Eric Horvitz, Distinguished Scientist at Microsoft and former President of AAAI, gave a talk at the Yahoo Labs Santa Clara campus recently about how learning and reasoning from massive amounts of data can aid in a wide range of domains, including transportation, energy and healthcare, potentially providing extraordinary value to individuals and to society. He spoke broadly about learning and reasoning, but presented a number of compelling concrete examples using Bayesian inference systems of the sort developed by his research team. Horvitz himself has been on the leading edge of such technology and its applications for well more than two decades. Using Bayesian inference, Horvitz and his team have been developing and testing a traffic forecasting system that’s based on data from the past 5-6 years from sensors on major highways around Seattle. The system correlates traffic data with non-traffic data such as weather, incident reports, and the times of major sporting events, and uses algorithms to predict the future of traffic, making its forecasts accessible through mobile devices. The forecasting method takes into account past traffic patterns as well as feedback from drivers on what surprises them (e.g., a bottleneck on a road where there is almost never a bottleneck at that time, traffic flowing easily at places where congestion usually happens at rush hour). This goes well beyond conventional real-time traffic reporting, and allows people to avoid bottlenecks in order to get to their destinations more quickly, using what Horvitz refers to as “surprise forecasting.” In the area of healthcare, Horvitz describes how data is critical in “evidence-based healthcare.” One specific challenge he discussed is helping predict the probability of patients “bouncing back” to hospitals (and needing to be admitted) soon after they are discharged from emergency room visits, a phenomenon that is extraordinarily expensive. Years of data and labs results can help with prediction and decision on discharge and follow-up care, ultimately saving potentially huge amounts of money and enhancing the quality of healthcare. In terms of the environment and climate changes, Horvitz explains that computation can lead to energy efficiency through calculations that can bring down carbon emissions and optimize programs such as rideshare initiatives and other “good old-fashioned conservation” efforts. Horvitz concluded his talk by stressing that systems that sense, learn and reason from streams of data are still in their infancy, but promise to provide extraordinary value to people and society in the near future.