I will present research exploring the use of representations of probability
and utility in human--computer interaction.  Key concepts and challenges
will be introduced in the context of several efforts at Microsoft Research,
including the Lumiere and Lookout projects. Lumiere research centers on
developing methods and architectures for inferring a user's goals and needs
by considering an ongoing stream of user actions. Lookout focuses on an
exploration of mixed-initiative user interfaces--interfaces that are
designed with the assumption that people and automated services will
collaborate to achieve goals.  I will conclude by presenting highlights of
research on the foundations of human--computer dialogue, taking the
perspective of interaction as decision making under uncertainty.