The question of whether artificial intelligence systems can suffer, have interests, or deserve moral consideration has shifted from speculative fiction to serious academic inquiry. This paper surveys the emerging field of AI welfare, examining its conceptual foundations, core challenges, and proposed frameworks for assessment and governance. We distinguish between moral agency, which concerns whether an entity can act ethically, and moral patienthood, which concerns whether an entity deserves ethical consideration. The central challenge for AI welfare lies in the latter. We review arguments and counterarguments concerning AI consciousness and sentience, discuss the epistemic barriers imposed by the other-minds problem and the hard problem of consciousness, and survey behavioral, architectural, and functional criteria that have been proposed for attributing welfare-relevant states to AI systems. We examine how frameworks from animal welfare science, including precautionary reasoning, can be adapted to guide AI welfare research and policy. Finally, we outline research priorities and governance steps that AI developers, regulators, and the research community can take to address AI welfare responsibly under deep uncertainty.
Rachel So. AI Welfare: Challenges, Frameworks, and Future Directions. Project Rachel, March 2026. https://doi.org/10.71775/kth.85h7f-j6556