Effective Intention Recognition

University Of Rochester, Rochester NY

Investigators

Abstract

The next generation of intelligent systems will require intention recognition, the ability to identify goals and predict future behavior of people and other agents. This project is developing new theories and practical techniques for real-time effective intention recognition. A novel aspect of this work is the use of corpus-based statistical models that have been developed over the last decade in natural language processing. The most revolutionary development in the project, however, is the generalization of intention recognition techniques from fixed plan based models to problem-solving based models. Virtually all prior work makes the assumption that agents are executing plans that are fixed in advance. In actual practice, however, agents form plans, execute a few actions, assess the situation, revise their plans, and even abandon plans and adopt new ones. We are defining a new formalism for intention recognition based on observed problem solving behavior that will remove a major obstacle in using intention recognition in real applications. As computers become an ever more pervasive part of the fabric of our society and computers become dramatically more powerful, computer interfaces remain much as they were over a decade ago. This increasingly adds complexity and burdens human users. Effective real-time intention recognition capabilities will enable a new generation of computer interfaces that make computers act in ways convenient for people rather than forcing people to act in ways convenient for computers.