2. IIT: the definition-level claim, and the critical literature

#2.1 What IIT actually claims

• Tononi, An information integration theory of consciousness, BMC Neuroscience 5:42 (2004), doi 10.1186/1471-2202-5-42. The original definition-level claim: consciousness "corresponds to the capacity of a system to integrate information"; Φ is *"the amount of causally effective information that can be integrated across the informational weakest link of a subset of elements"*; a complex has Φ>0 and a higher Φ than any of its parts; and the quality of an experience is determined by the relationships among the complexes specified by the values of effective information among them. The paper claims it can be tested against neurobiological observations (thalamocortical vs cerebellar involvement, unconscious processing, loss in sleep/generalised seizures, time requirements).

• Albantakis, Barbosa, Findlay, Grasso, Haun, Marshall, Mayner, Zaeemzadeh, Boly, Juel, Sasai, Fujii, David, Hendren, Lang, Tononi, *Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in physical terms, arXiv 2212.14787. Verbatim: "In principle, the postulates can be applied to any system of units in a state to determine whether it is conscious, to what degree, and in what way. IIT offers a parsimonious explanation of empirical evidence, makes testable predictions, and permits inferences and extrapolations."* New in 4.0: *"a more accurate translation of axioms into postulates and mathematical expressions, the introduction of a unique measure of intrinsic information that is consistent with the postulates, and an explicit assessment of causal relations."*

On falsifiability, IIT 4.0 claims "testable predictions" but does not, in the abstract, address the unfolding argument or the substrate-invariance objection. I could not retrieve the full 53-page text (PDF only; no ar5iv rendering succeeded), so I cannot verify whether falsifiability is discussed in the body. Treat that as unresolved.

#2.2 Aaronson's expander/Vandermonde objection (primary source retrieved in full)

Scott Aaronson, "Why I Am Not An Integrated Information Theorist (or, The Unconscious Expander)", Shtetl-Optimized, scottaaronson.blog/?p=1799.

His own framing of the target. He distinguishes the Hard Problem from what he names the "Pretty-Hard Problem": *"to tell us which physical systems are associated with consciousness and which aren't, purely in terms of the systems' physical organization"*, tested against common-sense cases. His verdict, verbatim:

"In my view, IIT fails to solve the Pretty-Hard Problem because it unavoidably predicts vast amounts of consciousness in physical systems that no sane person would regard as particularly 'conscious' at all: indeed, systems that do nothing but apply a low-density parity-check code, or other simple transformations of their input data. Moreover, IIT predicts not merely that these systems are 'slightly' conscious (which would be fine), but that they can be unboundedly more conscious than humans are."

The construction: let S = F_p, V an n×n Vandermonde matrix over F_p, f(x) = Vx. Because every submatrix of V is full-rank, *"the normalized information integration has the same value—namely, the maximum value!—for every possible bipartition"*, so Φ is undefined (ambiguity over which bipartition to use). He then replaces V with W (first n/2 rows of V, each repeated twice) to force a unique minimising bipartition, giving Φ = (n/2)·log₂p:

"I've shown that my system—the system that simply applies the matrix W to an input vector x—has an enormous amount of integrated information Φ. Indeed, this system's Φ equals half of its entire information content. So for example, if n were 10¹⁴ or so—something that wouldn't be hard to arrange with existing computers—then this system's Φ would exceed any plausible upper bound on the integrated information content of the human brain. And yet this Vandermonde system doesn't even come close to doing anything that we'd want to call intelligent, let alone conscious!"

He also flags a robustness problem: to make Φ well-defined he had to decrease the intuitively-integrated information by a factor of 2 — *"I hope I'm not alone in fearing that this illustrates a disturbing non-robustness in the definition of Φ."* And he conjectures that approximating Φ is NP-hard (his own analysis; he notes the best upper bound he can show is in AM). In the comments he concedes the expander realisation has physical-space problems, but turns that around: *"whatever geometric difficulties there are in physically realizing expander graphs, those difficulties apply to the brain just as much as to artificial systems."*

I did not find a published, peer-reviewed rebuttal by IIT proponents to the Vandermonde/expander argument specifically. (I did not search exhaustively; see §7.)

#2.3 The unfolding argument

Doerig, Schurger, Hess, Herzog, *The unfolding argument: Why IIT and other causal structure theories cannot explain consciousness, Consciousness and Cognition* 72 (2019) 49–59, doi 10.1016/j.concog.2019.04.002.

⚠️ Correction to the brief: the arXiv ID you gave, 1810.11143, is not this paper. It resolves to "Smell Pittsburgh: Community-Empowered Mobile Smell Reporting System" (Hsu et al.), arXiv 1810.11143. The unfolding argument paper appears to have no arXiv preprint; the DOI above is the canonical locator. I retrieved the abstract via OpenAlex, not the full text.

Abstract, verbatim, including the punchline:

"Certain theories suggest that consciousness should be explained in terms of brain functions, such as accessing information in a global workspace, applying higher order to lower order representations, or predictive coding. These functions could be realized by a variety of patterns of brain connectivity. Other theories, such as Information Integration Theory (IIT) and Recurrent Processing Theory (RPT), identify causal structure with consciousness. For example, according to these theories, feedforward systems are never conscious, and feedback systems always are. Here, using theorems from the theory of computation, we show that this relation between causal structure and consciousness is either false or outside the realm of science."

The argument structure (as the title and abstract state it): any recurrent causal structure can be unfolded into a purely feedforward system that is functionally equivalent; therefore causal-structure theories must either (a) say the unfolded system is conscious too — abandoning the recurrence criterion, or (b) say it is not — which, given functional equivalence, makes consciousness behaviourally/epistemically undetectable, i.e. outside science.

#2.4 Making the dilemma concrete

Hanson & Walker, Formalizing Falsification for Theories of Consciousness Across Computational Hierarchies, arXiv 2006.07390 (Neuroscience of Consciousness 2021, niab014). Verbatim: they give *"a simple example of functionally equivalent machines realizable with table-top electronics that take the form of isomorphic digital circuits with and without feedback" and show "how IIT is simultaneously falsified at the finite-state automaton (FSA) level and unfalsifiable at the combinatorial state automaton (CSA) level." Their general criterion: "to avoid being unfalsifiable or already falsified scientific theories of consciousness must be invariant with respect to changes that leave the inference procedure fixed at a given level in a computational hierarchy."*

Kleiner & Hoel, Falsification and consciousness — referenced in the title of the reply below; I did not independently verify its arXiv ID, so I do not assert one. The reply is verified:

Ganesh, No Substitute for Functionalism — A Reply to 'Falsification & Consciousness', arXiv 2006.13664. Verbatim: *"we will prove that substitutions do not exist for a very broad class of Level-1 functionalist theories, rendering them immune to the aforementioned substitution argument." Note the arXiv comment field records that this replaced an earlier version titled "C-Wars: The Unfolding Argument Strikes Back"*.

Tsuchiya, Andrillon, Haun, *A reply to "the unfolding argument": Beyond functionalism/behaviorism and towards a science of causal structure theories of consciousness, Consciousness and Cognition* (2020), doi 10.1016/j.concog.2020.102877. Two IIT-affiliated authors (Tsuchiya, Haun) plus Andrillon. Abstract not available via OpenAlex — I have only the title, which already tells you the shape of the defence: it accuses the unfolding argument of presupposing functionalism/behaviourism. I could not verify the substance.

#2.5 The 2023 "pseudoscience" controversy — verified

"The Integrated Information Theory of Consciousness as Pseudoscience", 2023, published as a preprint on PsyArXiv, doi 10.31234/osf.io/zsr78.

• First author listed as "IIT-Concerned" (a collective pseudonym; OpenAlex records the affiliation as Department of Psychology, McGill University) — with named co-authors including Stephen M. Fleming, Chris D. Frith, Melvyn Goodale, Hakwan Lau, Joseph E. LeDoux, Alan L. F. Lee, Matthias Michel, Adrian M. Owen, Megan A. K. Peters, Heleen A. Slagter.

• Abstract in full, verbatim: *"The media, including news articles in both Nature and Science, have recently celebrated the Integrated Information Theory (IIT) as a 'leading' and 'empirically tested' theory of consciousness. We are writing as researchers with some relevant expertise to express our concerns."*

• The OSF landing page returned only a shell ("OSF") to web_fetch, so I could not read the body or verify the full signatory list or whether it was formally submitted to a journal. Treat the item as a real preprint with a verified title/DOI/author list, but do not treat its detailed contents as verified.

• Keith Frankish's response is verified: *Integrated Information Theory: Pseudoscience or appropriately anomalous science?*, 2023, doi 10.31234/osf.io/uscwt. Verbatim abstract: *"The integrated information theory of consciousness (IIT) has recently been branded pseudoscience. Its methodology is certainly anomalous, but does that make it pseudoscience or an appropriately anomalous science of an anomalous phenomenon? This article considers the question and argues that the accusation has merit. At best, IIT is a metaphysical theory mispresented as science."*

#2.6 The empirical verdict against IIT (and against GNWT)

Ferrante, Gorska-Klimowska, Henin, Hirschhorn, Khalaf, Lepauvre, Liu, Richter, Vidal, Bonacchi, Brown, Sripad, Armendariz, Bendtz, Ghafari, Hetenyi, Jeschke, Kozma, … Blumenfeld, Boly, Chalmers, Devore, Fallon, de Lange, Jensen, Kreiman, Luo, Panagiotaropoulos, Dehaene, Koch, Pitts, Mudrik, Melloni, *Adversarial testing of global neuronal workspace and integrated information theories of consciousness, Nature* (2025), doi 10.1038/s41586-025-08888-1. Open access; abstract retrieved in full. Pre-registered adversarial collaboration, n=256, fMRI + MEG + iEEG, 6 theory-impartial laboratories.

Verbatim results and interpretation:

"Our results align with some predictions of both IIT and GNWT, while substantially challenging key tenets of both theories. For IIT, a lack of sustained responses within the posterior cortex contradicts the claim that network connectivity specifies consciousness. GNWT is challenged by general lack of ignition at offset and limited representation of certain conscious dimensions in prefrontal cortex. These challenges extend to other theories of consciousness that share the predictions tested here. Beyond the theories, we present an alternative approach to advance cognitive neuroscience through principled, theory-driven, collaborative research and highlight the need for a quantitative framework for systematic testing and theory building."

Also relevant: Corcoran, Haun, Dorman, Tononi, Friston, Pennartz, INTREPID Consortium, *Integrated information and predictive processing theories of consciousness: An adversarial collaborative review*, arXiv 2509.00555 (Neuroscience & Biobehavioral Reviews 187:106742, 2026) — IIT vs Neurorepresentationalism vs Active Inference, with pre-specified outcomes that would "support, refute, or challenge" each theory.

★ Assessment: IIT is the theory the critical literature most clearly marks as unusable as an engineering target. Three independent lines converge: (i) Aaronson's construction shows large Φ in systems nobody would call conscious and shows the Φ definition is non-robust in its normalization; (ii) the unfolding argument + Hanson & Walker show causal-structure theories are either falsified or unfalsifiable depending on the level of description; (iii) a large group of senior consciousness researchers publicly labelled the theory's presentation as pseudoscientific, and an IIT-sympathetic philosopher (Frankish) agreed the accusation "has merit" while reframing IIT as metaphysics. Add the practical problem that Φ is not computable for realistic systems (Aaronson's NP-hardness conjecture), and the conclusion is: do not build Darlin's test suite on Φ or on any "causal structure" proxy. IIT 4.0's own framing is the strongest counterpoint — it claims to be formulated purely operationally, and Phua's negative result (below) is a concrete caution against porting PCI-A to engineered agents.

---