Advising Your Adaptive Software Agent

Jude Shavlik

Imagine you are observing the decision-making of a software agent and notice some imperfections. What should you be able to say to it? What would be a good "advice language" that (a) allows one to provide, in a natural, flexible, and imprecise manner, new information to the agent, while (b) ensuring that the agent’s learning mechanism can use the advice? We discuss these questions about broadening the "information pipeline" between human teacher and machine learner, briefly describe an implemented system that provides initial answers, and describe preliminary work on an advice-taking, adaptive Internet wanderer that searches for pages of interest to the user.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.