There's a particular kind of intellectual courage required to submit your theory to a process specifically designed to break it. Integrated Information Theory has been the most mathematically ambitious framework in consciousness science for years — and it's now entered a formal adversarial collaboration that will, by design, try to find where it fails.
That process is worth understanding carefully, because it's genuinely unusual in a field where competing theories tend to talk past each other rather than directly clash.
What the Adversarial Collaboration Actually Is
A new review published in Neuroscience & Biobehavioral Reviews lays out the structure of the INTREPID Consortium's adversarial collaboration, which pits three theories against each other: Integrated Information Theory (IIT), Neurorepresentationalism, and Active Inference (the predictive processing framework associated with Karl Friston). The consortium includes researchers from the University of Amsterdam, University of Glasgow, University of Wisconsin-Madison, Monash University, Columbia, and several other institutions — and notably, the architects of the competing theories are themselves participants.
The review is careful about what adversarial collaboration can and can't do. It outlines key hypotheses to be tested across multi-site experiments, describes what kinds of observations would support or challenge each theory, and — this is the methodologically interesting part — proposes a formal mechanism for quantitatively integrating evidence across disparate experiments and their replicates. The goal is a running score of evidential support, not a single knockout result.
This matters because consciousness research has a long history of theories that are unfalsifiable in practice even when they claim to be falsifiable in principle. The adversarial structure forces each camp to specify, in advance, what would count as evidence against them. That's a harder ask than it sounds.
IIT's Specific Claims and Their Vulnerabilities
IIT's core proposition is that consciousness is identical to integrated information — mathematically formalized as Φ (phi), a measure of how much a system's causal structure exceeds the sum of its parts. A recent analysis in Frontiers in Computational Neuroscience engages directly with what it calls IIT's ontological overreach: the theory's claim that systems lacking sufficient Φ don't merely lack consciousness but don't "truly exist" in any intrinsic sense. The author, Camilo Signorelli at Oxford, argues this creates problems for how IIT handles embodiment — the theory's mathematical formalism tends to treat the brain as a bounded computational system, which sits awkwardly with evidence that cognition is distributed across brain, body, and environment.
This isn't a fringe critique. It points to a genuine tension in IIT's architecture: the theory is powerful precisely because it's mathematically precise, but that precision requires drawing a boundary around the system being analyzed. Where you draw that boundary turns out to matter enormously for what Φ you calculate — and there's no theory-internal way to determine where the boundary should go.
The adversarial collaboration will test specific predictions that bear on this. The INTREPID review describes experiments designed to distinguish IIT's predictions from those of Active Inference, which handles the brain-body-environment relationship very differently. Active Inference treats perception and action as a unified process of minimizing prediction error, which generates different predictions about where in the brain consciousness-relevant activity should appear and when.
Why This Connects to Anomalous Cognition Research
Here's where this becomes relevant to readers of this newsletter beyond pure neuroscience interest: IIT's mathematical framework is, in principle, substrate-neutral. Φ can be calculated for any physical system with causal structure. That's what makes IIT philosophically provocative — it implies that consciousness is a feature of certain kinds of information integration, not of biological neurons specifically.
This has obvious implications for questions about non-human intelligence, but it also creates a framework for thinking about anomalous cognition claims more rigorously. If consciousness is a function of integrated information rather than a byproduct of specific neural hardware, then questions about extended or distributed cognition — the kind that psi researchers invoke when discussing remote perception or non-local correlations — at least become formally askable within IIT's framework, even if the theory doesn't endorse those phenomena. The question becomes: what would the Φ structure of a system capable of anomalous information transfer look like, and is that structure physically realizable?
I'm not suggesting IIT validates psi research. I'm suggesting that having a mathematically precise theory of consciousness is a prerequisite for asking precise questions about anomalous cognition — and that the adversarial collaboration now underway is the most rigorous attempt yet to determine whether IIT's precision is actually tracking something real.
The INTREPID review frames the whole enterprise as an exercise in meta-science as much as neuroscience: what does it mean for a theory to make progress, and how do you measure that progress across experiments that weren't designed to be compared? Those are questions worth watching regardless of which theory wins.
Watch for the first multi-site experimental results from the INTREPID Consortium — the review describes the experimental design but the empirical data aren't yet published. When they arrive, the question won't just be which theory predicted the results, but whether any of the three theories can actually accommodate what they didn't predict.
