basedexplanationsimproveA I

Neural Information Processing Systems 

Our explanations help users more accurately reject AI'swrong decisions than all other tested methods.