Integrated Information Theory

← Encyclopedia

A neurologist standing at a hospital bed faces a question older than any instrument on the cart. The patient does not move, does not track a face, does not answer — and the staff need to know whether anyone is still inside. Behavior has failed as a guide; a body can be awake without an occupant, or occupied without a sign. For decades the only honest answer was a clinical guess. The theory examined here was built to replace that guess — and it starts nowhere near the bedside.

Integrated Information Theory, proposed by the neuroscientist and psychiatrist Giulio Tononi in 2004 and developed since with Christof Koch, Larissa Albantakis, Marcello Massimini, and others, makes a claim more radical than it first sounds. Most accounts of consciousness ask how the brain gives rise to experience. IIT refuses the question’s shape. Its central thesis is an identity: a conscious experience does not arise from, emerge out of, or get produced by the brain’s activity — it is a particular mathematical object, the maximally irreducible cause-effect structure that a physical substrate specifies in its current state. The wording is load-bearing, and its proponents defend it fiercely. Consciousness is not what integrated information does. It is what integrated information, of the right kind, is.

The route to that claim inverts the usual order of inquiry. Rather than starting from neurons and working toward experience, IIT starts from experience and works toward the physics it would require. It takes a handful of properties that every experience self-evidently has — that it exists for itself; that it is structured into parts; that it is the specific experience it is and no other; that it is unified, not splittable into independent experiences; that it is definite, this much and no more — and treats these as axioms. From each axiom it infers a property the underlying substrate must possess: intrinsic cause-effect power, composition, specificity, irreducibility, and a maximum. Anil Seth, who defends the theory without subscribing to it, calls this “a consciousness-first, rather than a brain-first approach… it turns the standard neuroscientific game upside down.” The most disputed move sits exactly at the hinge — the inference from experience is unified to the substrate has irreducible intrinsic cause-effect power. Critics, including the Internet Encyclopedia of Philosophy’s careful survey, hold that this translation from phenomenology into physics is not forced; nothing compels the leap from a felt fact to that particular formalism.

What the formalism yields is a number, and that number is where the theory’s ambition and its trouble both concentrate. Each part of a system, in its current state, constrains what could have caused it and what it will cause; the irreducibility of the whole web of these constraints — how much is lost when the system is cut at its weakest seam — is measured by Φ, “phi.” The quantity of consciousness tracks the value of Φ; the quality, the specific way an experience feels, is carried by the shape of the whole structure, which the theory’s 2023 fourth edition makes explicit by folding in the relations among a system’s parts rather than reporting Φ as a lone scalar. There have been several non-equivalent definitions of Φ across the theory’s versions, from the 2004 original through the 2014 third edition to the fourth, and skeptics treat the moving target as a weakness. But the deeper limit is plainer, and IIT’s honest expositors state it outright: Φ is, for any real brain, incomputable in practice. Exact computation requires testing every way of partitioning the system, and the count explodes past all possibility. The roundworm C. elegans has just 302 neurons, yet the number of ways its network can be cut is, in Koch’s illustration, ten followed by 467 zeros. No one has ever computed Φ for a human brain, or expects to. The theory’s quantity of consciousness is, at the scale that matters most, a number that cannot be reached.

This is why the bedside answer, when it came, did not come from Φ. It came from a proxy built on the same intuition — that consciousness lives where the brain is both highly differentiated and highly integrated. The method is blunt and elegant: zap the cortex with a single magnetic pulse and record the echo across the scalp with dense electroencephalography, then zip — compress the spatiotemporal response. An unconscious brain answers a pulse with a simple, local, stereotyped wave that compresses easily. A conscious brain answers with a complex, widespread, non-repeating pattern that resists compression. The Perturbational Complexity Index reads out that resistance. Anton Casali and colleagues introduced it in 2013, calibrated across wakefulness, sleep, several anesthetics, and brain injury; in 2016 Silvia Casarotto’s team set an empirically validated threshold and found that some patients judged unresponsive at the bedside crossed it — carrying the brain complexity of consciousness with no behavioral way to show it. Covert awareness, detected by physics where the clinical exam was blind. It is IIT’s strongest applied result, and it must be stated with care: the index does not measure Φ. It is an inspired surrogate for the integration-plus-differentiation idea, not the theory’s incomputable core made flesh. The practical triumph and the untested center are two different things, and the theory’s credibility depends on not confusing them.

On the brain’s own map, IIT makes a prediction that has become its signature. The cerebellum holds roughly 69 billion of the brain’s 86 billion neurons — the overwhelming majority — yet a person born without one can be conscious, while damage to a far smaller patch of cerebral cortex can extinguish consciousness entirely. Neuron count cannot explain this; architecture can. The cerebellum is built as a vast bank of parallel, feed-forward, modular circuits that barely integrate — low Φ — while the cortex is densely re-entrant, looping back on itself, the kind of structure that integrates richly. IIT further locates conscious content not in the prefrontal cortex but in a posterior temporo-parietal-occipital “hot zone,” a claim that sets it directly against Global Neuronal Workspace Theory, its leading rival, which centers the front of the brain and a moment of global “broadcast.”

Those two theories met in an unusual arena. In an adversarial collaboration — a design in which proponents of both camps, refereed by a neutral consortium, agree in advance what each theory predicts and what would count as failure — the Cogitate group recorded 256 people with three methods at once and published the verdict in Nature in 2025. It was a split decision, not a coronation. The results, the authors wrote, “align with some predictions of IIT and GNWT, while substantially challenging key tenets of both.” Conscious content proved decodable from posterior cortex, and the prefrontal involvement the rival theory needs was not necessary — points for IIT’s map. But the sustained synchronization across posterior cortex that IIT’s connectivity claim requires did not appear, which struck at a core mechanism. Neither theory won; both bled. To report this as a confirmation of IIT, as some coverage did, is to misread it.

The same year’s headlines carried a second story, harder to weigh. In September 2023, 124 scholars signed an open letter calling IIT “pseudoscience,” arguing that its commitment to a widespread, fundamental consciousness was extravagant and that the theory as a whole outran what could be tested — and that media attention had outrun the evidence besides. It caused, Nature reported, an uproar. The rebuttals came fast and from serious quarters. Seth called the charge “inflammatory,” noting that strange or even untestable consequences do not make a research program pseudoscientific — general relativity implies untestable singularities, and quantum mechanics its uninterpretable interpretations; what matters is whether the program generates testable predictions over time, which IIT plainly does. Koch answered that the theory makes its assumptions clear and is therefore falsifiable in principle. Liad Mudrik, a leader of the very collaboration above, put it most sharply: “Not only did we test it, we managed to falsify one of its predictions.” A field survey found only a small minority who fully endorsed the pseudoscience label. The fairest reading is the unglamorous one — IIT is mainstream-adjacent and genuinely contested, neither debunked nor established, a live theory drawing live fire.

What the theory implies about the rest of the world is where the traditions documented here lean in. If any system with nonzero Φ has some flicker of experience, and nonzero Φ can occur outside brains, then consciousness is graded and widespread — Tononi and Koch grant it “the same status as mass or electric charge.” This is panpsychism’s near neighbor, and the resemblance to the old doctrine of an ensouled cosmos is not faint. But the theory draws a hard line the classical view never did: experience requires the right architecture, so not everything is conscious. A heap of sand has none; a feed-forward network has none; a digital computer simulating a brain, however convincing its behavior, has effectively zero Φ and so, by this account, no inner life at all. That last consequence cleaves IIT cleanly from the assumption that a machine acting conscious must be conscious — and it is a different theory entirely from the Penrose-Hameroff proposal that consciousness arises from quantum events inside neuronal microtubules, with which it shares the word consciousness and no machinery.

Set against the longer history this archive keeps, IIT marks an unfamiliar position. The traditions held the primacy of consciousness as vision; here it appears as a peer-reviewed hypothesis with a clinical instrument attached, asserting that the felt fact is the firm one and the brain the thing to be inferred from it. The convergence is worth noticing and worth not overstating: a theory that puts experience first sits among live rivals, owing nothing to the traditions and proving nothing for them. What is settled is narrow. A magnetic pulse, an echo, a number that resists compression: with these a physician can now find, in a body that gives no other sign, the trace of someone present.

→ Related: Panpsychism · Hard Problem Of Consciousness · Orch Or · Global Consciousness Project · Quantum Entanglement