Engage in disambiguation or gracefully degrade the AI system’s services when uncertain about a user’s goals.
Limit disruptive changes when updating and adapting the AI system’s behaviors.