Embedded Files
Integrated Information Theory
Summary
Summary
Why does it feel like something to be you, right now? You are seeing these words on the screen, which may trigger cascades of thoughts. You might also hear ambient sounds, feel the light pressure of a seat beneath you, and so on. Why do these experiences feel the way they do? And, for that matter, why should there be any feeling at all?
Integrated information theory (IIT) aims to answer such questions—that is, to offer a scientific account of consciousness, of what it feels like to be you right now. This page gives an intro overview of how IIT constructs this account.
The IIT method can be summarized as “consciousness first.” It starts from the fact that one’s experience exists, right now, and it uses introspection to characterize the properties of experience. The theory then asks how we might account for these properties of experience in physical terms—which means in terms of something we can observe and manipulate “out there.” IIT offers an answer in what is called a complex and the associated Φ-structure ("phi structure"), which will become clearer as you read on.
As regards empirical validation, IIT provides several predictions and explanations with regard to consciousness in the human brain. The theory also offers a powerful basis to make inferences about, for example, consciousness in non-human animals and in artificial systems, and the place of consciousness in nature.
Contents
Contents
Technical Summary (click)
Technical Summary (click)
IIT aims to account for consciousness (or experience) in physical terms. The theory starts from phenomenology to identify the essential properties of experience (axioms); it then formulates these as physical properties (postulates)—understood as cause–effect power—and provides a mathematical formalism for assessing these properties. The tools of IIT allow us to assess the cause–effect power of any substrate of units in a state (e.g., a system of neurons, some firing and some silent), and to fully unfold the cause–effect structure which that substrate specifies.
In this way, IIT accounts for consciousness through its explanatory identity: an experience is identical to a specific cause–effect structure in both quality and quantity. This identity can be applied to account for why specific experiences feel the way they do (e.g., why space feels extended and time flowing). IIT offers a parsimonious explanation of empirical evidence, makes testable predictions, and permits inferences and extrapolations—for example, about the place of consciousness in nature and about ontology, free will, and ethics.
What does it mean to explain consciousness scientifically?
What does it mean to explain consciousness scientifically?
We all know consciousness intimately—so intimately, in fact, that we may struggle to define it. It is what fades every night when we fall into dreamless slumber, and what reappears every morning when we awake. Consciousness is our “inner world” in its entirety; it encompasses everything it is “like to be you” [1]—every experiential state you have had, are having now, and will have. Consciousness is experience itself, no matter what that experience is [2].
In IIT’s view, to explain consciousness scientifically means to answer two questions:
Why is experience present vs. absent?
Why do specific experiences feel the way they do?
The first question can be subdivided. First, why is consciousness present in humans when awake or vividly dreaming, and why does it fade in dreamless sleep or general anesthesia? Second, why should consciousness depend on some parts of our brain and not others?
As for the second question, if consciousness is present, it is always present in a specific way. Thus, a full account of consciousness cannot settle for mere presence vs. absence; it needs to explain why, say, the sound of a bell feels the unique way it does, and why it does not feel like, say, the taste of vanilla.
In approaching these questions, IIT carefully distinguishes three things: what needs to be explained, how it is explained, and how the explanation is validated [3]:
As outlined already, what needs to be explained is experience—its presence and its contents. This is a unique scientific challenge since science is presumed to be the objective study of objective phenomena. But in a science of consciousness, what must be explained is—by definition—not objective; it is subjectivity itself (see this FAQ). We may indeed discover objective correlates of consciousness—be they neural, functional, or behavioral. For example, my experience of a light on a screen may be accompanied by activity in visual cortical areas (neural correlate), by my directing attention to the light source (functional correlate), and by my pressing a button when I see it (behavioral correlate). However, these objective correlates are not what needs to be explained and cannot be swapped in as proxies. What needs to be explained, as said already, is rather the fact that I experience the bright light in the first place, and why exactly it feels “bright” and “light” as opposed to, say, the smell of lavender.
In contrast, how consciousness is explained, in IIT, must be in terms of objective, measurable properties and lead to testable predictions. Hence, IIT formulates the essential properties of experience in objective, physical terms, and then turns to empirical neuroscience to validate these conjectures.
How the explanation is validated must start in humans who can introspect and report their experiences. Only then can we confidently use the theory to extrapolate to unresponsive patients, non-human animals, and artificial systems, and ultimately to draw conclusions about the place of consciousness in nature.
Footnotes
[1] This is an oft-cited way of defining consciousness from Nagel, T. 1974. What is it like to be a bat? The Philosophical Review, 83(4), 435–450. Also see FAQ: What do you mean by consciousness?[2] In IIT, we often use the terms consciousness and experience interchangeably, in the sense that being conscious is synonymous with having an experience. But some nuances are worth mentioning. Consciousness refers to having experiences regardless of their content, while experience is always of something specific. Hence we describe a patient as “losing consciousness” but not as “losing experience,” and we might say “my experience changed” (from one content to another) but not “my consciousness changed.” Also note that some people use terms such as “subjective experience” or “phenomenal experience.” However, we prefer simply experience, since there is no such thing as an experience that is not subjective and phenomenal. Also see FAQ: What do you mean by "consciousness"?[3] For the philosophically minded, we must distinguish the explanandum, the explanans, and the validation.
The IIT Method
The IIT Method
The IIT method can be outlined in six steps. It starts from phenomenology to characterize the properties of experience, and then aims to account for these properties in physical terms. IIT builds its explanation of consciousness from first principles, and then turns to empirical neuroscience for validation. If its account of consciousness is correct, the IIT framework can be seen not just as a theory of consciousness but as an intrinsic ontology, offering principled answers to many perennial questions in metaphysics.
Click the steps on the image to learn more, and find longer explanations below.
Image sources
Robot image by James Sutherland from Pixabay (modified by cropping to isolate head and torso)Fetus image by Weslley Carvalho de Souza Weslleycs97 from Pixabay (slightly modified by reducing length of umbilical cord)Octopus Image by themarioman56 from DeviantArtComa patient image generated by OpenAI DALL·E, 24 October 2024Visual cortex with popout from Song et al. 2017, reproduced with permissionPosterior hotzone image from Siclari et al. 2017, reproduced with permission All other images are original.
1. Introspect to characterize phenomenal structure
1. Introspect to characterize phenomenal structure
The IIT method begins with recognizing that there is something it is like to be you: you are conscious, and your consciousness is immediately and irrefutably present for you (see 0th axiom). You can use introspection to notice and highlight properties of a given experience—a single moment, ”right here, right now” [1]. Step 1 in the IIT method is to use introspection to recognize which properties are true of every experience you have ever had or could ever have.
For instance, imagine you are the person in the image, lying on a bed with a book in your hands. This experience has countless features that distinguish it from other experiences. Yet it also has some properties that are common to every experience. For example, every experience is integrated: though you experience your body, the book, the bed, etc. as different components, your experience of them is at the same time a unitary whole. We call the properties that are true of every experience essential phenomenal properties. According to IIT, there are five such properties, captured in the five axioms of the theory. These axioms guide us in characterizing any experience as a phenomenal structure, which is what we aim to then account for scientifically (see previous section).
Some argue that introspection is “unscientific”—that it is a “subjective” technique that has no place in the “objective” practice of science. But the science of consciousness is unlike any other area of science in that the thing to be explained is experience itself. The “object” of study, so to speak, simply is subjectivity. Hence, introspection is our indispensable first step.
Footnotes
[1] Of course, introspection is an imperfect tool since it requires reflection and reasoning, which necessarily take us out of the experience “right here, right now.” Despite this limitation, introspection is our only option to begin to characterize the properties of experience. For more, see FAQ: If introspection requires reflection, how can we be sure that non-reflective experiences exist at all?
2. Formulate phenomenal properties in physical (causal) terms
2. Formulate phenomenal properties in physical (causal) terms
For each essential property of experience, IIT asks, What could account for this property in physical terms? Here, physical is meant in parsimonious, operational terms of cause–effect power—the ability to “take and make a difference.” We uncover cause–effect power through manipulating and observing things in the world—for example, a substrate of neurons, represented by iAbcdo here [1].
Step 2 is to formulate each essential phenomenal property (the axioms) as an essential physical property (as a postulate)—that is, as a property of the cause–effect power of the substrate of that experience. We use the term formulate in this step because it is not a formal deduction but rather an inference to a good explanation, and whether it is indeed “good” depends on various factors.
For example, we saw in step 1 that integration is an essential phenomenal property—that every experience is a unitary whole. In step 2, therefore, we ask, What physical (i.e., causal) property might account for this? A natural answer is that the cause–effect power of a substrate must also be a unitary whole. We can probe this by “cutting” the system in different ways to figure out whether it causally “hangs together” as a unitary whole [2]. These sorts of insights have been formalized in mathematical terms in the technical IIT papers (see IIT 4.0).
The upshot of step 2 is the notion of a Φ-structure (pronounced “phi structure”). Also called a cause–effect structure, a Φ-structure is a mathematically precise way to account for the properties of an experience in terms of the cause–effect power of the substrate of that experience (called a complex). The image here, for example, depicts the Φ-structure unfolded from the four-unit substrate Abcd (a subset of iAbcdo above).
Φ-structure unfolded from Abcd
Footnotes
[1] Just as we aim to introspect a single experience, we analyze the cause–effect power of systems in a given state—for example, in the figure here, unit A is ON (uppercase) and all others are OFF (lowercase). [2] More precisely, we assess whether a system is irreducible—whether a partition makes any difference to the intrinsic information specified by the system in its current state.
3. Establish the identity of a phenomenal structure and a Φ-structure
3. Establish the identity of a phenomenal structure and a Φ-structure
Step 3 is less of a “step” than a moment to pause, zoom out, and make a conjecture based on steps 1 and 2. This conjecture is the explanatory identity of IIT: an experience and a Φ-structure should correspond one-to-one in every respect. The quality of experience corresponds to the specific “shape” of the Φ-structure; in IIT, all quality is structure. And the quantity of experience corresponds to the amount of integrated information (Φ value) of the structure.
The explanatory identity between an experience (left) and a Φ-structure (right), unfolded from the neural substrate of consciousness, thought to be found in posterior cortex
The fundamental identity between an experience and a Φ-structure is what allows IIT to answer the two key why questions of a theory of consciousness (as outlined above): 1) An experience should be present when the integrated information of the Φ-structure specified by its substrate is high, and it should vanish when integrated information breaks down. 2) A specific experience feels the way it does because the Φ-structure specified by its substrate is composed the particular way it is.
If IIT is right, the neural substrate of consciousness in the human brain unfolds into a Φ-structure of unfathomable richness, as illustrated schematically here.
4. Account for contents of experience
4. Account for contents of experience
The fundamental identity of IIT is a mere conjecture unless it can be demonstrated that all properties of an experience indeed correspond one-to-one with the properties of its corresponding Φ-structure. Hence step 4 is to demonstrate how the tools of IIT allow us to account not only for the essential properties of experience but also for its accidental properties—those that vary across different experiences. For example, in the sample experience depicted above, dominant features include the sight of various objects (e.g., the walls, bed, book) occupying visual space (which has a left side, a right side, etc). All such accidental properties of experience need to be explained, and IIT proposes that we account for these properties as sub-structures in the corresponding Φ-structure.
Hence, step 4 can be thought of as iterating through steps 1–3, but now for the accidental properties of experience. (These iterations are indicated by the large circular arrow in the main graphic). For example, recent work has accounted for why space feels the way it does by analyzing the Φ-structure unfolded from a grid-like substrate (such as that found in posterior cortex). In a similar vein, current IIT projects aim to account for the experiences of time and objects, and future projects aim to account for local qualities such as color.
Each of these accounts demonstrates the explanatory power of the fundamental identity and bolsters the IIT framework as an overall good explanation of consciousness. For more, see Accounting for Contents of Experience.
5. Validate empirically in humans
5. Validate empirically in humans
To validate the fundamental identity scientifically, IIT then turns to the brains of humans who can introspect and report their experiences. The theory offers both explanations of well-established facts about the brain and predictions that can be empirically tested.
For example, IIT gives a principled explanation of a long-standing paradox: the cerebral cortex (or a part of it) is tightly linked to consciousness while the cerebellum is not, despite its having the lion's share of neurons. IIT answers this by contrasting the specific causal properties of the two brain architectures: parts of the cortex are organized in such a way that suggests it would unfold into a rich Φ-structure corresponding to an experience, while the cerebellum is not organized in such a way—it would rather unfold into many trivially small Φ-structures. As regards predictions, the most basic one is that integrated information should be high when consciousness is present and negligibly low when consciousness is seemingly absent. This prediction has found substantial evidence using proxy measures of Φ, opening a path toward developing a bedside “consciousness meter” to detect consciousness in unresponsive patients.
Some of IIT’s predictions are quite counterintuitive. For example, IIT predicts that neurons that are inactive (but not inactivated [1]) can contribute to an experience. This may be surprising since inactive neurons are usually assumed to contribute nothing to consciousness. Such predictions are currently being explored empirically.
As indicated by the dotted arrow in the main method graphic above, insights from step 4 about neural connectivity and physiology may indeed feed back into the previous steps to help refine the theory. For example, in our account of spatial experience, we make use of the knowledge that visual cortical areas are arranged as stacks of grids with relatively strong nearest-neighbor connections. Such insights not only help validate the account of space but have also been a useful intuition pump for how to construct the account in the first place.
Footnotes
[1] To say that the neurons are inactive but not inactivated means that they are not currently firing but they could fire if stimulated. Inactivated refers to neurons that will not fire even if stimulated.
6. Extrapolate beyond humans
6. Extrapolate beyond humans
As empirical evidence grows, the IIT account may become increasingly robust as an overall good explanation of both the quantity and quality of consciousness. Only then can we make inferences from a good explanation: principled inferences about consciousness beyond human subjects who can report their experience.
For example, the theory would guide us in assessing consciousness in unresponsive patients (with the support of adequate technology). The principles of IIT also let us predict that typical computers would not be conscious [1] (but an artifact that fulfills the postulates might be). Of course, any conclusions about consciousness beyond humans who can report their experiences will always remain an inference. However, those inferences become more compelling the more thoroughly IIT fulfills the criteria of a good explanation.
Finally, the theory opens the way for principled positions on millennia-old questions about, for example, the place of consciousness in nature, meaning, causation, free will, and even ethics.
Footnotes
[1] Findlay G, Marshall W, Albantakis L, Mayner WGP, Koch C, Tononi G. Dissociating Intelligence from Consciousness in Artificial Systems – Implications of Integrated Information Theory. In: Proceedings of the 2019 Towards Conscious AI Systems Symposium, AAAI SSS19; 2019 and forthcoming.
Frequently Asked Questions
Frequently Asked Questions
Ask or upvote a question about the IIT method
Ask or upvote a question about the IIT method
Please keep the discussion relevant to the contents of this page of the IIT Wiki. You'll find discussion threads at the bottom of the other pages too.
Cite this page
Cite this page
Hendren, Jeremiah, Matteo Grasso, Bjørn Erik Juel, and Giulio Tononi. "Overview: IIT as a Theory of Consciousness." IIT Wiki. Center for Sleep and Consciousness, University of Wisconsin–Madison. Last modified December 19, 2024. https://www.iit.wiki/overview (DOI: https://doi.org/10.5281/zenodo.14160283).
Archived versions of this page:30 June 2024
Page updated
Google Sites
Report abuse