On Assessing the Risk of Nuclear War

What are the risks of nuclear war in all its potential manifestations? This is not an easy question to answer, and I do not propose to answer it here. Rather, the more tractable question is whether the process of studying it could yield policy-relevant insights even if it is unlikely to lead to a precise determination of the actual risks of nuclear weapons use. In this chapter, I summarize the current state of analysis regarding the likelihood of nuclear war, focusing on The Lugar Survey on Proliferation Threats and Responses, the Bulletin of the Atomic Scientists’ Doomsday Clock, and a sampling of analysts’ estimations of the likelihood of interstate nuclear war and nuclear terrorism. These estimations differ widely and are all of questionable validity because they are either fundamentally intuitive or based on very simple—even simplistic—analyses. Can we improve on this state of analysis by using more structured and more comprehensive approaches to provide a sounder basis for policies that will inevitably be based on imperfect analyses of the likelihood of nuclear war?

Case studies are useful in analyzing infrequent events because they can assess “close calls” in which such events could have occurred, as well as those instances in which they actually occurred. Nuclear weapons have been used twice, but there have been many more close calls. This chapter outlines an agenda for using case studies to assess the risks of nuclear weapons use. First, it identifies twelve cases in which leaders used, seriously contemplated using, or might have considered using nuclear weapons. Second, it notes thirteen cases of close calls of accidental or unauthorized detonation of a nuclear weapon. Third, it assesses three possible paths toward the use of nuclear weapons by non-state actors, none of which as yet has had any known close-call incidents. The chapter then briefly assesses how the historical risks of nuclear weapons use might change as the world evolves toward a larger number of nuclear weapons states. Finally, the chapter develops policy-relevant questions on the risks of nuclear weapons use that can be addressed through case studies, including the behavior of new nuclear weapons states, the likelihood of nuclear weapons use by field commanders versus that by national command authorities, the safety trade-offs of dispersed versus centralized nuclear weapons sites, and the differences between contemporaneous and historical evaluations of nuclear risks. These contributions are unlikely to lead to clear point estimates of nuclear risks, but they may help identify which paths toward possible nuclear weapons use deserve more attention and how risks on these paths can be reduced.

Every decision and problem solution involves the use of knowledge gained from the experiences and thought processes of humans. Even for data-rich problems, humans influence how data are gathered, interpreted, modeled, and analyzed. For data-poor problems, such as those assessing risks of never-seen, rare, or one-of-a-kind events, knowledge from experts may be the sole available source of information. Assessing the risk of nuclear deterrence failure is an ill-posed problem that falls into the data-poor category. As a result, experts are needed (1) to supply the information and knowledge for the risk assessment and (2) to define and structure the deterrence problem. These two uses of elicited expert knowledge are discussed. For both, formal elicitation methods for bias minimization are recommended and briefly described. Formal elicitation also involves planning and the use of methods for obtaining the best-quality information from the experts’ thinking and problem solving. This formalism includes the characterization of uncertainties, which are prevalent in the deterrence problem, and the analysis of the elicited information, which is necessary for assessing the likelihood and consequence constituents of risk.

Probabilistic risk assessment (PRA) can provide a quantitative estimate of risk for catastrophes that have not yet occurred by analyzing sequences of events that can lead to that event—in our case a major nuclear war. PRA is also useful for reducing that risk by identifying potential paths to nuclear weapons use that otherwise might escape attention. While PRA has been embraced in nuclear power, spaceflight, and other engineering fields, there are significant challenges to transferring that experience to the risk of nuclear deterrence failing. In-depth PRA of nuclear deterrence holds promise but requires significant further research. Fortunately, a simple approach can be used to show that the risk of nuclear deterrence failing currently appears to be on the order of 1 percent per year. It is hoped that this surprising result will cause society to invest in the larger efforts required for in-depth analysis, both to estimate and to reduce the risk of a major nuclear war.

Even if the US Cold War nuclear deterrence system could be regarded as a triumphant success because no nuclear war occurred between the United States and the Soviet Union, the strategic nuclear deterrence system of today must contend with a geopolitical landscape far more complicated than that of the Cold War. Seven acknowledged nuclear powers exist, and others, including transnational organizations, are attempting to join their ranks. Multiple nuclear states, nuclear capabilities that vary widely in technological sophistication, and different levels of stockpiles and security implementations all suggest that the nuclear deterrence landscape is far more uncertain in its risk of failure than at any other time in history. These components also suggest that the nuclear deterrence system has features that are consistent with the formal definition of complex systems; therefore, complex systems theory is most appropriate for addressing fundamental questions of risk. We explore these features and discuss failures from the points of view of accidents and human error or missteps, drawing on treatments of complex systems in general and the Cuban missile crisis in particular. We suggest how fundamental research in complex systems theory can contribute to assessing the risk of failure of nuclear deterrence. Whether formal modeling of nuclear deterrence systems can provide practical utility in a multipolar nuclear world has yet to be determined. We suggest construction of simplified mathematical models as a first step in grappling with the complexities of systems of nuclear deterrence. We propose that assessment of the risk of failure of nuclear deterrence associated with the close calls in the Cuban missile crisis would be a practical test of preliminary understanding of this complex problem.

The considerable body of knowledge on the consequences of nuclear weapons use—accumulated through an extensive, sustained, and costly national investment in both testing and analysis over two-thirds of a century—underlies all operational and policy decisions related to US nuclear planning. We find that even when consideration is restricted to the physical consequences of nuclear weapons use, where our knowledge base on effects of primary importance to military planners is substantial, there remain very large uncertainties. These uncertainties exist in no small part because many facets of the issue, such as the effects on the infrastructures that sustain society, have not been adequately investigated. Other significant uncertainties in physical consequences remain because important phenomena were uncovered late in the nuclear test program, have been inadequately studied, are inherently difficult to model, or are the result of new weapon developments. Nonphysical consequences, such as social, psychological, political, and full economic effects, are even more difficult to quantify and have never been on any funding agency’s radar screen. As a result, the physical consequences of a nuclear conflict tend to have been underestimated, and a full-spectrum all-effects assessment is not within anyone’s grasp now or in the foreseeable future. The continuing brain drain of nuclear scientists and the general failure to recognize the post–Cold War importance of accurate and comprehensive nuclear consequence assessments, especially for scenarios of increasing concern at the lower end of the scale of catastrophe, do not bode well for improving this situation. This paper outlines the current state of our knowledge base and presents recommendations for strengthening it.

Analyses of the effects of nuclear weapons have traditionally focused on the physical destruction they produce, especially their human toll, devastation of cities, and damage to the environment. To the extent that nonphysical effects are taken into account, strategists have emphasized the influence of nuclear weapons on national decision-making, particularly whether a limited strike would escalate to an all-out nuclear exchange. Yet, the range of nonphysical weapon effects is much broader, encompassing social, psychological, political, and economic impacts that would reverberate long after a nuclear attack. In a limited-use scenario, these ramifications—designated as “intangible” effects in this analysis—may greatly surpass the physical damage incurred, just as the cost and scope of the response to September 11 dwarfed the direct effects of the attacks. Moreover, unlike physical phenomena, many of these intangible effects are the result of human decisions and are thus theoretically controllable. Given that the limited use of nuclear weapons is probably more likely than a massive nuclear war, there is a pressing need to understand these intangible effects and identify practical steps to minimize them.

For multifaceted problems such as assessing the risk of nuclear deterrence failure, data, information, and knowledge can emerge from many different sources involving diverse subject areas and in myriad qualitative or quantitative forms. Often the amounts of data, information, and knowledge are limited, apply to rare events or events that have never occurred, or both, necessitating the combined use of all sources. For example, sources include historical data on past events; expertise from authorities in different subject areas; and knowledge about past and current cultures, human behaviors, sociology, politics of people and states, as well as the theory or rules governing politics. Regardless of source and form, available knowledge has uncertainty attached. Some uncertainties can be significant, and the uncertainties themselves can be of different types. Depending on the type of uncertainty, quantification may not be feasible or the appropriate mathematical theory for it may be difficult to apply. Nonetheless, decision- and policy-makers need a final or top-level answer about nuclear deterrence failure accompanied by an understandable uncertainty. Knowledge integration methods address these needs and provide ways to tackle other difficulties encountered when combining all available data, information, and knowledge and their associated uncertainties to produce an assessment of risk. Some of the integration principles and methods are described in this chapter, especially those related to the challenges in assessing the risk of nuclear deterrence failure—a problem of significant uncertainties and poor data, information, and knowledge.

Motivated by the importance of the perceived risk of nuclear deterrence failure in national security policy formulation, we began our study by asking whether more structured analytic approaches could improve on the highly intuitive manner by which the risk of deterrence failure has generally been assessed. For the likelihood dimension of risk, each of the approaches included in this book—case study, elicitation of expert judgment, probabilistic risk assessment, and application of complex systems theory—has something unique to offer. However, none of these approaches can do the job by itself. Rather we have reinforced the notion that multiple disciplines can each shed limited light on the question. We must extract from each of them whichever valuable insights they offer and do our best to synthesize these insights, using the art and science of knowledge integration, into a policy-relevant assessment. However daunting this task, discernible research paths hold significant promise. As for the physical consequences of nuclear use, it is clear that our knowledge base, derived primarily from concern about the military effectiveness of nuclear weapons, is inadequate to assess the potential consequences from the broader array of nuclear uses that now appear possible or from intangible consequences that could exceed even the physical consequences. This lack of knowledge is easier to address from an analytic perspective but requires an adequately funded research program. The dim prospects of such a program are yet another consequence of the complacency induced by our intuitive sense that nuclear weapon risks have largely abated.