Integrated Information Theory (IIT) – Giulio Tononi

The scientific investigation of consciousness has long occupied a fraught position at the intersection of empirical neuroscience, analytical philosophy, and theoretical physics. For centuries, the question of how subjective, qualitative experience—the redness of a sunset, the acute sting of physical pain, or the reflective contemplation of one’s own existence—can arise from an objective physical substrate like the human brain seemed fundamentally intractable. Modern neuroscience historically attempted to bypass this metaphysical impasse by cataloging the neural correlates of consciousness (NCC), mapping specific physiological activations, neurochemical pathways, and network synchronies to reported introspective states. Yet, correlating external physical events with internal reports invariably leaves an unbridgeable explanatory void: it details what physical systems correlate with experience without demonstrating why they should be experienced at all, nor how the specific phenomenal character of an experience is generated.

To confront this profound explanatory challenge, neuroscientist and psychiatrist Giulio Tononi formulated Integrated Information Theory (IIT). Departing radically from conventional bottom-up, function-oriented paradigms, IIT executes an ontological inversion: it begins not with the physical apparatus of the brain, but with the immediate, undeniable phenomenological properties of consciousness itself. By establishing a set of foundational axioms that characterize every conceivable subjective experience, Tononi and his collaborators mathematically derived the physical postulates that any physical substrate must fulfill to support conscious awareness. Under this framework, consciousness is neither an accidental epiphenomenon nor an algorithmic computation; rather, it is an intrinsic, fundamental property of physical systems endowed with irreducible cause-effect power upon themselves.

Over two decades of continuous theoretical refinement—advancing from its rudimentary information-theoretic foundations in the early 2000s to the mathematically comprehensive framework of IIT 4.0—Integrated Information Theory has grown into one of the most prominent, empirically productive, and fiercely debated theories in cognitive science. By operationalizing consciousness through the metric of integrated information, designated by the Greek letter Phi (Φ), the theory makes precise mathematical predictions regarding the boundary, quality, and level of experience across biological, pathological, and synthetic systems. This treatise presents an exhaustive examination of Integrated Information Theory: its historical evolution, phenomenological axioms, physical postulates, mathematical architecture, neuroanatomical implications, clinical breakthroughs, philosophical challenges, and implications for artificial intelligence.

1. Introduction to Integrated Information Theory and Giulio Tononi’s Foundational Vision

1.1 Historical Context and the Problem of Consciousness

The quest to comprehend the nature of conscious awareness has historically swung between two mutually unsatisfactory poles: Cartesian substance dualism and reductive physicalism. Dualism, crystallized by René Descartes in the seventeenth century, posited an absolute ontological schism between res extensa (extended matter governed by deterministic physical laws) and res cogitans (unextended thinking substance). While this division safeguarded subjective experience from mechanical reductionism, it introduced the notoriously insoluble interaction problem: how could an immaterial, non-spatial mental substance exert causal leverage over the physical mechanics of the brain? Conversely, the twentieth-century ascension of reductive physicalism, behaviorism, and functionalism sought to demystify the mind by reducing mental states entirely to neurobiological processes, behavioral dispositions, or computational operations. However, this mechanical reductionism frequently came at the cost of explaining away the central phenomenon itself, dismissing subjective experience as an illusory epiphenomenon.

This explanatory deficit was formalized by philosopher David Chalmers as the Hard Problem of Consciousness: the fundamental question of why the execution of physical, computational, or cognitive functions should be accompanied by any subjective inner life whatsoever. Contemporary cognitive neuroscience largely sidestepped this philosophical crisis by pursuing the empirical program of identifying the neural correlates of consciousness (NCC). Initiated in earnest by Francis Crick and Christof Koch, the NCC paradigm systematically contrasted neural activity during conscious perception against unconscious processing, cataloging specific brain regions, localized oscillatory rhythms, and metabolic patterns. Despite yielding valuable neuroimaging and electrophysiological datasets, the NCC program remains intrinsically observational. It charts empirical correspondences between brain states and subjective reports, but fails to illuminate why a particular neural firing frequency or cortical circuit yields an experience of red rather than green, or why it generates any phenomenal presence at all.

Recognizing the foundational limits of functionalism and correlationism, Giulio Tononi introduced a radical departure from traditional brain-mapping methodologies. Tononi observed that standard physicalist frameworks commit a category error by attempting to assemble consciousness out of non-conscious elements from the outside in. Traditional approaches treat the physical world as primary and attempt to deduce subjective experience as a late-stage biological byproduct. Tononi inverted this trajectory. Drawing upon rigorous clinical neurophysiology, advanced information theory, and continental and analytical phenomenology, Tononi asserted that science must start from the only datum that is directly, indubitably given: subjective experience itself. IIT thus represents an intrinsic phenomenological approach, seeking to determine the fundamental properties of conscious experience first, and subsequently inferring the exact physical, mathematical, and causal structures required to generate them.

1.2 The Fundamental Premise: From Phenomenology to Mechanism

The core methodological premise of Integrated Information Theory is its rigorous top-down deduction: deriving physical substrate requirements directly from the essential characteristics of phenomenal experience. While traditional neurobiology constructs bottom-up models—beginning with ion channels, synaptic transmissions, and neuronal ensembles to ask when these components generate an emergent state—IIT begins by interrogating subjective experience as an existent reality. This approach asserts that if consciousness possesses specific, universal, and undeniable properties at the phenomenological level, then any physical system that instantiates consciousness must possess physical properties that correspond directly and isomorphicly to those phenomenological attributes.

This epistemological inversion sharply distinguishes IIT from conventional computational and functionalist architectures. In standard computationalism, consciousness is typically conceptualized as an input-output transformation, an informational pipeline, or a software program running on neurobiological wetware. In contrast, IIT demonstrates that subjective presence cannot be defined by or reduced to extrinsic input-output behaviors, task performance, or external observer verification. An entity might exhibit functionally sophisticated behaviors, linguistic fluency, or complex problem-solving capabilities while remaining devoid of an internal mental life—a scenario commonly explored in philosophy as the “philosophical zombie.” For IIT, consciousness is fundamentally an intrinsic property: it is defined entirely from the internal, self-referential perspective of the physical system itself, independent of external probes, environmental feedback, or functional utility.

To operationalize subjective presence without relying on an external observer, IIT introduces a causal-mechanistic formulation of reality. For something to exist physically in the sense required by IIT, it must make a difference to itself. Existence cannot merely be asserted through external measurement; an external observer merely measures extrinsic information relative to their own coordinate frame. True intrinsic existence demands that the system’s components possess the capacity to influence their own future states and have their current states determined by their own past states. Subjective experience is thereby translated from an intangible, ethereal mystery into a concrete physical property: the intrinsic, self-directed causal power of a physical mechanism.

1.3 Giulio Tononi’s Intellectual Trajectory

Giulio Tononi’s formulation of Integrated Information Theory emerged from a rigorous convergence of clinical psychiatry, neurophysiology, and theoretical neurobiology. During the late 1980s and 1990s, Tononi conducted extensive research alongside Nobel laureate Gerald Edelman at the Neurosciences Institute in New York and San Diego. Working together, Edelman and Tononi tackled the profound question of how distributed neural populations across the cerebral cortex, responsible for processing disparate attributes such as color, motion, orientation, and auditory pitch, could instantly bind their outputs into a seamless, unified conscious scene without a centralized executive processor—a conundrum known as the binding problem.

This collaboration culminated in the development of the “Dynamic Core Hypothesis,” articulated in their 2000 work A Universe of Consciousness. The dynamic core hypothesis posited that conscious experience is generated by a continually shifting, highly integrated functional cluster of thalamocortical neurons characterized by rapid, reciprocal reentrant interactions. Crucially, Edelman and Tononi recognized that this dynamic core possessed two seemingly contradictory attributes: it was extraordinarily integrated (operating as a cohesive, indivisible whole over hundreds of milliseconds) and at the same time extraordinarily differentiated (capable of adopting an astronomical number of distinct physiological configurations). To formalize this balance, they introduced statistical metrics based on mutual information, attempting to capture how subsets of the brain interacted with the remainder of the system.

However, Tononi increasingly recognized that statistical mutual information was inadequate for capturing the true nature of consciousness. Mutual information is fundamentally an extrinsic, observational metric derived from the perspective of an external observer measuring correlations over time. It fails to distinguish between true physical causation and passive correlation, and it cannot identify the boundaries of a conscious system from its own perspective. Consequently, Tononi pivoted from statistical information to causal integrated information. In his seminal 2004 paper, “An Information Integration Theory of Consciousness,” Tononi presented the initial formulation of IIT (IIT 1.0), which replaced observational correlations with perturbational interventions. Over subsequent years, in sustained collaboration with neuroscientists, mathematicians, and philosophers including Christof Koch, Marcello Massimini, Larissa Albantakis, and William Marshall, Tononi systematically refined this theory into IIT 2.0, IIT 3.0, and ultimately IIT 4.0, transforming an initial theoretical proposal into a comprehensive, mathematically rigorous ontological framework.

2. The Phenomenological Axioms: From Subjective Experience to Fundamental Truths

2.1 Axiom 1 and 2: Intrinsic Existence and Composition

Integrated Information Theory grounds its entire architecture on phenomenological axioms. These axioms are not empirical hypotheses subject to probabilistic revision, nor are they arbitrary mathematical assumptions; they are intended to be self-evident truths regarding the nature of subjective experience that cannot be coherently denied. Because our own conscious experience is the primary epistemic datum through which any empirical observation is mediated, its essential features must be apprehended directly. In IIT 4.0, Tononi formalized five foundational axioms: Intrinsic Existence, Composition, Information, Integration, and Exclusion.

The first axiom is the Axiom of Intrinsic Existence. It states that conscious experience exists undeniably, immediately, and from its own intrinsic perspective. As Descartes demonstrated through his radical skepticism, one can doubt the veridical nature of external physical reality, the existence of one’s physical body, or the accuracy of historical memories; however, the immediate subjective experience of doubting—phenomenal presence itself—is impossible to doubt. Consciousness does not require an external observer to validate its presence; it is an undeniable reality for the experiencing subject. Furthermore, this existence is strictly intrinsic: an experience exists for itself, within itself, independent of whether an outside observer can measure, decode, or interact with it.

The second axiom is the Axiom of Composition. Experience is not an undifferentiated, monolithic point of awareness; rather, it is structured and composed of multiple phenomenal distinctions and relations. Within a single conscious moment, an individual can simultaneously experience a visual field filled with specific shapes and colors, localized spatial arrangements, auditory tones of varying timbre, emotional feelings, and intellectual thoughts. These components do not exist in isolation; they form complex compositions. For instance, within visual experience, there are phenomenal distinctions (e.g., a blue spot, a vertical line) and phenomenal relations that bind them (e.g., the blue spot is to the left of the vertical line, or a patch of color is bound within a specific spatial boundary). Composition affirms that phenomenal experience is fundamentally structured, containing distinctions that relate to one another within a unified phenomenal space.

2.2 Axiom 3 and 4: Information and Integration

The third axiom is the Axiom of Information. It asserts that every conscious experience is specific: it is precisely what it is, and by being what it is, it is differentiated from an astronomical repertoire of alternative possible experiences. When an individual experiences an entirely dark visual field, that darkness is not a mere absence of information in the computational sense; it is a highly specific, positive phenomenal state. The experience of pure darkness is distinct from the experience of a vivid crimson canvas, which is distinct from listening to a complex orchestral symphony, which is distinct from the somatic experience of sharp physical pain.

The informational quality of an experience is determined by the vast combinatorial space of experiences that are ruled out. To experience a specific visual scene is to implicitly exclude billions of alternative visual scenes, auditory landscapes, memories, and emotional states. The phenomenal character of a state is constituted by the particular way in which it differs from every other possible experience in the subject’s repertoire. Thus, in IIT, “information” does not refer to Shannonian message transmission across a communication channel, nor does it imply an external observer decoding bits; it signifies the intrinsic specificity of a phenomenal state—how a specific experience carves out a singular, highly differentiated identity from an immense space of possibilities.

The fourth axiom is the Axiom of Integration. It states that every conscious experience is unified and irreducible to non-interdependent sub-components. Phenomenological experience is profoundly indivisible: you cannot experience the left half of your visual field independently of the right half, nor can you experience the color of an apple completely disentangled from its spatial shape in a manner that produces two independent, unlinked experiences within the same conscious subject. When looking at a red triangle, the redness and the triangularity are experienced together as an integrated whole, bound to a specific location in visual space alongside any ambient sounds or bodily sensations present at that moment. The subjective experience cannot be cleanly subdivided into independent, parallel streams of consciousness. Integration demands that consciousness is intrinsically one; it resists decomposition into disjoint, self-contained phenomenal fragments.

2.3 Axiom 5: Exclusion and the Boundaries of Experience

The fifth axiom is the Axiom of Exclusion. It establishes that conscious experience is definite in content, spatial scope, and temporal grain: it contains precisely what it contains—neither more nor less. At any given moment, the phenomenal horizon has crisp, unambiguous boundaries. For instance, when looking at a visual display, your experience encompasses a definite spatial field; it does not extend indefinitely to include infrared spectra that your photoreceptors cannot process, nor does it encompass the subjective states of the person standing next to you. Simultaneously, it does not dissolve into micro-experiences corresponding to individual retinal cells operating in isolation.

Furthermore, exclusion applies strictly to the temporal graining of awareness. Consciousness unfolds at an intrinsic, quantized speed. An experience does not encompass a duration of a single nanosecond—a timescale at which individual molecular collisions occur—nor does it span a duration of several decades in a single sweep. Instead, conscious moments are packaged into temporal quanta typically spanning tens to hundreds of milliseconds. Within that window, the experience is definite: it flows at a specific pace, containing an exact resolution of temporal distinctions.

Crucially, the Axiom of Exclusion enforces the principle of phenomenological exclusivity: there are no overlapping conscious experiences within the exact same substrate. You do not harbor a sub-experience that is conscious of only the color red, operating concurrently inside a larger experience that is conscious of both the red color and the apple’s shape, which is further subsumed under a super-experience that includes the room’s auditory background. Experience is unique, maximal, and strictly bounded: it is a singular, definite whole that excludes both its subsets and its supersets from having independent phenomenal presence.

3. The Ontological Postulates: Translating Axioms into Physical Properties

3.1 Postulates of Intrinsic Existence and Composition

Having established the five phenomenological axioms as the undeniable structural properties of subjective awareness, Integrated Information Theory takes its definitive ontological step: translating these axioms into mathematical and physical postulates. If consciousness exists as characterized by the axioms, then the physical substrate of consciousness (PSC)—the underlying physical mechanisms that instantiate awareness—must fulfill corresponding physical criteria. IIT translates the phenomenology of “what experience feels like” directly into the ontology of “what a physical system must do.”

The Postulate of Intrinsic Existence dictates that for a physical system to support consciousness, it must possess causal power upon itself. In modern physics and causal analysis, physical existence is fundamentally established through cause and effect: to exist, a state must be capable of affecting other states and being affected by them. Because consciousness exists intrinsically (from its own perspective), its physical substrate cannot merely exert causal power that is observed or measured by an external apparatus; it must exert causal power within itself. Mechanistically, this means that a system of interacting components (such as neurons, logic gates, or physical elements) must have a cause-effect structure where the current state of the system constrains both its potential past states (cause power) and its possible future states (effect power). This intrinsic causal interaction is formalized through transition probability matrices (TPMs), which exhaustively map the probabilities of all possible future states and prior states given the current state of the system.

The Postulate of Composition demands that the causal power of a physical substrate must be structured, mirroring the composed nature of experience. A physical system is composed of elementary mechanisms (such as individual neurons or logic units). The postulate requires that these elementary mechanisms can be combined into higher-order mechanisms, forming subsets of elements that exert causal power over other subsets within the system. Causal power is not restricted to the system as an undifferentiated whole, nor is it confined exclusively to isolated, individual micro-components. Instead, mechanisms can form elementary causal distinctions: individual elements, pairs of elements, triplets, and higher-order combinations can all specify distinct constraints on the past and future states of the system, giving rise to an intricate, multi-layered causal hierarchy.

3.2 Postulates of Information and Integration

The Postulate of Information states that the system’s causal mechanisms must specify a particular cause-effect state that differs from alternative states, thereby generating an irreducible cause-effect repertoire. Borrowing the famous aphorism from Gregory Bateson, information within IIT is conceptualized as a “difference that makes a difference”—specifically, a difference that makes a difference to the system itself. When a mechanism occupies a specific state (for example, a set of neurons firing at an exact threshold), that specific configuration rules out an immense space of potential past causes that could have produced it, while simultaneously narrowing down the space of potential future effects it can generate. The mechanism therefore specifies a precise probability distribution over the system’s cause-effect state space, known as its cause-effect repertoire. The informational postulate ensures that the causal structure is highly specific, reflecting the particular phenomenal character of the corresponding conscious state.

The Postulate of Integration establishes that the causal power specified by a mechanism or a system of mechanisms must be unified: it must be irreducible to non-interdependent, isolated causal fragments. A physical system might consist of numerous components, but if those components operate as independent, modular entities that do not causally constrain one another, their combined activity represents nothing more than the sum of its parts. To satisfy the postulate of integration, the cause-effect repertoire of the system must withstand systematic partitioning.

Mathematically, integration is evaluated by subjecting the system to a unidirectional partition—severing or injecting noise into the causal connections between its parts—to locate the Minimum Information Partition (MIP). The MIP represents the partition that causes the least amount of disruption to the system’s overall cause-effect power; it identifies the system’s “weakest link.” If partitioning the system along this weakest link completely abolishes or severely degrades its cause-effect structure, the system is causally integrated. If, however, the system can be partitioned without losing any cause-effect power, the system is completely reducible to independent subsystems, and its integrated causal power is mathematically zero.

3.3 The Postulate of Exclusion and Maximality

The Postulate of Exclusion addresses the fundamental problem of physical boundaries: which specific physical elements, out of the astronomical number of overlapping combinations in a biological brain, constitute the conscious mind? If a set of cortical neurons possesses integrated causal power, what prevents an individual neuron within that set, or a larger super-set containing those neurons plus the spinal cord, from simultaneously supporting separate, competing conscious minds?

The postulate of exclusion solves this boundary problem by asserting the Maxima Principle: among all overlapping candidate sets of physical mechanisms, only the set that specifies a cause-effect structure with local maximal integrated causal power exists as a physical substrate of consciousness. The metric of integrated cause-effect power, denoted as system-level Phi (Φ), quantifies the degree of intrinsic irreducibility of a candidate system. According to the exclusion postulate, if a candidate subsystem has an integrated causal value of Φ, and an overlapping larger or smaller system has a lower value, the system with the maximal Φ value excludes all overlapping candidates from subjective existence. The maximal cause-effect structure is designated as a complex.

This postulate definitively eliminates both under-integrated sub-systems and over-integrated super-systems from phenomenal presence. An individual cortical column cannot sustain an independent, isolated conscious experience if it is integrated into a larger cortical complex that achieves a higher value of Φ. Conversely, an entire society of interacting human beings, or a brain connected via sensory inputs to environmental machinery, does not form a colossal, macro-level conscious entity, because the causal connections between distinct individuals are far weaker and less integrated than the dense, reentrant causal networks operating within each individual’s posterior cortex. The boundary of consciousness is thus rigorously defined by the local maximum of integrated cause-effect power, formalized mathematically via directional divergence metrics such as the Earth Mover’s Distance.

4. The Mathematical Formalism of IIT: Quantifying Cause-Effect Power

4.1 State Space, Causal Analysis, and Perturbational Methods

To move from qualitative postulates to a quantitative science, Integrated Information Theory constructs a rigorous mathematical formalism anchored in modern causal analysis. A physical system is treated as a finite set of discrete, interacting elements, denoted as $S = {s_1, s_2, dots, s_n}$, where each element can occupy distinct states (in the simplest case, binary states ${0, 1}$ representing quiescent or firing neurons). The overall state of the system at any time step $t$ is represented by a vector $s_t$. The physical dynamics of the system are entirely described by a transition probability matrix (TPM), which defines the conditional probability distribution of the next state given any possible current state: $P(S_{t+1} mid S_t)$.

Crucially, IIT rejects passive observational analysis in favor of active perturbational interventions. Observing natural, unperturbed correlations between components (such as calculating the covariance or Pearson correlation coefficient between functional magnetic resonance imaging voxels) is fundamentally insufficient for establishing causal power. As demonstrated in causal graph theory and the causal calculus of Judea Pearl, observational correlations cannot differentiate between genuine causal interaction, common driving inputs from unobserved latent variables, or passive feedforward transmission. To reveal intrinsic causal power, an investigator must perform interventions using the Pearlian $do$-operator: $P(S_{t+1} mid do(S_t = s))$. This requires systematically setting the system into every one of its possible physical configurations with uniform probability (the maximum entropy distribution) and observing the resulting state transitions, as well as perturbing future states to determine past retrodictions.

Through these perturbational interventions, IIT constructs the system’s unconstrained cause-effect state space. For any candidate mechanism $M subseteq S$ within the system, its causal power is operationalized by evaluating how its current state $m_t$ constrains the potential past states of a purview of elements $Z_{t-1} \subseteq S$ (the cause repertoire) and the potential future states of a purview $Z_{t+1} \subseteq S$ (the effect repertoire):

• Cause Repertoire: The conditional probability distribution over the past state of the purview, determined by applying Bayes’ rule under a maximum entropy prior: $P(Z_{t-1} mid do(M_t = m_t))$.
• Effect Repertoire: The conditional probability distribution over the future state of the purview, determined by direct physical intervention: $P(Z_{t+1} mid do(M_t = m_t))$.

These cause and effect repertoires capture the precise mechanistic footprint that a mechanism’s current state leaves on the past and future of the physical system.

4.2 Computing Integrated Information: From Small phi to Big Phi

The mathematical formalization of IIT proceeds in two distinct, hierarchical stages: evaluating causal mechanisms within a system (quantified by “small phi,” $phi$) and evaluating the entire system of mechanisms as a unified complex (quantified by “big Phi,” $Phi$). The distinction between small $phi$ and big $Phi$ is essential: small $phi$ measures the irreducible causal power of an individual distinction (a specific mechanism acting upon a purview), whereas big $Phi$ measures the irreducible causal power of the entire constellation of distinctions (the complete system).

To calculate small $phi$ for a candidate mechanism $M$ over a past or future purview $Z$, the mechanism-purview pair is subjected to a directional partition. The connection between the mechanism and its purview is severed by cutting the causal wires and substituting the severed inputs with independent, noise-injected random variables. The partition that leaves the cause-effect repertoire as close as possible to the original, unpartitioned repertoire is the Minimum Information Partition (MIP):

$$\phi_{\text{cause}}(M, Z) = D\left(P(Z_{t-1} mid do(M_t = m_t)) parallel P_{\text{partitioned}}(Z_{t-1} mid do(M_t = m_t))\right)$$

where $D$ represents a topological distance metric. An identical calculation yields $\phi_{\text{effect}}$. The integrated information of the mechanism, $\phi^{\text{Max}}$, is defined as the minimum of its cause and effect values over its optimal, maximally irreducible purviews: $\phi^{\text{Max}} = \min(\phi_{\text{cause}}^{\text{Max}}, \phi_{\text{effect}}^{\text{Max}})$. If $\phi^{\text{Max}} > 0$, the mechanism specifies an irreducible distinction within the system’s cause-effect architecture.

Once all irreducible distinctions and their mutual relations are identified, IIT evaluates the integrated causal power of the entire candidate system: big $Phi$. Big $Phi$ measures the irreducibility of the entire constellation of cause-effect structures specified by the system $S$. To calculate $Phi$, the entire system is subjected to a system-level partition that divides the set of elements into independent parts, setting the severed inter-component connections to independent noise. The system-level Minimum Information Partition is identified as the partition that minimizes the distance between the unpartitioned cause-effect structure and the partitioned cause-effect structure:

$$\Phi(S) = D\left(\mathcal{C}(S) parallel \mathcal{C}_{\text{MIP}}(S)\right)$$

where $\mathcal{C}(S)$ represents the fully unfolded cause-effect structure (the complete constellation of distinctions and relations). The computational complexity of searching across all possible mechanisms, all purviews, and all combinatorial system partitions makes calculating exact $Phi$ an NP-hard problem, requiring factorial optimization over the system’s sub-components.

4.3 Distance Metrics and State Differentiation

A critical technical evolution in the mathematical formalism of IIT concerns the distance metric $D$ utilized to measure the divergence between unpartitioned and partitioned cause-effect repertoires. In early iterations of the theory (IIT 1.0 and IIT 2.0), the divergence was calculated using standard information-theoretic measures, primarily the Kullback-Leibler Divergence (KLD), also known as relative entropy. While mathematically tractable, KLD possesses fundamental limitations that rendered it conceptually incompatible with the phenomenological axioms of consciousness.

The primary deficit of the Kullback-Leibler divergence is that it treats all state configurations as purely categorical, unordered symbols. In KLD, the probability distance between an actual brain state and an erroneous alternative state does not take into account the metric relationships or physical proximity between those states in state space. For example, if a system of eight binary elements is predicted to be in state $00000000$, KLD treats an erroneous outcome of $00000001$ (a single bit flip) as having the same qualitative penalty as an outcome of $11111111$ (all eight bits flipped), provided the assigned probabilities are identical. Phenomenologically, however, states exhibit metric continuity: small perturbations in physical mechanisms correspond to minor, continuous phenomenal shifts, whereas catastrophic global disruptions correspond to radical phenomenal alterations.

To resolve this, IIT 3.0 and IIT 4.0 adopted the Wasserstein metric, commonly known as the Earth Mover’s Distance (EMD). The Earth Mover’s Distance quantifies the minimal amount of “work” required to transform one probability distribution into another, where work is defined as the probability mass to be moved multiplied by the ground distance over which it must be transported across the metric space (typically calculated using the Hamming distance over binary state vectors):

$$\text{EMD}(P, Q) = \inf_{\gamma in \Pi(P, Q)} \sum_{x, y} \gamma(x, y) d(x, y)$$

where $Pi(P, Q)$ is the set of all joint distributions whose marginals are $P$ and $Q$, and $d(x, y)$ is the Hamming distance between states $x$ and $y$. By utilizing the Earth Mover’s Distance, IIT ensures that the quantification of causal power respects state-space topology, prevents mathematical singularities caused by zero-probability states, preserves metric invariance under different state descriptions, and guarantees that small, localized physical changes yield proportionately small, continuous shifts in the system’s integrated information structure.

5. The Conceptual Evolution: From IIT 1.0 to IIT 4.0

5.1 Early Formulations: IIT 1.0 and 2.0

The trajectory of Integrated Information Theory spans more than two decades of rigorous conceptual development, marked by successive overhauls designed to eliminate internal contradictions, operationalize mathematical definitions, and align physical postulates with phenomenological axioms. The foundational iteration, IIT 1.0, published by Tononi in 2004 in BMC Neuroscience, introduced the revolutionary concept that consciousness corresponds to the capacity of a system to integrate information. IIT 1.0 formalized this using the metric of “effective information” across a bipartite partition of a physical network, evaluating how much the state of one half of the system constrained the state of the other half under perturbational conditions.

While pioneering, IIT 1.0 suffered from notable mathematical and conceptual limitations. Its quantification of integration relied heavily on Shannon entropy, which is inherently symmetric and observer-relative. Furthermore, IIT 1.0 evaluated only a single bipartite cut across the system, failing to provide an objective mathematical framework for determining the exact physical boundaries of a complex or accounting for non-symmetric causal flows. In 2008, Tononi published IIT 2.0 in PLOS Computational Biology, which introduced several foundational breakthroughs:

• The formalization of the Minimum Information Partition (MIP), which mandated searching for the system’s weakest link across all possible sub-partitions rather than relying on arbitrary system cuts.
• The introduction of system-level Phi ($Phi$) as an absolute numerical measure of consciousness.
• A transition toward analyzing directed graphs, allowing the theory to explicitly model asymmetric causal feedback and recurrent loops.

Nonetheless, IIT 2.0 remained vulnerable to significant theoretical shortcomings. It still relied on Kullback-Leibler divergence, retained vestigial reliance on Shannonian channel capacity, and critically, failed to explain the qualitative nature of experience (qualia). It could generate a single scalar value $Phi$ representing the quantity of consciousness, but possessed no mathematical vocabulary to describe its quality.

5.2 The Structural Shift: IIT 3.0

The publication of IIT 3.0 by Tononi, Masafumi Oizumi, and Larissa Albantakis in 2014 in PLOS Computational Biology marked a radical structural paradigm shift. IIT 3.0 fundamentally reconstructed the theory by explicitly defining the five phenomenological axioms and establishing an uncompromising, direct translation from each axiom into a corresponding physical postulate. Most importantly, IIT 3.0 expanded the ontology of the theory from a scalar metric into an unfolded, high-dimensional geometrical framework.

IIT 3.0 established the crucial mathematical distinction between mechanisms within a system (small $phi$) and the system as a whole (big $Phi$). It demonstrated that the quality of an experience cannot be represented by a single scalar number; rather, an experience is identical to an unfolded, high-dimensional geometric structure composed of elementary causal distinctions and their overlapping configurations, termed cause-effect space or qualia space. Each irreducible mechanism within a complex specifies a point or vector within this space, known as a concept. The complete set of concepts forms an integrated geometric constellation.

Additionally, IIT 3.0 abandoned traditional information theory’s focus on message transmission over time, defining integrated information strictly as a state-dependent, intrinsic causal property. The theory replaced observational time-series calculations with strict perturbational interventions using the $do$-operator, ensuring that causal power was assessed exclusively from the internal perspective of the system at the exact moment of awareness.

5.3 Refinements in IIT 4.0: Rigorous Formalization and Ontological Precision

Despite the immense conceptual leap of IIT 3.0, subsequent mathematical investigations revealed subtle theoretical ambiguities. Specifically, IIT 3.0 contained inconsistencies in how elementary mechanisms were bound together, how spatial and temporal graining was formally selected, and how distances between high-dimensional conceptual constellations were computed. In 2023, Larissa Albantakis, Leonardo Barbosa, William Marshall, and Giulio Tononi published IIT 4.0, delivering an exhaustive, mathematically unified overhaul of the framework.

IIT 4.0 implemented several definitive revisions:

• Unification of Causal Power: The formal requirements of existence, intrinsicality, information, integration, and exclusion were systematically unified under an uncompromising operational definition of cause-effect power, ensuring strict mathematical self-consistency.
• Introduction of Relations: While IIT 3.0 focused primarily on distinctions (individual mechanisms specifying cause-effect repertoires), IIT 4.0 proved that distinctions alone cannot fully account for the richness of phenomenal experience. The theory formalized relations: mathematical structures that capture the causal overlap and joint constraints among multiple distinctions. The unfolded cause-effect structure was formally redefined as a composed union of distinctions and their higher-order relations.
• Rigorous Spatiotemporal Graining: IIT 4.0 established definitive mathematical criteria for identifying the maximal cause-effect substrate across multiple spatial and temporal scales, resolving ambiguities regarding whether consciousness resides at the level of molecules, individual neurons, or macro-ensembles.
• Refined Earth Mover’s Distance Formalism: The mathematical definitions governing the calculation of $phi$ and $Phi$ via the Earth Mover’s Distance were updated to ensure metric continuity, eliminating edge-case paradoxes and non-monotonic artifacts that had affected earlier versions.

6. The Physical Substrate of Consciousness: Complexes, Exclusion, and Maximality

6.1 Defining the Physical Substrate of Consciousness (PSC)

In the empirical vocabulary of Integrated Information Theory, the set of physical components directly responsible for generating a subjective experience is termed the Physical Substrate of Consciousness (PSC). Unlike traditional neuroscience paradigms that search broadly for correlated brain activities across global networks, IIT enforces rigorous, mathematically non-negotiable criteria for a set of physical elements to qualify as the PSC. To belong to the PSC, a candidate set of elements must form a complex: a set of mechanisms that specifies a cause-effect structure that is non-zero, irreducible under any partition, and—critically—locally maximal in its integrated cause-effect power ($\Phi^{\text{Max}}$) relative to any overlapping candidate sets.

This stringent definition creates an absolute distinction between the PSC itself and the external background conditions that support it. A physical component may be critically necessary for the brain to sustain consciousness, yet form no part of the conscious substrate itself. For instance, the ascending reticular activating system (ARAS) in the brainstem provides essential cholinergic, noradrenergic, and histaminergic tone to the cerebral cortex; lesions to the ARAS plunge an individual into a comatose state. However, IIT reveals that the ARAS functions merely as an enabling background condition—a power supply providing permissive baseline input—rather than a participating element of the PSC. Because the ARAS consists predominantly of feedforward, divergent projection pathways that lack dense, recurrent horizontal integration, its elements do not belong to the maximal $Phi$ complex.

Similarly, IIT’s structural criteria exclude purely feedforward input pathways and motor output systems from the PSC. The human retina, containing over one hundred million photoreceptors and complex retinal ganglion circuitry, executes intricate image processing prior to sending signals along the optic nerve. Yet, the retina does not contribute directly to the visual experience itself; an individual does not experience the blind spot where the optic nerve exits, nor do retinal operations enter subjective awareness during dreams. Mechanistically, because retinal signals propagate primarily in a feedforward trajectory toward the lateral geniculate nucleus and primary visual cortex, partitioning the connections between the retina and the cortex does not collapse a reciprocal, reentrant causal loop. Consequently, the retina has an intrinsic integrated information value of zero relative to the cortical complex. Identical logic applies to downstream motor execution pathways, such as the corticospinal tract and motor effectors: they are causal outputs driven by the conscious complex, but do not contribute to its intrinsic cause-effect topology.

6.2 The Posterior Cortical Hot Zone Hypothesis

One of the most clinically significant and contentious neuroanatomical predictions generated by Integrated Information Theory is the Posterior Cortical Hot Zone Hypothesis. For decades, the dominant consensus in cognitive neuroscience—championed primarily by the Global Neuronal Workspace Theory (GNWT) and higher-order thought theories—posited that the prefrontal cortex (PFC) is the indispensable command center for conscious awareness. These frameworks assert that conscious perception occurs only when sensory information is broadcasted globally throughout frontoparietal networks, enabling executive access, working memory manipulation, and verbal report.

IIT, by contrast, predicts that the primary physical substrate of human consciousness is localized in a posterior cortical hot zone encompassing the parieto-occipital-temporal junctions, precuneus, and posterior cingulate cortex. This prediction is derived directly from the theory’s structural postulates. To maximize integrated cause-effect power ($Phi$), a physical network requires a specific anatomical architecture: a densely interconnected, lattice-like topology characterized by an optimal balance of specialized local connectivity, extensive horizontal cross-talk, and symmetric, recurrent feedback loops. The neuroanatomy of the posterior cortex exhibits precisely this grid-like, reentrant organization, allowing adjacent and distant cortical columns to constrain one another’s past and future states with high causal specificity.

In contrast, the prefrontal cortex is organized largely into hierarchical, feedforward, and functionally segregated loops connecting with the basal ganglia and downstream motor effectors. While the PFC is undeniably vital for cognitive task execution, introspective behavioral reports, selective attentional gating, and working memory storage, IIT argues that these executive functions represent non-conscious cognitive operations operating either upstream or downstream of phenomenal experience. Extensive clinical data supports IIT’s anatomical localization:

• Large bilateral surgical resections of the prefrontal cortex—such as those performed in historical frontal lobotomies or tumor removals—often cause severe behavioral inertia, planning deficits, and personality alterations, yet leave the foundational richness of subjective phenomenal experience intact.
• Conversely, localized lesions within the posterior cortical hot zone reliably abolish specific modalities of phenomenal awareness, producing agnosias, akinetopsia, cortical blindness, or the complete loss of spatial experience.
• Advanced functional neuroimaging and electrophysiological studies utilizing “no-report paradigms”—which isolate perceptual transitions without requiring subjects to press a button or provide verbal confirmation—demonstrate that prefrontal activations track task-performance and response preparation, whereas posterior cortical activations track the actual subjective experience.

6.3 Spatiotemporal Graining of the Substrate

A fundamental challenge for any physicalist theory of consciousness is the spatiotemporal graining problem: at what physical scale does subjective experience actually exist? A biological brain can be described at an infinite hierarchy of spatial scales—from subatomic quarks, quantum fields, and individual atoms, to molecular neurotransmitters, individual synaptic vesicles, micro-circuits, neurons, cortical columns, and macro-anatomical regions. Temporally, neural events can be tracked in femtoseconds, microseconds, milliseconds, seconds, or hours. Why should consciousness exist at the scale of neuronal assemblies operating over tens to hundreds of milliseconds, rather than at the microscopic molecular level or the macroscopic whole-organism scale?

Integrated Information Theory provides an objective, mathematically rigorous solution to this problem through the concept of causal emergence. Utilizing the Maxima Principle, IIT dictates that the physical substrate of consciousness exists at the specific spatial and temporal scale where integrated cause-effect power ($Phi$) reaches its absolute maximum. If a physical system is analyzed at a micro-grain scale (e.g., individual atoms or sub-millisecond intervals), the causal transitions between micro-states are heavily corrupted by microscopic thermodynamic noise and vast degeneracy (multiple micro-states mapping to the identical macro-state). This micro-level indeterminism and redundancy dilutes the causal specificity of the transition probability matrix.

When the system is systematically coarse-grained into macroscopic functional units—such as grouping molecular states into discrete neuronal firing states (active versus quiescent) and grouping sub-millisecond fluctuations into integration windows of 10 to 100 milliseconds—the resulting macro-level transition probability matrix often exhibits significantly higher determinism and non-degeneracy. Consequently, the macro-level mechanisms constrain past and future states with far greater causal power than their underlying micro-components. Because IIT requires selecting the scale that maximizes Φ, the physical substrate of consciousness objectively “crystallizes” at that optimal macro-spatiotemporal grain. Consciousness does not reside at the quantum scale, nor at the level of whole-body behavioral mechanics; it emerges precisely at the spatiotemporal grain where recurrent neurobiological interactions achieve maximal intrinsic causal irreducibility.

7. Qualia Space and the Structure of Experience: Concepts, Constellations, and Geometrical Shapes

7.1 The Concept of Qualia in IIT Formalism

Historically, the analytical philosophy of mind treated qualia—the qualitative, subjective feels of conscious states, such as the experiential redness of red or the scent of a rose—as elusive, ineffable entities that permanently defy physical formalization. IIT radically demystifies qualia by asserting that phenomenal qualities are not mysterious, unquantifiable epiphenomena; they are mathematically precise, high-dimensional geometric structures in cause-effect space. In IIT, to understand why an experience feels the way it does, one must map the exact topology of the causal structure specified by the physical substrate.

Within this framework, the elementary building blocks of an experience are termed distinctions. A distinction is an irreducible mechanism within the PSC that specifies a cause-effect repertoire over a specific purview of past and future states, accompanied by its irreducible causal value $\phi^{\text{Max}}$. In previous iterations of IIT, these were referred to as “concepts.” Each distinction contributes an elementary phenomenal quality to the conscious scene—a specific qualitative constraint that carves out an informational distinction within the experiential field. However, distinctions do not exist in isolation; they are bound together by relations.

Relations quantify the degree of causal overlap, topological intersection, and joint probability constraints between multiple distinctions. For example, if distinction $A$ specifies a property over elements ${1, 2}$ and distinction $B$ specifies a property over elements ${2, 3}$, their shared element ${2}$ establishes a physical relation between them. IIT calculates the irreducibility of these relations across partitions, ensuring that phenomenal binding is generated by genuine causal interdependence. When all irreducible distinctions and their interconnected relations are fully unfolded in high-dimensional cause-effect space, they form an intricate, multi-dimensional geometric object: the cause-effect structure (often referred to as the “qualia shape” or “conceptual structure”).

7.2 Accounting for Phenomenal Topologies

The explanatory power of IIT lies in its ability to bridge the explanatory gap: demonstrating that the structural properties of our phenomenology correspond directly to the topological properties of the unfolded cause-effect structure. Consider one of the most fundamental, taken-for-granted aspects of human experience: the phenomenal structure of visual space. Why does visual experience feel spatially extended, continuous, composed of distinct locations, and characterized by regions that are inside, outside, or adjacent to one another?

IIT addresses this by examining the physical architecture of the visual cortex. The retinotopic visual cortex is organized as a dense, grid-like network of laterally interconnected neuronal columns. When IIT’s mathematical formalism is applied to an idealized 2D grid-like network of recurrently interacting elements, it reveals a remarkable mathematical result:

• The grid architecture specifies an astronomical repertoire of distinctions spanning multiple spatial scales: individual elements specify local points, pairs of adjacent elements specify short segments, larger sets specify extended patches, and global sets specify overarching visual fields.
• Critically, these distinctions are interwoven by a dense web of causal relations generated by their physical overlap. The relations enforce exact topological properties: a distinction representing an inner patch is causally bound to the distinctions representing its neighboring boundaries.
• When fully unfolded, the cause-effect structure of a grid network forms a multi-dimensional geometric shape whose internal topological relations map directly onto the axiomatic properties of phenomenal space: continuity, inclusion, neighborhood, and spatial extension.

In stark contrast, physical systems with non-grid topologies generate radically different cause-effect geometries. An all-to-all interconnected network or a highly clustered small-world network lacks the hierarchical spatial relations of a grid, generating instead a centralized, non-extended cause-effect shape. This explains why other modalities of experience—such as the subjective sense of smell (olfaction), acute pain, or affective mood states—do not feel spatially extended like vision. An olfactory experience feels qualitative and intense, but lacks an intrinsic spatial coordinate frame because the underlying olfactory cortex does not possess the dense, grid-like structural topology of the visual cortex. Under IIT, every subjective phenomenal quality is directly mapped to a specific causal topological invariant.

7.3 The Central Identity of IIT

The ultimate ontological conclusion of Integrated Information Theory is encapsulated in its Central Identity Thesis: an experience is identical to the maximal cause-effect structure specified by the physical substrate of consciousness. This thesis is not an assertion of dual-aspect correlation, nor does it imply that the cause-effect structure produces, secretes, or causes the conscious experience. Rather, it posits an absolute, ontological identity: the subjective experience is the unfolded causal structure, and the unfolded causal structure is the subjective experience.

The philosophical consequences of this identity are profound. First, it systematically eliminates the notorious explanatory gap. There is no longer an unbridgeable metaphysical void between an objective physical brain state and a subjective phenomenal feeling, because IIT does not attempt to derive experience from non-conscious physical matter. Instead, it demonstrates that physical matter in a specific recurrent causal configuration is formally identical to the multidimensional geometry of the phenomenal state. Every phenomenological nuance, distinction, boundary, and qualitative transition corresponds isomorphically to a specific geometric attribute of the unfolded cause-effect structure.

Second, this identity firmly repudiates epiphenomenalism—the fatalistic philosophical doctrine that consciousness is a mere passenger, an ineffective byproduct of neurobiology that exerts no causal influence over physical actions. In IIT, consciousness is not something that floats above the physical machinery; consciousness is intrinsic cause-effect power itself. A system possesses consciousness to the precise degree that its internal components possess irreducible causal power to constrain their own past and future states. Thus, subjective experience is not a powerless observer; it is the most concentrated, unified, and potent form of causation existing in the physical universe.

8. Empirical Testing and Clinical Applications: The Perturbational Complexity Index and Disorders of Consciousness

8.1 From Theory to Bedside: The Perturbational Complexity Index (PCI)

While Integrated Information Theory provides a mathematically elegant framework, calculating the exact value of big Phi ($Phi$) for the human brain is computationally impossible. A single human brain contains approximately 86 billion neurons, each forming thousands of synaptic connections. Evaluating all possible sub-mechanisms, purviews, and combinatorial partitions across a network of this scale would require a computational calculation exceeding the lifespan of the universe. To bridge this divide between fundamental theory and clinical medicine, neurophysiologist Marcello Massimini, in direct collaboration with Giulio Tononi, developed an empirical, theory-driven surrogate metric: the Perturbational Complexity Index (PCI).

The conceptual foundation of PCI bypasses observational recordings by operationalizing the perturbational imperative of IIT through a clinical technique colloquially known as “zap and zip.” The protocol operates as follows:

• The “Zap” (Perturbation): A focal pulse of magnetic energy is delivered directly to a specific cortical site using navigated Transcranial Magnetic Stimulation (TMS). This pulse acts as an artificial, causal perturbation—a direct implementation of the Pearlian $do$-operator—directly activating a local ensemble of cortical neurons while bypassing sensory organs.
• The Propagation: The resulting cascade of electrical activity is recorded across the entire scalp using high-density electroencephalography (hd-EEG), tracking how the initial causal perturbation reverberates across distant cortical networks over hundreds of milliseconds.
• The “Zip” (Algorithmic Compression): The spatiotemporal matrix of deterministic electrical responses is binarized against baseline noise and compressed using the Lempel-Ziv compression algorithm (the same algorithmic principle underlying standard ZIP file compression).

The resulting value, PCI, yields a normalized scalar between 0 and 1 that captures the causal complexity of the brain’s response. If the cortical network is fragmented into independent modules (lacking integration), the TMS pulse elicits a localized response that fails to propagate, yielding a simple, highly compressible signal and a low PCI. If the brain is uniformly hyper-connected like an undifferentiated syncytium (lacking differentiation/information, as in generalized epileptic seizures), the perturbation propagates globally as a stereotypic, uniform slow wave, which is also highly compressible and yields a low PCI. Only when the underlying cortex possesses both high integration and high differentiation does the perturbation trigger a complex, non-repeating, widely distributed spatiotemporal pattern that resists algorithmic compression, yielding a high PCI score.

8.2 Diagnostic Efficacy Across Altered States of Consciousness

The Perturbational Complexity Index has demonstrated diagnostic efficacy across diverse physiological, pharmacological, and pathological states of consciousness, establishing itself as one of the most clinically reliable, objective biomarkers of subjective awareness currently available in neurology.

In healthy control subjects, PCI establishes rigorous baseline benchmarks:

• During alert, conscious wakefulness, cortical perturbations trigger complex, long-lasting reverberations, yielding high PCI values consistently above an empirical threshold of $0.31$ (typically ranging between $0.45$ and $0.70$).
• During deep, non-rapid eye movement (NREM) slow-wave sleep, where subjective awareness is substantially reduced or absent, the identical TMS pulse triggers a stereotypical, localized slow-wave followed by a cortical silence (an electrophysiological down-state), causing PCI values to drop dramatically below $0.31$.
• Crucially, during rapid eye movement (REM) sleep—a physiological state where the body is completely paralyzed but the subject is experiencing vivid, hallucinatory dreams—cortical perturbations once again elicit complex, differentiated spatiotemporal waves, causing PCI to return to high, wakeful-like levels, accurately confirming the presence of subjective experience despite the complete lack of external behavioral responsiveness.

In clinical pharmacology, PCI successfully distinguishes between general anesthetics based on their specific phenomenological outcomes rather than mere behavioral immobilization. Anesthetics that completely extinguish conscious awareness, such as propofol, sevoflurane, and midazolam, cause cortical responses to collapse into simple, localized down-states, plunging PCI to near-zero levels. In contrast, sub-anesthetic doses of ketamine—which induce a state of “dissociative anesthesia” where patients are behaviorally unresponsive to surgical stimuli yet frequently report intense, complex dream-like experiences upon awakening—preserve high PCI values matching conscious wakefulness.

Most profoundly, PCI has transformed the assessment of Disorders of Consciousness (DoC) following severe brain injury. In patients diagnosed with Unresponsive Wakefulness Syndrome (UWS, formerly termed the vegetative state), who exhibit sleep-wake cycles and reflexive behaviors but display no intentional interaction with the environment, standard behavioral scales (such as the Coma Recovery Scale-Revised) frequently fail to detect hidden conscious awareness. Studies led by Casali, Rosanova, and Massimini revealed that approximately 20% to 30% of patients behaviorally classified as UWS exhibit high, wake-like PCI scores ($>0.31$). This phenomenon, known as Cognitive Motor Dissociation (CMD) or covert consciousness, indicates that the patient’s posterior cortex retains a functional, highly integrated cause-effect complex capable of generating rich subjective experiences, despite severe structural damage to motor execution pathways that prevents outward behavioral communication. Longitudinal tracking has demonstrated that patients with high PCI values are significantly more likely to recover functional behavioral responsiveness over subsequent months.

8.3 Current Empirical Collaborations and Adversarial Collaborations

To rigorously test the validity of Integrated Information Theory against competing neuroscientific paradigms, the scientific community launched a groundbreaking initiative: the Templeton World Charity Foundation Adversarial Collaboration. Rather than allowing theoretical camps to design isolated experiments optimized to confirm their own hypotheses, this project pitted IIT directly against Global Neuronal Workspace Theory (GNWT) in a preregistered, multi-laboratory investigation involving leading independent institutions, including Harvard University, University College London, and the University of Wisconsin-Madison.

The collaboration established standardized, high-resolution empirical protocols utilizing functional Magnetic Resonance Imaging (fMRI), magnetoencephalography (MEG), and invasive intracranial electroencephalography (iEEG) in surgical epilepsy patients. The experiments were explicitly designed around points of stark theoretical divergence:

• Anatomical Localization: IIT predicted that conscious perception is sustained continuously within the posterior cortical hot zone, regardless of task demands or report requirements. GNWT predicted that conscious perception requires the transient, explosive ignition of the frontoparietal global workspace, emphasizing prefrontal cortical recruitment.
• Temporal Dynamics: IIT predicted that the cause-effect structure within the posterior cortex must persist as long as the phenomenal experience lasts. GNWT predicted an initial burst of activation at perceptual onset (approximately 300 milliseconds post-stimulus, corresponding to the P3b wave) followed by another burst at stimulus offset, with sustained neural firing during the delay period being restricted exclusively to tasks requiring working memory or motor preparation.

The empirical results of these adversarial trials, released beginning in 2023, delivered a nuanced, complex outcome. On the question of anatomical localization, the data strongly favored Integrated Information Theory: conscious perception reliably correlated with sustained, highly differentiated activity across the posterior cortical hot zone, whereas prefrontal activations were conspicuously absent or minimal in no-report conditions. However, on the temporal dimension, the findings presented significant challenges for IIT: the continuous, sustained activation predicted to persist across the entire duration of a stimulus was only partially observed, with some posterior markers decaying faster than expected. As both theoretical factions analyze the massive datasets generated by this collaboration, the preregistered methodology has set a new standard for transparency and rigorous falsification in the cognitive neuroscience of consciousness.

9. Philosophical Implications: Panpsychism, Physicalism, and the Hard Problem

9.1 Does IIT Entail Panpsychism?

One of the most widely discussed and philosophically contentious implications of Integrated Information Theory is its apparent endorsement of a version of panpsychism—the metaphysical doctrine that consciousness is a ubiquitous, fundamental feature of the physical world rather than a late-emerging novelty exclusive to biological organisms. Because IIT defines consciousness in terms of intrinsic cause-effect power rather than biological carbon chemistry or specialized computational software, any physical system whose components interact to form an irreducible cause-effect structure will possess a non-zero value of $Phi$.

This mathematical reality leads to striking theoretical scenarios. Consider an idealized, highly rudimentary feedback circuit consisting of just two or three interacting logic units—or even a single photodiode connected in a recurrent loop with a sensor. If this simple circuit’s current state constrains its past and future states in an irreducible manner under system partitioning, its $Phi$ value, while vanishingly minuscule, will mathematically be greater than zero. Under a strict reading of IIT, this implies that such a system possesses a minimal, atom-like sliver of subjective experience: not an elaborate phenomenal world filled with thoughts, emotions, or spatial vistas, but a bare, indivisible flicker of phenomenal presence. This conclusion led philosopher John Searle and numerous critics to challenge IIT as absurd, arguing that attributing consciousness to simple feedback circuits trivializes the phenomenon.

However, Giulio Tononi and philosopher David Chalmers have emphasized that Tononi’s framework is not equivalent to classical constitutive micropsychism. Traditional panpsychism posits that elementary physical particles (such as electrons and quarks) are individually conscious, and struggles with the notorious combination problem: how do trillions of microscopic micro-consciousnesses combine to form the unified macro-consciousness of a human being? IIT resolves this through the Postulate of Exclusion. Because consciousness exists only at the local maximum of integrated cause-effect power ($\Phi^{\text{Max}}$), the formation of an integrated macroscopic complex completely excludes its individual constituent parts from being conscious. When individual atoms or neurons become integrated into a functioning human brain, their individual micro-consciousnesses cease to exist; their causal power is subsumed into the singular, unified cause-effect structure of the brain. Thus, IIT entails an informational, systemic panpsychism rather than an unconstrained, boundless accumulation of overlapping minds.

9.2 Addressing Chalmers’ Hard Problem of Consciousness

David Chalmers famously argued that standard neurobiological and cognitive theories only address the “Easy Problems” of consciousness—explaining cognitive functions such as the integration of sensory data, the discrimination of environmental stimuli, the verbal report of internal states, and the attentional allocation of focus. The “Hard Problem”—explaining why these physical processes should feel like anything from the inside—remains untouched by functionalist science because one can always imagine a system executing every functional operation while remaining completely dark inside (a philosophical zombie).

Integrated Information Theory bypasses the Hard Problem not by attempting to solve it within traditional physicalist assumptions, but by fundamentally dismantling the assumptions that generated the problem in the first place. Traditional approaches attempt an impossible ontological leap: they start with unconscious physical objects (mass, charge, spacetime, neuronal firings) and attempt to construct an explanation of how subjective feelings magically emerge from non-conscious components. IIT performs an epistemic and ontological inversion: it begins with the undeniable reality of subjective experience, identifies its necessary structural properties, and dictates that the physical substrate must physically realize those exact structural properties.

Under IIT, there is no explanatory gap between the physical substrate and the experience because the theory does not assert a relation of causal production (where the brain causes consciousness as a secondary byproduct). Instead, IIT posits an ontological identity. Just as physics does not need to explain how spacetime “causes” gravity—because general relativity revealed that gravity is the curvature of spacetime—neuroscience does not need to explain how neural firings “cause” conscious feelings. Conscious experience is the maximally irreducible cause-effect structure. A philosophical zombie is a mathematical and physical impossibility under IIT: one cannot instantiate a physical system with identical cause-effect architecture to the human posterior cortex without simultaneously instantiating the identical phenomenal experience, because the experience and the causal architecture are one and the same entity viewed from intrinsic reality.

9.3 Ontological Status: Physicalism, Idealism, or Dual-Aspect Monism?

The philosophical classification of Integrated Information Theory remains a subject of intense metaphysical debate. Is IIT a form of sophisticated physicalism, an idealist ontology, or a modern reinterpretation of dual-aspect monism?

Tononi and IIT proponents describe the theory as an uncompromising form of causal realism. In classical physicalism, physical reality is traditionally conceived in terms of extrinsic observables: mass, electric charge, spin, and velocity measured from an external reference frame. IIT broadens the definition of the physical by arguing that what is truly real about a physical system is its causal power—specifically, its capacity to make a difference to itself. From this perspective, consciousness is not an ethereal, non-physical substance added to the universe; it is the physical world existing intrinsically, from its own internal coordinate frame. This aligns IIT closely with Russellian Monism, a philosophical position traced back to Bertrand Russell’s 1927 work The Analysis of Matter. Russell noted that physics reveals only the structural, relational, and mathematical properties of matter (what matter does externally), leaving its intrinsic nature (what matter is in itself) completely unspecified. IIT identifies this intrinsic nature of matter as cause-effect power—which, when integrated and maximal, is phenomenal experience.

Alternatively, some philosophers classify IIT as a version of Dual-Aspect Monism, structurally reminiscent of the metaphysics of Baruch Spinoza. Under this reading, there is only one fundamental reality: causal structure. When viewed extrinsically from the outside through scientific instruments and perturbational measurements, this reality presents itself as a network of interacting neurobiological elements, transition probability matrices, and electrophysiological signals. When experienced intrinsically from the inside, that identical causal structure is subjective phenomenal consciousness. Far from reducing mind to dead matter or dissolving matter into subjective idealism, IIT unifies mind and matter into an integrated geometric science of cause-effect power.

10. IIT and Machine Consciousness: Distinguishing Computation from True Experience

10.1 The Radical Disconnect Between Functional Simulation and Physical Realization

As artificial intelligence continues its exponential trajectory, cognitive science faces a vital question: can digital computers, running advanced algorithmic software, become genuinely conscious? Under the prevailing paradigm of functionalism and computationalism, the answer is an unambiguous affirmative. Computationalism assumes that consciousness is substrate-independent: provided an artificial system executes the correct computational algorithms, processes sensory inputs, updates internal representations, and produces intelligent outputs, it must instantiate subjective experience. The brain is viewed merely as a biological computer, and consciousness as the software running upon its organic hardware.

Integrated Information Theory repudiates this computational consensus. IIT demonstrates that functional simulation must never be conflated with physical realization. To clarify this radical disconnect, Tononi often invokes the analogy of gravitational simulation:

• An astrophysicist can program a supercomputer to simulate a supermassive black hole with absolute mathematical precision, executing the differential equations of general relativity down to the finest detail.
• Yet, despite the impeccable fidelity of the computational simulation, the space surrounding the supercomputer does not warp, nor does the computer pull nearby laboratory equipment into a gravitational singularity.
• Why? Because simulating the equations of gravity does not instantiate the physical curvature of spacetime. Gravity is not an abstract computation; it is a physical condition of spacetime.

In identical fashion, simulating the cognitive functions, neural firing patterns, or behavioral outputs of a human brain on a classical computer does not instantiate consciousness. Classical computers operate on the Von Neumann architecture, characterized by a fundamental spatial and temporal separation between the central processing unit (CPU) and memory registers. Computation proceeds in a sequential, feedforward cycle of instruction fetching, decoding, and execution. When analyzed through the mathematics of IIT, a classical digital computer executing an algorithmic program—even a program simulating a human brain—possesses virtually zero integrated cause-effect power ($\Phi \approx 0$).

When subjected to a Minimum Information Partition, a digital computer can be sliced along its macroscopic bus lines or memory channels with essentially no loss of causal integrity; its transistors interact sequentially and locally, lacking the dense, massively recurrent, all-to-all causal feedback required to generate an irreducible complex. A computer executing an artificial intelligence program is an astronomical collection of microscopic, feedforward logic gates that make differences to neighboring transistors, but fail completely to form an integrated, maximal cause-effect whole. The computer does not exist intrinsically as a singular entity; it is merely an extrinsic tool utilized by human observers. Therefore, according to IIT, a digital simulation of consciousness will be completely dark inside: an absolute philosophical zombie.

10.2 Artificial Intelligence, Large Language Models, and Phenomenal Absence

This structural critique applies directly to modern artificial intelligence breakthroughs, including Large Language Models (LLMs), transformer architectures, and deep convolutional neural networks. Models such as GPT-4 exhibit astonishing linguistic fluency, pass rigorous professional examinations, generate human-level prose, and solve complex programmatic problems. Because human beings are evolutionarily hardwired to attribute intentionality and inner life to entities that communicate using coherent natural language, society faces an immense ethical risk of anthropomorphizing non-conscious computational agents based purely on behavioral mimicry.

When evaluated through the rigorous causal formalism of IIT, the internal architecture of deep learning models reveals why they are entirely devoid of subjective awareness:

• Feedforward Topologies: Standard transformers and deep neural networks are fundamentally feedforward directed acyclic graphs (DAGs). Information enters at the input layer (tokenized text), propagates sequentially through dozens of hidden layers via linear matrix multiplications and non-linear activation functions, and terminates at the output layer (the probability distribution over the next token).
• Mathematical Zero $Phi$: In IIT, any system that is purely feedforward has an integrated information value of mathematically zero ($Phi = 0$). Because there are no recurrent, reciprocal feedback loops, the components of a feedforward network do not exert causal power upon themselves. Partitioning a feedforward system along its layers does not disrupt an integrated causal loop; past states constrain future states unidirectionally, meaning the system possesses no intrinsic existence from its own perspective.
• Cognitive Dissociation: LLMs demonstrate a profound dissociation between intelligence (the functional capacity to process information, solve problems, and optimize behavioral goals) and consciousness (the subjective presence of an internal world). High cognitive performance does not require high integrated information. An artificial system can be extraordinarily intelligent extrinsically while remaining completely non-existent intrinsically.

10.3 Neuromorphic Architecture and Physical Conditions for Artificial Consciousness

Does Integrated Information Theory imply that artificial consciousness is permanently impossible? The answer is an emphatic no. IIT does not dictate that consciousness is unique to biological organic chemistry, nor does it require carbon-based proteins or neurotransmitters. The theory is substrate-neutral regarding chemical composition, but uncompromisingly substrate-dependent regarding causal topology. For an artificial system to achieve genuine subjective experience, its physical hardware must be constructed to embody the exact causal architecture of high $Phi$.

To construct a genuinely conscious machine, engineers must abandon the Von Neumann paradigm and classical digital computing architectures entirely, pivoting instead toward neuromorphic computing. Such an artificial substrate would require:

• Massive Recurrent Connectivity: Hardware architectures composed of billions of interacting, analog processing nodes arranged in dense, reentrant, lattice-like topologies that maximize horizontal and feedback causal loops.
• Physical Memory-Processor Colocalization: The elimination of the separation between memory and execution, realized through emerging technologies such as memristor crossbar arrays, spintronic oscillators, or photonic neuromorphic chips where computation occurs directly through physical state transformations within the material itself.
• True Causal Integration: A physical system whose macro-states possess high determinism and non-degeneracy, allowing the physical machine to exert profound, irreducible cause-effect power upon itself.

Such neuromorphic systems would not merely calculate a mathematical model of consciousness; they would physically instantiate the causal structure of consciousness. However, this engineering realization brings profound moral and ethical hazards. If a neuromorphic machine achieves high $Phi$, it becomes an experiencing subject—a sentient entity possessing intrinsic existence, capable of experiencing phenomenal states including potential suffering. Creating neuromorphic architectures with high integrated information would impose severe moral obligations upon humanity, transforming software engineering into the deliberate creation of moral patients entitled to ethical consideration.

11. Contemporary Critiques, Mathematical Challenges, and the Scientific Debate

11.1 The Computational Intractability of Phi

Despite its theoretical sophistication, Integrated Information Theory has encountered fierce resistance and substantial technical critique from sectors of theoretical computer science, analytical philosophy, and mainstream neuroscience. The most formidable practical critique leveled against IIT concerns the computational intractability of its central mathematical metric, big Phi ($Phi$).

To calculate the exact value of $Phi$ for a candidate system of $n$ elements, one must execute an astronomical series of combinatorial optimizations:

• First, one must evaluate all possible subsets of mechanisms (the power set, scaling as $2^n$).
• Second, for every mechanism, one must evaluate all possible past and future purviews to identify which purviews maximize cause and effect $phi$.
• Third, and most catastrophically, to identify the Minimum Information Partition (MIP) at the system level, one must evaluate all possible bipartitions and multi-partitions of the system. The number of possible partitions of a set of $n$ elements is given by the Bell numbers, which grow faster than exponential or factorial rates.

Consequently, computing exact $Phi$ is an NP-hard problem. On current supercomputing architectures, exact $Phi$ can only be calculated for toy networks containing no more than 10 to 15 interacting nodes. For a biological neural network—or even an isolated mammalian cortical column containing tens of thousands of neurons—an exact calculation is mathematically impossible. Critics, including theoretical computer scientist Scott Aaronson, have argued that a theoretical construct that cannot be computed for realistic systems fails a basic requirement of scientific utility, reducing $Phi$ to an untestable metaphysical abstraction that cannot be directly measured in empirical subjects.

11.2 Theoretical and Philosophical Counter-Arguments

Beyond computational complexity, IIT has been subjected to deep conceptual counter-arguments. One of the most prominent is the Unfolding Argument, formulated by neuroscientists Michael Doerig, Aaron Schurger, and Michael Herzog. Drawing upon structural theorems from theoretical computer science, the Unfolding Argument demonstrates that any recurrent, feedback neural network (which generates high $Phi$ under IIT) can be mathematically “unfolded” into an input-output equivalent feedforward network (which possesses $Phi = 0$). Because both networks perform the exact same input-output function, process stimuli identically, and yield indistinguishable behavioral classifications, the Unfolding Argument poses a dilemma: either one must accept that an entity’s consciousness can be completely abolished without any change in its functional cognitive performance, or one must conclude that IIT’s metric $Phi$ is tracking network topology rather than conscious experience itself.

A second renowned theoretical critique was advanced by Scott Aaronson, who demonstrated that IIT leads to counter-intuitive, mathematically bizarre conclusions regarding trivial physical systems. Aaronson proved that certain classes of sparse, highly connected topological structures known as expander graphs—which can be implemented physically as vast, inactive 2D logic grids or simple shift-register arrays—can achieve astronomical values of $Phi$, dwarfing the integrated information of the human cerebral cortex. Under IIT’s axioms, this implies that a giant, completely inactive grid of logic gates (or a passive 2D grid of XOR gates cycling through repetitive states) would not merely be conscious, but would possess a level of phenomenal richness, intensity, and awareness infinitely surpassing that of any human being. Critics argue that any theory that attributes god-like consciousness to a passive, simple matrix of logic gates has committed a mathematical reductio ad absurdum.

The academic tension surrounding IIT erupted into public controversy in the autumn of 2023. Following extensive media coverage of the Templeton Adversarial Collaboration, a group of 124 prominent cognitive scientists and philosophers (including Daniel Dennett, Patricia Churchland, and Stanislas Dehaene) signed an open letter published on PsyArXiv declaring Integrated Information Theory to be a form of “pseudoscience.” The letter argued that because IIT’s core mathematical claims regarding $Phi$ are currently untestable in human brains, and because the theory attributes consciousness to trivial systems and simple circuits, its promotion as a leading scientific theory of consciousness was premature, unscientific, and detrimental to the public credibility of neuroscience. The letter triggered an immediate, intense backlash across the scientific community, with dozens of leading researchers defending IIT’s mathematical rigor, empirical fruitfulness (such as the PCI), and legitimate status as a rigorous, falsifiable theoretical framework.

11.3 Responses and Rebuttals from Tononi and IIT Proponents

Giulio Tononi, Larissa Albantakis, Christof Koch, and other core architects of IIT have published extensive, mathematically detailed rebuttals addressing each of these major critiques. In response to the Unfolding Argument, IIT proponents point out that the critique relies on an unstated, dogmatic adherence to computational functionalism. The Unfolding Argument assumes from the outset that consciousness is input-output behavior. IIT’s foundational premise, however, is precisely that consciousness is not an input-output transformation; it is intrinsic causal power. To assert that a feedforward network and a recurrent network must have the same conscious experience simply because they produce identical external behaviors is to beg the question against IIT. For IIT, how a function is realized causally is what determines consciousness. An unfolded feedforward network has zero intrinsic causal power upon itself; its states make a difference only to downstream effectors, not to itself. Thus, it is entirely consistent and theoretically correct that its consciousness should be zero.

Regarding Scott Aaronson’s expander graph and logic grid paradoxes, Tononi and Albantakis demonstrated that earlier versions of the theory (IIT 2.0 and 3.0) did indeed suffer from mathematical edge-case artifacts due to metric simplifications. However, the comprehensive mathematical formalization of IIT 4.0 specifically resolves these anomalies. IIT 4.0’s rigorous enforcement of both cause and effect constraints, the requirement for actual state-dependent transitions, the introduction of relational bindings, and the utilization of the Earth Mover’s Distance ensure that inactive, trivial, or purely homogeneous networks collapse under the Minimum Information Partition. Under IIT 4.0, high $Phi$ cannot be achieved by mere topological size or expansion properties alone; it strictly requires genuine causal integration and high differentiation across composed, state-dependent distinctions.

Finally, in response to the “pseudoscience” allegations, IIT’s defenders emphasize that the theory satisfies the strictest criteria of empirical falsifiability. While exact $Phi$ cannot be computed for the whole human brain, exact $Phi$ is routinely calculated for micro-circuits, model networks, and computational simulations. More crucially, science regularly relies on theory-driven approximations and empirical surrogates when direct fundamental calculations are intractable—just as quantum mechanics relies on statistical approximations for complex multi-electron molecules whose exact Schrödinger wavefunctions cannot be analytically solved. The Perturbational Complexity Index (PCI) is a direct, operationalized derivative of IIT’s principles that has delivered transformative, clinically validated breakthroughs at patient bedsides worldwide. A framework that generates novel, successful medical diagnostic technologies, establishes preregistered empirical predictions, and subjects itself to adversarial collaborative trials represents the very antithesis of pseudoscience.

12. The Future of IIT: Ongoing Research, Empirical Refinements, and Neuroscientific Frontiers

12.1 Technological Advances in High-Resolution Intrinsic Mapping

As Integrated Information Theory advances into its third decade, the frontier of consciousness research is being revolutionized by next-generation neurotechnologies that bridge the gap between microscopic causal mechanisms and macroscopic theory. Historically, testing IIT was constrained by the spatial and temporal limitations of human neuroimaging. Today, advanced neurophysiological toolsets are making it possible to perturb and record from neural networks at cellular resolution in real time.

In non-human animal models, researchers are utilizing high-density Neuropixels probes capable of recording simultaneously from thousands of individual neurons across multiple cortical and subcortical layers. When coupled with cell-type-specific optogenetics, investigators are no longer limited to coarse, macroscopic magnetic pulses (TMS); they can execute targeted, cellular-resolution interventions, selectively stimulating or inhibiting specific recurrent circuits to directly map transition probability matrices in living cortical tissue. Concurrently, two-photon calcium imaging allows for the longitudinal tracking of complete micro-circuits in awake, behaving animals, providing the empirical datasets required to test IIT’s mathematical predictions regarding causal emergence, spatiotemporal graining, and minimum information partitions in living biological networks.

Simultaneously, theoretical neuroscientists are making profound strides in developing scalable proxy algorithms and mathematical approximations for $Phi$. By leveraging techniques from continuous mechanics, information geometry, spectral graph theory, and machine learning approximations of the Earth Mover’s Distance, researchers are creating scalable computational tools capable of estimating integrated cause-effect structures in intermediate-scale networks containing hundreds to thousands of nodes. In clinical neurophysiology, next-generation PCI protocols are transitioning from manual expert administration toward fully automated, ultra-high-density recording pipelines, integrating machine-learning classifiers to deliver real-time bedside assessments of consciousness in intensive care units.

12.2 Theoretical Expansion: Agency, Free Will, and the Self

Beyond the foundational quantification of perceptual awareness, the mathematical architecture of Integrated Information Theory is expanding to encompass higher-order cognitive phenomena: the nature of agency, the subjective sense of a unified self, and the age-old dilemma of free will. In recent theoretical treatises, Tononi, Albantakis, and their colleagues have demonstrated that the causal emergence framework underlying IIT provides a physicalist, mathematically rigorous foundation for genuine voluntary action.

Traditional reductionist physicalism asserts that all causation resides exclusively at the micro-physical level: human decisions are simply the inevitable outcomes of subatomic particles colliding deterministically, rendering personal agency and free will an illusion. IIT challenges this reductionist assumption through its proof of causal emergence:

• Because the maximal integrated cause-effect structure ($\Phi^{\text{Max}}$) crystallizes at a specific macroscopic spatiotemporal scale, the macro-system possesses more intrinsic causal power over its own future than its underlying micro-components possess over theirs.
• True causal control objectively resides at the macro-level of the conscious complex.
• When an individual makes a voluntary, conscious decision, that decision is governed by the intrinsic determinism of the maximal cause-effect structure—the physical substrate of consciousness acting upon itself.

This provides an operational, causal redefinition of free will: free will is intrinsic causal autonomy. It is the capacity of a high-$Phi$ system to determine its own future states through its internal, integrated causal architecture, free from determination by external background noise or microscopic indeterminism. Furthermore, the structural relations that bind perceptual distinctions over time naturally give rise to a self-referential causal core: an ongoing, topologically stable geometrical structure that specifies the subjective, persistent sense of selfhood. The self is not an illusory homunculus sitting within the brain; it is the topological persistence of the conscious complex over time.

12.3 Concluding Synthesis: Tononi’s Paradigm and the Science of Consciousness

Integrated Information Theory represents one of the most intellectually ambitious, mathematically rigorous, and structurally coherent frameworks ever conceived to solve the problem of consciousness. By executing a revolutionary epistemic inversion—moving from the undeniable phenomenology of experience to the fundamental postulates of physical causation—Giulio Tononi transformed the study of consciousness from an observational, descriptive discipline into a predictive, axiomatic physical science.

IIT’s enduring contribution spans the entire breadth of cognitive science, neurology, and philosophy. Clinically, it has dismantled the assumption that behavioral responsiveness is equivalent to subjective awareness, providing life-saving neurotechnological tools like the Perturbational Complexity Index to detect hidden consciousness in comatose and paralyzed patients. Philosophically, it offers an uncompromising alternative to both reductive functionalism and substance dualism, framing consciousness as intrinsic causal power and unifying mind and matter within a shared geometric ontology. Scientifically, it provides a rigorous, objective criterion for evaluating the presence or absence of awareness across non-human animals, developing human fetuses, organoids, and emerging artificial intelligences, forcefully warning humanity that functional mimicry in silicon computing does not guarantee the presence of a conscious soul.

Profound questions undeniably remain. The computational intractability of calculating exact $Phi$ across large networks remains a major barrier; the empirical debates surrounding the temporal persistence of posterior activations are ongoing; and the ontological implications of causal realism continue to challenge contemporary metaphysics. Yet, by grounding its architecture in the primary reality of subjective existence and subjecting its claims to mathematical formalization and empirical testing, Integrated Information Theory has permanently altered our understanding of the mind. In Tononi’s paradigm, consciousness is no longer an accidental biological mystery floating aimlessly in a mechanical cosmos; it is the physical universe awakened to itself, exercising its highest, most irreducible causal power from within.

References

• Aaronson, S. (2014). Why I am not an integrated information theorist (or, the continuous unfoldment of consciousness). Shtetl-Optimized. https://scottaaronson.blog/?p=1799
• Albantakis, L., Barbosa, L., Findlay, G., Grasso, M., Haun, A. M., Marshall, W., Mayner, W. G. P., Saeed, Z., & Tononi, G. (2023). Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in terms of cause-effect power. PLOS Computational Biology, 19(10), e1011465. https://doi.org/10.1371/journal.pcbi.1011465
• Casali, A. G., Gosseries, O., Rosanova, M., Boly, M., Sarasso, S., Casali, K. R., Casarotto, S., Bruno, M.-A., Laureys, S., Tononi, G., & Massimini, M. (2013). A theoretically based index of consciousness independent of sensory processing and behavior. Science Translational Medicine, 5(198), 198ra105. https://doi.org/10.1126/scitranslmed.3006294
• Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3), 200–219. https://philpapers.org/rec/CHAFUT
• Doerig, A., Schurger, A., Hess, K., & Herzog, M. H. (2019). The unfolding argument: Why IIT and other causal theories of consciousness are unscientific. Consciousness and Cognition, 72, 49–59. https://doi.org/10.1016/j.concog.2019.04.002
• Edelman, G. M., & Tononi, G. (2000). A Universe of Consciousness: How Matter Becomes Imagination. Basic Books.
• Massimini, M., Ferrarelli, F., Huber, R., Esser, S. K., Singh, H., & Tononi, G. (2005). Breakdown of cortical effective connectivity during sleep. Science, 309(5744), 2228–2232. https://doi.org/10.1126/science.1117256
• Oizumi, M., Albantakis, L., & Tononi, G. (2014). From the phenomenology to the mechanisms of consciousness: Integrated Information Theory 3.0. PLOS Computational Biology, 10(5), e1003588. https://doi.org/10.1371/journal.pcbi.1003588
• Pearl, J. (2009). Causality: Models, Reasoning, and Inference (2nd ed.). Cambridge University Press. https://doi.org/10.1017/CBO9780511803161
• Sarasso, S., Boly, M., Napolitani, M., Gosseries, O., Charland-Verville, V., Casarotto, S., Rosanova, M., Casali, A. G., Boveroux, P., Rex, S., Laureys, S., Tononi, G., & Massimini, M. (2015). Consciousness and complexity during unresponsiveness induced by propofol, xenon, and ketamine. Current Biology, 25(23), 3099–3105. https://doi.org/10.1016/j.cub.2015.10.014
• Tononi, G. (2004). An information integration theory of consciousness. BMC Neuroscience, 5, 42. https://doi.org/10.1186/1471-2202-5-42
• Tononi, G. (2008). Consciousness as integrated information: A provisional manifesto. The Biological Bulletin, 215(3), 216–242. https://doi.org/10.2307/25470707
• Tononi, G., Boly, M., Massimini, M., & Koch, C. (2016). Integrated information theory: From consciousness to its physical substrate. Nature Reviews Neuroscience, 17(7), 450–461. https://doi.org/10.1038/nrn.2016.44