Timely State Exposure for the Coevolution of Mental Models and Dynamic Systems

The computational systems we interact with are increasingly intelligent and dynamic, as they learn from user interactions and are updated over time. Principles of good design highlight the importance of understanding user mental models and providing feedback to expose the internal state of the system. In the context of intelligent systems situated in the real world, this notion of feedback is highly temporal and requires a deep investigation of when feedback should be provided to users. In this chapter, we first motivate the need for timely state exposure in the design of such dynamic systems. We then lay out the design space of timing strategies for providing feedback, highlighting the frequency of state exposure and the information necessary for each strategy. We report on a preliminary exploration of this design space using our news article recommender system. We find that one feasible approach may be to provide feedback only after failures when the system behavior is incorrect. Finally, we discuss open challenges that need to be studied in future work, including methods for effective evaluation of user mental models over time, anticipation of breakdowns, and detection of system failures.