Computational Intractability and the Falsifiability Problem in IIT 4.0

The central computational requirement of IIT 4.0 generates a falsifiability problem that mathematical refinement cannot dissolve. Integrated Information Theory 4.0 (IIT) makes an explanatory identity claim. Consciousness *is* the Φ-structure unfolded from a maximal substrate, the candidate substrate whose system integrated information, computed at its minimum partition, is maximal (Albantakis et al., 2023). The number of partitions of an n-element set is the Bell number B(n), which grows super-exponentially. For a 302-neuron *Caenorhabditis elegans* nervous system, B(302) is approximately 4.79 × 10^457, which exceeds the 10^120 elementary operations the universe can have performed over its history (Lloyd, 2002) by hundreds of orders of magnitude. The count passes physical scale well below that: at n = 103, B(n) ≈ 1.26 × 10^120 already exceeds the bound, so any n at cortical scale is past the bound on any estimate of it. A separate gap sits beside the computational one: the exclusion postulate stipulates exactly one maximal substrate, and Albantakis et al. (2023) give no proof that the φs maximization across candidate substrates has a unique global solution, which would remain open even for a system small enough to compute exhaustively. Four consequences follow. Exact Φ values cannot be computed for any biological nervous system of more than 102 neurons. Every published claim that IIT has been empirically tested in a real brain rests on heuristic approximations that no publication has validated against ground-truth Φ. When a prediction conflicts with measurement, the gap between approximation and theoretical Φ provides an indefinite range of post hoc adjustments that protect the theory from refutation. On IIT's own physical-realism commitment, no physical substrate can instantiate the Φ-structure as a real property of itself, since the object's specification requires resources no physical system can supply. The paper formalizes this argument, examines IIT's own treatment of approximation, and concludes that IIT 4.0 is empirically untestable for any system on which a substantive theory of consciousness is required to apply. Drawing on the comprehensive critique in *Phi Without Phenomenology: The Compositional Failures of Integrated Information Theory (of "Consciousness")* (Stewart, 2026o), the analysis treats this as a structural property of the theory's mathematical commitments. **Keywords:** integrated information theory, IIT 4.0, Phi, Bell number, combinatorial explosion, computational intractability, falsifiability, approximation, proxy measures, partition lattice, physical realism, intrinsicality, theories of consciousness

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24409491
Primary Topic
Philosophy and Theoretical Science
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Computational Intractability and the Falsifiability Problem in IIT 4.0

Arthur Stewart
Zenodo (CERN European Organization for Nuclear Research)
Philosophy and Theoretical Science
preprint

Computational Intractability and the Falsifiability Problem in IIT 4.0

Arthur Stewart
preprint en

Abstract

The central computational requirement of IIT 4.0 generates a falsifiability problem that mathematical refinement cannot dissolve. Integrated Information Theory 4.0 (IIT) makes an explanatory identity claim. Consciousness *is* the Φ-structure unfolded from a maximal substrate, the candidate substrate whose system integrated information, computed at its minimum partition, is maximal (Albantakis et al., 2023). The number of partitions of an n-element set is the Bell number B(n), which grows super-exponentially. For a 302-neuron *Caenorhabditis elegans* nervous system, B(302) is approximately 4.79 × 10^457, which exceeds the 10^120 elementary operations the universe can have performed over its history (Lloyd, 2002) by hundreds of orders of magnitude. The count passes physical scale well below that: at n = 103, B(n) ≈ 1.26 × 10^120 already exceeds the bound, so any n at cortical scale is past the bound on any estimate of it. A separate gap sits beside the computational one: the exclusion postulate stipulates exactly one maximal substrate, and Albantakis et al. (2023) give no proof that the φs maximization across candidate substrates has a unique global solution, which would remain open even for a system small enough to compute exhaustively. Four consequences follow. Exact Φ values cannot be computed for any biological nervous system of more than 102 neurons. Every published claim that IIT has been empirically tested in a real brain rests on heuristic approximations that no publication has validated against ground-truth Φ. When a prediction conflicts with measurement, the gap between approximation and theoretical Φ provides an indefinite range of post hoc adjustments that protect the theory from refutation. On IIT's own physical-realism commitment, no physical substrate can instantiate the Φ-structure as a real property of itself, since the object's specification requires resources no physical system can supply. The paper formalizes this argument, examines IIT's own treatment of approximation, and concludes that IIT 4.0 is empirically untestable for any system on which a substantive theory of consciousness is required to apply. Drawing on the comprehensive critique in *Phi Without Phenomenology: The Compositional Failures of Integrated Information Theory (of "Consciousness")* (Stewart, 2026o), the analysis treats this as a structural property of the theory's mathematical commitments. **Keywords:** integrated information theory, IIT 4.0, Phi, Bell number, combinatorial explosion, computational intractability, falsifiability, approximation, proxy measures, partition lattice, physical realism, intrinsicality, theories of consciousness

Zenodo (CERN European Organization for Nuclear Research)
Philosophy and Theoretical Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon