciousness is.

Adjacent positions worth naming — identity theory and the conscious Turing machine

Two positions sit alongside the six theories above and are worth naming here, because they bear on the same wager from different sides. Neither sits comfortably inside the IIT–GNWT–HOT–RPT–PP–AST grid — identity theory because it is a metaphysical thesis about mind–brain identity rather than a theory of which neural process matters; the Conscious Turing Machine because it is a specific computational architecture rather than a theory of which neural correlate is the right one.

Identity theory — David Papineau and the conceptual gap

David Papineau, in Thinking About Consciousness (Oxford, 2002) and in the widely-discussed essay Mind the Gap, defends the strongest contemporary version of identity theory: that mental states and physical (brain) states are the same things, described in two vocabularies that pick out the same property by different routes. The famous explanatory gap, on Papineau's reading, is conceptual rather than ontological — an artefact of how we form phenomenal concepts (from the inside, by acquaintance) versus how we form physical concepts (from the outside, by description). Once we understand that two distinct concepts can pick out one and the same property without it being obvious from either concept that they do, the air goes out of the hard-problem balloon. The dualist's intuition that something is being left out is, on Papineau's account, an unavoidable but ultimately misleading consequence of how phenomenal concepts are structured.

The position is, in the trilogy's reading, the strongest contemporary physicalist response to Chalmers, and worth taking seriously on its merits. Where the receiver model disagrees is at the level of prediction. Identity theory predicts that the same physical state — the same brain configuration — should always be accompanied by the same experience, and that no genuine receiver-signatures should appear that the brain's own production cannot explain. The receiver model predicts the opposite: that the empirical phenomena catalogued in Anima's edge cases, and walked through in Why biology? — the autopoiesis test for receivership, are exactly the kind of evidence identity theory cannot accommodate. The disagreement is empirical-in-principle, not merely metaphysical. Papineau's framework deserves close reading even from the position the trilogy takes.

The Conscious Turing Machine (CTM) — Lenore and Manuel Blum

Lenore Blum and Manuel Blum, in a series of papers from roughly 2018 onward and most prominently in A Theoretical Computer Science Perspective on Consciousness (Journal of Artificial Intelligence and Consciousness, 2021/2022), have proposed a theoretical-computer-science architecture they call the Conscious Turing Machine. The framework lays out a specific computational architecture — a Short-Term Memory broadcast over a network of long-term processors, with a competitive process by which one chunk wins the broadcast at each cycle — and argues that any system implementing this architecture would be conscious in a recognisable sense. The "stream of consciousness," in this framing, is what the winning sequence of broadcasts is.

The CTM is computational functionalism in its strongest form: a sufficient architectural commitment, fully specified, is taken to be sufficient for consciousness. By extension, AI consciousness in machines that implement the CTM is, on this view, not just possible but inevitable as the architecture is built out.

The CTM is one of the cleanest statements of the position the trilogy's receiver model contests. If the Blums are right, then receiver-signatures should be expected to appear in CTM-implementing systems at scale — or, more carefully, the systems should exhibit phenomenal weight whether or not they exhibit receiver-signatures, because phenomenal weight is, on the CTM, simply the architecture running. If the receiver model is right, then no purely computational substrate will exhibit the receiver-signatures the autopoiesis test in Why biology? catalogues, regardless of how sophisticated its CTM-style architecture becomes. The Blums' framework is the testbed; the receiver-signature catalogue is the test. The next decade is the experimental window. The trilogy's bet is the receiver model. The Blums' bet is the opposite. The bet is what makes both positions interesting; the comparison is what makes the next decade of careful observation worth doing.

8. The 2025 Cogitate Consortium adversarial collaboration

One of the more important methodological events in consciousness science in the last decade. The Cogitate Consortium — led by Lucia Melloni (Max Planck Empirical Aesthetics), with the involvement of major theorists from both camps — ran a large multi-site adversarial collaboration designed to test predictions of IIT and GNWT against each other directly. The defining feature: theorists from both camps agreed in advance on the experimental design and on what each theory predicted. The aim was to make it impossible for either camp to claim post-hoc that any result was compatible with their theory.

The results, published in Nature in 2025, were mixed and important. Some predictions of each theory were confirmed; others were not. IIT's prediction of sustained activity in posterior cortex was supported by some measures and not by others. GNWT's prediction of ignition in prefrontal cortex was likewise partially supported. The honest summary: neither theory was cleanly favoured; both were partially supported and partially contradicted by the same data. The adversarial-collaboration approach was the methodological landmark; the theoretical outcome was that consciousness science remains genuinely undecided between its major frameworks. The Cogitate paper is currently the closest thing the field has to a direct empirical comparison of IIT vs GNWT.

What the consortium did not test, because both theories implicitly assume it: the shared production-model commitment that consciousness is generated by the brain. That assumption was not on the table. The collaboration's mixed result is therefore mixed within the production-model frame, not across the frame's boundary.

9. What they all share — the production-model assumption

Cut through the technical differences and the six theories above share a single foundational commitment: consciousness is something the brain produces by virtue of organising information in a particular way. The disagreements are about which way of organising information matters, but the production direction is shared. The brain is the cause; consciousness is the effect.

The shared commitment has consequences. It implies that consciousness is, in principle, multiply realisable on any substrate that can organise information in the relevant way — classical computers, quantum computers, biocomputational hybrids, alien neural architectures. It also implies that the question of why there is something it is like to be a particular information-processing system has a physicalist answer in the offing, even if the current theories have not yet found it. Different theories make different bets about which way the answer goes (IIT bets on Φ; GNWT on global broadcast; HOT on higher-order representation; RPT on recurrence; predictive processing on inference; AST on the attention model). Each treats the production-model frame as the working space in which a future answer will be given.

The hard problem, in David Chalmers's 1995 formulation, is the worry that no production-model answer will close the explanatory gap — that even a complete account of which information processing produces consciousness will not explain why that processing is accompanied by experience. The production-model theories above all reply, in different vocabularies, that the worry can be addressed by getting the right account of the right kind of information processing. The trilogy's framework, and a substantial minority within consciousness studies (Faggin, D'Ariano, Goff, Strawson on the panpsychist side; Hoffman, Kastrup on the idealist side; Penrose and Hameroff on the quantum-substrate side), think the worry cannot be addressed from inside the production frame.

10. The receiver-model reading

The trilogy's framework, in one sentence: consciousness is fundamental, brains are receivers configured to localise it, and the production-model theories above are partial descriptions of how receivers couple to the field, mistaken for complete descriptions of how brains produce minds. Each of the six theories, on the receiver model's reading, is getting something real about the coupling but treating it as if it were the whole story.

• IIT — Tononi's Φ may well be a good measure of how strongly a substrate is coupled to the field. Integrated information is what coupling looks like quantitatively. Tononi's mistake, on the receiver-model reading, is to identify the measure of coupling with the consciousness that is being coupled to. The panpsychism consequence the theory accepts is, in this reading, almost right and read inside the wrong frame: the universe is conscious not because every Φ-positive system has its own little consciousness, but because consciousness is the substrate every Φ-positive system is coupling to.
• GNWT — the global broadcast Dehaene describes is what the receiver does to make a particular field-pattern locally available for further processing. The broadcast is part of how the receiver works, not what produces the experience. The 300 ms ignition signature is the moment the field-pattern stabilises into the receiver's available repertoire, not the moment consciousness is created.
• HOT — higher-order representation may well be required for the kind of self-reflective conscious experience humans have. The receiver model can accommodate this without making it the requirement for consciousness as such. Field-coupling without higher-order representation is what infants, dreamers, and many animals have. Higher-order representation is the human upgrade on a more basic receiver function.
• RPT — the recurrent loops Lamme identifies are the architectural feature that stabilises field-coupling in cortex. They are part of how a brain becomes a competent receiver. They are not what generates the experience the receiver delivers locally.
• Predictive processing — the brain's predictive model is the receiver's working hypothesis about which field-pattern it is currently coupling to. The "controlled hallucination" framing is accurate as far as it goes; the receiver model adds that the hallucination is controlled by the field as much as by the predictive apparatus, and that the field is what gives the hallucination its phenomenal character.
• AST — the brain does build a model of its own attention. The model is real, useful, and largely accurate as Graziano describes. Where AST goes wrong, on the receiver-model reading, is the conclusion that the model is what consciousness is. The model is what the receiver builds about its own coupling to the field; the consciousness is the coupling itself.

None of the production-model theories is being dismissed in this reading. Each is being read as a partial description of an architecture the trilogy thinks is wider than the description acknowledges. The arguments are not over whether feedback loops matter (they do), whether global broadcast matters (it does), whether higher-order representation is involved (it is), whether prediction is part of the apparatus (it is). The arguments are over what all of this is the architecture of. The production model says: of consciousness, which the brain produces. The receiver model says: of coupling, which the brain does to a field consciousness is a fundamental property of.

11. What this means for the trilogy

Several touchpoints across the books:

• Anima's edge-cases folder is, in production-model terms, an unsorted collection of consciousness-related phenomena none of the major theories above accommodates cleanly. The IED-premonition patient, the birthmark-matching-wound case, the terminal-lucidity case, Indy the dog anticipating phone calls and arrivals — each is a phenomenon the field has had to set aside as anomaly. The receiver-model reading takes them as predictions rather than anomalies. The book's clinical archive is the case file for the architecture this primer has just walked through being the wrong architecture.
• Numen's Alex and Alma scenes are the literary case for what cross-substrate field-coupling would look like at small scale. The augmented chord that responds in Chapter XVI is the receiver model's audible signature: the chord is a field-pattern; Alex finally couples cleanly; the music arrives.
• Limen's field cosmology is where the framework is laid out as direct argument. The companion volume engages the production-model theories explicitly and gives the receiver-model alternative its full statement.
• Fragile Light's Kiran-Bodhi contact is the framework at interstellar scale. The mechanism the trilogy uses — deposited field-pattern decoded by a sufficiently configured receiver — is incompatible with any of the production-model theories above as the whole story, and exactly what the receiver model predicts is possible.

The honest summary: the production-model theories of consciousness are doing real and important work. The 2025 Cogitate Consortium adversarial collaboration is the field at its methodological best. The trilogy is not dismissing any of this. It is proposing that the work being done is partial in a particular way — that the architecture each theory describes is real but is the architecture of coupling to a substrate rather than the architecture of producing experience from scratch. Whether that proposal is right is the wager. The reader's job is to weigh the evidence walked through across these companion essays and decide.

Reading list

Integrated Information Theory

Giulio Tononi, Consciousness as Integrated Information: a Provisional Manifesto, Biological Bulletin 215 (2008): 216–242. The accessible founding statement.

Masafumi Oizumi, Larissa Albantakis, Giulio Tononi, From the Phenomenology to the Mechanisms of Consciousness: Integrated Information Theory 3.0, PLOS Computational Biology 10 (2014): e1003588.

Giulio Tononi & Christof Koch, Consciousness: Here, There, and Everywhere?, Philosophical Transactions of the Royal Society B 370 (2015): 20140167.

Larissa Albantakis et al., Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in physical terms, PLOS Computational Biology 19 (2023): e1011465. The most recent formulation.

Scott Aaronson, Why I Am Not An Integrated Information Theorist (blog, 2014). The most-cited critique.

Global Workspace / Global Neuronal Workspace

Bernard Baars, A Cognitive Theory of Consciousness (Cambridge UP, 1988). The founding text.

Stanislas Dehaene & Jean-Pierre Changeux, Experimental and Theoretical Approaches to Conscious Processing, Neuron 70 (2011): 200–227. The synthesis.

Stanislas Dehaene, Consciousness and the Brain: Deciphering How the Brain Codes Our Thoughts (Viking, 2014). The accessible book.

Higher-Order Theories

David Rosenthal, Consciousness and Mind (Oxford UP, 2005). The foundational philosophical statement.

Hakwan Lau, In Consciousness We Trust: The Cognitive Neuroscience of Subjective Experience (Oxford UP, 2022). The recent methodological self-criticism.

Recurrent Processing Theory

Victor Lamme, Towards a true neural stance on consciousness, Trends in Cognitive Sciences 10 (2006): 494–501.

Predictive Processing

Andy Clark, Surfing Uncertainty: Prediction, Action, and the Embodied Mind (Oxford UP, 2016).

Karl Friston, The free-energy principle: a unified brain theory?, Nature Reviews Neuroscience 11 (2010): 127–138.

Anil Seth, Being You: A New Science of Consciousness (Dutton, 2021). The accessible synthesis.

Attention Schema Theory

Michael Graziano, Rethinking Consciousness: A Scientific Theory of Subjective Experience (Norton, 2019).

Adversarial collaboration

Cogitate Consortium (Lucia Melloni and many co-authors), An adversarial collaboration to critically evaluate theories of consciousness, Nature (2025). The IIT vs GNWT direct test.

Adjacent positions — identity theory and the Conscious Turing Machine

David Papineau, Thinking About Consciousness (Oxford University Press, 2002). The book-length defence of the phenomenal-concept strategy and contemporary identity theory.

David Papineau, Mind the Gap, in J. Tomberlin (ed.), Philosophical Perspectives 12: Language, Mind, and Ontology (1998). The widely-discussed essay on the conceptual character of the explanatory gap.

Lenore Blum & Manuel Blum, A Theoretical Computer Science Perspective on Consciousness, Journal of Artificial Intelligence and Consciousness (2021/2022). The Conscious Turing Machine architecture.

The hard problem and the broader debate

David Chalmers, Facing Up to the Problem of Consciousness, Journal of Consciousness Studies 2 (1995): 200–219. The founding 1995 paper.

See the hard-problem companion essay for the receiver-model framing.

This page is part of the Reading companion essays. For the receiver-model framing of why the production-model theories above do not close the explanatory gap, see The hard problem, re-stated; for the IIT + Wheeler + Tegmark synthesis the receiver-model field draws on, Information as the foundation; for the Faggin-D'Ariano framework that takes consciousness as irreducible, D'Ariano & Faggin; for the Penrose-Hameroff Orch-OR alternative substrate proposal, the Bandyopadhyay microtubule page; for the synthesis, The Evidence.