Global Broadcast Fails as a Theory of Consciousness

Abstract

Global Workspace Theory has failed to prove that broadcast is necessary or sufficient for human-like reportable access consciousness. Either broadcast is just availability to reasoning, memory, action and speech, in which case it is a renaming and not a mechanism, or it is a workspace architecture that has not been shown necessary for that availability. To resolve this, I construct direct routes to all four functions without any workspace. Existing experiments associate late distributed activity with reportable access, but no-report studies and a preregistered adversarial test challenge canonical claims about prefrontal ignition. I then prove insufficiency. To mean and report a content, a speaker must model what the listener will infer the speaker intends, which requires a model of the listener's model of the speaker. I prove that every report policy lacking that state fails under identifiable decoder mismatch. GWT routes content into speech but does not require such a model, so a GWT machine can report while human-like access consciousness is absent. Conversely, direct routing combined with a model of the listener's model of the speaker yields human-like reportable access without any workspace. Global broadcast is therefore neither necessary nor sufficient for human-like reportable access consciousness.