Integrated Information Theory: What It Claims and Why It Is Contested

IIT does not ask first what the brain can report. It asks what any conscious experience must be like, then tries to derive the physical structure capable of being that experience. The ambition is enormous. So are the objections.

Start with the experience, not the brain

Look at the room around you.

Your experience exists.

It has a particular structure.

The lamp is not the wall.

The wall is not the sound outside.

Yet the whole scene belongs to one experience.

It contains distinctions.

It has boundaries.

It is specific.

And from your own point of view, it exists for itself.

Integrated Information Theory, or IIT, begins here.

Not with behavior.

Not with report.

Not even initially with neurons.

It asks:

What properties are necessarily true of experience simply because experience exists?

Then it attempts something unusually ambitious.

It tries to derive what kind of physical cause-and-effect structure would have to exist for those experiential properties to be real.

If Global Workspace Theory asks what information becomes globally usable,

IIT asks what physical system could intrinsically be an experience at all.

The theory works from axioms to postulates

IIT 4.0 formalizes several proposed axioms of experience.

The terminology is technical, but the intuition can be simplified.

Experience is said to have properties including:

Intrinsic existence

Experience exists from its own perspective, not merely for an outside observer.

Composition

Experience contains structured distinctions and relations.

Information

An experience is specific. Seeing this scene excludes countless other possible scenes.

Integration

Experience is unified. It is not merely a bag of independent mini-experiences.

Exclusion

A conscious experience has definite borders and a definite grain rather than being every possible overlapping combination at once.

IIT then proposes corresponding physical requirements.

A physical system capable of consciousness must possess a matching intrinsic cause-effect structure.

This is where the theory becomes mathematical.

The famous symbol is phi

IIT is associated with Φ, pronounced phi.

Popular summaries often say:

“Phi measures consciousness.”

That is too simple.

Across different generations of IIT, Φ-like quantities have attempted to characterize forms of irreducibility or integrated information.

IIT 4.0 is more elaborate than one scalar number.

The theory identifies a complex: a set of interacting units whose intrinsic causal structure is maximally irreducible under the theory’s rules.

The detailed cause-effect structure of that complex is then proposed to correspond to the quality of experience.

So the claim is not only:

more integration = more consciousness.

The stronger claim is:

the structure of experience is identical, in the relevant explanatory sense, to a particular intrinsic cause-effect structure.

That is much more radical.

IIT tries to solve a problem other theories often postpone

Many neuroscience theories identify mechanisms correlated with conscious access.

Critics can then ask:

Why should that mechanism produce experience?

Why should global broadcasting feel like anything?

Why should recurrent activity feel like anything?

IIT attempts to avoid that explanatory gap by starting with phenomenology itself.

If the physical structure is derived from what experience necessarily is, the theory hopes not merely to correlate brain activity with consciousness but to identify the physical substrate with the experience.

Supporters see this as one of IIT’s deepest strengths.

Critics see it as the source of some of its deepest problems.

The posterior cortex became a major empirical prediction

IIT proponents have argued that the neural substrate of many conscious contents should be located primarily in a posterior cortical “hot zone,” involving temporal, parietal and occipital regions.

This contrasts with strong frontally centered versions of Global Neuronal Workspace Theory.

The distinction made empirical comparison possible.

If conscious visual content depends on intrinsic posterior causal structure, certain patterns should persist there while the experience persists.

That prediction became central to the 2025 adversarial collaboration.

The 2025 experiment did not crown IIT

The Cogitate Consortium preregistered predictions from both IIT and GNWT.

For IIT, one major prediction involved sustained connectivity within posterior cortical areas as conscious visual content remained present.

The study found some results compatible with IIT.

But a key predicted pattern of sustained synchronization within posterior cortex was not observed.

The authors stated that this challenged a central tested tenet of the theory’s proposed neural implementation.

That is scientifically healthy.

A theory should meet data capable of hurting it.

But the result did not test every philosophical or mathematical claim made by IIT.

This is important because theory testing can target different layers.

A failed neural prediction may challenge an implementation while leaving the abstract formalism untouched.

That flexibility is both normal in science and a source of controversy.

The testability problem sits at the center of the debate

A theory can survive a failed prediction by revising an auxiliary assumption.

All scientific theories do this sometimes.

The danger appears when the central theory becomes insulated from possible failure.

Critics of IIT argue that some of its strongest claims are difficult to connect to decisive empirical tests.

If the formal identity between experience and intrinsic causal structure can survive almost any neural result by changing the proposed implementation, what would falsify it?

IIT proponents respond that the theory does generate empirical predictions and has already motivated experiments.

This disagreement is not merely personal.

It is about what level of a theory must be empirically vulnerable to count as scientific explanation.

Then came the 2023 “pseudoscience” controversy

In September 2023, an open letter signed by 124 researchers argued that IIT should be labeled pseudoscience.

The accusation was explosive.

The signatories objected to what they saw as media overstatement, insufficient testability and extraordinary implications not adequately constrained by evidence.

Other consciousness researchers, including some who were critical of IIT itself, objected strongly to the label.

They argued that “pseudoscience” was disproportionate, polarizing or conceptually unhelpful.

Nature covered the dispute because it exposed a broader problem in consciousness science.

Researchers were no longer arguing only about a theory.

They were arguing about the boundary of science itself.

DarkBrain should preserve that distinction.

Some researchers called IIT pseudoscience. The field did not reach a consensus that IIT is pseudoscience.

Why the implications worry critics

IIT is substrate neutral in an unusual way.

A system does not need to behave intelligently to possess intrinsic causal structure under the theory.

This can generate counterintuitive implications.

Some simple networks may receive nonzero consciousness attribution.

Conversely, a sophisticated digital system could in principle perform impressive computations without the kind of integrated intrinsic causal structure IIT requires.

This means intelligence and consciousness can come apart sharply.

The theory therefore has implications for:

• artificial intelligence;
• brain organoids;
• animals;
• altered brains;
• simple engineered systems.

Those implications matter ethically.

But they should not be treated as established facts simply because they follow from one theory.

IIT is often associated with panpsychism, but the relation needs precision

Panpsychism proposes, broadly, that consciousness or consciousness-like properties are fundamental and widespread.

IIT is sometimes described as panpsychist because it can attribute consciousness to systems far simpler than humans.

But IIT is not identical to every form of panpsychism.

Its criterion is not:

everything is conscious.

It is:

systems with the relevant intrinsic cause-effect structure possess consciousness to the extent and in the form specified by that structure.

Still, the wide distribution of possible consciousness is one reason IIT intersects philosophically with panpsychist debate.

That is why DBK-067 is a natural Related Article.

The combination problem reappears in a new form

If small systems have some integrated information, why do they not each produce separate consciousnesses inside your brain?

IIT uses exclusion principles to identify a maximally irreducible complex.

The dominant complex excludes overlapping alternatives at that grain.

This is meant to prevent an explosion of simultaneous overlapping consciousnesses.

Whether that solution succeeds is heavily debated.

The technical structure answers the problem internally.

Critics can still ask whether the formal rule is independently justified by experience and empirical evidence.

Computing Φ is not a simple brain scan

Another popular misunderstanding is that scientists can put someone in an fMRI scanner, calculate one phi number and read off consciousness.

For realistic systems, exact IIT calculations can become computationally prohibitive.

Empirical studies often use measures inspired by complexity or integration rather than literal full IIT 4.0 calculations.

For example, perturbational complexity measures used in anesthesia and disorders of consciousness are sometimes discussed near IIT.

But a useful complexity marker is not the same thing as validating the full theory.

This distinction is critical.

Evidence for complexity as a correlate of consciousness is not automatically evidence for every IIT postulate.

Why supporters remain interested

IIT does several things many rival theories avoid.

It takes phenomenology seriously.

It tries to explain unity.

It distinguishes intrinsic causal power from mere input-output behavior.

It offers a formal vocabulary.

It produces unusual predictions.

It forces researchers to ask whether a perfect behavioral simulation could still lack consciousness.

Even critics often agree these are valuable questions.

A theory can be productive without being correct.

Why critics remain unconvinced

The main objections are varied.

Empirical testability

Can the theory’s core identity claim be falsified?

Phenomenology-to-physics inference

Why should the proposed axioms uniquely imply IIT’s physical postulates?

Counterintuitive attributions

Should simple systems really count as conscious?

Computational feasibility

Can the relevant quantities be calculated for realistic brains?

Theory flexibility

Do failed biological predictions genuinely reduce confidence in the core theory?

These are serious objections.

None can be answered by saying the theory has impressive mathematics.

Mathematical precision and empirical truth are separate achievements.

The 2025 adversarial study gave both sides something uncomfortable

IIT supporters could point to findings that were not a victory for GNWT.

GNWT supporters could point to failed IIT predictions.

Critics of both could point out that neither framework received decisive confirmation.

That may be the most important result.

Consciousness science is moving from:

Which theory has the most papers?

toward:

Which theory makes risky predictions when competitors agree on the test?

That is progress even when the answer is “neither won.”

The DarkBrain model: phenomenology, formal bridge, prediction, vulnerability

Phenomenology

What properties of experience does IIT begin from?

Formal bridge

How are those properties translated into physical postulates?

Prediction

What measurable differences follow?

Vulnerability

What result would genuinely reduce confidence in the theory?

The last question should follow every theory in this Collection.

Not only IIT.

What the evidence does not support

The evidence does not support these universal claims:

• IIT has been proven;
• IIT has been disproven;
• phi is a simple consciousness meter;
• complexity measures clinically validate IIT;
• the 2023 pseudoscience label represents scientific consensus;
• the 2025 adversarial study confirmed IIT;
• the 2025 study destroyed IIT;
• IIT proves current AI systems are conscious;
• IIT proves simple objects are conscious;
• mathematical elegance establishes metaphysical truth.

IIT remains one of the field’s most ambitious and contested frameworks.

The answer

What does IIT claim?

That the essential structure of consciousness can be derived from experience itself and matched to an intrinsic, irreducible cause-effect structure in a physical system.

Why is it contested?

Because connecting that extraordinary identity claim to decisive empirical tests is difficult, its implications are sometimes counterintuitive, and important preregistered predictions have faced challenges.

That is not a reason to stop studying IIT.

It is the reason the debate matters.

Continue exploring

Next: Higher-Order Theories: Does Awareness Require Representation?

The next family of theories asks a deceptively simple question: is a mental state conscious because the brain represents the world, or because the brain also represents itself as being in that state?