The AI Arms Race: A Deadly Rivalry Between the USA and China

The rapid advancement of artificial intelligence (AI) in recent years has not only unlocked transformative economic potential, but also ushered in a new era of great power competition. The United States and China, the world‘s leading tech superpowers, are now locked in an intense strategic rivalry to gain a decisive edge in AI—an advantage that could determine the global balance of power for decades to come.

At the heart of this rivalry lies an increasingly unstable AI arms race, as both nations pour tens of billions into militarizing AI to achieve battlefield superiority. The Pentagon‘s unclassified investments in AI swelled from $600 million in FY2016 to nearly $2.5 billion in FY2023, while China‘s estimated military AI spending grew over 20x from 2017 to 2022.

The implications of this escalating competition are profound and potentially catastrophic. Many leading AI experts fear that without proper safeguards and international cooperation, the headlong pursuit of artificial intelligence dominance could not only undermine global stability, but even pose existential risks to humanity itself.

Artificial Intelligence as the New Nuclear Weapon

The AI arms race has often been compared to the Cold War nuclear arms race between the United States and Soviet Union. Both are characterized by a destabilizing action-reaction cycle driven by mutual fears of inferiority. And like nuclear weapons, analysts warn that AI could have a "strategic latency"—the potential for apparently benign systems to be rapidly converted to military purposes in a crisis.

However, in many ways AI is even harder to control than nuclear weapons. The knowledge and materials needed to build a nuclear bomb can at least theoretically be monitored and restricted. But according to Andrew Moore, former head of Google Cloud AI, the research and talent needed to build cutting-edge AI is already thoroughly diffused across the commercial sector and academia. "It‘s very, very hard to stop everybody in the world from building AI," he said in a 2019 interview.

Moreover, the underlying drivers of private sector AI development are economic, not military. Breakthroughs achieved by tech companies and AI labs in areas like machine learning and computer vision are rapidly dual-purposed for military aims. In the words of Chinese general Li Minghai, "The barriers between peacetime and wartime, between military and civilian, will all be broken."

This means that even defensive efforts to maintain a cutting edge in AI are inherently destabilizing, as innovations can be weaponized by rivals or stolen through espionage. An open-source AI research ecosystem that is tremendously beneficial for global innovation is also impossible to fully secure.

The Militarization of Artificial Intelligence

Artificial intelligence is being applied to virtually every aspect of warfare by the US and Chinese militaries, from intelligence gathering and analysis to logistics and weapons systems. While much attention has focused on lethal autonomous weapons, or "killer robots," this is only one alarming use case of military AI. Other key application areas include:

Intelligence, surveillance, and reconnaissance (ISR): AI can sift through massive amounts of data from drones, satellites, and other sensors to identify targets and piece together enemy force dispositions. The US National Geospatial-Intelligence Agency is developing an AI-powered "living map" fusing real-time data feeds to give commanders an unprecedented "pattern of life" understanding of the battlefield.

Command and control: AI-enabled decision support systems can rapidly analyze courses of action, model complex scenarios, and recommend optimized battle plans to commanders. DARPA‘s "Gamebreaker" program is developing AI to discover and exploit vulnerabilities in enemy systems. China‘s PLA believes AI will compress the decision-action loop and enable "intelligenized" warfare.

Cyberwarfare: AI can power a new generation of autonomous cyberweapons able to penetrate networks, identify targets, and self-propagate to maximize damage. In 2022, Chinese researchers unveiled an AI-powered "hybrid cyber warfare" system claimed to be "impossible to defend against using conventional cybersecurity protocols."

Information warfare: AI enables the industrialized production and micro-targeting of disinformation, including realistic fake videos of enemy leaders generated by AI. China‘s SenseTime is developing AI systems that can "read" human behavior for signs of anti-government sentiment. AI could be the ultimate psychological warfare tool.

Training and simulation: AI enables unprecedented war-gaming and simulation to identify weaknesses and zero-day exploits in weapons systems. DARPA‘s AlphaDogFight program used AI to defeat a human F-16 pilot in simulated dogfights 5-0. China‘s "Qinggong" (‘Skillful Work‘) system is an AI-powered war-gaming platform to design joint operation plans.

The US and China are investing heavily across all these application areas. Though specific programs are often classified, the overall trend lines are unmistakable. For instance, the US Joint Artificial Intelligence Center (JAIC) saw its budget increase from $89 million in FY2019 to nearly $278 million in FY2022. Investment in the Air Force‘s flagship AI program, the Advanced Battle Management System (ABMS), grew from $35 million in FY2018 to $302 million in FY2022.

Risks of Military AI: Unintended Escalation and Instability

The pursuit of decisive advantage through AI is not only expensive, but also profoundly destabilizing. AI could undermine existing military doctrines and blur the line between peace and war. The allure of an "AI offset" that could neutralize an adversary‘s conventional forces creates temptations for risk-taking and pre-emptive action in a crisis.

AI also introduces risks of unintended escalation due to the speed of interaction between two AI-enabled forces. Writing in the Belfer Center Paper on "The Impact of Artificial Intelligence on Strategic Stability and Nuclear Risk," researchers Horowitz, Scharre and Velez-Green warn, "The speed of interaction between two military systems with a high degree of autonomy could surpass the ability of human cognition…creating a ‘flash war‘ before political and military leaders have time to understand and react."

The black box nature of AI systems, often described as "brittle" and "unpredictable," adds further risks of miscalculation. In the fog and friction of conflict, even extensively tested AI systems could fail in unexpected ways. A noted AI researcher and professor, Yejin Choi, put it bluntly: "The notion of a 100 percent trustworthy, responsible AI system is an impossible dream."

These risks are especially acute in the nuclear domain, where the US and China are exploring AI for early warning and command and control. A 2020 RAND Corporation study, "How Might Artificial Intelligence Affect the Risks of Nuclear War?", outlines several concerning scenarios for accidental escalation, such as AI spoofing an enemy attack or recommending a pre-emptive strike based on false patterns in training data. Even if AI objectively improves the accuracy of warning systems, overconfidence in AI could cause leaders to stumble into nuclear war.

Yet despite these inordinate risks, international efforts to restrain military AI have made little progress. The UN Group of Governmental Experts on Lethal Autonomous Weapons Systems has met annually since 2017 without agreeing on even a basic definition of autonomous weapons, let alone binding regulations. The US insists on a non-legally binding "code of conduct", while Russia and China oppose any restrictions that would limit their military AI programs.

Mitigating Risks: A Multi-Stakeholder Challenge

Forestalling an unconstrained AI arms race is a profound challenge with no easy solutions. From my perspective as an AI researcher, it will require a multi-pronged and multi-stakeholder effort across at least three key fronts.

First and most urgently, the US and China need to initiate a serious dialogue on AI in the nuclear domain specifically. This should include concrete proposals for AI-related confidence building measures and crisis management mechanisms to reduce risks of unintended escalation. Though any formal arms control agreement remains highly unlikely, improving mutual understanding of each side‘s "redlines" and engaging in joint risk assessment is critical.

Second, leading AI companies and researchers need to more proactively shape the normative and ethical frameworks around AI development. Google‘s 2018 decision not to renew its contract for the Pentagon‘s Project Maven and employee protests against the JEDI program show the influence that key talent can have on the direction of military AI. In 2023, over 1200 AI researchers signed an open letter calling for a UN ban on AI systems that "pose an unacceptable level of risk." Scientists should refuse to work on AI weapons and insist the goal of their work is strictly peaceful, commercial innovation.

Finally, global civil society needs to urgently raise awareness of AI risks and build international coalitions to pressure governments. Groups like the Campaign to Stop Killer Robots and the Future of Life Institute are doing vital work, but a much broader public mobilization is needed. Only coordinated pressure from scientific, industry and civil society leaders can change national calculations around AI superiority.

Conclusion: Steering the Future of AI

The stakes of the US-China AI rivalry could not be higher. If left unchecked, the pursuit of an artificial intelligence arms race threatens to poison our politics, derail the global innovation ecosystem, and in the worst case, destroy any hopes for a beneficial long-term future with transformative AI systems.

Yet amidst this competition, it is more critical than ever that we hold onto a vision of AI as a tremendous force for good—a technology that could help solve global challenges from clean energy to pandemic response and space exploration. We must build international frameworks for the responsible development of AI that enhance, rather than undermine, our shared security and prosperity.

The path forward will be arduous and require sustained engagement from all stakeholders. But we have overcome similar challenges before in the nuclear domain and we can do so again with AI. The future of artificial intelligence is still ours to shape, if we have the wisdom and foresight to steer it in a peaceful and beneficial direction. We must ensure that in the final balance, AI remains a tool for human flourishing, not a weapon for our destruction.