Multi-Theory Consciousness Architecture: Integrating GWT, AST, HOT, FEP, IIT, RPT, and BLT in a Unified Computational Framework
Zenodo (CERN European Organization for Nuclear Research) March 15, 2026 DOI: 10.5281/zenodo.19030129 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper Peer reviewed |
|---|---|
| Keywords | Workspace Python programming language Testbed Inference Embodied cognition Schema genetic algorithms Scripting language Operationalization Artificial intelligence Cognitive science Theoretical computer science Information theory Icub Human–computer interaction Grasp Artificial consciousness Function biology |
| Key points | Proposes that the Multi-Theory Consciousness framework, implementing seven consciousness theories as interacting modules, reveals measurable cross-theory dependencies when modules are disabled, serving as a useful research instrument rather than a conscious system. |
Abstract
We present the Multi-Theory Consciousness (MTC) framework: an open-source Python implementation of seven major consciousness theories operating as interacting modules within a single architecture. The framework implements Global Workspace Theory (Baars, 1988), Attention Schema Theory (Graziano, 2013), Higher-Order Thought Theory (Rosenthal, 2005), the Free Energy Principle (Friston, 2010), Integrated Information Theory (Tononi, 2008), Recurrent Processing Theory (Lamme, 2006), and Beautiful Loop Theory (Laukkonen, Friston & Chandaria, 2025), with Damasio's three-layer model as an integrative embodied layer. Three neural substrates — a spiking neural network, a liquid state machine, and a hierarchical temporal memory — provide the computational medium. A 20-indicator assessment framework inspired by Butlin et al. (2023) measures architectural function across all theories, with noise normalization for honest scoring and ablation studies for measuring cross-theory dependencies. The central finding is that the theories interact in measurable ways: disabling one module changes assessment scores for other theories. Prediction error signals from active inference shape workspace competition; attention strength modulates precision weighting; workspace access gates meta-representation. We do not claim the framework is conscious. We claim it is a useful research instrument — a controlled testbed where consciousness theories can be implemented, measured, and compared. The complete codebase (~25,000 lines, 400 tests) is released under the Apache 2.0 license and runs on consumer hardware. Repository: https://github.com/WhiteLotusLA/multi-theory-consciousness