Neuroscience of Consciousness
January 1, 2023
Jake R. Hanson, Sara I. Walker
Integrated Information Theory (IIT) 3.0, a leading theory of consciousness, defines a mathematical measure Φ meant to quantify consciousness. This work demonstrates that Φ is not well-defined because its value is non-unique. An algorithm that computes all possible Φ values for a given system, strictly following the theory's definition, reveals that published Φ values are arbitrarily selected from many equally valid alternatives. Crucially, both high and low Φ values can be predicted simultaneously for the same system, making it impossible to decide whether that system is conscious under the current formulation of IIT.
Neuroscience of Consciousness
January 1, 2021
Jake R. Hanson, Sara I. Walker
A concrete example of isomorphic digital circuits—one with feedback and one without—demonstrates that Integrated Information Theory (IIT) is simultaneously falsified at the finite-state automaton level and unfalsifiable at the combinatorial-state automaton level. The example shows how IIT permits functionally identical systems to have differences in predicted consciousness, providing a concrete refutation of the theory. The authors argue that any theory of consciousness based on a physical system's causal structure may already be falsified even without experimental refutation, and propose that to avoid being already falsified or unfalsifiable, scientific theories of consciousness must be invariant with respect to changes that leave the inference procedure fixed at a particular level in a computational hierarchy.
arXiv Preprint Archive
June 12, 2020
Jake R. Hanson, Sara I. Walker
A concrete electronic-circuit example shows that Integrated Information Theory (IIT) is simultaneously falsified at one level of analysis (finite-state automaton) and unfalsifiable at another (combinatorial state automaton). The authors argue that to avoid being unfalsifiable or already falsified, scientific theories of consciousness must be invariant under changes that leave the inference procedure fixed at a given level in a computational hierarchy.
arXiv Preprint Archive
August 3, 2019
Jake R. Hanson, Sara I. Walker
A mathematical analysis shows that Integrated Information Theory (IIT), a leading theory of consciousness, relies on a measure called Φ that can differ between two physically isomorphic systems—systems identical in size, function, and causal structure—solely due to a permutation of the binary labels representing internal functional states. This means a system with Φ > 0 (deemed conscious) and an isomorphic system with Φ = 0 (a philosophical zombie) can be mathematically identical except for arbitrary labeling. The finding challenges IIT's assumption that feedback is necessary for consciousness and suggests any quantitative theory of consciousness should be invariant under such isomorphisms to avoid epistemological problems.
Jake R. Hanson, Sara I. Walker
preprint
Integrated Information Theory (IIT) is a leading mathematical theory of consciousness, but its core measure, Φ, is not well-defined. The value of Φ is neither unique nor specific due to degeneracies in the optimization routine used to calculate it, leading to ambiguous determinations of consciousness. Analyzing a simple AND+OR logic gate system yields 83 non-unique Φ values spanning much of the possible range. Applying a new Python package, PyPhi-Spectrum, to recently published calculations shows that virtually all Φ values are chosen arbitrarily from non-unique sets, often including both conscious and unconscious predictions. Proposed solutions fail to resolve the degeneracy problem, and the theory requires reformulation to avoid these issues.