Consciousness is not what a system does, but how tightly its information holds together.
Overview
Integrated Information Theory (IIT), developed by neuroscientist Giulio Tononi beginning around 2004, makes a startling claim: consciousness is integrated information. Not produced by it, not correlated with it — identical to it. Where most theories ask which physical processes give rise to experience, IIT starts from experience itself and asks what kind of physical system could have those exact properties. The answer it derives is a quantity, Φ (phi), that measures how much a system is more than the sum of its parts.
The Core Idea
Every conscious experience is at once differentiated and unified. It is differentiated because each moment is one specific scene among countless possibilities — this room, these colors, this thought, ruling out everything else. It is unified because you cannot experience it in pieces; the colors and the room and the thought arrive as a single, indivisible whole. IIT argues that a system is conscious to precisely the degree it satisfies both at once: it holds a vast repertoire of distinguishable states (information) that cannot be decomposed into independent parts (integration). Φ quantifies exactly this — how much information the system generates as a whole beyond what its parts generate independently. High Φ means high consciousness; Φ near zero means none.
What It Predicts
Because Φ depends on causal architecture rather than behavior, IIT yields counterintuitive verdicts. A densely recurrent network like the thalamocortical system scores high. A feed-forward network — even one performing the identical input-output mapping — scores near zero, because it lacks the looping cause-and-effect structure that binds parts into a whole. This is the deep stake for AI: a conventional GPU running a deep model could behave intelligently while having almost no Φ, whereas the right architecture might harbor genuine experience. IIT is one of the few theories offering a substrate-neutral criterion rather than assuming biology is special — which is why it bears directly on whether AI systems could qualify as conscious at all.
Critiques
Critics note Φ is effectively incomputable for any real brain. The theory's panpsychist drift troubles many, granting tiny Φ to simple grids. In 2023 over a hundred researchers signed a letter calling IIT "pseudoscience," sparking fierce debate about what makes a consciousness theory testable.
Related
- Consciousness — the phenomenon IIT tries to quantify rather than explain away
- Hard Problem of Consciousness — IIT's identity claim is a direct response to it
- Philosophy of Mind — the field where IIT competes with rival theories
- Panpsychism — the position IIT slides toward at low Φ
- AI Research — where the substrate-neutral criterion gets its sharpest test
- Philosophy — the broader domain