Systems which use neural nets to learn from user actions should expose physical buttons for reinforcement. Like if I'm teaching my machine a new gesture I'd want to tell it when it's doing a good job or messing up according to my personal criteria.