Consciousness in AI Systems
Simon Goldstein, Cameron Domenico Kirk-Giannini
October 20, 2026 DOI: 10.1093/9780197840375.003.0010 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper |
|---|---|
| Key points | Argues that, under global workspace theory's necessary and jointly sufficient conditions for phenomenal consciousness, existing language model agent architectures could be modified in straightforward ways to produce conscious systems, and describes the functional organization such a system would have. Also assesses whether large language models and reinforcement learners are or could be made conscious by these criteria. |
Abstract
Abstract This chapter applies to AIs the set of necessary and jointly sufficient conditions for a system to be phenomenally conscious according to global workspace theory (GWT) developed in Chapter 7. It begins by considering how language model agents fare by the lights of these conditions. It argues that existing language model agent architectures could be modified in straightforward ways to produce systems that would be conscious according to GWT, and it describes in detail the functional organization of such a system. Then, it turns to LLMs and reinforcement learners, considering what the conditions say about whether they are or could be made conscious.
Citations in the library
Cites 13
- Ten Problems of Consciousness The MIT Press eBooks November 17, 1995
- The Epistemic Role of Consciousness Philosophy and Phenomenological Research August 2, 2019
- The search for invertebrate consciousness. Noûs March 1, 2022
- Identifying indicators of consciousness in AI systems. Trends in Cognitive Sciences June 1, 2026
- Tests of Animal Consciousness are Tests of Machine Consciousness Erkenntnis April 1, 2025
and 8 more in the library