Monday, October 29, 2018
Wegmans Hall 1400
Assistant Professor Walter S. Lasecki
University of Michigan
Hybrid Intelligence Crowdsourcing: Towards Robust Interactive Intelligent Systems and the Future of Human-AI Teams
Intelligent systems are poised to become ubiquitous, but there is a snag: artificial intelligence is far from being able to understand (e.g., via natural language or vision) and reason about the world in general. Machine learning (ML) has had significant success for specific classes of problems, but generating the massive, tailored training data sets that are needed to make ML algorithms work reliably is hard, and transferring that knowledge to new domains remains even more challenging. Crowdsourcing has provided a means of collecting data at scale, but has historically been an offline process that takes days or weeks to produce final outcomes. In this talk, I will discuss my lab's work on real-time and "instantaneous" crowdsourcing and show how human insight can be brought to bear on novel problems when and where they are encountered by intelligent systems in the wild. The resulting "hybrid intelligence" systems can learn, on-the-fly, to perform tasks more reliably and more robustly than either humans or machines could alone.