Integrated Information Theory is testable enough to count as science, and unfalsifiable enough to make that answer uncomfortable. It generates predictions that researchers have run experiments on; its central identity claim — that consciousness is integrated information — is far harder to put at risk. In September 2023, 124 researchers signed a letter calling it pseudoscience.
Neither the letter nor the replies ended the argument, and a major 2025 experiment left it open. Most writing on this picks a side. What follows says what the disagreement is actually about — then applies the same test to this framework, which fares worse than IIT.
What does integrated information theory actually claim?
Start with what is not in dispute. IIT was proposed by the neuroscientist Giulio Tononi in a 2004 paper in BMC Neuroscience. Its central claim is unusually strong. Consciousness is not produced by integrated information, and is not merely correlated with it — it is integrated information. The theory defines a quantity, φ (phi), measuring how far a system's cause-effect structure is irreducible to the structures of its parts. A system whose whole carries more than its partitions do has high φ; one that decomposes without loss has φ of zero and, on the theory, no experience at all. More φ is meant to mean richer experience. That identity is what makes IIT a scientific candidate rather than one more metaphysical position: it converts "is this thing conscious?" into "what is this system's φ?" — a question about physical structure with, in principle, a number for an answer.
Two things follow. First, IIT is a physicalist theory: a physical system, arranged the right way, constitutes an experience. Explanation runs from structure to consciousness — the opposite direction from anything proposed on this site, and the Internet Encyclopedia of Philosophy's entry on IIT sets out the formalism in full. Second, because φ is a property of structure rather than of biology, the theory does not restrict consciousness to brains; anything with the right cause-effect architecture qualifies. That is what critics reached for, and where the trouble starts. The wider map of who claims what sits in the main theories of consciousness compared.
Why did 124 researchers call IIT pseudoscience?
On 16 September 2023 a preprint titled "The Integrated Information Theory of Consciousness as Pseudoscience" was posted to PsyArXiv by Stephen Fleming, Chris Frith, Hakwan Lau, Joseph LeDoux, Megan Peters, Heleen Slagter and colleagues, and signed by 124 researchers, many of them leading experimentalists. The charge has two parts, both worth stating at full strength.
The first is testability. What gives IIT its content is the identity between consciousness and φ — and φ, defined over the partitions of a system, cannot in practice be computed for a human brain. So the experiments that get run test downstream predictions, and a prediction several theories share does not put the identity claim at risk. The core is insulated: the parts that can be checked are not the parts carrying the claim.
The second is the reductio. If φ is what matters, systems nobody expects to be conscious can be conscious on the theory's own terms — an inactive grid of logic gates, or plants. The critics read this not as a fascinating implication but as a sign that something has gone wrong upstream, the way an absurd conclusion normally counts against a premise. That is legitimate reasoning, not a rhetorical trick.
How does the defense answer the pseudoscience charge?
The most prominent published reply came from the neuroscientist Anil Seth in Nautilus, on 27 September 2023, under the title "The Worth of Wild Ideas". His defense is not that IIT is true, but that IIT is science — and the distinction is the whole point.
The argument is Lakatosian: research programs are not refuted one experiment at a time, but judged over years by whether they keep generating new, testable predictions. Measured that way, IIT has produced a great deal of empirical work, some of it directly adversarial. Seth's sharpest line is also his most concessive: "even if it is very wrong, it can still be very useful." And "pseudoscience" is a heavy word, carrying an implication of bad faith an ambitious formal theory in an immature field may not have earned.
Be exact about what the reply does not do. It does not refute the letter: the testability objection is real, and Seth reframes the criterion rather than computing the number the critics say cannot be computed. The letter, equally, settled nothing — a signature is not a finding.
Did the 2025 Cogitate experiment settle it?
No. On 30 April 2025 the Cogitate Consortium published an adversarial test of global neuronal workspace theory and integrated information theory in Nature. The format is the interesting part: proponents of the two rival theories agreed in advance what the experiment should show if each were right — close to the opposite of the unfalsifiable practice the letter describes.
The result was inconclusive, with challenges for both. Neither IIT nor the global neuronal workspace account developed from Bernard Baars's and Stanislas Dehaene's work came out vindicated, and anyone reporting a winner is reporting something the paper does not say. What it established is narrower but real: IIT can be put in a room with a rival, committed in advance to predictions, and scored. That sits awkwardly beside a flat charge of untestability — and it leaves the sharper objection untouched, which was never that IIT makes no predictions but that its core identity claim is not what those predictions test.
If following an argument to where it stops being comfortable is the kind of thinking you want more of, the full case for this framework is set out in one place.
So is IIT science or pseudoscience?
This is where the genuinely open debate sits, and it is not about consciousness — which is why it has not resolved. The two camps are not looking at different evidence; they are applying different criteria for what counts as science — and the demarcation problem, the question of where that line falls, is itself unsettled among philosophers of science.
Run the falsifiability test inherited from Karl Popper — can this claim be exposed to a result that would sink it? — and IIT's identity thesis looks bad, because the quantity carrying it cannot be computed for the systems in dispute. Run the Lakatosian test — is the program generating new work and surviving contact with data? — and IIT looks like ordinary, ambitious science. Both verdicts are honestly reached.
The most defensible reading is a split one. IIT behaves like science in everything checkable: a published formalism, derived predictions, preregistered adversarial testing. Its identity claim behaves like metaphysics — a proposal about what consciousness is, which experiments illuminate but do not adjudicate. The pseudoscience charge is strongest against any suggestion that the identity thesis has been empirically supported, weakest against the program built on top of it.
Does the same falsifiability challenge land on Holopsychism?
Yes — and harder. This is where the piece crosses from reporting a dispute to stating a position, so the line should be visible: what follows is a proposal, not a result.
Holopsychism treats consciousness as pure potential — a structureless field of what can take form — and awareness as selection, the act that settles potential into one stable configuration. The brain, on this reading, is a receiver, filter and translator rather than a generator. That is the opposite direction of explanation from IIT, which is why IIT's difficulties are not this framework's gain: a physicalist theory in trouble is evidence about a physicalist theory, nothing more. And the panpsychist-sounding consequences the critics press are a reductio aimed at IIT, not a vote for anything argued here. Borrowing them would be intellectual shoplifting.
So apply the critics' own test. Holopsychism computes nothing — no φ, no derived prediction, no preregistered experiment, no number that could come back wrong. On a strict falsifiability criterion it is not a scientific theory at all, but a metaphysical proposal about the substrate, in the same category as idealism or cosmopsychism. It should not be graded on IIT's scale, or borrow IIT's credibility. Whether that category is worth occupying is taken up in is consciousness fundamental.
That does not leave the framework immune. The honest version of "how could this fail" has three parts. If illusionism is right — if phenomenal experience is a representational artifact rather than a fact needing explanation — the framework loses its subject matter, which is the strongest case against everything argued here. If a physicalist theory ever does what IIT is attempting — specifying from structure alone which systems are conscious, and getting it right — the framework becomes an unnecessary extra story: IIT succeeding is a defeat condition here, not a victory. And there is a standing debt. Nothing in the framework says which physical configurations count as receivers, or what makes one tune more finely than another — the very thing φ was built to quantify. IIT has a number and an argument about what it means; this has an argument and no number.
What the argument is really tracking
The label is the least useful part of this fight. "Science" and "pseudoscience" are verdicts on a whole enterprise; the thing worth measuring is exposure — how much of a theory stands where a result could hit it, and whether its advocates behave as though they want to find out. By that measure IIT's distinction is not that it is right, since nobody has shown that, but that it built a quantity, staked predictions on it, and let a rival camp help design the test. Holopsychism has not done that and cannot yet, and the claim to occupy the argued middle means nothing unless the framework accepts the scorecard it hands everyone else.
Frequently asked questions
Is integrated information theory accepted by scientists? It is among the most actively tested theories of consciousness, and among the most contested. In September 2023, 124 researchers signed a preprint calling it pseudoscience; defenders including Anil Seth replied that it is legitimate science. No consensus exists.
What is phi (φ) in integrated information theory? φ is IIT's measure of how far a system's cause-effect structure is irreducible to its parts. The theory identifies consciousness with it: high φ means richer experience, φ of zero means none at all.
Why do critics say IIT is unfalsifiable? Because the claim carrying the theory's content — that consciousness is φ — rests on a quantity that cannot in practice be computed for a human brain, so the experiments run test downstream predictions, not the identity claim.
Does IIT say everything is conscious? Not quite. IIT ties consciousness to cause-effect structure, so it implies unexpected systems — an inactive grid of logic gates, or plants — could have non-zero φ, which critics press as a reductio. It is not panpsychism, which places mentality in the micro-level constituents of matter.
Did the 2025 Cogitate experiment prove IIT wrong? No. The adversarial collaboration published in Nature on 30 April 2025 returned inconclusive results, with challenges for both IIT and global neuronal workspace theory, and named no winner.
If you want the whole argument laid out with its failure conditions attached rather than hidden, that is what the guide is for.