Giulio Tononi introduces a theory linking consciousness to a system’s capacity to integrate information. The proposal aims to explain both how much consciousness a system has (quantity) and what kind (quality) using a measure called Φ (phi). It also frames consciousness as tied to a system’s internal causal organization, not just its outward behavior.
Researchers led by Adenauer Casali and Marcello Massimini (with Tononi and others) introduce the Perturbational Complexity Index (PCI), using TMS-EEG to measure the complexity of the brain’s response to a controlled perturbation. PCI is motivated by IIT-like ideas: conscious states should show responses that are both integrated and differentiated. Although PCI is not the same as computing Φ directly, it becomes an important practical tool in clinical assessment of consciousness.
A major update, often called IIT 3.0, reformulates the theory around five phenomenological axioms (features of experience) and corresponding physical postulates (requirements for a system’s causal structure). This version emphasizes “cause–effect power” inside a system and adds a more explicit method for identifying the system’s main conscious “complex” (the subset with maximal irreducibility). The shift helps make IIT more systematic, but also increases its mathematical and interpretive complexity.
Computer scientist Scott Aaronson publishes a widely discussed critique arguing that, under some readings, IIT can assign very high Φ to systems that seem intuitively non-conscious (such as certain networks of logic gates). The critique becomes influential because it highlights how sensitive IIT’s conclusions can be to modeling choices and formal details. This exchange helps push IIT supporters to clarify assumptions and strengthens the role of “toy models” in the debate.
Tononi and Christof Koch publish a high-profile review presenting IIT as a theory that starts from the structure of experience and derives constraints on physical systems. The paper argues that feedforward systems (no feedback loops) are not conscious, while certain simple recurrent systems could have minimal consciousness. It also stresses the theory’s controversial implications for machines and simulations, helping bring IIT into broader philosophy-of-mind and AI discussions.
Tononi, Boly, Massimini, and Koch summarize IIT for a neuroscience audience and connect it to evidence about brain networks and levels of consciousness. The article also emphasizes that IIT makes strong claims about the “physical substrate of consciousness,” especially the idea that consciousness corresponds to a maximum of intrinsic cause–effect power. This consolidates IIT as a major contender in consciousness science while acknowledging its counterintuitive predictions.
Barrett and Mediano argue that, in some formulations, IIT’s Φ measure is not well-defined for general physical systems and that multiple mathematically consistent choices can satisfy the axioms while giving different Φ values. This critique focuses less on intuitions and more on whether IIT has a unique, stable mathematical core. It becomes part of a broader push for clearer definitions, better invariances, and more consistent measurement standards.
Doerig, Schurger, Hess, and Herzog publish what becomes known as the Unfolding Argument, claiming that a recurrent system’s input–output behavior can be replicated by an unfolded feedforward system. If IIT says feedforward systems are not conscious, then matching behavior does not settle the consciousness question, which critics argue makes IIT hard to test in the usual scientific way. Supporters respond that IIT is not intended to be behavior-based, sharpening a central methodological dispute.
Merker, Williford, and Rudrauf publish a detailed critique claiming IIT’s Φ (as used in practice) tracks properties like global information transfer or network efficacy rather than consciousness itself. They frame this as a problem of construct validity: even if Φ is measurable, it may not measure what IIT says it measures. The debate highlights the gap between formal definitions and how measures are interpreted in real neuroscientific contexts.
Albantakis and collaborators (including Tononi) present IIT 4.0, updating how IIT translates axioms into mathematics and introducing a revised measure of intrinsic information. The paper also makes causal relations more explicit in the formalism, aiming to reduce ambiguity and better match IIT’s “intrinsic perspective” goals. IIT 4.0 becomes the reference point for many newer technical discussions, including attempts to make computation more tractable.
A group of scholars releases a public letter arguing that IIT has received outsized attention despite what they see as limited empirical support and unclear testability. The letter’s call to label IIT “pseudoscience” triggers strong pushback from other researchers, intensifying disputes about what counts as evidence in consciousness science. The episode becomes a turning point in how the field debates theory evaluation and scientific norms.
A large, preregistered, multi-lab “adversarial collaboration” tests competing predictions of Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT) using common experimental methods and agreed-upon analyses. The published result is not a clean win for either side, and parts of each theory’s preregistered expectations face challenges. As a closing outcome, the study helps shift the debate toward clearer predictions, stronger preregistration, and theory refinement rather than reliance on broad claims.
Integrated Information Theory (Giulio Tononi) and critical responses (2004–present)