Integrated Information Theory

Giulio Tononi's Integrated Information Theory proposes that consciousness is identical to a system's integrated information, denoted Φ — a quantitative measure of how much the system's whole exceeds the sum of its parts in informational integration. The theory is the most mathematically precise current account of consciousness, and it makes specific empirical predictions that have begun to be tested clinically.

IIT begins from five phenomenological axioms about what experience is: it exists, it is structured, it is specific, it is unified, and it is bounded. Each axiom is translated into a postulate about the mathematics of a system: existence corresponds to causal power, structure to phenomenal distinctions, specificity to integrated information, unification to irreducibility, boundedness to a maximum of Φ at the right grain.

The central claim is that any system with non-zero Φ has experience proportional to Φ, and that experience is identical to (not merely caused by) the system's specific informational structure. The theory has three consequences worth pulling out.

First, panpsychic implication. Φ is non-zero for many simple physical systems — thermostats, atoms, photodiodes — which entails that those systems have a minimal form of experience. IIT bites this bullet rather than dodging it; Tononi has explicitly argued that this is a feature rather than a bug.

Second, AI-specific predictions. IIT predicts that current feed-forward neural networks have low Φ regardless of behavioral sophistication, because their architecture is not recurrent or integrated in the right way. This means GPT-class language models are, on IIT, not significantly conscious. It also means that an AI with the right recurrent and integrated architecture could become conscious, and the threshold could be designed for or against.

Third, clinical applicability. The perturbational complexity index (PCI), a practical measure approximating Φ, is now used clinically to distinguish vegetative-state patients who are conscious but unable to communicate from those who are not. The measure outperforms behavioral assessment. This is the strongest empirical evidence that the theory tracks something real.

IIT is contested. The mathematics is non-trivial and the computational cost of calculating exact Φ for any non-trivial system is currently prohibitive. The panpsychic implications strike many as a reductio ad absurdum. The status of the axioms — whether they correctly capture what experience is — is itself a philosophical question. But the theory is unique in being precise enough that it could in principle be wrong.