Nuclear Non-Proliferation, Satellite Images, AI, and Risk

Abstract

Satellite images are one type of data used by analysts who monitor states to verify compliance with nuclear non-proliferation and arms control agreements (citizen-based monitoring & verification, or “CBM&V”). The people who engage in this work use data from a variety of sources, including satellites and other publicly available information, often referred to as open source intelligence or OSINT. The work of obtaining and analyzing satellite image data with the goal of identifying activities such as the development of weapons facilities requires the work of highly skilled analysts. This poster will present preliminary results from a July 2024 workshop and gallery walk with 14 satellite image analysis experts. This study received approval from the author’s Institutional Review Board. I explore here the following research questions: 1. How and in what ways do experts in satellite image analysis use machine learning and/or artificial intelligence (“ML/AI”) in their work? 2. How and in what ways do those experts understand current and/or future applications of ML/AI in their work? 3. How do those experts understand risks associated with their work in light of recent developments in ML/AI? Preliminary findings suggest that experts in satellite image analysis for CBM&V understand that ML/AI has a useful role to play in the management and processing of digital information, but that they tend not to find it directly useful for their current image analysis work. Participants were able to discuss a range of potential applications for ML/AI, with the caveat that the inherently risky nature of their work would demand greater transparency than is currently available. This is a community of data users for whom trust in the source of data, in the work of expert data users, and in the ability of publics to understand and believe in the work of those experts is crucial at all stages of work. This poster examines the ways in which those experts understand the impact of ML/AI on their high-stakes work.