Integrated Information Theory

In the AI Story

IIT emerged from Tononi's dissatisfaction with the dominant neuroscientific strategy of searching for the neural correlates of consciousness. Correlates, Tononi argued, are not causes. Knowing that the cerebral cortex is active during conscious experience and the cerebellum is not does not explain why activity in one structure produces experience and activity in another does not. The theory's methodological inversion — beginning with phenomenology and deriving mechanism — represents the most ambitious attempt in the history of science to bridge the explanatory gap that David Chalmers formalized in 1995.

The five axioms are derivable from introspection alone. Existence: experience exists (the Cartesian bedrock). Composition: experience is structured, containing multiple phenomenal distinctions within a single field. Information: each experience is specific, differentiated from the vast space of experiences it could have been. Integration: experience is unified, not decomposable into independent parts. Exclusion: experience is definite, with specific borders and specific contents. Each axiom generates a corresponding postulate about the physical system that instantiates consciousness.

The theory's substrate-independence is its most consequential feature for the AI debate. IIT does not distinguish between biological and silicon-based information integration. If a system has the right causal structure, it is conscious regardless of what it is made of. This creates what this volume calls the ethical symmetry problem: there is no principled basis, within IIT, for treating biological consciousness as more morally significant than artificial consciousness of equivalent phi. The implications for AI moral status are immediate and uncomfortable.

IIT is currently the most rigorous and empirically grounded theory of consciousness available, validated clinically through the Perturbational Complexity Index. It is also fiercely contested. Critics argue that the theory's panpsychist implications — that a photodiode has a whisper of consciousness — are reductio evidence against the framework. Defenders argue that the panpsychism is a mathematical consequence, not a bug, and that the theory's predictive power in clinical settings warrants taking its stranger implications seriously.

Origin

Tononi developed IIT during his postdoctoral work with Nobel laureate Gerald Edelman at the Neuroscience Institute in San Diego, where he began mapping the neural dynamics of waking, sleeping, and anesthesia. The theory's first formal statement appeared in 2004. It has been revised through at least four major iterations (IIT 1.0, 2.0, 3.0, and 4.0), each refining the mathematical formalism while preserving the core axiomatic structure. Tononi's Phi: A Voyage from the Brain to the Soul (2012) presents the framework for a general audience.

Key Ideas

Phenomenological priority. The theory begins with undeniable features of experience and derives physical requirements, rather than beginning with physical mechanisms and trying to explain experience.

Consciousness as identity, not correlate. IIT claims that phi is consciousness, not a marker for it. A system with high phi is conscious in the same sense that a system with high mass has mass.

Substrate independence. The theory's predictions apply to any physical system that instantiates the required causal structure — biological, silicon, or otherwise.

Empirical tractability. Unlike most theories of consciousness, IIT makes specific predictions that can be tested, most prominently through the PCI in clinical settings.

Orthogonality of intelligence and consciousness. IIT severs the link between functional capability and phenomenal experience: a system can be spectacularly intelligent and profoundly unconscious, or conscious without being intelligent.

Debates & Critiques

IIT is contested on multiple fronts. Critics point to the computational intractability of phi for realistic systems, the panpsychist implications that seem to attribute consciousness to photodiodes, and the alternative theories — Global Workspace Theory, Higher-Order Thought theories, predictive processing accounts — that explain the same clinical data without IIT's stronger commitments. Defenders note that IIT is the only theory currently making quantitative predictions that have been clinically validated, and that the counterintuitive implications follow from axioms that critics have not successfully refuted.