Abstract

Sophisticated AI's will make decisions about how to respond to complex situations, and we may wonder whether those decisions will align with the moral values of human beings. I argue that pessimistic worries about this value alignment problem are overstated. In order to achieve intelligence in its full generality and adaptiveness, cognition in AI's will need to be embodied in the sense of the Embodied Cognition research program. That embodiment will yield AI's that share our moral foundations, namely coordination, sociality, and acknowledgement of shared resources. Consequently, we can expect a broad moral alignment between human beings and AI's. AI's will likely show no more variation in their values than we find amongst human beings.