AI is stabilizing.

From a nuclear policy perspective, the majority of scholarship addressing the implications of AI, with a few important exceptions,8 argues that military applications of AI will have a destabilizing effect on strategic stability and global security. Areas regularly cited as problematic include the potential for deep fakes, human bias carried over into AI programming, cyber vulnerabilities, exacerbating great power rivalry, application to nuclear command and control, security dilemmas, and potential for loss of human control.9 This focus creates a rather defeatist paradigm, particularly for policymakers confronted with rapid advances in technology and limited thinking on how it might be applied in a stabilizing way. Whereas many of these studies treat AI as a monolith, we argue that the opportunities and risks associated with AI depend on its application, and considering their impact on nuclear weapons policies requires a greater degree of granularity. Can some military applications of AI strengthen deterrence, for example? Might it facilitate advances in arms control, at a time when the international community desperately needs new ideas and new verification techniques?

Despite the potential risks inherent with adopting AI in nuclear systems, many of the potential benefits to both deterrence and arms control have been overlooked or ignored. Additionally, many in the nuclear policy community rely on the most extreme negative examples to make the case that AI is too dangerous to adopt into nuclear policies and practices. These views are short-sighted and ignore the reality that AI will increasingly play a role in all aspects of our civil and military policies—and that potential great power competitors such as China and Russia are investing heavily in AI-related military technology.10 The broader nuclear policy community should therefore start thinking now about how to harness the power of AI as a tool for greater stability, transparency, and security as well as steps that can be put in place to mitigate the most significant risk factors.

Potential benefits of AI to deterrence and arms control have been overlooked or ignored

We argue that considering AI’s impact on nuclear weapons—both in the field of nuclear deterrence and for strategic arms control—requires a more nuanced approach, whereby it offers both opportunities as well as challenges for strategic stability. Because the risks of AI are increasingly familiar and discussed in increasing scope and detail by experts and scholars, we choose here to focus our exploration on the potential benefits of AI on nuclear policy, which have received far less attention. We do not mean to diminish the potential risks associated with increasing reliance on AI, but rather to offer a more optimistic picture of how AI might also contribute to nuclear deterrence and arms control in unique and helpful ways in order to answer this question: how can decision-makers pursue potentially stabilizing applications of AI while managing the risks associated with AI and nuclear systems?

More Electricity than Lightbulb—What Is “AI”?

Trying to define “AI” has become somewhat of a parlor trick. Is it software or hardware? Killer robots or drone swarms? Director of Perry World House, a center at the University of Pennsylvania, Michael Horowitz, offers a helpful framework wherein AI can either be applied broadly as “systems that select and engage targets on their own,” such as computer-guided precision weapons, or narrowly as “intelligent machines capable of cognitive judgements on par with humans.”11 According to the US Air Force and NATO, AI “refers to the ability of machines to perform tasks that normally require human intelligence—for example, recognizing patterns, learning from experience, drawing conclusions, making predictions, or taking action—whether digitally or as the smart software behind autonomous physical systems.”12 AI is often associated with lethal autonomous weapons (LAWS), which are weapons systems that can be activated to select and engage targets independently; however, we are interested in a broader range of AI military applications.

While AI systems may behave autonomously, they are not necessarily automatic. For decades, militaries have relied on automatic technologies, such as missile defense interceptors or perimeter security systems, that do not rely on AI. AI is also not synonymous with “unmanned” systems like unmanned aerial vehicles (UAVs) or underwater drones. What makes AI and machine learning unique from automated or unmanned systems is the ability to process massive amounts of data and make an independent analysis to achieve a desired objective. There are, of course, varying degrees of autonomy within AI, depending on a systems’ ability to move, take/provide direction, and make determinations.13 But whereas automatic systems are reactive, AI is more proactive. It can sift, sort, identify, and offer an initial conclusion on what something means.14 AI is particularly valuable in a society that creates ever-increasing amounts of data—for example, in 2020, the world was estimated to produce 44 trillion gigabytes of digital data, an amount humanly impossible to synthesize.15

To be clear, AI, machine learning, and deep learning are distinct but often confused concepts that fall under data synthesis.16 AI is the umbrella term describing the capability of a machine to imitate intelligent human behavior, such as complex decision-making. Machine learning is a method of AI whereby computers have the ability to learn without being specifically programmed for a particular task, and deep learning is a subset of machine learning where a computer has the ability to learn more complex relationships between datasets.

Of course, AI goes well beyond military applications. Other applications include facial recognition, which has been used in China for years and was deployed for the first time in the UK in early 2020 against a watchlist of violent offenders.17 And law firms have begun exploring the use of AI for contract and document reviews to identify relevant information and reduce human error.18 Kevin Kelly, a leading technology expert and founder of Wired magazine, posits that technologies such as AI are “inevitable” and sees AI as akin to electricity.19 AI is not the lightbulb or a single application of electricity—it is the electricity itself, whereby it is a fundamental enabler of civilian and military systems and technologies. In short, AI is an unavoidable technology for governments, societies, and militaries.

AI is an unavoidable technology for governments, societies, and militaries

Whether or not AI is as problematic as many people fear thus depends on politics more than the technology itself. In a recent article in The Washington Quarterly, researcher James Johnson cautioned that AI could create instability but also acknowledged that “AI is unlikely to be a strategic game-changer. Instead, it will likely reinforce the destabilizing effects of existing advanced conventional—especially counterforce—capabilities, thereby increasing the speed of warfare and compressing the decision-making timeframe.”20

The impact of AI will also depend on its application in relation to other technologies. For example, if AI is adopted by a nuclear-armed nation with an aggressive nuclear neighbor, AI could have a stabilizing effect by improving early warning systems, reducing the pressure for early launches, and preventing miscalculations. Conversely, given the current lack of methods to test the safety and reliability of machine learning and other AI tools in nuclear weapons technology, integrating AI into nuclear systems may have a destabilizing effect, potentially undermining confidence in deterrence capabilities or creating additional vulnerabilities in nuclear systems and decision-making.

Nuclear Deterrence—A Role for AI in Strategic Stability

Traditionally, nuclear deterrence relies on a country having credible nuclear capabilities plus the political will or resolve to use them if necessary (and only if necessary), therefore persuading an adversary to not attack. Integrating AI into nuclear systems can enhance nuclear decision-making, improving perceptions of both capability and resolve.

Early Warning and Detection: Avoiding Unnecessary Use and Escalation

Most significantly, integrating AI into intelligence, surveillance, and reconnaissance (ISR) systems and using AI as part of the analytical tool kit for early warning and detection could improve target identification, prevent false positives or close calls, and increase understanding of adversary actions. AI can be used to increase a nation’s ability to understand the nature of its environment as well as speed of analysis and development of courses of action in a conflict, thus improving its overall capabilities in the nuclear space and reducing the risk of unintended escalation in the midst of a crisis with a nuclear dimension.

In a nuclear exchange, national leadership may have just minutes to assess a situation and determine whether to launch its own nuclear response. Nuclear weapons states such as the United States have sought to increase this decision-making time to reduce the potential for misunderstanding or miscalculation in a crisis, given the immense consequences of using nuclear weapons.21 Incorporating AI into early warning and decision-making analysis may improve both the speed and quality of information processing as well as eliminate potential biases from military leadership, which will be crucial in a crisis to allow time for de-escalation and tension reduction between sides.