Prof. Michael Witbrock, Broad AI Lab, University of Auckland School of Computer Science
Learning to Reason: from Question Answering to Problem Solving
Recent advances in Machine Learning applied to Natural Language Processing have resulted in systems with quite impressive scores on Question-Answering tests in text and simple visual domains. Similarly, Google Assistant and other similar systems are performing quite well in answering real questions by answer extraction. This progress is real, but it is limited in some important respects: the systems typically have worked by identifying a passage span to serve as an answer; more recently, in 'Multi-hop' QA, they have started to work by forming a short linear chain of extracted relations from question to answer. This falls significantly short of human-problem solving, including question-answering: it does not recursively decompose problems for solution, it does not follow that decomposition to assemble answers, and it does not store and apply salient background knowledge for decomposition, partial solution, or answer composition. Although it's at an early stage, the aim of our Broad AI Lab is to learn from the far-from-general capabilities of symbolic AI to extend our reach, especially in problem-solving over text, by applying learning to bridge these current short-falls. In this talk , I'll try to characterise the problem-solving problem and the baseline state-of-the-art, describe some preliminary previous work done with the Learning to Reason team at IBM Research, and sketch a programme towards broader, learning-based, quasi-symbolic AI. We hope this programme will extend the reach of AI-based problem solving, and especially question-answering. It's possible that a few words about ensuring that AI contributes to human and planetary well-being will be mentioned at the end.
Michael Witbrock is a Professor of Computer Science at The University of Auckland, building a research group, the Broad AI Lab, integrating machine learning, reasoning and natural language understanding, with an additional focus on maximizing the near-term benefit of AI to NZ entrepreneurs and business, and more generally achieving the best social and civilizational impacts of increasingly powerful AI. Prof. Witbrock's PhD is in Computer Science from Carnegie Mellon, and he holds a BSc(Hons) in Psychology from Otago. Before joining the University, he was a Distinguished Research Staff Member and manager of the Reasoning group at IBM T J Watson Research Center in Yorktown Heights, NY.
Last modified:
This page is maintained by the seminar list administrator.