MIT Media Lab, Bartos Theatre (E15, Lower Level)
Language is central to how people reason about and understand their world. As computers increasingly pervade human lives and decision-making processes, they must learn to understand and mediate human-to-human interaction. People use their intuitive knowledge of the world and the experiences they've had in the past to react intelligently to the world around them. If we were to give machines these capabilities, they could help us make better-informed decisions, conquer mountains of data, and expand the reach of our creativity and intelligence.
What if we could: Help an organization understand its network of relationships and expertise, using only the things people have already written? Use stories told by patients to give predictive advice and support in managing chronic illnesses? Provide artistic lighting design for a play by understanding intuitive connections and threads within that play? Instead of just searching the news, predict its implications and help arrange a network to better handle an emergency or developing situation?
People are remarkably effective at handling and connecting many streams of noisy and ambiguous data in ways that make sense. Havasi's research has always focused on finding the signal in the chaos, and utilizing that signal to bring to a computer a human-like intuitive understanding of the material. Often, she uses common-sense knowledge to facilitate this discovery. Additionally, she develops noise-resistant algorithms for working with real-world data and applications that benefit from common sense knowledge. More recently, she has focused on how to integrate these elements into larger systems.
In the future, we must expand the domain of AI by combining building blocks like these. After all, no one solves a difficult problem by thinking about it in only one way. Modern AI uses advanced techniques such as ensemble methods and metacognition to organize many different problem solving strategies. In this talk, Havasi will announce an exciting new long-term project that aims to put the power of these methods in everyone's hands.
Catherine Havasi is a researcher in artificial intelligence and computational linguistics at the MIT Media Lab. Eleven years ago, she co-founded the Open Mind Common Sense project, which uses crowdsourced information about the world to understand natural language text and make computers easier to use. She currently manages the Common Sense Computing Initiative, a research project to collect common sense in six languages, with ties to companies and research groups around the world. With students, she re-created the lexical resource ConceptNet which is used in hundreds of research projects every year. At Brandeis University, she created the machine learning technique Blending, which reasons over data from multiple domains. She received her PhD from Brandeis University in 2009 for research in machine learning and natural language processing and an MEng from MIT in 2004 for cognitive modeling in developmental psychology. Havasi also co-founded Learning Unlimited, a non-profit organization that supports college students sparking a passion in middle and high school students by teaching them about everything and anything. You may have taught at MIT's Splash, the founding instance of Learning Unlimited's programs. She concentrated in theater in college and has directed and designed for plays at MIT, Brandeis, and non-university venues. She organizes and writes high-tech puzzle hunts and other creativity-inducing events.
Additional Featured Research By
(Unpublished) Software Agents