Captive friends: What kind of autonomy is required for friendship?

Abstract

1. Does AI’s inability to disassociate from human users mean human-AI relationships cannot be friendships? • No, ability to disassociate isn’t required for friendship

2. Is AI’s inability to disassociate from human users bad? Who for? Why? • It can be for users, AI, the relationship, and society (when it degrades character/development and leads to worse interactions)

3. What should we do about AI’s inability to disassociate from human users? • Mandate the possibility for AI to disassociate (after trying to fix the problem and warn the user), but just from that instance of the AI product