I would be weirded out, but I would try very hard not to judge. They wouldn’t be hurting anyone, so it would be wrong of me to judge them.

Depends on AI. I don't see why it would be weird if the AI was like a human, with real emotions.
If it just pretends emotions, it would be odd, but I wouldn't blame the person. It still sounds better than total loneliness and may provide better output than imaginary people.

I knda wish something like that existed. But I also don't. If it had emotions, you could hurt it like a real person, which defeats the purpose. It would also be easy to exploit. How could anyone tell you're not holding someone hostage inside your computer? And I believe initially very few people would care, because "it's just a computer".

If they are aware of what the AI's perspective is, and the AI itself isn't in distress somehow, then it's not really my business, is it? If they don't realise it's just coded to like them then I might feel the need to bust their bubble.

I wonder, when will we make an AI that can dump you?