A theory of consciousness that can't be tested isn't a theory — it's a metaphysical preference. That charge has followed Integrated Information Theory for years, and it sharpened considerably in 2022 when philosopher Ned Bartlett argued that two of IIT's signature predictions about "silent neurons" were unfalsifiable by design. A new paper published this August in Neuroscience of Consciousness pushes back directly, and the argument is worth following carefully — because the question of whether IIT is testable determines whether the entire research program around consciousness measurement has scientific legs.
The Specific Predictions That Were Called Untestable
IIT makes a genuinely strange claim about inactive neurons. According to what the paper calls the "silent brain" (SB) prediction, rendering all active neurons inactive in the brain's primary conscious substrate doesn't eliminate consciousness — because those neurons retain the capacity to fire, and that capacity is what matters to IIT's measure of integrated information, phi. The second prediction, the "disabled neuron" (DN) case, holds that permanently disabling a subset of silent neurons can alter the qualitative character of experience in counterintuitive ways, even though those neurons weren't firing to begin with.
Bartlett's objection was that any evidence you could gather for either prediction would simultaneously undermine the conditions required to test it. You can't confirm the silent brain state while also getting a report of consciousness from the subject — because the neural mechanism producing that report would itself violate the silence condition. It's a neat trap.
The new paper, by S. Ponce de Leon at UC Merced, argues the trap has an exit. For the SB case, the key move is identifying a neural mechanism outside the main conscious complex that could generate a report of consciousness while the complex itself remains silent — preserving the test conditions rather than collapsing them. For the DN case, the paper distinguishes between two causal frameworks: an IIT-consistent "dispositionalist" account, where disabled neurons affect consciousness through their dispositional properties, and a more conventional "actualist" account, where causal influence requires actual firing. The distinction matters because it determines what kind of evidence would count as confirmation versus disconfirmation.
Drawing on Imre Lakatos's philosophy of science, Ponce de Leon frames the broader issue as one of research program structure: the difficulty in resolving these competing explanations reflects something fundamental about why consciousness theories are so hard to adjudicate, not a flaw unique to IIT.
Why This Matters Beyond the Philosophy Seminar
The practical stakes here are real. IIT has been one of the primary theoretical frameworks driving attempts to measure consciousness in clinical contexts — including anesthesia research, where the goal is distinguishing genuine unconsciousness from awareness under paralysis. If the theory's core predictions are untestable, the measurement tools derived from them rest on shakier ground than their proponents acknowledge.
The testability debate also intersects with a broader moment in consciousness science. A competing framework has just arrived from MIT, where researchers at the Picower Institute published a review in The Journal of Neuroscience arguing that cognition and consciousness arise from analog computations performed by traveling brain waves of different frequencies — a fundamentally different mechanistic account than IIT's information-integration story. Picower Professor Earl K. Miller, the paper's senior author, puts it plainly: "The brain generates waves, and wave dynamics are a highly efficient way to coordinate and perform computation." Where IIT focuses on the structure of causal relationships among neurons, the analog-computation theory emphasizes the dynamic coordination role of neural oscillations. These aren't necessarily incompatible, but they make different predictions about what to measure and where.
Meanwhile, researchers are actively applying consciousness theories to AI systems — and IIT is among the frameworks in play. A Dimensional Consciousness Model described in recent coverage weights evidence from ten scientific theories simultaneously to estimate the probability that large language models have some form of experience, with IIT contributing to that assessment. Whether IIT's predictions are testable in biological systems has direct bearing on whether its application to artificial ones means anything at all. As one framing of the problem notes, the more you examine the intricacies of the biological brain, the richer and more dynamic it appears compared to silicon-based systems — which is precisely why getting the biological theory right matters before exporting it.
The Honest Accounting
What Ponce de Leon's paper does well is refuse the easy exits. It doesn't argue that IIT is confirmed — it argues that the theory is more testable than its critics claimed, and it maps the specific conditions under which tests could be run. That's a methodological contribution, not an empirical one, and the paper is careful about the distinction.
What remains genuinely open: whether the experimental setups required to test the silent neuron predictions are achievable with current neuroscience tools, and whether the Lakatosian framework Ponce de Leon invokes actually resolves the underdetermination problem or just names it more precisely. The paper was published in August 2026; replication and critical response haven't had time to accumulate.
The most useful thing to watch for in the next year is whether IIT researchers attempt to operationalize the testable versions of these predictions in actual experimental protocols — particularly in anesthesia contexts where the silent-neuron question has direct clinical relevance. The gap between a theory surviving a testability challenge and a theory generating confirmed experimental results is exactly the kind of distance this newsletter exists to track. A theory that clears the methodological bar is still just a theory. The next step is the experiment.
