More than three decades of research on human-computer interaction (HCI) have resulted in significant advances in theories, tools, and methods to enhance collaboration between humans and computing systems. However, current user interfaces still lack the ability to attribute mental states to their users, i.e. they fail to understand their intentions and to anticipate their actions. This is despite the fact that these abilities are essential for natural, effortless, and seamless human-machine collaboration.
The overall goal of the proposed project is to contribute to the development of a new generation of user interfaces that pro-actively adapt to users’ future actions by predicting their interaction intentions. In HCI, intention prediction is the task of inferring users’ high-level goals and needs from sequences of their low-level actions, e.g. mouse or keyboard input or search queries. In interactive systems this task exhibits significant uncertainty and is thus highly challenging. Specifically, actions cannot be assumed to manifest a single intention currently held by the user. Instead, interactions are characterised by highly idiosyncratic and often exploratory sequences of user actions, potentially serving multiple, concurrent intentions. Also, users may unintentionally achieve intentions through these exploratory actions, abandon their intentions at any time, or dynamically adopt new intentions based on information shown on the user interface. The current project will provide an important complementary perspective to ongoing efforts in the host groups by formulating and studying intent prediction in a fully Bayesian framework. This approach will not only enable future interfaces to quantify and adapt to the inherent uncertainty of the intent predictions themselves, but will also leverage this information to improve subsequent predictions of users’ actions.
|Project Name||Bayesian Intent Prediction for Human-Machine Collaboration|
|Project Leader||Andreas Bulling
|Project Members||Lei Shi, PostDoc Researcher|