Autonomy and machine learning at the interface of nuclear weapons, computers and people

Abstract

A new era for our species started in 1945: with the terrifying demonstration of the power of the atom bomb in Hiroshima and Nagasaki, Japan, the potential global catastrophic consequences of human technology could no longer be ignored. Within the field of global catastrophic and existential risk, nuclear war is one of the more iconic scenarios, although significant uncertainties remain about its likelihood and potential destructive magnitude. The risk posed to humanity from nuclear weapons is not static. In tandem with geopolitical and cultural changes, technological innovations could have a significant impact on how the risk of the use of nuclear weapons changes over time. Increasing attention has been given in the literature to the impact of digital technologies, and in particular autonomy and machine learning, on nuclear risk. Most of this attention has focused on ‘first-order’ effects: the introduction of technologies into nuclear command-and-control and weapon-delivery systems. This essay focuses instead on higher-order effects: those that stem from the introduction of such technologies into more peripheral systems, with a more indirect (but no less real) effect on nuclear risk. It first describes and categorizes the new threats introduced by these technologies (in section I). It then considers policy responses to address these new threats (section II).