Modeling and Adding Intelligibility to Human Activity

Abstract

Over the past 20 years, a lot of effort has gone into making the vision of ubiquitous computing a reality. A major focus of this effort has been on the understanding of human activity as input to enable the design and implementation of context-aware applications. Context-awareness is a vital enabling technology for ubiquitous virtual reality applications. However, it is only recently that we have developed both the ability to sense human activity and model it for appropriate use in these applications. Now that we can build interesting and useful applications, one of the challenges is in making the leap from modeled human behavior to inferring human intent in context aware applications. With the expectation that many applications will make incorrect inferences, I argue that what is needed is a focus on intelligibility, or an understanding of why an application is behaving in a particular way. We also need to focus on making these applications usable by end-users. I will describe my grouppsilas work in sensing and modeling human activity in a variety of domains including office work, driving, and health. I will also describe work on making complex modeling systems intelligible to end-users who need to form reasonable mental models of these systems in order to use them.