Enable the user to access an explanation of why the AI system behaved as it did.
Maintain short-term memory and allow the user to make efficient references to that memory.
Limit disruptive changes when updating and adapting the AI system’s behaviors.