Abstract

Most scholars who argue that users cannot have genuine friendships with their AI companions attempt to show this by arguing that no current AI can be a friend back to its user. Lott and Hasselberger (2025) argue, instead, that the reason such friendships aren’t possible is that the user cannot be a friend to an AI. In short, the user cannot care for the AI in the way required for friendship, since the AI lacks a good of its own. Here, we argue that Lott and Hasselberger have missed various ways in which AI companions can indeed have interests of their own in the teleological sense employed in their argument, and, further, that, intuitively, non-delusional individuals can genuinely care for an artifact for that artifact’s own sake. Therefore, those who deny that users can be friends with their AI companions should either argue that the notion of “to care for” that’s relevant to befriending is based on a kind of morally-relevant welfare that doesn’t reduce to teleological interests, or simply take the more typical line of showing that AI companions can’t be friends back to their users.