This project looks at a novel method that can help intelligent systems alert users who are multitasking. Consider semiautonomous vehicles, for example: For handover and takeover situations, a system needs to understand what the driver is paying attention and what might the consequences (disruption, safety, comfort etc) if an alert was sounded. We study a new way that takes into account a driver's beliefs, capabilities, and the external situation. Theoretically, our work is based on POSG (partially observable stochastic game) that uses a simulator to plan the AI assistant's action. New to our work is the inclusion of user models in such simulator. The work is carried out together with UMPC/CNSR (Gori).