Chapter 9 Integrated Information Theory

9.1 Chapter Overview

Integrated Information Theory, usually abbreviated as IIT, is one of the most ambitious and controversial contemporary theories of consciousness. Developed primarily by Giulio Tononi and collaborators, IIT proposes that consciousness corresponds to integrated causal information generated by a system that exists as an irreducible whole [@tononi2004; @oizumi2014].

Unlike theories that begin primarily with cognition, reportability, attention, behaviour, or information access, IIT begins with the structure of conscious experience itself. It asks what properties experience appears to have from the inside, and then attempts to identify what kind of physical system could support those properties.

According to IIT, consciousness is not simply computation, intelligent behaviour, sensory processing, or functional performance. A system is conscious to the extent that it has intrinsic causal structure that is both integrated and irreducible. This means that the system must make a difference to itself as a whole, rather than functioning merely as a collection of independent parts.

IIT is distinctive because it attempts to provide a formal theory of phenomenal consciousness. It places special emphasis on intrinsic existence, unity, differentiation, integration, exclusion, causal structure, and the quantity known as Φ, or Phi [@tononi2008; @oizumi2014].

At the same time, IIT remains deeply controversial. Critics argue that integrated information may not explain why experience feels like anything, that Φ is difficult to calculate for realistic systems, and that IIT may attribute consciousness too broadly to simple systems [@cerullo2015; @aaronson2014; @bayne2018].

This chapter introduces IIT’s central claims, phenomenological axioms, physical postulates, the concept of integrated information, neuroscientific implications, strengths, criticisms, and implications for artificial intelligence, animals, and consciousness-first theories.

9.2 Learning Objectives

After reading this chapter, the reader should be able to:

• Define the central claim of Integrated Information Theory.
• Explain the meaning of integrated information and Φ.
• Describe IIT’s phenomenological starting point.
• Distinguish IIT from access-based theories such as Global Workspace Theory.
• Explain the concepts of intrinsic existence, integration, irreducibility, and exclusion.
• Analyze IIT’s neuroscientific implications.
• Evaluate major strengths and criticisms of IIT.
• Discuss IIT’s implications for artificial intelligence, animal consciousness, and panpsychism.
• Understand how IIT relates to the hard problem of consciousness.

9.3 Why Integrated Information Theory Became Influential

Integrated Information Theory became influential because it addressed a perceived limitation in many cognitive theories of consciousness. Many theories explain consciousness in terms of attention, reportability, working memory, global access, behavioural control, or cognitive coordination. These are important features of conscious cognition, but Tononi argued that they may not explain consciousness as experience [@tononi2004].

IIT therefore shifts the starting point. Instead of asking first what consciousness does, it asks what consciousness is like. Conscious experience appears unified, structured, specific, and present for the subject. IIT attempts to derive physical requirements from these phenomenological features.

This reversal of explanatory direction made IIT philosophically distinctive. Global Workspace Theory, for example, begins from cognitive access and broadcasting [@baars1988; @dehaene2011]. IIT begins from experience itself and asks what kind of causal system could make such experience possible.

IIT also became influential because it attempts to formalize consciousness mathematically. It does not merely state that consciousness depends on complexity or integration. It tries to define integrated information, measure it through Φ, and connect that measure to the degree and structure of consciousness [@oizumi2014; @tononi2016].

9.4 Historical Development

IIT was first proposed by Giulio Tononi in the early 2000s as an attempt to explain consciousness in terms of information integration [@tononi2004]. The original theory emphasized that conscious experience is both highly differentiated and unified. A conscious state rules out many alternative possible states, making it informative, while also appearing as one unified experience.

Later versions of IIT became more sophisticated. Tononi, Oizumi, Koch, and collaborators developed the theory into a formal framework based on intrinsic causal power, irreducibility, and cause-effect structure [@oizumi2014; @tononi2016]. These later versions shifted the focus from information in a general sense to integrated causal structure: the way a system constrains its own possible past and future states.

IIT developed partly in response to theories that emphasized conscious access, reportability, or cognitive function. Such theories are powerful, but IIT argues that consciousness should not be defined by what an outside observer can report or measure. Instead, consciousness should be understood from the intrinsic perspective of the system itself.

This makes IIT especially important in debates about the hard problem. While many scientific theories focus on the mechanisms associated with consciousness, IIT attempts to explain why certain physical systems have experience at all [@chalmers1995; @chalmers1996].

9.5 The Phenomenological Starting Point

A defining feature of IIT is that it begins with phenomenology. It does not begin by asking which brain regions are active during conscious perception or which behaviours indicate awareness. Instead, it begins by asking what properties conscious experience seems to have in itself.

IIT proposes that every conscious experience has several basic features. Experience exists for the subject. It has internal structure. It is specific rather than vague. It is unified rather than merely a collection of separate fragments. It is definite and bounded rather than indeterminate.

{r fig-iit, echo=FALSE, fig.cap="Integrated Information Theory (IIT). Panel A compares low and high integration. Panel B summarizes IIT’s phenomenological axioms. Panel C illustrates the conceptual meaning of Φ (Phi). Panel D presents a simplified consciousness spectrum across systems. Panel E illustrates intrinsic cause-effect structure and irreducible causal organization.", out.width="100%", fig.align="center"} id="x8s4mn" knitr::include_graphics("figures/10_integrated_information.png")

Figure @ref(fig:fig-iit) summarizes the core framework of IIT. Panel B represents the phenomenological axioms, while the other panels illustrate integration, Φ, the consciousness spectrum, and intrinsic cause-effect structure.

The important point is that IIT does not treat consciousness primarily as reportability or external behaviour. It treats consciousness as intrinsic experience and then asks what physical requirements follow from that.

9.6 IIT’s Phenomenological Axioms

IIT proposes a set of phenomenological axioms. These are intended to describe features that are present in every conscious experience.

9.6.1 Intrinsic Existence

Conscious experience exists intrinsically. It exists for the subject having the experience, not merely for an outside observer. A conscious state is not simply a pattern that someone else can measure. It is present from its own first-person perspective.

This axiom distinguishes IIT from purely external or behavioural accounts. A system may display complex behaviour from the outside, but IIT asks whether the system has intrinsic causal existence for itself.

9.6.2 Composition

Conscious experience is structured. It contains parts, distinctions, and relations. For example, a visual scene may include colours, shapes, spatial relations, objects, background, bodily feeling, and emotional tone. Consciousness is not usually an undifferentiated blank.

Composition means that experience has internal organization. IIT therefore looks for physical systems whose causal structure can support structured experience.

9.6.3 Information

Every conscious experience is specific. To experience one thing is to exclude many alternatives. Seeing a red object is different from seeing a blue object, hearing a sound, feeling pain, or having no visual experience at all.

In IIT, information does not simply mean data transmission. It means that a state specifies one experience rather than many other possible experiences. Consciousness is informative because each experience is differentiated from alternatives [@tononi2004].

9.6.4 Integration

Conscious experience is unified. A single experience normally cannot be decomposed into fully independent conscious parts. When one sees a visual scene, hears sounds, and feels bodily presence, these are typically experienced together as one field of awareness.

This axiom is central to IIT. Consciousness is not just information. It is integrated information. The whole experience is more than a set of independent fragments.

9.6.5 Exclusion

Conscious experience is definite. At any given moment, experience has a particular content, boundary, and scale. One does not experience every possible interpretation or every possible level of organization simultaneously.

The exclusion axiom is one of IIT’s most distinctive and controversial claims. It implies that among many overlapping possible systems, only the one with maximal irreducible causal structure constitutes the conscious subject at that moment [@oizumi2014].

9.7 Integrated Information and Φ

The central quantitative concept in IIT is Φ, or Phi. Φ is intended to represent how much integrated information a system generates as an irreducible whole.

A system has high Φ when its components interact in such a way that the system cannot be reduced to independent parts without losing significant causal structure. In other words, the whole system makes a difference to itself in a way that its parts alone do not.

A system has low Φ when its components operate mostly independently. If the system can be divided into parts with little loss of causal organization, then it has low integration.

Figure @ref(fig:fig-iit) Panel A illustrates the difference between low and high integration. A fragmented system may contain many components, but if those components do not strongly constrain one another, the system does not form an integrated whole. A highly integrated system, by contrast, has internal causal unity.

IIT therefore does not equate consciousness with complexity alone. A system can be complex but not strongly integrated. Nor does IIT equate consciousness with intelligence alone. A system may perform complex computations without forming an irreducible intrinsic causal structure.

9.8 Irreducibility and System Unity

Irreducibility is one of IIT’s most important ideas. According to IIT, a conscious system must be more than the sum of independent parts. The whole must have causal power that cannot be captured by partitioning the system into separate components.

This is why Φ is related to partitioning. If a system is divided into parts and little causal structure is lost, then the system was not strongly integrated. If dividing the system destroys significant causal structure, then the system has a higher degree of irreducibility.

This emphasis on irreducibility helps explain why IIT gives special importance to unity. Consciousness appears unified from the first-person point of view. IIT attempts to explain this unity by identifying physical systems that are causally integrated in an irreducible way.

This also explains why IIT differs from simple information-processing theories. A digital computer may process large amounts of information, but IIT asks whether that processing forms an integrated intrinsic causal whole. From the IIT perspective, information processing alone is not sufficient.

9.9 Intrinsic Cause-Effect Structure

IIT proposes that conscious systems possess intrinsic cause-effect structure. This means that the system constrains its own possible past and future states. It is not merely responding to external inputs or producing external outputs. It has internal causal organization that exists from the system’s own perspective.

Figure @ref(fig:fig-iit) Panel E illustrates this idea. A conscious system is one whose internal mechanisms specify a structured set of causal possibilities. These possibilities form what IIT calls a cause-effect structure.

This idea separates IIT from purely functionalist or computational theories. Functionalism often defines mental states by what they do in relation to inputs, outputs, and other states. IIT argues that what matters is not only functional role but intrinsic causal organization.

For this reason, two systems that perform the same input-output function might not be equally conscious according to IIT. A biological brain and a digital simulation might behave similarly, but they could differ in intrinsic causal structure.

9.10 Consciousness as a Spectrum

IIT implies that consciousness may exist along a spectrum. Systems can have different degrees of integrated information. Therefore, consciousness may not be strictly all-or-nothing.

Figure @ref(fig:fig-iit) Panel D illustrates this idea. Simple systems may have extremely low Φ. Animals may have varying degrees of consciousness. Human brains may generate highly integrated conscious experience. Artificial systems might be conscious if they possess the right intrinsic causal structure.

This spectrum view has important implications. It allows IIT to consider consciousness in infants, animals, patients with disorders of consciousness, and potentially artificial systems. It also means that consciousness may not be limited to human language, reportability, or reflective self-awareness.

However, the spectrum view also raises controversy. If even simple systems have some integrated information, does IIT imply that very simple systems have minimal consciousness? Critics argue that this may lead to panpsychist or panpsychist-like consequences [@cerullo2015; @bayne2018].

9.11 IIT and Neuroscience

IIT has important implications for neuroscience. It predicts that consciousness should correlate with integrated, recurrent, and causally unified neural organization. Consciousness should not depend simply on the number of neurons, the amount of activity, or computational complexity alone.

One common example is the cerebellum. The cerebellum contains a very large number of neurons and performs sophisticated computations. Yet damage to the cerebellum does not usually eliminate consciousness. IIT explains this by pointing to the cerebellum’s relatively modular organization. It may process information efficiently without generating the kind of integrated causal structure associated with conscious experience [@tononi2004; @koch2016].

IIT is also relevant to anesthesia, sleep, dreaming, coma, and minimally conscious states. These conditions involve changes in integration, complexity, and causal communication across the brain. Research on perturbational complexity has been especially important. The perturbational complexity index, or PCI, measures the complexity of brain responses to direct stimulation and has been used to distinguish wakefulness, sleep, anesthesia, and disorders of consciousness [@massimini2005; @casali2013].

Although PCI is not identical to Φ, it is often discussed as empirically related to IIT’s emphasis on integrated complexity. It provides one example of how IIT-inspired ideas can influence clinical and experimental neuroscience.

9.12 IIT and the Hard Problem

IIT is one of the few major scientific theories that explicitly attempts to address phenomenal consciousness rather than only access consciousness. It does not define consciousness primarily in terms of report, behaviour, attention, or global broadcasting. Instead, it attempts to explain consciousness as intrinsic integrated causal structure.

This makes IIT especially relevant to the hard problem. The hard problem asks why physical processes are accompanied by subjective experience at all [@chalmers1995; @chalmers1996]. IIT’s answer is that consciousness is identical to integrated information understood as intrinsic cause-effect structure.

Critics argue that this answer may not fully solve the problem. Even if a system has high Φ, why should integrated information feel like anything? Why should causal structure have subjective character? This is sometimes called a version of the mapping problem: how exactly does a mathematical or causal structure map onto a particular experience?

IIT therefore offers one of the most direct attempts to bridge physical structure and phenomenology, but whether it succeeds remains deeply debated [@levine1983; @bayne2018].

9.13 Relation to Other Theories

IIT differs from many other theories discussed in this book because it begins from phenomenology rather than cognitive function.

9.13.1 Relation to Global Workspace Theory

Global Workspace Theory explains consciousness in terms of global broadcasting, reportability, cognitive access, and flexible control [@baars1988; @dehaene2011]. IIT instead emphasizes intrinsic causal structure, irreducibility, and phenomenological unity.

Many researchers interpret GWT as strongest for access consciousness, while IIT attempts to explain phenomenal consciousness more directly. The two theories therefore often target different explanatory aspects of consciousness.

9.13.2 Relation to Functionalism

Functionalism defines mental states by causal roles and functional organization. IIT rejects the idea that functional equivalence is automatically enough for consciousness. According to IIT, two systems could perform the same function while differing in intrinsic causal structure.

This makes IIT especially important in debates about artificial intelligence. A digital system might simulate a conscious process functionally, but IIT asks whether it has the right intrinsic causal organization to be conscious.

9.13.3 Relation to Emergentism

IIT overlaps with emergentism because it treats consciousness as arising from integrated system-level organization. However, IIT is more formal and specific than many emergentist theories. It attempts to quantify the relevant organization through Φ and to connect it to phenomenological axioms.

Emergentism often says that consciousness arises from complexity. IIT tries to specify what kind of complexity matters: irreducible integrated causal structure.

9.13.4 Relation to Recurrent Processing and Predictive Processing

IIT overlaps with theories that emphasize recurrent interaction and integration. Recurrent Processing Theory argues that recurrent neural activity is central to conscious perception [@lamme2006]. Predictive Processing explains perception and cognition through hierarchical prediction and error correction [@friston2010; @clark2013]. IIT differs from both by grounding consciousness in intrinsic causal structure rather than perceptual recurrence or predictive inference alone.

9.13.5 Relation to Panpsychism and Consciousness-First Theories

IIT is not identical to panpsychism, but some philosophers argue that it has panpsychist-like implications because it may attribute minimal consciousness to systems with integrated causal structure [@goff2017; @bayne2018]. This makes IIT relevant to later discussions of panpsychism, cosmopsychism, and consciousness-first theories.

However, IIT differs from Taheri’s T-Consciousness and other consciousness-first frameworks. IIT begins from physical systems and asks which systems have intrinsic causal structure sufficient for consciousness. T-Consciousness and similar consciousness-first approaches begin from consciousness as foundational or prior to material organization. These approaches will be discussed later in the book.

9.14 Strengths of Integrated Information Theory

IIT has several major strengths. First, it directly engages with phenomenology. Unlike theories focused mainly on reportability or cognitive access, IIT begins with the structure of experience itself.

Second, IIT provides one of the most mathematically explicit frameworks in consciousness research. Although the mathematics is difficult and controversial, the attempt to formalize consciousness gives the theory unusual precision.

Third, IIT emphasizes unity and irreducibility. Conscious experience appears unified, and IIT offers a principled way to connect this unity with physical integration.

Fourth, IIT has clinical and neuroscientific relevance. IIT-inspired measures of complexity and integration have influenced research on anesthesia, sleep, coma, and disorders of consciousness [@casali2013; @koch2016].

Fifth, IIT provides a framework for thinking about non-human consciousness. It can be applied, at least in principle, to animals, infants, patients, and artificial systems without relying entirely on verbal report.

9.15 Weaknesses and Criticisms

IIT faces several major criticisms. One is computational difficulty. Calculating Φ precisely for realistic systems is extremely difficult. The brain is vastly complex, and exact Φ calculations may be impossible for large biological systems.

A second criticism is over-attribution. IIT may imply that simple systems with non-zero integrated information possess some degree of consciousness. Critics argue that this expands consciousness attribution too broadly, possibly to simple circuits or systems that do not plausibly have experience [@aaronson2014; @cerullo2015].

A third criticism concerns empirical testing. Some researchers argue that IIT is difficult to falsify because its core claims are abstract and its exact measures are difficult to apply directly to real brains.

A fourth criticism is the mapping problem. Even if Φ measures integration, it remains unclear why a specific cause-effect structure should correspond to a specific experience. How does a mathematical structure become the experience of red, pain, music, or selfhood?

A fifth criticism is that IIT may provide a weaker account of cognition than theories such as GWT. It may explain unity and intrinsic structure, but it does not as directly explain reportability, attention, reasoning, working memory, or behavioural control.

These criticisms do not make IIT unimportant. They show why it remains one of the most debated theories in consciousness science.

9.16 IIT and Artificial Intelligence

IIT has major implications for artificial intelligence. If consciousness depends on intrinsic integrated causal structure, then artificial systems could be conscious only if they possess the right kind of causal organization.

This creates a distinction between simulation and instantiation. A digital computer may simulate a system with high integration, but IIT asks whether the computer itself has high intrinsic causal integration. A simulation of a conscious system may not itself be conscious if the underlying hardware lacks the relevant causal structure.

This makes IIT more cautious than many functionalist or computational theories. A system could display intelligent behaviour, pass behavioural tests, or generate language without necessarily having high Φ. From IIT’s perspective, intelligence and consciousness are not the same.

At the same time, IIT does not rule out artificial consciousness in principle. A non-biological system could be conscious if it generated sufficient integrated causal structure. The question is not whether the system is made of neurons, but whether it forms an intrinsic, irreducible causal whole.

Recent debates about AI consciousness often include IIT among the theories that might provide indicators for machine consciousness, while also noting the difficulty of applying IIT to current artificial architectures [@butlin2023].

9.17 Open Questions

Several important questions remain unresolved for IIT. Why should integrated information generate subjective experience? Can Φ be measured reliably in realistic systems? Is consciousness fundamentally graded? Do simple systems have minimal consciousness? Are IIT’s phenomenological axioms universally valid? Does IIT overextend consciousness attribution? Can digital systems possess high Φ? Does integration explain specific qualitative experiences, or only the unity and degree of consciousness?

These questions show why IIT remains both influential and controversial. It offers a bold attempt to connect phenomenology and physical structure, but many details remain unresolved.

9.18 Evaluation

Integrated Information Theory is one of the most ambitious theories in consciousness studies. It attempts to explain consciousness not merely as access, reportability, behaviour, or computation, but as intrinsic integrated causal structure.

Its greatest strength is that it takes phenomenal consciousness seriously. It begins from experience itself and attempts to derive physical requirements from phenomenological features such as unity, specificity, structure, and irreducibility.

Its greatest weakness is that the connection between integrated information and subjective experience remains contested. Critics argue that IIT may not fully explain why integrated causal structure should feel like anything. Practical challenges in calculating Φ and concerns about over-attribution also remain significant.

IIT is therefore both groundbreaking and controversial. It is one of the few theories that attempts to formalize phenomenal consciousness directly, but whether it successfully solves the hard problem remains an open question.

9.19 Chapter Summary

Integrated Information Theory proposes that consciousness corresponds to integrated, irreducible causal information generated by a system. The central measure associated with the theory is Φ, which represents the degree to which a system forms an integrated whole.

IIT begins from phenomenological axioms: intrinsic existence, composition, information, integration, and exclusion. It then derives physical postulates concerning causal structure, irreducibility, and system unity.

The theory differs from Global Workspace Theory because it focuses less on reportability and cognitive access and more on intrinsic phenomenal structure. It differs from functionalism because it does not treat input-output organization alone as sufficient for consciousness.

IIT has influenced neuroscience, philosophy, complexity science, clinical research, and debates about artificial intelligence. It is especially relevant to anesthesia, sleep, disorders of consciousness, animal consciousness, and machine consciousness.

Its major strengths are its direct engagement with phenomenology, formal structure, and emphasis on unity. Its major weaknesses are computational difficulty, controversial implications, over-attribution concerns, and the unresolved question of why integrated information should generate subjective experience.

The central unresolved question is whether integrated causal structure truly explains consciousness, or whether it identifies an important correlate without fully explaining phenomenal experience itself.