Help the user understand how often the AI system may make mistakes.
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.