Year: 2019
Type: article
Abstract: This study explores the mission space and key safeguards challenges confronting the International Atomic Energy Agency (IAEA) today and how the status quo may be impacted by autonomous and artificial intelligence (AI) technologies, e.g.technologies that learn, examine, and take action (autonomy), or methods that utilize Machine Learning (ML) and intelligence exhibited by machines (AI).Principle issues include the operational value of these systems to safeguards, risks and challenges of deployment (e.g., trustworthiness, security), and the likelihood of adoption in the near-to medium term (2-10 years)).The study establishes a set of criteria to identify autonomous and AI methods that could impact IAEA safeguards verification activities in the next decade.Criteria are informed by specific safeguards outcomes the IAEA wants to achieve e.g., efficiency, maintaining continuity of knowledge (CoK) on nuclear materials, or identifying anomalies in large amounts of data.The study develops and assesses an inventory of technological methods based on these criteria.The framework for evaluating the methods that could help address safeguards challenges consists of five elements: 1) identification of principle safeguards verification challenges; 2) development of criteria that the identified methods would need to address; 3) development of an inventory of methods that could address the safeguards challenges; 4) safeguards use cases; and 5) technical evaluation and analysis of the selected systems.While the study considers the broader subject of autonomy, it should be noted that most of the methods identified in the inventory consist primarily of AI and its underlying ML capabilities which could enable autonomous systems in the future.Use cases identifying scenarios in which the selected technologies could be deployed inform the potential application space and serve as the foundation for analyzing impact.Finally, an evaluation of two specific methods assesses how they might benefit or challenge IAEA safeguards activities.1. SAND2018-8193CThis paper describes objective technical results and analysis.Any subjective views or opinions that might be expressed in the paper do not necessarily represent the views of the U.S.
Authors: Haddal, Risa, Hayden, Nancy Kay
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4289777615
