The Principal–Agent Alignment Problem in Artificial Intelligence