Help the user understand how often the AI system may make mistakes.
Enable the user to access an explanation of why the AI system behaved as it did.
Limit disruptive changes when updating and adapting the AI system’s behaviors.