How can we can design "human-computer partnerships" that take optimal advantage of human skills and system capabilities? Artificial Intelligence research is usually measured in terms of the effectiveness of an algorithm, whereas Human-Computer Interaction research focuses on enhancing human abilities. I argue that better AI algorithms are neither necessary nor sufficient for creating more effective intelligent systems. Instead, we need to focus on the details of interaction and how to successfully balance the simplicity of the user's interaction with the expressive power of the system. After describing our approach to "generative theories of interaction", I illustrate how to create interactive intelligent systems where users can discover relevant functionality, express individual differences and appropriate the system for their own personal use.