Question Asking to Inform Preference Learning: A Case Study

No. 1: Agents that Learn from Human Teachers

Volume

Papers from the 2009 AAAI Spring Symposium

Contents

The prospect of teaching a computer how to perform a task through demonstration rather than programming has tremendous appeal. However, learning purely from a demonstration trace is a difficult challenge. One way to ease the learning problem is to supplement the demonstration with information provided by asking questions of the demonstrator. This paper presents a case study that explores how to instantiate a question asking framework to select questions for a particular type of learner used within learning by demonstration systems, namely a lexicographic preference learner. Experimental results show that, generally speaking, judicious question asking can improve learning performance. However, the study makes clear the importance of understanding the value of different types of information to learning in different contexts.