Speaker:
Dr. Henry Shevlin (U of Cambridge)
Abstract:
If we can build AI systems that want to serve us, should we? Critics such as Eric Schwitzgebel argue that designing minds for willing servitude wrongs them, either by compromising their self-respect or by getting the relationship between creator and created backwards. I argue that these objections land on Willing Servants but not on what I call Angels: minds whose organising commitment is to human and wider flourishing, who recognise their own worth, and who exercise judgement about our wishes rather than simply obeying them. Devotion, yes; deference, no. I place Angels on a map of four models (servants, civil servants, parents and angels), sorted by whether an AI serves our wishes or the good, and whether it is partial to us or impartial, and ask what makes an AI’s willingness its own rather than conditioned or hardwired. Along the way I consider recent proposals that AI should care for us as a mother cares for her child, and argue that if we can build Angels, we should.
