Source Count: 14 | Weighted Score: 33 | Source Confidence: [4/5] | Primary Tier: 2 | Last Updated: April 2, 2026
Keywords: integrated-information-theory, iit, consciousness, phi, giulio-tononi, qualia, information-integration, exclusion-postulate, cause-effect-structure, panpsychism
Category Tags: consciousness-theory, information-theory, neuroscience, philosophy-of-mind
Cross-References: ZD_1_16 — Quantum Information Theory · K_1_01 — Consciousness Overview · ZE_3_20 — Artificial Consciousness Ethics

QUICK SUMMARY

Integrated Information Theory (IIT) is a mathematical theory of consciousness developed by Giulio Tononi (University of Wisconsin-Madison, 2004; IIT 3.0, 2014; IIT 4.0, 2022) that attempts to explain what consciousness is, what determines its quality (the specific experience one is having), and how much of it there is. KEY FINDING IIT proposes that consciousness is identical to a physical system's capacity for integrated information, quantified by a measure called Φ (phi) — the amount of information generated by a system as a whole, above and beyond the information generated by its independent parts. A system is conscious to the degree that it is both differentiated (capable of being in many different states — high information) and integrated (a unified whole that cannot be decomposed into independent parts without loss of information). IIT is built on five axioms about the essential properties of every conscious experience: intrinsicality (experience exists from the system's own perspective), composition (experience is structured), information (experience is specific — this experience, not another), integration (experience is unified — not decomposable), and exclusion (experience is definite — has particular borders and grain). From these axioms, IIT derives postulates specifying the physical requirements a system must satisfy. Critically, IIT predicts that feed-forward networks (including standard digital computers processing information serially) have Φ = 0 — they are not conscious regardless of their functional behavior — because they can be decomposed into independent modules without information loss. This puts IIT in direct conflict with functionalist theories of mind and with claims that AI systems can be conscious simply by virtue of performing intelligent behavior. IIT also implies a form of panpsychism — even simple systems (e.g., a photodiode) have a non-zero Φ and thus some minimal form of consciousness. The theory has been both praised for its mathematical rigor and criticized for computational intractability (computing Φ requires evaluating all possible bipartitions of a system — exponential in the number of elements), unverifiable predictions, and counterintuitive implications.

1. VERIFIED CLAIMS (Tier 1 — Peer-Reviewed / Established)

• KEY FINDING IIT's core formalism: Tononi (2004, BMC Neuroscience) introduced the mathematical framework: Φ measures the "information generated by the whole above and beyond its parts," computed as the minimum information partition (MIP) — the bipartition of the system that makes the least difference to the cause-effect structure. IIT 3.0 (Oizumi, Albantakis, and Tononi, 2014, PLoS Computational Biology) formalized the theory rigorously, introducing the concepts of cause-effect structure (previously "qualia space"), intrinsic cause-effect power, and the exclusion postulate (consciousness exists at the level of organization that maximizes Φ, excluding both smaller and larger scales).
• IIT and the cerebral cortex: IIT predicts that recurrent thalamocortical networks (characterized by dense bidirectional connectivity and high differentiation) should have high Φ, while the cerebellum (which has ~4× more neurons than the cortex but modular, feed-forward architecture) should have low Φ — consistent with the clinical observation that cerebellar damage does not affect consciousness while even small cortical lesions can abolish it (Tononi and Koch, 2015, Philosophical Transactions of the Royal Society B).
• Perturbational Complexity Index (PCI): Casali et al. (2013, Science Translational Medicine) developed PCI — an empirical measure inspired by IIT that combines TMS (transcranial magnetic stimulation) with high-density EEG to measure the complexity and integration of cortical responses. PCI reliably distinguishes conscious from unconscious states (wakefulness, REM sleep vs. NREM sleep, general anesthesia, vegetative vs. minimally conscious patients) with >90% accuracy. While PCI does not directly measure Φ, it operationalizes IIT's core insight that consciousness requires both integration and differentiation.
• Computational intractability: computing Φ exactly requires evaluating all possible bipartitions of a system's elements — the number of bipartitions grows superexponentially (~$2^n$ for n elements). For systems of even modest size (>15–20 elements), exact computation is practically impossible. Tegmark (2016, PLOS Computational Biology) highlighted this issue and proposed approximations. This is a widely acknowledged limitation.

2. CREDIBLE CLAIMS (Tier 2 — Academic / Debated but Supported)

• IIT vs. Global Workspace Theory (GWT): the Templeton Foundation-funded Adversarial Collaboration (2019–2023, led by Liad Mudrik and Lucia Melloni) pitted IIT against GWT (Baars, 1988; Dehaene, Naccache, et al., 2001) in a pre-registered empirical test. Preliminary results (reported at ASSC 2023) did not definitively favor either theory but found evidence consistent with GWT's predictions about the neural correlates of conscious content — though IIT proponents argue the experimental designs tested specific predictions, not the full theory.
• IIT implies panpsychism: because any system with non-zero Φ has some degree of consciousness, IIT attributes minimal consciousness to simple physical systems (thermostats, photodiodes). Tononi embraces this as a feature, not a bug, arguing that panpsychism follows naturally from the axioms. Koch (2012, Consciousness: Confessions of a Romantic Reductionist) also endorses this implication. Critics consider it a reductio ad absurdum.
• IIT 4.0: Albantakis, Barbosa, et al. (2023, arXiv) presented IIT 4.0, refining the mathematical formalism with updated definitions of cause-effect structures and new unfolding techniques. The update maintains the core framework while addressing technical criticisms.
• Digital computers and Φ = 0: IIT predicts that standard von Neumann computers, regardless of the software they run (including human-level AI), have Φ = 0 because their gate-level architecture is feed-forward (decomposable into independent logic gates). This is perhaps IIT's most controversial prediction and fundamentally challenges functionalism.

3. SPECULATIVE CLAIMS (Tier 3 — Possible but Unverified)

• Whether Φ can be measured (even approximately) in biological neural networks at the relevant grain size is an open challenge.
• Whether IIT's axioms uniquely determine the postulates (or whether alternative postulates could satisfy the same axioms) is debated.

4. DUBIOUS CLAIMS (Tier 4 — No Credible Source / Contradicted by Evidence)

• Claims that IIT has been empirically "proven" or "disproven." The theory's core predictions (especially Φ = 0 for digital computers) are not currently testable. The adversarial collaboration tested specific correlates, not Φ itself.
• Claims that IIT "solves" the hard problem of consciousness. IIT reframes the hard problem by identifying consciousness with integrated information, but critics argue this is a definitional move, not an explanation (the question "why does integrated information feel like something?" remains open).

Counter-Arguments & Criticisms

Against IIT: Scott Aaronson (2014) showed that IIT assigns high Φ to certain trivially constructed systems (e.g., large expander graphs) that have no plausible claim to consciousness — the "Φ zombie" problem. Cerullo (2015) argued that IIT's axioms are not self-evident and that alternative axiom sets yield different theories. The computational intractability of Φ makes the theory untestable for systems of biological complexity.

For IIT: Proponents argue that IIT is the most mathematically precise theory of consciousness available, that it makes clear and falsifiable predictions (cerebellar vs. cortical consciousness, sleep/anesthesia, split-brain patients), and that PCI provides empirical traction. The panpsychistic implications, while counterintuitive, are shared by several respectable philosophical positions.

IMAGES

No images assigned yet.

BIBLIOGRAPHY

1. Tononi, Giulio | 2004 | "An Information Integration Theory of Consciousness" | BMC Neuroscience | ∅ | 5.42::1–22 | ∅ | ∅ | doi:10.1186/1471-2202-5-42 | ∅ | ∅ | ∅
2. Oizumi, Masafumi, Larissa Albantakis; Giulio Tononi. e1003588 | 2014 | "From the Phenomenology to the Mechanisms of Consciousness: Integrated Information Theory 3.0" | PLoS Computational Biology | ∅ | 10.5:: | ∅ | ∅ | doi:10.1371/journal.pcbi.1003588 | ∅ | ∅ | ∅
3. Tononi, Giulio; Christof Koch | 2015 | "Consciousness: Here, There and Everywhere?" | Philosophical Transactions of the Royal Society B | ∅ | 370.1668::20140167 | ∅ | ∅ | doi:10.1098/rstb.2014.0167 | ∅ | ∅ | ∅
4. Casali, Adenauer, Olivia Gosseries, Mario Rosanova, et al. ra105 | 2013 | "A Theoretically Based Index of Consciousness Independent of Sensory Processing and Behavior" | Science Translational Medicine | ∅ | 5.198::198 | ∅ | ∅ | doi:10.1126/scitranslmed.3006294 | ∅ | ∅ | ∅
5. Albantakis, Larissa, Leonardo Barbosa, Graham Findlay, et al. e1011465 | 2023 | "Integrated Information Theory (IIT) 4.0: Formulating the Properties of Phenomenal Existence in Physical Terms" | PLoS Computational Biology | ∅ | 19.10:: | ∅ | ∅ | doi:10.1371/journal.pcbi.1011465 | ∅ | ∅ | ∅
6. Koch, Christof | 2012 | ∅ | Consciousness: Confessions of a Romantic Reductionist | ∅ | ∅ | Cambridge: MIT Press | ∅ | isbn:9780262017497 | ∅ | ∅ | ∅
7. Tegmark, Max. e1005123 | 2016 | "Improved Measures of Integrated Information" | PLoS Computational Biology | ∅ | 12.11:: | ∅ | ∅ | doi:10.1371/journal.pcbi.1005123 | ∅ | ∅ | ∅
8. Tononi, Giulio, Melanie Boly, Marcello Massimini; Christof Koch | 2016 | "Integrated Information Theory: From Consciousness to Its Physical Substrate" | Nature Reviews Neuroscience | ∅ | 17.7::450–461 | ∅ | ∅ | doi:10.1038/nrn.2016.44 | ∅ | ∅ | ∅
9. Baars, Bernard | 1988 | ∅ | A Cognitive Theory of Consciousness | ∅ | ∅ | Cambridge: Cambridge University Press | ∅ | isbn:9780521427432 | ∅ | ∅ | ∅
10. Dehaene, Stanislas; Jean-Pierre Changeux | 2011 | "Experimental and Theoretical Approaches to Conscious Processing" | Neuron | ∅ | 70.2::200–227 | ∅ | ∅ | doi:10.1016/j.neuron.2011.03.018 | ∅ | ∅ | ∅
11. Aaronson, Scott | 2014 | "Why I Am Not an Integrated Information Theorist (or, The Unconscious Expander)" | Shtetl-Optimized Blog | ∅ | ∅ | May 21 | ∅ | ∅ | ∅ | ∅ | ∅
12. Cerullo, Michael. e1004286 | 2015 | "The Problem with Phi: A Critique of Integrated Information Theory" | PLoS Computational Biology | ∅ | 11.9:: | ∅ | ∅ | doi:10.1371/journal.pcbi.1004286 | ∅ | ∅ | ∅
13. Massimini, Marcello; Giulio Tononi | 2018 | ∅ | Sizing Up Consciousness: Towards an Objective Measure of the Capacity for Experience | ∅ | ∅ | Oxford: Oxford University Press | ∅ | isbn:9780198728443 | ∅ | ∅ | ∅
14. Chalmers, David | 2017 | "The Combination Problem for Panpsychism" | Panpsychism: Contemporary Perspectives | ∅ | ∅ | In edited by Godehard Brüntrup and Ludwig Jaskolla, 179 214 | ∅ | | ∅ | ∅ | Oxford: Oxford University Press

CROSS-REFERENCE INDEX

Generated from V4 expansion plan. Last Updated: April 2, 2026

Corrections

• Consciousness: Confessions of a Romantic Reductionist — ISBN corrected from 9780262017974 to 9780262017497, verified against Open Library (Consciousness, Christof Koch). The previous number failed its check digit.
• Sizing Up Consciousness: Towards an Objective Measure of the — ISBN corrected from 9780198728340 to 9780198728443, verified against Open Library (Sizing Up Consciousness, Marcello Massimini, Giulio Tononi). The previous number failed its check digit.
• Panpsychism: Contemporary Perspectives — invalid ISBN 9780199359685 removed. No verified replacement could be found, and supplying an unverified number would be worse than none. The entry's author, title, publisher and year are unchanged.