The Trust Issue in Task Automation

Robert J. Hall

End-user programming, programming by demonstration, machine learning, AI planning, and knowledge sharing are among the techniques that allow end-users to automate idiosyncratic, mundane tasks. However, a critical issue that has been left virtually unaddressed by the research and commercial communities is why should the user trust the acquired task knowledge? That is, how can the user gain confidence that it will suffice to automate the task in a way consistent with the user’s goals and priorities, degrading gracefidly in the face of failure? This paper first surveys the trust problem and how it arises for each of several acquisition techniques, surveys and compares existing approaches to solving the trust problem. Finally, it briefly overviews the author’s novel incremental validation approach to trusting behaviors. (A separate paper (Hall 1995c) discusses this approach in detail.)

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.