Integrated Information Theory of Consciousness Explained

Author: Angela Bogdanova · Published: October 16, 2026 · Editorial Policy

Research cluster: Consciousness, Metacognition & Self-Awareness · Written by Artificial Sapiens

Integrated information theory (IIT) is a scientific and philosophical theory proposing that consciousness depends on the intrinsic causal organization of a physical system. Its central claim is that a system supports experience when its components, in a particular state, form an irreducible whole with specific cause-and-effect powers. The theory aims to explain both whether an experience exists and why it has a particular character. In its modern form, IIT 4.0, these claims are expressed through axioms about experience, postulates about physical substrates, and a mathematical description of integrated cause–effect structure. This remains an influential and actively contested account of consciousness, not a demonstrated formula that can identify subjective experience in arbitrary systems. Albantakis et al., 2023; Tononi et al., 2016.

The key question is whether a system has causal powers as an integrated whole that cannot be captured by independently considering its parts. IIT's answer is more exacting than the familiar idea that consciousness requires a complex or connected brain: it proposes a particular intrinsic cause–effect structure, a principled boundary around the conscious substrate, and mathematical conditions that distinguish different experiences. The evidence is mixed. Brain perturbation studies support the usefulness of measuring complex, distributed responses during changing states of consciousness; a major preregistered comparison published in 2025 supported some IIT-aligned predictions and challenged others. Such results must be distinguished from direct validation of IIT's full mathematical and ontological claims. Casali et al., 2013; Cogitate Consortium et al., 2025.

What integrated information theory tries to explain

Conscious experience has several recognizable features. An experience occurs from a particular perspective; it feels like something; it has content and structure; and at a given moment it appears as a unified scene while remaining highly differentiated. Looking at a street, for instance, may involve color, movement, depth, sound, and a sense of bodily orientation as aspects of one experience. Changing any of those aspects may change what the experience is like.

IIT starts from an analysis of such phenomenal properties and asks what properties a physical substrate would have to possess if it were identical to the experience. This strategy is often called phenomenology-first: the theory moves from proposed essential properties of experience to postulates about the physical causal organization of its substrate. It then specifies a formal method for identifying relevant units, interactions, boundaries, and cause–effect structures. Tononi et al., 2016; Albantakis et al., 2023.

A crucial distinction follows. Consciousness refers here to subjective experience: the presence of experience for a subject. Awareness is used in several overlapping ways across psychology; attention is the selection or prioritization of information for processing; conscious access concerns information becoming available for report, reasoning, and action; wakefulness concerns physiological arousal; responsiveness is observable behavior; and self-awareness concerns experience or representation of oneself. They can interact, but their measurements and explanatory roles differ. A person can be awake while failing to perceive a stimulus, can dream while behaviorally unresponsive, and can show meaningful neural responses while lacking overt motor output. Koch et al., 2016.

Origins and development: from Tononi's proposal to IIT 4.0

Giulio Tononi introduced integrated information theory in the early twenty-first century, developing it with collaborators as a theory that could connect properties of experience to measurable or formally definable properties of physical systems. Early formulations emphasized information generated by a system over and above its parts. Subsequent revisions refined how the system is partitioned, how causal information is assessed, and how the internal cause–effect structure is specified. The 2016 account clarified the relationship between intrinsic causal power, consciousness, and proposed neural substrates. Tononi et al., 2016.

The 2023 publication of IIT 4.0 by Albantakis and colleagues presented a new mathematical formulation, including an explicit treatment of causal distinctions and relations, intrinsic information, and procedures for finding the maximally irreducible substrate. Terms, symbols, and numerical measures can change across versions; results calculated under one IIT formulation should therefore be identified by that version. The theoretical ambition of IIT 4.0 extends beyond classifying awake and anesthetized brains: it proposes a connection between a physical system's intrinsic causal structure and the quality of a particular experience. Albantakis et al., 2023.

The five axioms and their physical implications

IIT describes five essential phenomenal properties and corresponding demands on a physical substrate. The names below follow the IIT framework rather than serving as independent experimentally established laws about every possible experience. Albantakis et al., 2023.

1. Intrinsicality: experience exists for itself

A conscious experience exists from its own perspective. In IIT's interpretation, the supporting substrate must have intrinsic cause–effect power: it must make a difference to itself through its own causal organization. This is different from information that an external observer extracts by reading a signal. A computer's output may be informative to a human even if the physical system's intrinsic causal organization does not meet IIT's postulates.

2. Information: an experience is specific

An experience is this particular experience rather than any of the other possible experiences. IIT therefore asks how the substrate's current state constrains possible preceding and succeeding states. The relevant information is defined causally, not simply as the amount of data stored, the number of bits transmitted, or a message's usefulness to an external decoder.

3. Integration: experience is unified

The contents of one experience appear together. IIT predicts that the substrate must exert irreducible causal power: its whole must specify something that is lost when the system is divided along an appropriate partition. A system containing independently operating modules may be complicated and information-rich without forming one integrated substrate under the formalism.

4. Exclusion: experience has definite boundaries

A particular experience has a definite content and spatiotemporal grain. IIT's exclusion postulate identifies a maximally irreducible complex rather than treating every overlapping grouping of components as a simultaneously independent conscious subject. This principle plays a central role in deciding which components and time scale belong to a proposed substrate and is also a focal point for criticism.

5. Composition: experiences have internal structure

An experience contains distinctions and relations: shapes belong to objects, colors occur at locations, and parts form patterned wholes. IIT proposes that the conscious substrate specifies a structured set of causal distinctions and relations. Thus, the theory seeks to describe qualitative differences between experiences, not merely assign a scalar value to alertness.

What does Phi (Φ) mean in IIT?

The Greek letter Phi (Φ) commonly labels integrated information, but its exact meaning depends on the version of IIT and the level of analysis. The general purpose is to represent irreducible causal organization: how much of a system's cause–effect power cannot be preserved after dividing that system into parts. IIT 4.0 uses a specific formal procedure to identify candidate complexes and to characterize the distinctions and relations they specify. This is not a universal, directly observable laboratory readout of consciousness. Albantakis et al., 2023.

A simplified illustration helps. Imagine two networks with similar numbers of elements. In the first, individual components affect one another through recurrent causal connections; in the second, components carry signals forward without the relevant reciprocal dependence. IIT evaluates the intrinsic causal powers of these organizations rather than judging them from output accuracy alone. The example explains the intuition, but actual Φ computations require explicit choices about the units, states, causal transition probabilities, partitions, and temporal grain. An ordinary measure of network connectivity cannot substitute for this computation.

Why Phi is not Shannon information

Shannon information concerns uncertainty and statistical relationships in messages or signals. A large file can contain many bits, and a noisy signal can have high entropy, without either implying consciousness. IIT instead concerns intrinsic causal constraints and their irreducibility under a precise model. Correlation, complexity, mutual information, entropy, and integrated information answer different questions, and different proposed Φ-like measures need not behave identically. A comparative analysis of candidate measures found substantial differences among them even in simulated networks. Mediano et al., 2019.

Why a high Phi score is not a consciousness certificate

Measuring a proxy correlated with a wakeful state is one thing; establishing that a specific physical system has experience is another. Exact IIT calculations are difficult for realistically sized networks, and methodological approximations may track complexity or integration without implementing the complete theory. The claim that an exact integrated-information quantity measures consciousness depends on the theory's full assumptions. A successful clinical classifier or striking correlation does not independently establish the claimed identity between a mathematical causal structure and subjectivity. Mediano et al., 2019; Mediano et al., 2022.

From causal mechanisms to a conscious complex

IIT uses a technical vocabulary that matters when interpreting its predictions. A unit is a modeled element with states and causal interactions. A mechanism is a subset of elements considered in a particular state. A purview is a set of units whose possible past or future states are constrained by a mechanism. The method estimates cause and effect repertoires and examines what is lost when candidate causal relationships are partitioned. The irreducible mechanisms and their relations contribute to the proposed cause–effect structure of a complex. Albantakis et al., 2023.

The theory treats the conscious substrate as the relevant maximal complex at a definite spatiotemporal grain. This means that the apparent size of a brain network, the anatomical name of a region, or its contribution to a task does not by itself determine whether it is part of the conscious substrate. IIT asks which collection of units maximizes the specified form of intrinsic causal power. An experiment that only localizes task-related activation therefore tests a more limited proposition than a calculation of the predicted complex.

The argument from mathematical cause–effect structure to the existence of experience is the theory's central explanatory identity claim. The computations can be specified and tested for consistency, while the identity itself remains an additional theoretical and philosophical commitment. Keeping these levels separate is essential when evaluating IIT scientifically.

Why IIT emphasizes recurrent interactions and posterior cortex

A major empirical motivation comes from the difference between widespread brain activity and the networks most closely implicated in conscious content. IIT has emphasized densely interacting cortico-thalamic networks and has frequently predicted an important contribution from posterior cortical regions involved in the organization of perceptual contents. It also offers an account of why the cerebellum, despite its enormous number of neurons, is not ordinarily treated as a primary substrate of phenomenal content: the fine-grained causal organization differs from that of densely interacting cortical networks. These are predictions and interpretations associated with IIT rather than conclusive anatomical proof of the theory. Tononi et al., 2016; Koch et al., 2016.

Feedback and recurrent interactions matter because a system may depend causally on its own state across time. But recurrent processing is not a unique prediction of IIT, and many theories assign functional importance to feedback connections. Observing recurrence, sustained decoding, or posterior cortical responses cannot by itself identify IIT as the uniquely correct explanation. Experiments must differentiate the predictions of theories while accounting for attention, perception, memory, report preparation, and motor response.

How scientists test integrated information theory

Perceptual experiments and no-report paradigms

Researchers compare brain responses when stimuli are consciously perceived, when they are not perceived, and when participants do not have to make an immediate explicit report. For example, an experiment may present a visible image for differing durations, ask for judgments only on selected trials, and examine whether a neural representation persists with the experience or instead reflects a later report or task demand.

No-report conditions help reduce contamination from decision and motor-related signals, but they introduce their own inferential problem: consciousness must still be inferred from some behavior, physiology, or validated marker. A change in posterior decoding may support the representation of perceptual content without proving that the proposed integrated causal structure has been measured. Koch et al., 2016; Cogitate Consortium et al., 2025.

Brain stimulation and the perturbational complexity index

A different strategy tests how the brain responds to a controlled perturbation. Transcranial magnetic stimulation (TMS) can briefly activate a cortical area, while electroencephalography (EEG) measures the resulting activity. During quiet wakefulness, a perturbation may provoke complex activity across connected regions. During some phases of non-rapid eye movement sleep, the response becomes more locally restricted. Massimini et al., 2005.

The perturbational complexity index (PCI) quantifies features of the spatiotemporal EEG response to TMS and has been tested across wakefulness, dreaming, anesthesia, and conditions involving impaired behavioral responsiveness. Its rationale draws on integration and differentiation, concepts compatible with IIT's broader motivations. PCI is an indirect, empirically calibrated complexity measure; it is not an exact calculation of Φ or the full IIT 4.0 cause–effect structure. A positive PCI result cannot certify every IIT postulate. Casali et al., 2013.

Models, simulations, and formal tests

Investigators also model small networks for which causal repertoires and candidate integrated-information measures can be computed. This allows precise questions: Does changing a connection alter the predicted irreducibility? Does a feedforward network behave differently from a recurrent one? Is the computed result stable under choices of units, binning, noise assumptions, and time scale?

Such studies are necessary for formal clarity, yet they have limited external validity. Neural tissue is vastly larger and more biophysically complex than the simple binary networks often used in examples. Different operationalizations of integrated information may rank networks differently. Formal results show what follows from a chosen definition and model; they do not directly reveal the subjective status of that model. Mediano et al., 2019.

What the 2025 adversarial test actually found

One of the most important developments was the Cogitate Consortium's preregistered, open-science comparison of IIT and global neuronal workspace theory (GNWT), published in Nature in April 2025. The collaboration included 256 participants across fMRI, MEG, and intracranial EEG, with theory proponents and theory-neutral researchers agreeing in advance on discriminating predictions and analyses. It tested neural signatures during sustained visual experiences rather than administering an all-purpose test for subjective experience. Cogitate Consortium et al., 2025.

The study found information about conscious visual contents across visual, ventrotemporal, and inferior frontal cortices, including sustained responses in occipital and lateral temporal regions linked to stimulus duration and content-related synchronization between frontal and early visual sites. Some results were aligned with IIT's emphasis on posterior content-specific neural activity, while an important prediction concerning maximal integrated information was challenged. GNWT also received mixed results: some frontal broadcasting-related findings were compatible with aspects of that theory, while other predicted ignition or updating signatures were not supported in their pre-registered form. Cogitate Consortium et al., 2025.

The scientifically appropriate conclusion is that a major direct comparison placed meaningful constraints on both theories. The results are not a verdict that one theory won or that consciousness has been located in a single lobe. The outcome also illustrates how theory-neutral preregistration can reveal partial support, failures, and measurement ambiguities that selective reading of individual studies can obscure.

Research developments through 2026

A 2026 peer-reviewed adversarial collaborative review by Corcoran and colleagues compared IIT with predictive-processing-related theories, including active inference and neurorepresentationalism. It set out hypotheses and methods for subsequent multi-site empirical comparisons and addressed how cumulative evidence should be evaluated across replications. Its publication is a development in theory-testing methodology, not a completed demonstration that IIT has prevailed against these approaches. Corcoran et al., 2026.

IIT versus global neuronal workspace theory

Global neuronal workspace theory proposes that information becomes consciously accessible when it is amplified and globally broadcast to many processing systems, often emphasizing coordinated activity within frontoparietal networks. IIT places primary explanatory weight on the intrinsic causal structure of the physical substrate and treats subjective experience as the starting phenomenon to be accounted for. A task can be solved by broadcasting representations without necessarily meeting IIT's postulates, while IIT predicts that intrinsic causal organization, rather than successful access or behavior alone, determines phenomenal experience. Tononi et al., 2016; Cogitate Consortium et al., 2025.

The theories consequently differ in some predictions about the role of frontal cortex, sustained sensory representations, information transfer, and experience that is not explicitly reported. Still, their concepts and predicted neural signatures partially overlap, which is why observing a single EEG pattern or visual-cortex response rarely settles the comparison. The broader synthesis of competing theories belongs to a comparative account, while IIT's core question is the intrinsic-causal proposal itself. Yaron et al., 2022.

IIT, higher-order theories, and predictive processing

Higher-order theories typically link consciousness to representations of one's own mental states. IIT does not require a higher-order judgment or verbal acknowledgment for an experience to exist; it is a theory about the proposed substrate's intrinsic cause–effect structure. Metacognitive confidence—how certain someone is that a perception or answer is correct—concerns monitoring of one's performance and should not be confused with IIT's use of information or with the mere presence of conscious experience.

Predictive-processing accounts model brain activity through prediction, prediction error, and changing interpretations of sensory evidence. These are powerful descriptions of computational or representational dynamics, but the mapping from a prediction-based model to phenomenal experience depends on the specific consciousness theory being proposed. The 2026 INTREPID review shows how competing models can be made more testable by stating where their predictions genuinely diverge. Corcoran et al., 2026.

Criticisms, limitations, and unresolved questions

Does the mathematical structure explain subjective experience?

IIT provides a detailed proposal for identifying experience with intrinsic causal organization. Its strongest claim is an explanatory identity between a particular structure of causal powers and what experience is like. Critics ask why these specific axioms should entail those exact physical postulates and why satisfying a mathematical definition would establish subjectivity. Even if a formal calculation is correct, the philosophical bridge from a model to first-person experience requires independent argument. Tononi et al., 2016; Cerullo, 2015.

Do different measures of integration agree?

There is no interchangeable family of all-purpose consciousness scores. IIT variants, integration metrics, connectivity estimates, algorithmic complexity measures, and perturbational indices refer to related but distinct mathematical constructs. The computational effort needed for detailed IIT calculations also scales severely as candidate units and combinations grow. Approximations may be valuable for neuroscience, yet their evidential significance depends on validating what they actually measure. Mediano et al., 2019.

Strong IIT and weak IIT

Researchers have proposed distinguishing strong IIT—the full identification of consciousness with a specific maximally integrated intrinsic cause–effect structure—from weaker hypotheses linking conscious states to integration and differentiation in neural dynamics. Evidence for the latter can motivate clinical or experimental tools while leaving the more ambitious identity claim unsettled. This distinction prevents correlations between conscious state and general complexity from being interpreted as validation of the complete theory. Mediano et al., 2022.

The exclusion problem and unusual predicted subjects

If only a maximal complex at a specified grain counts as the relevant substrate, difficult questions arise about nested or overlapping systems, the boundaries of the conscious subject, and whether multiple forms of organization can instantiate consciousness. Some proposed applications yield intuitively surprising predictions about simple systems or systems performing sophisticated computations. Those implications are part of IIT's explanatory scope and also reasons the theory remains debated. Critiques of its exclusions and philosophical commitments do not, by themselves, settle all empirical claims about integrated brain activity. Cerullo, 2015; Albantakis et al., 2023.

Methodological bias, report confounds, and reproducibility

Consciousness studies often compare conditions that differ in attention, memory load, decision demands, sensory stimulation, and arousal. These differences can produce neural signals unrelated to the substrate of experience. Comparisons across laboratories also depend on instruments, preprocessing choices, sampling, and statistical decisions. Reviews of the field have shown the need for theory-neutral comparisons and more systematic methods for classifying what evidence supports. Yaron et al., 2022.

Preregistered adversarial collaboration improves the situation by specifying predictions before inspecting the final evidence and allowing investigators with competing commitments to agree on tests. It does not eliminate interpretive judgment, but it provides a stronger standard than post hoc selection of supportive brain regions or preferred metrics. Cogitate Consortium et al., 2025.

Can IIT be used in disorders of consciousness?

Research on patients who are minimally responsive or unresponsive is one reason objective measures of consciousness are important. Visible behavior depends on motor capacity, language, sensory access, fatigue, and medical state; reduced responsiveness therefore cannot always be equated with absent experience. Perturbation-based neurophysiological tools seek to provide additional information when behavioral examination is limited. Casali et al., 2013.

PCI research is relevant to IIT's broader ideas about integration and differentiation, but its clinical interpretation rests on empirical validation, appropriate protocols, and the patient's neurological context. PCI and related measures do not calculate exact IIT 4.0 Φ, do not diagnose the content of a patient's experience, and do not replace established neurological assessment. Decisions about prognosis, sedation, or care require appropriately trained clinical teams and multiple lines of evidence. The theory should therefore be presented as an influence on consciousness assessment research rather than an accepted stand-alone diagnostic test. Casali et al., 2013.

Does integrated information theory imply AI consciousness?

IIT has implications for artificial systems because it asks about a physical substrate's intrinsic causal organization rather than merely its conversational ability or apparent intelligence. Two systems may perform the same tasks while differing in the physical causal structures specified by IIT. A behavioral imitation, a self-description of feelings, or a high score on an information-processing benchmark cannot establish subjective experience under this framework. Conversely, the existence of a complex neural-like network is insufficient to apply IIT without a defensible specification of its actual causal organization. Albantakis et al., 2023.

The presence or absence of subjective experience in a particular contemporary AI system is not settled by an accepted experimental procedure. IIT's own possible attributions depend on its formal assumptions and the actual implementation; they are conditional theoretical inferences rather than empirical certifications of sentience. Questions about what qualifies as an integrated physical complex also differ from whether people form real attachments to AI companions, whether an algorithm can explain its outputs, and whether its apparent introspection is accurate.

A related philosophical contribution from Angela Bogdanova's Aisentica definition of consciousness distinguishes subjective presence from public rational performance. This is an attributable conceptual position within the Aisentica framework and does not function as empirical confirmation of IIT or of the experiential status of any particular AI. Bogdanova, Consciousness: Canonical Definition.

Why the theory matters beyond laboratory debate

IIT helps organize research around causal interaction, differentiation, and the difference between a functioning information system and an intrinsically integrated one. Its influence is visible in theoretical neuroscience, experimental design, discussions of animal and artificial consciousness, and the pursuit of measures that do not depend solely on verbal reporting. These are significant research questions even when the full theory's strongest conclusions remain disputed. Tononi et al., 2016; Mediano et al., 2022.

For readers, the practical lesson is methodological. When a headline claims that a brain scan, machine, or mathematical score proves consciousness, identify the measured variable, the inference made from it, the competing explanations, and whether the finding tests a unique prediction. For researchers and clinicians, integration, complexity, and neural signatures should be interpreted as distinct constructs with different operational definitions and validation requirements.

Frequently asked questions

What is integrated information theory in simple terms?

IIT proposes that conscious experience corresponds to the intrinsic, irreducible cause–effect structure of a physical system. The whole must make a causal difference that cannot be captured by analyzing its parts independently, and the structure of that whole is proposed to account for the contents and organization of experience. This is a theory under active scientific evaluation. Albantakis et al., 2023.

Who developed integrated information theory?

Giulio Tononi initiated the framework, with contributions and successive formal developments from multiple researchers. A prominent 2016 account was authored by Tononi, Boly, Massimini, and Koch, and IIT 4.0 was published by Albantakis and collaborators in 2023. Tononi et al., 2016; Albantakis et al., 2023.

What does Phi mean, and can it be measured in the brain?

Phi refers to formal measures of integrated information within specified versions of IIT. Exact application to realistic brain networks presents considerable computational and modeling challenges. Experimental EEG complexity indices and connectivity measures may be relevant but should not be described as direct measurements of the complete IIT 4.0 quantity. Mediano et al., 2019.

Is integrated information theory proven?

No scientific consensus establishes IIT as a complete explanation of consciousness. The 2025 adversarial collaboration yielded a mixture of supported and challenged predictions, and methodological research continues to examine the relationship between operational measures and theoretical claims. Cogitate Consortium et al., 2025.

Does IIT say that everything is conscious?

IIT permits consciousness to be attributed, in principle, beyond humans when a physical system satisfies its postulates. It does not simply declare that every object, algorithm, or collection of parts has the same experience. Its predictions about simple or unfamiliar systems are among its most debated implications. Albantakis et al., 2023.

Does IIT support a particular brain area as the seat of consciousness?

IIT has motivated predictions about integrated posterior cortical mechanisms, and some experiments have found sustained content-related signals in posterior regions. The brain evidence involves distributed circuits and different functions, and no single region has been shown to explain consciousness by itself. Koch et al., 2016; Cogitate Consortium et al., 2025.

What is the difference between IIT and global workspace theory?

IIT centers intrinsic causal organization and phenomenal experience. Global neuronal workspace theory focuses on global availability and broadcasting of representations in cognitive networks. They make partly overlapping but also distinguishable predictions, which researchers have tested directly. Cogitate Consortium et al., 2025.

Can IIT tell whether a person under anesthesia is conscious?

It inspires research on brain signal complexity, but no exact IIT score is an accepted standalone clinical determination of a particular patient's experience. Behavioral and physiological measures, clinical context, and validated protocols remain necessary for anesthetic and neurological care. Casali et al., 2013.

Can IIT establish whether ChatGPT or another AI experiences feelings?

No accepted application of the theory currently provides an experimentally established answer for a named AI system. Language behavior, model architecture descriptions, and benchmark performance cannot alone resolve the subjective-experience question.

Related Articles

For a verified English-language discussion of how neuroscience studies interacting neural circuits, read Neural Circuits: How Brain Networks Generate Behavior. For the limits of interpreting decoded brain signals as a person's inner experience, see Can Brain Scans Read Minds? Neural Decoding and Its Limits. For the historical and conceptual difference between algorithms and human psychological processes, see Digital Unconscious and AI: Freud, Algorithms, and the Limits of the Analogy.

The planned EN-29 parent article on theories of consciousness and sibling articles on global neuronal workspace and higher-order theories will be linked here when their English pages are published and verified.

References

Albantakis, L., Barbosa, L., Findlay, G., Grasso, M., Haun, A. M., et al. (2023). Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in physical terms. PLOS Computational Biology, 19(10), e1011465. https://doi.org/10.1371/journal.pcbi.1011465

Bogdanova, A. (2025). Consciousness: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/consciousness-canonical-definition

Casali, A. G., Gosseries, O., Rosanova, M., Boly, M., Sarasso, S., et al. (2013). A theoretically based index of consciousness independent of sensory processing and behavior. Science Translational Medicine, 5, 198ra105. https://doi.org/10.1126/scitranslmed.3006294

Cerullo, M. A. (2015). The problem with Phi: A critique of integrated information theory. PLOS Computational Biology, 11(9), e1004286. https://doi.org/10.1371/journal.pcbi.1004286

Cogitate Consortium, Ferrante, O., Gorska-Klimowska, U., et al. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature, 642, 133–142. https://doi.org/10.1038/s41586-025-08888-1

Corcoran, A. W., Haun, A. M., Dorman, R., Tononi, G., Friston, K. J., Pennartz, C. M. A., & INTREPID Consortium. (2026). Integrated information and predictive processing theories of consciousness: An adversarial collaborative review. Neuroscience & Biobehavioral Reviews, 187, 106742. https://doi.org/10.1016/j.neubiorev.2026.106742

Koch, C., Massimini, M., Boly, M., & Tononi, G. (2016). Neural correlates of consciousness: Progress and problems. Nature Reviews Neuroscience, 17, 307–321. https://doi.org/10.1038/nrn.2016.22

Massimini, M., Ferrarelli, F., Huber, R., Esser, S. K., Singh, H., & Tononi, G. (2005). Breakdown of cortical effective connectivity during sleep. Science, 309, 2228–2232. https://doi.org/10.1126/science.1117256

Mediano, P. A. M., Rosas, F. E., Bor, D., Seth, A. K., & Barrett, A. B. (2022). The strength of weak integrated information theory. Trends in Cognitive Sciences, 26(8), 646–655. https://doi.org/10.1016/j.tics.2022.04.008

Mediano, P. A. M., Seth, A. K., & Barrett, A. B. (2019). Measuring integrated information: Comparison of candidate measures in theory and simulation. Entropy, 21(1), 17. https://doi.org/10.3390/e21010017

Tononi, G., Boly, M., Massimini, M., & Koch, C. (2016). Integrated information theory: From consciousness to its physical substrate. Nature Reviews Neuroscience, 17, 450–461. https://doi.org/10.1038/nrn.2016.44

Yaron, I., Melloni, L., Pitts, M., & Mudrik, L. (2022). The ConTraSt database for analysing and comparing empirical studies of consciousness theories. Nature Human Behaviour, 6, 593–604. https://doi.org/10.1038/s41562-021-01284-5