After providing an overview of work at Microsoft Research, I will present research on harnessing representations of uncertainty and expected utility to enhance human-computer interaction. I will focus on the Lumiere and Qualia projects. Lumiere research centers on developing methods and architectures for inferring a user's goals and needs by considering the ongoing stream of user actions. An early Lumiere prototype served as the basis for the Office Assistant, the help system in the Office '97 suite of applications. In Qualia research, we are pursuing the use of expected utility to control graphics rendering under scarce resources.