I don't think the AI needs nor recognizes the distinction between the emotional vs logical side; it's just optimized for a task, and there's no theoretical limitation from it being optimized for de-escalation given enough time, training data and perhaps robot body if necessary