Dissertation Defense - Miles Brundage

Responsible Governance of Artificial Intelligence: An Assessment, Theoretical Framework, and Exploration

Abstract:

While artificial intelligence (AI) has seen enormous technical progress in recent years, less progress has occurred in understanding the governance issues raised by AI. In this dissertation, I make three contributions to the study and practice of AI governance. First, I provide an assessment of existing AI governance efforts from the perspective of responsible research and innovation (RRI), synthesizing for the first time a range of literatures and highlighting several limitations of extant efforts. Second, I justify and elaborate a theoretical framework for how to think about and enact AI governance, focusing on the unique and non-unique characteristics of AI and the deep uncertainty about its future. Third, I explore, through several case studies, the value of three different methods for conducting AI governance analysis: expert elicitation, scenario planning, and formal modeling. In each case, I explain why these particular methods were deployed, what they produced, and what lessons can be learned for improving the governance of AI in the future. I find that the three methods explored each have substantial but distinct and complementary value for AI governance, and identify areas for future work.

Committee:

David Guston - Chair
Joanna Bryson, University of Bath
Erik Fisher
Lauren Keeler

Location: 
Coor Hall, room 5536