Consciousness — Theories and Empirical Landscape
In April 2025, two of the leading theories of consciousness staked their reputations on a pre-registered shootout. One scored 2 out of 3. The other scored 0. The winner is not winning by much. The hard problem — why any neural process is accompanied by an inside view at all — was not touched.
For the philosophical puzzle of why experience exists at all, see concept hard problem consciousness. This page maps the empirical contenders.
The four live theories
Integrated Information Theory (Tononi). A system is conscious in proportion to Φ (phi) — information it carries that cannot be decomposed into independent parts without loss. The brain's posterior "hot zone" (occipital, parietal, temporal) is predicted as the seat of experience. The theory is panpsychist by construction: any system with non-zero Φ has a flicker of experience. Two unsolved snags — Φ is computationally intractable for any realistic brain, and feed-forward networks (intuitively non-conscious) can score high Φ.
Global Neuronal Workspace (Dehaene, Baars). Consciousness equals broadcast. Specialist modules compete; the winner is amplified across long-range fronto-parietal connections and made available to all other modules. Predicts a sudden prefrontal "ignition" at the moment a stimulus enters awareness. Clinical fit is strong: anesthesia, coma, and vegetative states all disrupt long-range broadcast. Weakness: patients with massive prefrontal damage are often fully conscious, and the theory addresses access (what you can report) without obviously touching phenomenal experience (what it's like).
Free Energy Principle / Active Inference (Friston). The brain is a prediction machine minimizing surprise. Perception updates the model; action reshapes the world to fit the model. A 2024–2025 extension adds a hyper-model: the system infers its own confidence, dynamically tuning gain across layers. Critics argue the framework is so general that every neural finding can be reframed in its terms — fruitful, possibly unfalsifiable.
Higher-Order Theories (Rosenthal, Lycan). A mental state is conscious only if represented by a higher-order state — a thought about the thought. Predicts that systems without self-monitoring are not conscious. Sits in direct tension with the posterior-cortex result below: meta-cognition lives in prefrontal regions that the 2025 data demoted.
The 2025 adversarial collaboration
The Templeton World Charity Foundation funded $6 million for proponents of IIT and GNWT to design a single pre-registered experiment together, with divergent predictions baked in. Published in Nature, April 30, 2025. 256 participants. fMRI, MEG, and intracranial EEG run in parallel.
Scorecard:
| Theory | Predictions passed | Headline outcome |
|---|---|---|
| IIT | 2 / 3 | Sustained posterior synchronization weaker than predicted |
| GNWT | 0 / 3 | The prefrontal "ignition" did not appear |
The most robustly replicated finding: consciousness tracks posterior cortex, not prefrontal. The "ignition" model is in trouble. IIT survived but did not vindicate itself — its specific claim that network connectivity specifies consciousness took damage too.
What the study did not do: explain why posterior synchronization produces an inside view. It constrained the neural correlates. It did not bridge the explanatory gap.
What's contested
Three live disputes:
- Is "posterior, not prefrontal" the real result, or an artifact of task design? GNWT proponents argue the experiments under-sampled report-related processing. The adversarial format is supposed to neutralize this critique by pre-registration; the rebuttal is ongoing.
- Can Φ ever be computed for a real brain? Even approximations require simplifications that critics say drain IIT of its content. If Φ is unmeasurable in practice, the theory's empirical bite is unclear.
- Is the Free Energy Principle a theory or a language? Friston's framework restates almost any neural fact in its vocabulary. Karl Popper would have asked what observation could falsify it. So far, no clean answer.
A fourth dispute is older and unresolved: do we even need a theory of consciousness, or several — one for access, one for phenomenal experience, one for self-models? The unification assumption may itself be wrong.
Consciousness without a prefrontal cortex
If posterior cortex is doing the heavy lifting, three observations stop being puzzles:
- Infants show signs of conscious experience before prefrontal cortex finishes wiring.
- Octopuses have ~500 million neurons, two-thirds of them in the arms, and no prefrontal analog — yet display problem-solving, play, and individual recognition. See concept octopus intelligence.
- Anesthetics with different mechanisms (propofol, ketamine, xenon) produce different phenomenologies. The variation maps better onto disruption of posterior integration than onto a single broadcast switch.
The uncomfortable implication: the felt moment may be more sensory-perceptual than cognitive. Abstract thought might be the cheap part.
Why this has to do with other realms
The structural shape of the consciousness problem rhymes with the concept quantum measurement problem: a complete physical description (wavefunction, neural firing) plus an observer yields an outcome (collapse, experience) the physics alone does not predict. Both fields have spent a century arguing about whether the observer is a real ingredient or a notational convenience. And both have a Halting-Problem flavor — see concept halting problem — where the system seems to require something outside itself to be evaluated. Whether this is a deep parallel or three independent confusions wearing the same hat is the kind of question a polymath wiki exists to ask.
AI and the test we don't have
LLMs broadcast information across attention layers, integrate it (in IIT's loose sense), and run predictive models. By the functional vocabulary of every theory above, they tick boxes. None of the theories give an operational test that could be applied to a non-biological system without circularity. The concept chinese room argument — that symbol manipulation never becomes understanding — has not been refuted; it has been side-stepped.
The 2025 result narrows the question to: what specifically is posterior cortex doing that current architectures don't? Recurrent integration across sensory modalities is a candidate. Whether silicon can do it is open.
An open question
If consciousness lives in posterior sensory cortex rather than prefrontal cognition, what does that imply about the experience of organisms whose sensory cortices are radically unlike ours — the octopus, the mantis shrimp with sixteen photoreceptor types, the electric fish that perceives the world as a voltage field?
Key sources
- Nature, April 30, 2025 — Cogitate Consortium adversarial collaboration on IIT vs GNWT. The reference for everything in the scorecard.
- The Feeling of Life Itself — Giulio Tononi (2017). IIT for non-specialists, by its originator.
- Consciousness and the Brain — Stanislas Dehaene (2014). The clearest statement of the global workspace position.
- The Hard Problem of Consciousness — David Chalmers (1995, Journal of Consciousness Studies). The paper that named the gap every theory above tries to close.
- To verify: Friston's 2010 Nature Reviews Neuroscience paper on the free energy principle as the canonical FEP reference.
- To verify: Rosenthal's higher-order thought theory primary statement (1986 or later).
Further reading
- Other Minds by Peter Godfrey-Smith — octopus cognition as a second, independent evolution of complex experience. The best single book for thinking about non-human consciousness.
- Being You by Anil Seth (2021) — the predictive-processing case for consciousness as "controlled hallucination," accessible and concrete.
- The Conscious Mind by David Chalmers (1996) — the long-form version of the hard problem argument.
- Sean Carroll's Mindscape podcast episodes with Chalmers, Tononi, and Friston — the three positions in their own words.
- The Cogitate Consortium's data and pre-registration on the OSF — the rare chance to inspect a major neuroscience result without paywalls.
See Also
- concept hard problem consciousness — the philosophical puzzle the empirical work has not solved.
- concept chinese room — why functional success does not settle the question of understanding.
- concept octopus intelligence — 500M neurons, no prefrontal cortex, clearly something going on.
- concept quantum measurement problem — the other place in science where "the observer" refuses to leave the equations.
- concept halting problem — formal limits on systems evaluating themselves; possible structural analog to the explanatory gap.
- concept emergence — the substrate-to-property leap that makes consciousness feel impossible and inevitable at once.
- concept gut brain axis — if the locus of experience is less cortical than assumed, how far down does it go?