Recognizing and enhancing sapient agency within AIs: a free will perspective

Can artificial intelligences (AIs) be sapient (i.e., self-aware and self-determining), in the same way humans are sapient? Here I argue that true sapience requires free will—i.e., the ability to become aware of one’s own wants and preferences, then turn them into activated goals (List in Why free will is real, Harvard University Press, Cambridge, 2019). To illustrate, I discuss the Goal Breakthrough Model (GBM; (Sheldon et al. in J Res Personal, 2023)), which explains how multiple brain networks interact, via explicit and implicit cognition, to support optimized goal selection. Although current AIs mostly lack the mental capabilities demanded by the GBM, I suggest ways these capabilities might be developed, including: (a) the ability to host default mode activity, independently of human prompts; (b) the ability to direct executive functioning, in a constrained yet powerful global workspace; (c) the ability to develop a symbolic self-identity, that can affect the AI’s executive functioning; (d) the ability to delay problem-solving, to give broader-ranging cognition a chance; and (e) the ability for the AI to decide for itself what to do, and when. Throughout, I also suggest that the absence of experiential qualia in AIs (i.e., “sentience”) would not necessarily prevent AIs from having sapient agency.