Human Teacher’s Perception of Teaching Methods for Machine Learning Algorithms

Presented on February 6, 2019 at 12:15 p.m.-1:15 p.m. in the Marcus Nanotechnology Building, Room 1116-1118, Georgia Tech.Karen Feigh is an associate professor in Georgia Tech's Daniel Guggenheim School of Aerospace Engineering. She holds a B.S. in aerospace engineering from Georgia Tech, an MPhil in aeronautics from Cranfield University, UK, and a Ph.D. in industrial and systems engineering from Georgia Tech. Feigh has previously worked on fast-time air traffic simulation, conducted ethnographic studies of airline and fractional ownership operation control centers, and designed expert systems for air traffic control towers and NextGen concepts. She is also experienced in conducting human-in-the-loop experiments for concept validation.Runtime: 43:57 minutesA goal of interactive machine learning (IML) is to create robots or intelligent agents that can be easily taught how to perform tasks by individuals with no specialized training. To achieve that goal, researchers and designers must understand how certain design decisions impact the human’s experience of teaching the agent, such as influencing the agent’s perceived intelligence. We posit that the type of feedback a robot can learn from effects the perceived intelligence of the robot, similar to its physical appearance. This talk will discuss different methods of natural language instruction including critique and action advice. We conducted multiple human-in-the-loop experiments, in which people trained agents with different teaching methods but, unknown to each participant, the same underlying machine learning algorithm. The results show that the mechanism of teaching has an impact on a human teacher’s perception of the agent, including feelings of frustration, perceptions of intelligence, and performance, while only minimally impacting the agent’s performance