Three conceptions of how biology constrains intelligence

Carlos Montemayor
San Francisco State University

Stevenson 1301
12:00 PM

Recent discussions on AI regulation and safety have highlighted the importance of developing an AI that is not only roughly aligned with human values, but also one that actually cares about humanity. For instance, Geoffry Hinton has recently proposed that one way to guarantee this care for humanity is for AI to have maternal instincts. In this talk, I address two issues with respect to such proposals. First, I explain why without the biologically grounded empathic needs humans, and at least mammals have, there is no realistic way to create AI with maternal instincts or any other care instincts. Second, I argue that the best way of understanding these needs is in terms of their temporal dimension and their symmetry with respect to other agents with similar needs. This temporal and social account contrasts with the two leading views about how to conceive biological constraints on intelligence, based on metabolism and our biological substrate, respectively.