People build their own models of systems they don’t understand, and may make unwarranted extrapolations of their capabilities—which can lead to disappointment and lack of trust. An effective intelligent system should be transparent, able to explain its own behavior in a way that connects to its users’ background and knowledge. An explanation is not a full trace of the process by which the system came to a conclusion. It must highlight important/surprising points of its process, and indicate provenance and dependencies of resources used. Systems that evolve through statistical learning must explain (and exemplify) categories it uses, and clarify for a user what properties makes a difference in a particular case. Such systems must not be single minded, hence should be interruptible and able to explain current goals and status. They should be able to to take guidance in terms of the explanation they have given. Artificial intelligence systems must not only understand the world, and the tasks they face, but understand their users, and most important—make themselves understandable, correctable, and responsible.