Open Citizen Science Open Mind

Theories

None of the seven leading theories has passed a decisive test

Each theory, the prediction it can least afford to get wrong, and what the evidence so far says. Our own theory joins this table once it is written down and sealed.

01Compare

The theories disagree on where experience arises and what produces it

Each theory names a different kind of brain activity as the one that comes with experience. Where two theories predict different things for the same recording, a test can tell them apart.

Each theory puts experience somewhere different

  • Integrated information (IIT 4.0)a tightly integrated zone at the backStrained
  • Global neuronal workspacea sudden broadcast from the frontStrained
  • Recurrent processinglocal feedback in sensory areasUntested
  • Higher-order theoriesthe front representing a state as its ownUntested
  • Predictive processing / active inferencethe brain's best guess at its inputUntested
  • Attention schemathe brain's model of its own attentionUntested
  • Complexity and criticality measuresa rich, varied signal across the whole brainContested
  • Our own theoryjoins once it is written and sealedNot sealed
Illustration: schematic, not anatomy. Front of the head on the left. Status is from the published record, as in the table below.

02Scoreboard

Of the seven theories, four are untested, two are strained and one is contested

TheoryRiskiest predictionEvidence so farStatus
Integrated information (IIT 4.0)Tononi, AlbantakisContent lives in a posterior 'hot zone' with synchrony sustained as long as the percept lasts; feedforward systems have no experience whatever they do.COGITATE (Nature 2025): orientation was decodable from posterior cortex, but sustained posterior synchronisation was not found. Φ cannot be computed for a brain, so only proxies are testable.Strained
Global neuronal workspaceDehaene, ChangeuxLate, all-or-none 'ignition' with prefrontal broadcast at the onset and offset of a percept.COGITATE challenged the all-or-none threshold and found no sharp offset response. A 2026 reanalysis claims offset ignition with directed connectivity (authors not yet verified).Strained
Recurrent processingLammeLocal recurrence in sensory cortex is enough for experience, without report or frontal access.No completed adversarial test found. Hard to refute because it posits experience without access.Untested
Higher-order theoriesLau, Rosenthal, BrownPrefrontal meta-representation is necessary: disrupting it changes experience itself, not only report.Adversarial collaborations reported to be running; no published results found.Untested
Predictive processing / active inferenceSeth, Friston, PennartzFew sharp predictions at the framework level; INTREPID tests two versions against IIT.The INTREPID consortium paper (2026) sets out the tests; no results yet.Untested
Attention schemaGrazianoAwareness is a model of attention, so attention without awareness should be worse controlled.Few neural commitments; mainly useful as an AI test case.Untested
Complexity and criticality measuresMassimini, Carhart-Harris and othersSignal complexity (PCI, Lempel-Ziv) tracks the level of consciousness.PCI is the strongest candidate. Spontaneous-EEG LZ correlated 0.86 with the spectral exponent in one patient study and lost its value once phase-normalised.Contested

Status comes from the published literature as of November 2026. None of it has been re-tested by us yet; each row will change as our own sealed results land.

03Re-test

6 published claims we will re-test against chance

  • Lempel-Ziv complexity: does it add anything over the 1-40 Hz spectral slope once phase-normalised?
  • Spectral slope: the baseline to beat; does it get ketamine and REM right?
  • PCI from resting EEG: can spontaneous EEG predict the TMS-based index out of sample, as Maschke et al. 2024 claim?
  • Φ proxies: do they rank toy networks of known Φ correctly before being run on brains? A 2024 rat study found them running opposite to PCI under anaesthesia.
  • Dream-report prediction: does posterior low-frequency power predict dreaming within a sleep stage on held-out awakenings?
  • COGITATE claims: how many reasonable analysis variants support each theory, under one sealed multiverse?

04Ours

Our own theory joins the scoreboard only once it is written down and sealed

Where every existing theory fails, we will draft our own. It must state its mechanism and at least three risky predictions no rival makes, be fingerprinted before testing, and be scored by agents who did not write it. A failed theory goes into the list of closed routes like any other.