When it comes to software, one size does not fit all. Different users employ the same software for a variety of different tasks. Yet regardless of whether a user does a task five times, or a thousand times, the software remains unchanged. Why can't software learn from the people using it, and customize itself for a better fit?

My work explores the use of artificial intelligence techniques to design adaptive interfaces. These intelligent interfaces leverage knowledge about users and their tasks to provide a better experience for users. Specifically, my work focuses on the use of programming by demonstration to enable end users to customize applications and automate repetitive tasks simply by demonstrating what the system should do.

However, adaptive user interfaces raise a new set of design challenges for HCI, especially concerning issues of trust and understandability. In this talk, I discuss the challenges in designing intelligent interfaces, illustrated with examples drawn from programming by demonstration systems.