Seminar by Ian Davidson: "If the AIs Are Going to Rule Us Then They Should At Least Be Nice To Us – Past and Future Work in AI Alignment"

"If the AIs Are Going to Rule Us Then They Should At Least Be Nice To Us – Past and Future Work in AI Alignment"

Abstract

In this talk I begin by arguing that it is inevitable that AIs will be making life altering decisions on all of us in the near future. Given this, we need to make sure that a human system of values is encoded into their behavior, a topic widely termed the "AI Alignment" problem. I will begin by discussing our prior work on fairness and explanation that include both theoretical and experimental results. However, most work on alignment only focuses purely on fairness and explanation which are just a small aspect of a machine being ethical. I discuss future work that moves beyond these topics.

Biography of Pr Davidson: Ian Davidson has been a Professor of Computer Science and the University of California - Davis since 2007. He has made basic research contributions on topics such as how to encode knowledge into machine learning algorithms focusing on both complexity and algorithmic results. He also works with domain experts to apply his work to problems of social significance in particular neuroscience applications. This work is funded by gifts from Google and grants from the National Science Foundation (NSF), The Office of Naval Research (ONR) and The National Institute of Health (NIH).

This seminar will be held in English.