The Larger World ArchiveClaims · Sources · Evidence
Mind, belief and memoryLive debateNeuroscience of consciousness · 2004 – present

IIT versus Global Workspace

One theory says consciousness is integrated information, wherever it occurs. The other says it is information broadcast widely in the brain. Both made predictions, and both were tested.

A high-density EEG cap. The adversarial collaboration that pitted integrated information t
A high-density EEG cap. The adversarial collaboration that pitted integrated information theory against global workspace theory used EEG, MEG and intracranial recordings. Image: Chris Hope · CC BY 2.0 — Wikimedia Commons.

The two theories

Integrated information theory (IIT), developed by Giulio Tononi from 2004, starts from experience and asks what physical system could have it. Its answer: a system with high integrated information, denoted Φ (phi) — information that is more than the sum of its parts. IIT predicts consciousness depends on the structure of the causal network, with the posterior cortex as the hot zone.

Global neuronal workspace (GNW), from Bernard Baars's workspace idea formalised by Stanislas Dehaene and colleagues, says conscious content is whatever gets broadcast by a network of long-range pyramidal neurons, especially prefrontal and parietal. Unconscious processing stays local; conscious processing goes global.

The adversarial test

The Cogitate consortium — IIT and GNW theorists plus independent labs — ran a preregistered study with fMRI, MEG and intracranial electrodes, comparing the theories' predictions on the same data. The results, published in 2025, were mixed on purpose: some predictions of each theory were supported and others were not.

GNW's prediction of prefrontal ignition found less support than expected; IIT's prediction of sustained posterior synchrony found some. Both camps claimed partial vindication, which is roughly what an honest adversarial collaboration produces.

The trouble with measuring consciousness theories

Theories must be falsifiable to be science. In 2023 a group of researchers published an open letter calling IIT 'pseudoscience' on the grounds that its core claim — that Φ measures experience — is not testable. The letter was widely criticised as overreach, but the underlying point about operationalising Φ is serious: computing it exactly is intractable for real brains, so tests use proxies.

Proxies weaken tests. When a theory's central quantity can only be approximated, a failed prediction can always be blamed on the approximation.

Both theories could be wrong. The adversarial result does not crown a winner, and neither theory explains why any of this is accompanied by experience — the 'hard problem' is untouched by both.

Why it is in the archive

Because this is the best current example of consciousness science trying to behave like science: preregistration, adversarial collaboration, predictions stated in advance.

The archive also keeps the fringe at arm's length: quantum consciousness (Orch-OR and its relatives) makes claims that this literature does not support.

Sources for this page

  • Wikipedia, “Integrated information theory” — claims and criticismen.wikipedia.org
  • Cogitate consortium — the adversarial collaboration's design and resultswww.arc-cogitate.com
  • Albantakis et al. (2023), PLOS Computational Biology — IIT 4.0doi.org
  • Nature news (2023), consciousness theory called pseudosciencewww.nature.com