Imagine how a truly intelligent assistant would operate on your behalf. The assistant would not only observe what and who you interacted with but learn which types of things are important, what is salient to you, what you are likely to forget, and what you need to know about, and could notify you when salient and significant changes occur. The assistant would be capable of these things because at its core was the ability to understand your intentions and goals and it could evaluate your actions and requests from others in terms of how they support these goals. In this talk, we present research that attempts to bridge the gulf between the lack of an AI that can understand your actions in the context of your goals and the status quo which simply observes and optimizes the individual actions you take toward those goals. Specifically we focus on several projects which provide contextual intelligence by: (1) presenting the information you want before you ask at the right time, place, and context; (2) and addressing your information needs in a context-specific way to the task you are doing. We present a summary of work that aims to identify the underlying goals which drive people’s behavior in memory recall, task planning, and meeting preparation and describe two prototypes which leverage these insights to better support people through intelligent context-aware interfaces. Finally we conclude with a discussion of research challenges on the path to goal-directed intelligence.
This talk presents joint work with Adam Fourney, Ryen White, Eric Horvitz, David Graus, Xin Rong, Qian Zhao, Susan Dumais, Adam Troy, Shane Williams and Anne Loomis Thompson as well as Ahmed Awadallah, Horaţiu Bota, Robin Brewer, Nirupama Chandrasekaran, Fernando Diaz, Cristina Garbacea, Nick Ghotbi, Marcello Hasegawa, Richard Hughes, Abhishek Jha, Ece Kamar, John Krumm, Merrie Morris, Rev Rameshkumar and likely many others.
Paul Bennett is the Principal Research Manager of the Information & Data Sciences group in Microsoft Research AI. His published research has focused on a variety of topics surrounding the use of machine learning in information retrieval – including ensemble methods and the combination of information sources, calibration, consensus methods for noisy supervision labels, active learning and evaluation, supervised classification and ranking, crowdsourcing, behavioral modeling and analysis, and personalization. Some of his work has been recognized with awards at SIGIR, CHI, and ACM UMAP. Prior to joining MSR in 2006, he completed his dissertation in the Computer Science Department at Carnegie Mellon with Jaime Carbonell and John Lafferty. While at CMU, he also acted as the Chief Learning Architect on the RADAR project from 2005-2006 while a postdoctoral fellow in the Language Technologies Institute.