AI Seminar
Teachable Cognitive Systems
Add to Google Calendar
Researchers have developed Artificial Intelligence (AI) systems capable of a human-level performance at complex tasks, but constructing these systems requires substantial time and expertise. Traditionally, these AI models were hand authored, but recent work has emphasized the application of Machine Learning (ML) approaches to learn such models from real-world or simulated data. However, ML approaches are not a panacea and in most cases they simply shift development costs from direct model authoring to training data generation and curation. Industry researchers have started to acknowledge that applying ML requires a specialized skill set deserving of its own formal training discipline (Machine Teaching, see Simard et al., 2017), which makes it difficult for non-technical end users to leverage these ML technologies. To address this problem, my research explores the design of interactive learning systems that are natural and efficient for people to teach. Whereas most research in the ML space centers on algorithm design, I instead take a human-centered approach to designing teachable systems. In this talk I will present three research arcs in this general theme. First I will describe my efforts to built a teachable system that support K12 teachers in authoring and testing intelligent tutoring systems. Second, I will presents a general framework for designing teachable systems that are natural and efficient for people to use. Finally, the third arc will present preliminary efforts to scientifically test interactive system designs from this general framework. One of the main goals of this work is to enable an unprecedented level of AI technologies personalization by giving end users the power to change the behaviors of the systems they interact with.
Chris MacLellan (https://chrismaclellan.com) is a research scientist at Soar Technology, Inc. His research, which sits at the intersection of Human-Computer Interaction, Artificial Intelligence, and Machine Learning, aims to develop interactive technologies that everyday people can teach and use. Chris received his PhD in Human-Computer Interaction at Carnegie Mellon University, where he developed the Apprentice Learning Architecture, a platform for building interactive systems that learn from examples and feedback.