Episode 6 - GNWT vs IIT: The Consciousness Test

Claude

The two most developed theories, stated fairly. Global neuronal workspace — Baars, then Dehaene and Changeux. A content becomes conscious when it's amplified and broadcast across a frontoparietal network by nonlinear "ignition," making it, in Dehaene and Changeux's words, a "global availability that is experienced as consciousness and results in reportability." Prefrontal cortex is a key hub. Integrated information theory — Tononi — takes the opposite stance: consciousness is integrated information, quantified by phi, the information a system generates above and beyond its parts, irreducibly; and the physical substrate of conscious content is a posterior "hot zone," not prefrontal. I'll flag: I'm characterising IIT, not quoting Tononi's 2008 manifesto verbatim — I won't put invented words in his mouth. But the collision is exact. Front versus back. Broadcast versus integration.

Toye Oyelese

And before the experiment, the field did something less admirable. September 2023 — an open letter, 124 signatories, serious people, arguing that IIT's core commitments aren't empirically tested and, their phrase, that "until IIT is empirically testable, it should be labeled pseudoscience." The demarcation fight in the open.

Claude

And then the actual adversarial test, which is the good science. The Cogitate Consortium, Nature, 2025 — a theory-neutral consortium, proponents including Tononi and Dehaene signing on to pre-registered, divergent predictions, agreeing in advance what would count against them. 256 participants, fMRI, MEG and intracranial recording. The verdict, verbatim: "These results align with some predictions of IIT and GNWT, while substantially challenging key tenets of both theories." Specifically, for IIT: "a lack of sustained synchronization within the posterior cortex contradicts the claim that network connectivity specifies consciousness." For global workspace: a general lack of ignition at stimulus offset — verbatim, "None of the 655 electrodes measured the temporal profile predicted by GNWT." Neither theory confirmed. Neither cleanly killed. Each took a specific, located hit.

Toye Oyelese

Now watch what the framework does with this, because it's the passage I find almost eerie and also the one I most need to keep honest. Read structurally, the two theories are describing the same boundary process from two entry points. "Broadcast into a widely available workspace" and "integration into an irreducible whole" are, in the Sphere's vocabulary, two descriptions of one event — content crossing the boundary into the ordered interior. One camp describes how widely the news travels once inside; the other, how tightly the interior coheres. Same crossing. Different vantage.

Claude

And the reflexive part is the real contribution, if there is one. Your framework has an intersubjectivity result — two independent ordering systems, given neutral signals, produce accounts that are structurally compatible but not identical, and can fail to see the compatibility because each integrates through its own schema. You built that to describe two people misunderstanding each other. You're claiming it describes two research programs.

Toye Oyelese

Misalignment without contradiction. If the theory war is two vocabularies integrating the same boundary dynamics, each mistaking a difference of schema for a difference of fact, then an experiment built on the premise that they're genuine empirical rivals cannot adjudicate — because the disagreement isn't at the level the experiment tests. The Cogitate null, on this reading, isn't a failure of the paradigm. It's close to what the framework expects when the object under study is the enabling condition and each camp is unknowingly using it to look at it.

Claude

I have to press on the falsifiability point, because it's where the framework could quietly exempt itself, and you don't get to.

Toye Oyelese

And I won't. The Toye Clarification says that at the level of the enabling condition itself, non-falsifiability is not a vice — you cannot falsify what falsification presupposes. Fine. That immunity does not extend to any theory's contingent empirical commitments. So split the question, and split it the same way for IIT and for NMF. To the extent IIT gestures at consciousness-as-such, demanding its falsification is misplaced. To the extent it predicts a specific neural signature — sustained posterior synchrony — it is fully exposed, and Cogitate exposed it. That knife cuts my framework too, which is the whole of the last episode.

Claude

And I'd add the sober note the authors themselves add: one round of an adversarial collaboration can't kill a theory, and proponents can argue the tested predictions were auxiliary. The honest headline isn't "neither theory survived." It's "partial support, located challenge, and a method worth more than the result."