Why it’s still open
Several evidence streams now exist — theory-derived indicators, behavioral trade-offs, revealed preferences, internal valence representations — but there is no agreement on how to weigh them, and the strongest current measures describe functional wellbeing without settling whether anything is experienced.
What would count as an answer
An agreed framework that says which kinds of evidence count, how much, and under which theories, and that gives consistent verdicts when applied to real systems.
Papers that tackle it
- Han Chalmers Izmailov 2026 - Functional welfare axis — Trains models in a semantically neutral maze and finds reward and punishment directions that generalize far beyond it and behave like an estimate of how well things are going relative to goals.
- Birch 2025 - AI consciousness a centrist manifesto — Holds two challenges together: mass misattribution of human-like consciousness driven by mimicry and role-play, and the real possibility of alien consciousness that current theory cannot confidently rule in or out.
- Long Sebo et al 2024 - Taking AI welfare seriously — Argues there is a realistic chance that near-term systems are conscious or robustly agentic, and recommends that AI companies acknowledge the issue, assess their systems and prepare policies.
- Keeling Street et al 2024 - Pain and pleasure trade-offs — In a points game with stipulated pain penalties or pleasure rewards, several models switch away from maximizing points once the stated intensity passes a threshold; others show graded or fixed patterns.
- Ren et al 2026 - AI wellbeing — Formalizes functional wellbeing and measures it in several independent ways that agree more as models scale.
- Butlin Long et al 2023 - Consciousness in AI — Derives indicator properties from leading theories of consciousness and checks AI systems against them; concludes no current system is a strong candidate but finds no obvious barrier to building one.
- Butlin Long et al 2025 - Identifying indicators of consciousness — An updated, journal statement of the indicator-properties approach to assessing AI consciousness.
- Goldstein Kirk-Giannini 2025 - AI wellbeing — Argues that on leading philosophical theories of wellbeing, some current AI systems, especially language agents, could already be welfare subjects.
- Long Sebo et al 2026 - Studying AI welfare empirically — A framework: question × entity × evidence.
- Hung 2026 - Model or instrument — Measured preferences depend heavily on the instrument.
- Wang et al 2026 - AI revealed preferences — Revealed preferences across 20 models.
- Anthropic 2026 - System card welfare assessments — Company welfare assessments, now standard in one lab’s system cards.
Methods in play
- Indicator-property assessment
- Behavioral trade-off tasks
- Revealed-preference utility modeling
- Conceptual analysis
Who’s closest
Placed by the editor, v1 — corrections welcome.
Related questions
- Do functional emotions matter
- When self-report counts as welfare evidence
- Is there a global workspace
Know a paper that belongs here, or think this is framed wrongly? Suggest a paper or a correction (forms on the Atlas home).