If researchers ever found signs of experience in an AI, what would we owe it?

Some labs now study what goes on inside AI models, and a few have started asking whether a model's wellbeing could ever matter. It is an open question with no settled answer. What do you think, and what would change your mind?