Formalizing consciousness: a comparative mathematical analysis of leading theories.
Tonglin Yan, Jules Tsukahara, Kenneth Williford, David Rudrauf, Grégoire Sergeant-Perthuis
Biological Cybernetics November 14, 2026 DOI: 10.1007/s00422-026-01068-y (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper Peer reviewed |
|---|---|
| Keywords | Active inference Computational modeling Consciousness theories Integrated information theory Projective consciousness model |
| Key points | Argues that Integrated Information Theory, Global Workspace Theory, Active Inference, higher-order representation approaches, and the Projective Consciousness Model can be reformulated in a shared mathematical language at the computational and algorithmic levels, enabling rigorous comparison. Contends that some apparent disagreements among these frameworks reflect differences in explanatory level rather than direct theoretical contradiction, and proposes common testbed models to study integration, broadcasting, higher-order access, and perspectival transformations within the same agent architecture. |
Abstract
This article provides a unified and accessible exposition of major mathematically articulated frameworks for consciousness: Integrated Information Theory (IIT), Global Workspace Theory (GWT), Active Inference (AIF), higher-order representation (HOR) approaches, and the Projective Consciousness Model (PCM). Building on probability theory, information theory, and geometric structures, we reformulate their core mechanisms in a shared mathematical language at the computational and algorithmic levels, following Marr's hierarchy. This enables rigorous comparison across theories that are often treated as conceptually incommensurable. Our analysis identifies shared formal tools, differences in explanatory scope, and specific questions for computational comparison. GWT emphasizes global availability; IIT emphasizes irreducible causal integration; Active Inference and HOR emphasize hierarchical inference and conscious access under uncertainty; PCM emphasizes the perspectival geometry of conscious experience. We assess computational tractability across frameworks and show that some apparent disagreements reflect differences in explanatory level rather than direct theoretical contradiction. We use this unified formal treatment to outline a route toward future computational comparisons: common testbed models in which integration, broadcasting, higher-order access, and perspectival transformations can be studied within the same agent architecture. The goal is to make overlaps, divergences, and empirical commitments across theories more explicit, thereby supporting more systematic model comparison in consciousness science.