How will innovations in artificial intelligence (AI) affect how states collect intelligence, wage warfare, and engage with rivals? What are the implications of this emerging disruptive technology for international stability? These are among the most critical questions in the fields of emerging technology and international politics with which academics and practitioners alike are grappling. And these questions form the crux of James Johnson’s timely book, AI and the Bomb: Nuclear Strategy and Risk in the Digital Age. In particular, Johnson tackles what is perhaps the most dangerous and important aspect of AI’s integration into military capabilities: its implications for nuclear deterrence and escalation.

As Johnson details, governments, especially military and intelligence organizations, are increasingly incorporating AI technologies into a range of capabilities and processes, with uncertain implications for international conflict and stability. Broadly defined, AI involves the use of computing capabilities to learn and make decisions in ways that simulate human intelligence. AI has the potential to revolutionize the volume, speed, and scale with which states gather and process information, with potentially transformative effects for decisions and capabilities that previously depended on human intervention, including those that pertain to the use of lethal—and potentially even nuclear—force. AI and the Bomb provides readers with a broad survey of the various ways in which AI may affect strategic stability, deterrence, and escalation. In doing so, the book raises a number of vexing and important research questions that other scholars are sure to pursue in future work. In this sense, AI and the Bomb lays an important foundation in the development of theories and hypotheses about how AI is changing longstanding concepts in international security.

There has been a great deal of writing about the implications of AI on warfighting and stability. However, Johnson’s book stands out in a number of important ways. For one, similar to other emerging technologies of warfare, such as cyber technologies, there is a significant amount of hype around AI. For example, some experts predict that, as AI technology evolves, it will entirely replace the role of human beings in warfighting—or humanity in general. Similarly, some expect that decisions about the use of force will become fully automatic, dramatically compressing the speed of conflict such that major wars could occur in mere moments. Yet, while AI and the Bomb is premised on the assumption that AI will have systemic and potentially even revolutionary effects on warfare, deterrence, and decision making, Johnson focuses on analyzing the current state of AI technology, rather than prognosticating about fantastical technological futures. At the same time, Johnson acknowledges the limitations of testing the arguments he offers in the book, in light of the dearth of available empirical examples. Therefore, he too finds himself exploring hypothetical futures and counterfactual scenarios to explore how various actors may apply existing AI technologies in prospective environments.

Johnson is also careful to point out AI’s limitations and constraints. The book opens with Johnson systematically detailing the limitations of AI, such as its brittleness when making predictions in new situations; its inadequacies in engaging in “System 2” thinking, that is, the complex and critical analysis that is central to human reasoning; the lack of sufficient quantities of good data against which to train AI tools; the opacity of algorithms, making it difficult to parse how a tool arrived at a particular decision; the vulnerability of AI to adversary manipulation and corruption; and so on. In this same discussion, Johnson also lays out the technical foundations of AI and the various ways AI is already being incorporated into military and intelligence missions, which is especially helpful to those readers who may be unfamiliar with the technology. Throughout the book, Johnson links AI technologies to the broader suite of digital capabilities that have emerged over the last several decades. AI, in Johnson’s conceptualization, is simply the latest innovation in a long series of information technology innovations. This is a refreshing contrast to some of the more hyperbolic writing about AI and the future of warfare.

Another important contribution of AI and the Bomb is its focus on the strategic implications of AI. Much of the literature on AI and warfighting focuses on the tactical and operational levels of analysis, such as the ways in which AI-enabled drones are changing battlefield dynamics, or whether integrating AI into offensive cyber operations will bring greater benefits to cyber attack versus defense. Johnson, however, takes a step back and examines the consequences of AI at the strategic level: whether theories that were developed during the height of the nuclear age in the Cold War—such as deterrence, strategic stability, the security dilemma, escalation—still make sense in an age of AI. In doing so, Johnson draws heavily on the cognitive psychology literature and the role that perception plays in these theories, especially with respect to causes and patterns of escalation. Rather than reject Cold War concepts of deterrence and escalation entirely, Johnson instead argues that they need to be modified to apply to the current technological environment and the nature and number of actors in the international system. With respect to escalation, for instance, Johnson borrows from Herman Kahn’s proverbial “escalation ladder” to argue that AI is creating new, potentially dangerous (inadvertent) escalatory rungs and pathways (Herman Kahn, On Escalation: Metaphors and Scenarios, 1965). Similarly, Johnson draws on cognitive and psychological aspects of the traditional security dilemma, extending work by Robert Jervis (e.g., Robert Jervis, Cooperation under the Security Dilemma, World Politics, 30 (2) 1978) and Barry Posen’s (1991) Inadvertent Escalation: Conventional War and Nuclear Risks to illustrate how an “AI-security dilemma” may be heightened, especially in light of the indistinguishability of offense and defense in an AI context, given the inordinate role of the private sector in developing these technologies. Throughout this analysis, Johnson peppers academic discussion with specific implications for policymakers, focusing on the rivalry between the United States and China.

AI and the Bomb covers a lot of ground—perhaps too much. The biggest limitation of the book is that it tackles what are complex, difficult, and nuanced topics in rather broad brush strokes. This is helpful for readers looking for a broad survey of the various arguments and hypotheses about AI and international stability through a nuclear lens. And for a new area of inquiry, such as AI, this type of contribution is important and valuable in its own right, and will surely be a key starting point for future research in the field. However, readers who are looking for more of a rigorous social scientific examination of particular research questions pertaining to AI and various outcomes of interest (i.e., deterrence, escalation, stability) may find themselves disappointed. While much of the arguments in AI and the Bomb are understandably speculative, given the novelty of this topic, the reader is ultimately left with little in the way of specifics about the conditions under which AI is likely to give rise to stability versus instability. There is also something of a disjuncture between the pragmatism about the limitations of AI, with which the book opens, and the grimmer picture Johnson paints throughout the bulk of the analysis. Instability appears to be inevitable—or, at least, overdetermined. Yet, Johnson might have considered exploring alternative hypotheses about AI and stability, rather than instability, in the discussion of counterfactual scenarios and the risks of catalytic escalation. Each of the four scenarios—cyber false-flag, cyber-terrorism, drone swarming, and deepfake disinformation—presumes AI technologies will trigger escalation. But one could plausibly construct other scenarios in which escalation is avoided, or not as acute, and unpacking these would help refine more specific hypotheses about AI.

Overall, AI and the Bomb offers a compelling introduction to exigent questions about AI and nuclear stability. It will surely serve as a springboard for future research on this important topic.

The views of the authors are personal and do not reflect the policy or position of any U.S. government entity or organization with which they are or have been affiliated.