TOM MITCHELL (Carnegie Mellon University, USA)
What if People Taught Computers?
Abstract: Whereas AIED focuses primarily on how computers can help teach people, this talk will consider how people might teach computers. Why? There are at least two good reasons: First, we might discover something interesting about instructional strategies by building computers that can be taught. Second, if people could teach computers in the same way that we teach one another, suddenly everybody would be able to program.
We present our ongoing research on machine learning by verbal instruction and demonstration. Our prototype Learning by Instruction Agent (LIA) allows people to teach their mobile devices by verbal instruction to perform new actions. Given a verbal command that it does not understand (e.g., “Drop a note to Bill that I’ll be late.”), the system allows the user to teach it by breaking the procedure down into a sequence of more primitive, more understandable steps (e.g., “First create a new email. Put the email address of Bill into the recipient field.”…). As a result, LIA both acquires new linguistic knowledge that enables it to better parse language into its intended meaning, and it learns how to execute the target procedure. In related work with Brad Meyers we are exploring combining verbal instruction with demonstration of procedures on the phone, to achieve “show and tell” instruction. In work with Shashank Srivastava and Igor Labutov, we are extending the approach to general concept learning (e.g., in order to teach “if I receive an important email, then be sure I see it before leaving work.” one must teach the concept “important email.”). This talk will survey progress to date, implications, and open questions. This work involves a variety of collaborations with Igor Labutov, Amos Azaria, Shashank Srivastava, Brad Meyers and Toby Li.
Tom M. Mitchell is the E. Fredkin University Professor at Carnegie Mellon University, where he founded the world’s first Machine Learning Department. Mitchell’s current research studies questions ranging from how machines might be able to learn from human instruction, to the role of self-reflection in learning. He is a member of the U.S. National Academy of Engineering, the American Academy of Arts and Sciences, and a Fellow of the American Association for the Advancement of Science (AAAS).
PAULO BLIKSTEIN (Stanford University, USA)
From the Skinner Teaching Machine to Multimodal Learning Analytics: utopias and dystopias for AI in Education
Paulo Blikstein is an assistant professor at Stanford University Graduate School of Education and (by courtesy) Computer Science Department, where he directs the Transformative Learning Technologies Lab. His academic research focuses on how new technologies can deeply transform the learning of science, engineering, and mathematics. He creates and researches cutting-edge educational technologies, such as computer modeling, robotics, and digital fabrication, creating hands-on learning environments in which children learn science and mathematics by building sophisticated projects and devices. Paulo is one of the pioneers of in the field of Multimodal Learning Analytics, a set of methods he uses to research complex learning in project-based environments. Blikstein has published widely in venues such as the Journal of the Learning Sciences, Journal of Learning Analytics, Journal of Educational Data Mining, and Nature Biotech. His work has been featured in the New York Times, ABC News, Scientific American, Wired, The Guardian, and several other outlets. A recipient of the AERA Jan Hawkins Early Career Award and the National Science Foundation Early Career Award, Blikstein holds a PhD. from Northwestern University and a MSc. from the MIT Media Lab.
MICHAEL THOMAS (University College London / Birkbeck University of London, UK)
Has the potential role of neuroscience in education been overstated? Can computational approaches help build bridges between them?
Abstract: In the first part of this talk, I will assess the progress of the field of educational neuroscience in attempting to translate basic neuroscience findings to classroom practice. While much heralded, has educational neuroscience yielded concrete benefits for educators or policymakers? Is it likely to in the future? Or is its main contribution merely to dispel neuromyths? In the second half of the talk, I will assess the role that computational approaches can play in this inter-disciplinary interaction. Is neuroscience best viewed as a source of inspiration to build better algorithms for educational AI? Or can neurocomputational models help us build better theories that link data across behaviour, environment, brain, and genetics into an integrated account of children’s learning?
Dr. Michael Thomas is a Professor of Cognitive Neuroscience at Birkbeck, University of London, UK, and Director of the University of London Centre for Educational Neuroscience. His current work in educational neuroscience includes understanding the role of inhibitory control in children’s science and math learning, investigating the influence of cell phone use on adolescent brain development, linking findings on sensitive periods in brain development to their educational implications, and building links between genetics, environment and education in children’s developmental outcomes. He also directs the Developmental Neurocognition Laboratory, whose goal is the multi-disciplinary study of cognitive variability. Michael employs computational modelling as one of his research methods, using population-level artificial neural network modelling to investigation the brain basis of intelligence and developmental disorders. In 2006, his research lab was the co-recipient of the Queen’s Anniversary Prize for Higher Education, for the project “Neuropsychological work with the very young: understanding brain function and cognitive development”. Michael is a Chartered Psychologist, Fellow of the British Psychological Society, and Fellow of the Association for Psychological Science.