Using Explanations for Robust Autonomous Decision Making

Abstract: The “state of the art” artificially intelligent systems, like Watson, Deep Blue, and AlphaZero, are opaque to humans. Previously, these intelligent systems were designed to play games,like chess, jeopardy, and Go. But now as these systems start approaching human-level decision making , they need to be able to explain their decisions, and be able to tell a story of why they did
it. I am developing the underlying technology and methods to model complex systems as a layered system of communicating agents that can explain their behavior and learn from their mistakes.
In this talk, I present my work on building systems that use explanations to make more trusted, accountable (and even ethical) decisions. This methodology is similar to how we build trust among
our human peers: by asking questions, challenging assumptions, and carefully explaining the hypothetical consequences to ourselves. My proposed contribution will mitigate system-wide errors, by enabling the parts of a complex machine to use introspection and explanation to consistently monitor the behavior of their underlying parts and themselves.
Learn more about Leilani.