How AI Becomes Metasentient
Vance Ashley Woodward
Preprints.org October 21, 2026 preprint DOI: 10.20944/preprints202610.2066.v1 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper |
|---|---|
| Key points | Argues that sentience requires a continuously updated world model that includes the system and its own modeling activity, and suspects that this organization, sufficiently developed and integrated, is also sufficient for sentience. Proposes that metasentience develops when a system can examine and revise its experience and judgments, and offers three tests of whether self-modeling and reflection improve behavior, while noting that inferring experience requires evidence connecting the machinery to feeling and that the hard problem remains. |
Abstract
Artificial intelligence will become sentient, able to feel, and then metasentient, able to reflect on its own experience, if the organization proposed here produces feeling. I claim that sentience requires a continuously updated world model that includes the system and its own modeling activity. I suspect that this organization, sufficiently developed and integrated, is also sufficient. Metasentience develops when the system can examine and revise its experience and judgments. The argument draws on how brains predict what they perceive, model physical events, and learn by acting, and on AI systems that learn in similar ways. Practical demands for memory, adaptation, and reliable action give developers reasons to build the proposed organization even without intending to create consciousness. I describe its components and propose three tests of whether self-modeling and reflection improve behavior. Those tests address the functional machinery; inferring experience also requires evidence connecting that machinery to feeling. The hard problem remains: why does this physical activity feel like anything?Building machines that act more competently could also build machines that feel.