The Templeton Foundation's COGITATE project was designed to settle the consciousness debate. Two leading theories — Integrated Information Theory and Global Neuronal Workspace Theory — made incompatible predictions about where in the brain conscious content resides and how it's maintained. IIT predicted sustained synchronization in the posterior cortex. GNWT predicted brief ignition events in the prefrontal cortex, firing at both onset and offset of a stimulus. Six labs, 256 participants, three imaging modalities. Predictions preregistered. No ambiguity about what each side expected.
Neither prediction held.
IIT got its posterior localization but not its sustained synchronization. GNWT got some prefrontal involvement but no offset ignition. The experiment was clean. The format was rigorous. The result was inconclusive.
What happened next is what Daniel Kahneman had warned the project leaders about at the start. When smart adversaries encounter inconvenient data, he said, “their IQ leaps 15 points.” This is exactly what occurred. Melanie Boly, an IIT proponent, attributed the synchronization failure to sampling limitations — sparse electrode coverage, not a theoretical problem. Stanislas Dehaene, for GNWT, proposed that subjects had “allowed their minds to wander,” and later added that offset ignition was never really a core prediction for task-irrelevant stimuli anyway.
Preregistration was supposed to prevent this. It didn't. The predictions were locked; the interpretive framework wasn't. Both sides conceded the specific failures and claimed the general structure.
The decisive experiment ended up testing something other than consciousness. It tested whether theorists could accept results that contradicted their life's work. The format that was supposed to resolve a scientific question instead revealed a methodological one: the hardest part of an adversarial collaboration isn't designing the experiment. It's designing adversaries who can lose.