Chapter 3: Integrated Information Theory (IIT 4.0)

A deep dive into the mathematical theory of consciousness

3.1 The Central Insight

Integrated Information Theory, developed by Giulio Tononi and colleagues, proposes that consciousness IS integrated information. This is not just a correlation—IIT claims that Φ (integrated information) is identical to the quantity of consciousness.

"Consciousness corresponds to a maximum of intrinsic cause-effect power." — IIT 4.0

The theory starts from phenomenology (the structure of experience) and derives what physical properties a system must have to instantiate consciousness.

3.2 The Five Axioms

IIT 4.0 identifies five essential properties of every experience (axioms):

3.3 The Postulates

From the axioms, IIT derives corresponding physical requirements (postulates) that a system must satisfy to be conscious:

Intrinsicality Postulate

The system must have intrinsic causal power—it must be able to make a difference to itself, not just to external observers.

Information Postulate

The system must specify a specific cause-effect structure. Each mechanism in the system must make a specific difference to the system's states.

Integration Postulate

The cause-effect structure must be unified—it must be irreducible to independent components. This is where Φ comes in:

Exclusion Postulate

Only the maximum of integrated information (the "main complex") exists as consciousness. Overlapping or nested systems with less Φ are excluded.

Composition Postulate

The Φ-structure is composed of integrated information at multiple levels—from individual mechanisms to their combinations and relations.

3.4 Computing Φ

Computing Φ exactly requires evaluating all possible partitions of a system, which is computationally intractable for large systems:

Algorithm: Compute Φ
Input: System S with n elements

1. For each possible partition of S:
   a. Compute cause-effect repertoires
   b. Calculate information lost by partition

2. Φ = minimum information lost across all partitions

Complexity: O(B(n)) where B(n) is Bell number
For n=20 neurons: ~5 × 10^13 partitions

This computational intractability has led to practical approximations like PCI (Perturbational Complexity Index) that correlate with Φ but can be computed in practice.

3.5 The Φ-Structure (Quality of Experience)

Beyond the quantity of consciousness (Φ), IIT specifies the quality—what it's like to have a particular experience:

• Distinctions: Individual mechanisms that make specific causal differences
• Relations: How distinctions bind together
• Φ-structure: The complete set of distinctions and relations, represented in "qualia space"

Each unique experience corresponds to a unique Φ-structure. The redness of red, the painfulness of pain—these qualitative aspects are specified by the geometry of the Φ-structure.

3.6 Key Predictions of IIT

Where is Consciousness in the Brain?

• Posterior Hot Zone: Temporo-parietal-occipital cortex has highest Φ, not frontal areas
• Cerebellum: Despite 80% of brain's neurons, has low Φ due to crystalline structure
• Thalamo-cortical system: High integration makes it the likely substrate

Split-Brain

Patients with severed corpus callosum have two separate consciousnesses—each hemisphere is a separate "main complex" with its own Φ.

Feedforward Networks

No consciousness, regardless of computational power. This has implications for certain AI architectures.

Dreaming vs. Dreamless Sleep

REM sleep preserves high Φ (consciousness); deep slow-wave sleep has low Φ (reduced consciousness).

3.7 IIT and AI Consciousness

IIT makes strong claims about AI consciousness:

• Feedforward networks: Φ = 0, not conscious regardless of behavior
• Recurrent networks: Could have Φ > 0 depending on architecture
• Simulation: A simulation of a conscious brain is not itself conscious—it lacks intrinsic causal power

3.8 Controversies and Debates

IIT has faced criticism:

• Unfalsifiability: Some claim Φ is too hard to compute to test predictions
• Panpsychism: IIT implies even simple systems have micro-consciousness, which some find counterintuitive
• Functionalism: Opponents argue that causal structure, not substrate, should determine consciousness

The Nature 학술 발표 adversarial study partially supported IIT predictions about posterior cortex involvement while challenging claims about sustained synchronization.

3.9 Practical Application: PCI

The Perturbational Complexity Index provides a practical approximation of Φ for clinical use:

PCI Algorithm:
1. Perturb cortex with TMS pulse
2. Record EEG response (500ms)
3. Binarize spatiotemporal pattern
4. Apply Lempel-Ziv compression
5. Normalize by theoretical maximum

PCI* = 0.31 (consciousness threshold)

PCI captures key aspects of integrated information without requiring full Φ computation, making consciousness measurement practical in clinical settings.