When would AGIs engage in conflict? — AI Alignment Forum