Machine Teaching for Question Answering in Jill Watson

Harsh Sikka, Vrinda Nandan, Ashok Goel
Karan Taneja

Machine Teaching is a collection of approaches explicitly aimed at solving the difficulties that lie in enabling domain experts to effectively teach machine learning systems. AI Systems, including agents like Jill Watson, face significant challenges in adapting to new domains. The cost for creating, training, and configuring Jill Watson was around 500 person-hours in 2016. In this research we focus on two main questions: 1. How can we rapidly teach Jill Watson to adapt to new domains i.e. new courses? 2. What sort of rich interfaces can be designed to facilitate Machine Teaching on the part of domain experts?

Ashok Goel
Ashok Goel, Keith McGreggor, Spencer Rugaber
Tesca Fitzgerald, David Joyner, Rochelle Lobo, Bryan Wiltgen, Gongbo Zhang

The Design & Intelligence Laboratory conducts research into human-centered artificial intelligence and computational cognitive science, with a focus on computational creativity. Current projects explore analogical reasoning in biologically inspired design, visual reasoning on intelligence tests, meta-reasoning in game-playing software agents, and learning about ecological and biological systems in science education.