Consciousness as Constrained Communication: Computer Study
Aman Chawla, Arun Raveendran, Salvatore Morgera
Zenodo (CERN European Organization for Nuclear Research) May 7, 2026 preprint DOI: 10.5281/zenodo.20060105 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper |
|---|---|
| Topics | Philosophy of mind |
| Key points | The authors report that varying noise strength in a competing-module workspace architecture produces achievable coherence-diversity pairs forming a concave hull, analogous to multiple access channel capacity regions. They propose this structure as a candidate signature of integrative cognition. |
Abstract
We study the application of Baar’s Global Workspace Theory (GWT) to large language model (LLM)-based systems. Specifically, we are interested in the joint behavior of two metrics - system coherence and system diversity. LLM-based systems are chosen since they mimic the human mind. We construct an architecture in which multiple specialist modules compete for access to a shared workspace. By introducing a disruptor (noise) module at varying strengths, we empirically characterize the system’s achievable (coherence, diversity)-pairs and find a concave hull structure analogous to multiple access channel (MAC) capacity regions studied in multiuser information theory. This characteristic structure is proposed as a candidate signature of integrative cognition.