AI and National Security: The Interior Dimension of Grand Strategy in the U.S.-China Tech War

By Matthew Waxman

Artificial Intelligence (AI) and national security relate in so many significant ways that it is hard to do any justice to the topic in a short essay. What follows is an effort to synthesize and organize some of the most important issues. This essay frames the issues primarily in terms of U.S.-China competition—an arena in which these issues are especially salient—though Europe is among the other major AI players, too, and AI has the potential to scramble international power distribution and empower other actors in ways that are difficult to predict. 

In outlining some of the major ways that AI affects state power, this essay explores two big questions: Is the U.S. system better suited and positioned than the Chinese system to compete in AI? And going forward, can the U.S. system do the things it needs to compete without, in the process, eroding its systemic virtues? The answer to both is a qualified “probably.”  

AI and National Power

AI is a general-purpose technology, often defined as an ability of machines to perform tasks through processes and features that resemble (or, increasingly, far surpass) human intelligence. Technology has always been a key ingredient of national power, though historically it was rarely immediately apparent how revolutionary technology would alter national power or which states would best harness it. Assessing how AI will affect U.S. power relative to other states is especially difficult for several reasons: it is developing fast—and accelerating; much of that advancement is driven by the private sector; and although advanced AI has high barriers to entry, much AI technology is widely dispersed and readily diffuses globally. 

As Paul Scharre details in his recent book Four Battlegrounds: Power in the Age of Artificial Intelligence, national AI strength is largely a product of (i) computing hardware, especially advanced chips, (ii) data, and the right kinds of it, (iii) human talent, and (iv) institutions capable of effectively applying and adapting AI.

A basic challenge for the United States is how to secure American advantages in or access to all four while continually checking China’s own. The United States has some advantages in the first and third ingredients, and it has tried to extend its advantage in the first with export controls and subsidies, but those edges may prove fleeting. The United States, and especially American companies, have some advantages in data—which is critical to developing and training advanced machine learning systems—though China has other data advantages. Both the United States and China are working hard to improve their capacity to absorb, apply and innovate with AI in the public and private sectors. 

It is undeniable that AI will dramatically impact state power, especially relative U.S.-China power balances. As AI technology proliferates, it will also empower non-state actors in ways that are hard to forecast.

AI and Economic Power: AI—along with developments in robotics, the Internet-of-Things, and other technologies—can propel rapid advancements in productivity and innovation across all economic sectors. It can improve public services and enable more efficient resource allocation. As AI augments human capabilities and creates new jobs in the AI-economy, however, it will also displace segments of the labor force as various types of work (unskilled as well as highly-skilled) are increasingly automated. Here, a key question is whether the United States’ more open, market-driven system or China’s more top-down controlled system can more effectively harness the economic promise of AI while avoiding dislocations.

AI and Hard Power: AI is directly important to military effectiveness in so many ways, as the United States works to maintain is qualitative edge and China works to sprint ahead. AI will enhance or enable new weapon capabilities, including increasingly autonomous ones. It will improve military logistics, planning, and targeting. It will be used at all levels of command to aid and speed decision-making. In intelligence, it will enable advances in sensing and analysis, as well as deception and counterintelligence activities. And it will be integrated to both offensive and defensive cyber-operations. As AI tech becomes more ubiquitous, these capabilities will diffuse globally. A critical question is not just who will develop and integrate AI applications faster but who will master concepts for making the best use of them—as well as for countering them. 

AI and Political-System Power: AI will propel advancements in information flows and analysis, with contrasting opportunities and dangers. Consider, for example, volumes of micro-targeted, autonomously-generated content (real or fake); or consider the possibilities of automated and ubiquitous monitoring of faces and speech. For the United States, a core challenge is protecting its open, representative democracy from misinformation and disinformation—including foreign disinformation—that can be generated and amplified at larger scale and speed. For China, this means protecting state control through more advanced surveillance and censorship. These contrasting assignments are important not only to the domestic health of U.S. and PRC political systems but to how effectively they can promote and defend their political models abroad—including China’s export of AI-driven surveillance capabilities. 

Five Big Policy Tasks

Although so much development of U.S. AI innovation and incorporation is market-driven, we are already seeing movement in five key overlapping areas of public policy. There are many others as well, but these are ones in which the United States must improve as it competes with China, which has declared its aim to lead the world in AI by 2030.

1. Regulation: The United States lags behind China (and the EU) in regulating AI, though that is not a bad thing—yet. Indeed, it is often said that whereas China has already promulgated several rounds of national AI regulations and the EU is intensely drafting union-wide legislation, the United States lacks any AI regulation, but that last part is not true. The U.S. government does regulate AI to some extent, but as in so many areas, that regulation is siloed across many agencies and levels of government. While slow, poorly coordinated, and cumbersome, that decentralized American system of governance also provides something of a testbed for AI regulation at a time when no one knows how best to protect against harms without suffocating innovation. Domestic pressure for federal AI regulation is intensifying, and the right balance of regulation can build public and consumer trust without choking commercial experimentation and advancement. Soon the United States will also need to put forward a more developed regulatory model, ideally one that can influence that of Europe and other democracies, before others push internationally their preferred frameworks. 

2. Industrial Policy: “Industrial policy” is a dirty word in some quarters, but markets alone will not sustain AI leadership. The Chinese government will pour more resources, and is less politically constrained in doing so, than the U.S. government into AI development. The PRC is also more practiced in centralized, top-down industrial policy, as it exerts vastly more direct and indirect control over Chinese companies than the United States ever could or should. That top-down control can help prioritize, but it often lacks agility and is prone to placing bad bets. Meanwhile, whereas the United States has major advantages over China in AI-oriented private capital and the global reach of major American technology companies, some degree of U.S. government intervention—including commercial subsidies, tax subsidies, and research and development initiatives and funding—is needed, too. As in some other critical technologies, the U.S. government must push and pull more industrial levers than it has in recent decades but accept (and even celebrate) that China will always out-plan it. 

3. Supply Chains: AI products and algorithms will spread globally, but the United States—in partnership with allies—can take steps to secure its advantage in equipment, namely advanced chips, while slowing China’s efforts to build indigenous design and production capacity. Key policy tools include subsidies, foreign investment restrictions, and especially export controls, to which the United States will soon likely add outbound-investment restrictions. One tension is that tightening controls on China’s access to foreign supply further spurs its efforts to develop domestic alternatives. Another is that some efforts to control global supply chains cause trade frictions with the allies whose support is important. In the short term, export control efforts are already biting China’s AI development capacity, but a big question is whether and how quickly China can compensate domestically for them.

4. Human Capital: At all levels of education and training vital to developing and working with AI, China engages in top-down planning and resource-surging that is impossible for the highly decentralized American system—and it draws on a population more than four times the size. Recent Reagan Institute reports on the U.S. National Security Innovation Base have concluded that the loosely federated ecosystem of private sector firms, government agencies, and universities currently excels in innovation leadership—including high-end research on critical technologies—but is failing to develop the broad base of talent across the workforce needed to put that innovation to use. ¹ And although the United States remains a magnet for talent in ways that China cannot be, it is still not sufficiently attracting and retaining foreign expertise (e.g. through needed R&D and high-skilled immigration programs). 

5. Military Adoption: The task of harnessing AI for military effectiveness is as much about organization and culture as technology, in large part because so many AI advances are coming from the private sector, not the other way around. The Defense Department has so far made some incremental but accelerating progress in reforming its organizational structures and acquisition processes to match the speed of technological development and to draw on private-sector innovation in areas like AI. The United States and Chinese militaries (the latter as part of its Military-Civil Fusion strategy) are racing to put in place new AI systems, processes, and capabilities—and, right now, the Ukraine war is catalyzing U.S. military learning—but it is not yet clear which will better be able to scale them and, at least as important, to embrace new warfighting concepts to take advantage of them.

Concluding Thoughts

In his book, In the Shadow of the Garrison State, Aaron Friedberg argues that Cold War pressures toward extreme American government interventionism, due to the Soviet threat and arms race, were counterbalanced by deeply-rooted American anti-statist pressures. Those counter-pressures not only helped to maintain the basic republican and free-market character of the American system, but they also checked more extreme levels of government concentration and control that would have stifled the American ingenuity so critical to successful long-term competition. Though the threats are (thankfully, and for now) lower today, and the security stresses less intense, similar anti-statist counter-pressures are at work again, too. 

Although comparisons to the Cold War can be misleading or dangerous, the broader lesson of Friedberg’s book is to mind what he calls this “interior dimension of grand strategy.” As in the Cold War geopolitical competition, effectively managing the national security dimensions of AI involves balancing impulses toward greater governmental intervention with core commitments to free enterprise and dispersed power—commitments that will continue to underlie some comparative American advantages in amassing and wielding AI power.