A Scruffy Approach to Commonsense Reasoning

Friday, March 15, 2019, 11:00 am - 12:00 pm PDTiCal
This event is open to the public.
AI Seminar
Henry Lieberman, MIT
Video Recording:

A dream of AI has long been understanding Commonsense knowledge, simple knowledge of everyday life and peoples' activities that underlies human intelligence. It has regained currency lately with intelligent conversational agents and more widespread use of AI in interactive applications.  Roger Schank talked about the diversity of approaches to AI, distinguishing between the "neats" (mathematically precise reasoning) and the "scruffies" (heuristic, approximate, uncertain reasoning).  Approaches to Commonsense reasoning in AI have also spanned this spectrum, from axiomatizing various Commonsense domains, to information extraction from the Web. I will talk about the approach to this problem we have developed at MIT, which leans to the "scruffy" side, motivated by the need for Commonsense reasoning in intelligent user interfaces. One of the results has been ConceptNet, an open-source Commonsense knowledge base that has been widely used. More recently, we have been using Commonsense knowledge in story understanding. I will argue that the essential mechanism for Commonsense reasoning is analogical inference, differing in some important respects from both logical inference and statistical reasoning. 

Henry Lieberman is a Computer Scientist at the MIT CSAIL and Media Lab in the fields of programming languages, artificial intelligence and human-computer interaction. He received the 2018 ACM Impact Award Intelligent User Interaction for work on mining affect from text, work that has been applied to the prevention of cyberbullying. His work in commonsense reasoning focused on crowdsourced construction of commonsense knowledge, and its application to intelligent user interaction. Dr. Liebermann authored books on end-user programming, programming by example and Semantic Web.
« Return to Upcoming Events