Whose interests will AI serve? Autonomous agents in infrastructure use

Abstract This article examines new challenges for sustainability presented by artificial intelligence (AI) in infrastructure megaprojects. We differentiate between the deliberative processes of infrastructure megaproject construction and the everyday uses of such infrastructure, focusing on the latter as both having major sustainability impacts and presenting distinct opportunities for AI intervention. While AI applications are often imagined supporting human decision-making processes, we argue that cases in which AI agents act autonomously without direct human intervention present novel questions about how to achieve sustainable outcomes. Specifically, we differentiate between the setting of AI objective functions in the development stages and the control of unanticipated behaviours once an AI system can act independently. After illustrating applications of AI in energy grid management and automated driving, we examine how such systems could operate either in support of shared sustainability interests or to advance their private interests.