Can we be friends with AI? What risks would arise from the proliferation of such friendships?

Abstract

In this paper we analyse friendships between humans and artificial intelligences, exploring the various arguments that have been or could be offered against the value of such friendships, and arguing that these objections do not stand up to critical scrutiny. As such, we argue that there is no good in-principle reason to oppose the development of human-AI friendships (although there may be some practical defeasible reasons to worry about such friendships becoming widespread). If we are right, there are important implications for how friendship is conceptualised and valued in modern times. Furthermore, if human-AI friendships are in principle valuable, the moral responsibilities for how governments and corporations should act in regards to AI friends are quite different to those generated by human-AI friendships being dis-valuable.