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.
Allow the user to globally customize what the AI system monitors and how it behaves.