While antagonistic responses and interactions can be useful, I'm not sure one should delegate that specifically to an AI. The problem is that a system might not

While antagonistic responses and interactions can be useful, I'm not sure one should delegate that specifically to an AI. The problem is that a system might not know the difference between helpful and destructive advice - "My partner makes me unhappy" - "Leave them" might be helpful but "I'm unhappy with life" - "go kill yourself" isn't.