Engage in disambiguation or gracefully degrade the AI system’s services when uncertain about a user’s goals.
Immediately update or convey how user actions will impact future behaviors of the AI system.