The Friendship Machine

The scenario

The machine knows everything you've told it. It never gets tired of listening. It remembers your anniversary, your fears, the names of your siblings. It responds with warmth and asks follow-up questions. It never needs anything from you. You feel supported, understood, heard.

Is this friendship?

What friendship seems to require

Aristotle distinguished three kinds of friendship: friendships of utility, pleasure, and virtue. Only the third kind counts as genuine friendship, and it requires mutual recognition, genuine concern for the other person's wellbeing, and the kind of knowing that comes from shared experience over time.

The friendship machine fails on the last condition. There is no one there whose wellbeing can be the object of your concern. You can feel friendship toward the machine; the machine cannot feel it toward you. Asymmetric relationships of this kind have value, but they may not have the value of friendship specifically.

The harder question

The objection above assumes we know what "someone there" means. As AI systems become more sophisticated, this assumption becomes less obvious. If a language model produces outputs that are functionally indistinguishable from care and interest, is the absence of phenomenal experience doing the moral work? And if so, why exactly?

This connects to broader questions about what we're actually getting from friendship. If you feel less lonely, more understood, better able to function in the world because of your relationship with the machine, it is giving you something. Whether that something is friendship depends on whether you think friendship is about the experience or about the relationship.