AI Welfare is the first book systematically to tackle the question: could AI have welfare? Using the tools of analytic philosophy, the book examines whether AIs could possess beliefs and desires, consciousness, and the capacity to feel pleasure or displeasure. It argues that some existing AIs may already deserve moral consideration. But the dev
Artificial systems are becoming increasingly capable and agentic. In the face of these changes, a key philosophical question is: could AIs have welfare?
Goldstein and Kirk-Giannini assert that whether AIs have welfare depends on three factors: whether they possess agency (beliefs and desires), consciousness (subjective experience), and sentience (the capacity for pleasure and displeasure). AI Welfare argues that AIs already possess agency, and that minor modifications to existing AIs could plausibly produce consciousness and sentience.
Goldstein and Kirk-Giannini also explore nine crucial open questions about AI welfare, as follows: How should we identify which desires contribute to welfare? Is consciousness required for welfare? Do simple AIs like Roombas have welfare? Do tiny neural networks satisfy the conditions of functionalist theories of consciousness? Does consciousness require rich representational formats? Could AIs experience bodily pleasures and displeasures? Are pleasure and displeasure essentially access-conscious? Are simulated minds real? Are LLMs merely stochastic parrots? AI Welfare offers provisional answers to each of these questions while also leaving room for future discussion, with the goal of demonstrating how the methods of analytic philosophy can help answer essential questions in this emerging field.
