China is going to develop superior artificial intelligence and take over the world, it seems—at least if one listens to what many commentators have been shouting from the rooftops. Indeed, over the past year, American security analysts, policymakers, and journalists alike have increasingly used a Cold War-era analogy to describe issues around the development of artificial intelligence (AI)—specifically characterizing the state of U.S.-China technological competition as an “AI arms race,”¹ whereby the United States and China are presumably locked in competition for artificial intelligence hegemony.² “Ultimately, we will win the race for AI,” President Trump’s Deputy Assistant for Technology Policy Michael Kratsios recently wrote.³ Talk of U.S.-China “algorithm battles” even made it into a 2018 U.S. National Security Council memo.⁴

Here, the framing is clearly winner-takes-all: One nation will reap the benefits of artificial intelligence, in fashions as wide-ranging as accelerated economic growth and enhanced military capability, while the other loses out and faces defeat. But this framing of artificial intelligence is not only wrong, it’s dangerous, and it seriously hampers the ways in which American policymakers approach the management of China’s technological rise and the development of artificial intelligence within our country. The United States needs to quickly address China’s growing AI development; artificial intelligence will have important influence over the balance of international power⁵ and the future world order⁶ —but, as this report will address, a winner-takes-all arms race framing is not the appropriate lens through which to view this strategic threat.

The first chapter argues why this winner-takes-all arms race framing treats AI development as if it occurs in vacuums within the United States and China. By ignoring the interconnection and interdependence of the sectors between the two countries, this framing causes American policymakers to overlook the many mutual benefits that could arise from the furthering of global artificial intelligence capabilities. As a result, policymakers in the United States risk causing damage to AI development, missing opportunities, and mishandling AI risks—when they should focus on engaging with China on AI projects without giving up critical expertise or technologies that could potentially enhance harmful applications of artificial intelligence.

The second chapter argues why this framing incorrectly treats artificial intelligence like one technology, rather than a catch-all term that alludes to a variety of technologies. This is blatantly wrong and overlooks the varying speeds and mechanisms of AI’s development in different application areas. Once again, American policymakers may mishandle AI risks and miss out on AI upsides as a result.

A winner-takes-all arms race framing is not the appropriate lens through which to view this strategic threat.

The third chapter explains why AI competition between the United States and China is critically important, even if the winner-takes-all arms race framing is wrong. Artificial intelligence is an increasingly major factor in state power—primarily economic growth and military capability. In an era of great power competition between the United States and China, AI will therefore play an important role in the balance of international power and the future world order. Further, the country with superior implementations of AI will set important global norms around the use of AI in society writ large, which could influence governance in a number of countries.

Finally, I will offer options for reframing AI competition—away from an arms race metaphor—in a fashion which maximizes AI development in the United States while avoiding potentially problematic outcomes and hopefully limiting the contributions of AI to Chinese power.

Interpreting U.S.-China AI development as an “arms race” or a winner-takes-all competition fundamentally misunderstands the transnational nature of AI development and technological interdependence. Policy prescriptions drawn from this “race” concept will thus be ill-fit to the goals they attempt to serve. This bad fit could result in, among other undesirable outcomes, damage to AI development, missed opportunities for AI development, and mishandling of real AI risks.

Yoshua Bengio, an early pioneer of modern AI techniques, has publicly disapproved of framing AI development as a race. “We could collectively participate in a race,” he told MIT Technology Review, “but as a scientist and somebody who wants to think about the common good, I think we’re better off thinking about how to build smarter machines and make sure AI is used for the well-being of as many people as possible.”⁷ Possible idealism aside, his point strikes at a fundamental flaw with this arms race analogy: AI is developed by a vast community of scientists, developers, and researchers who are not isolated within their respective countries. As such, the development of AI is not siloed within Chinese or American borders, and the benefits of artificial intelligence are not exclusive to either of those nations. Public and private entities in both countries can benefit from developments of artificial intelligence in areas that have wide reach between nations and positive impacts on economic growth and public well-being (e.g., skin cancer detection). In other words, the development of artificial intelligence is not winner-takes-all. And contrary to what some commentators seem to believe (or at least constantly and singularly discuss), not every significant application of AI will be a weapon.

Graham Webster eloquently provides context on this fact through his analysis of winner-takes-all U.S.-China rhetoric. “Unlike the US and USSR, in which science and technology developed on largely independent tracks, the US and China are part of a globally intertwined ecosystem,” he explains. As a result, “companies and innovators in both countries would suffer if international research, development, and manufacturing were to shut down,” and “even if the US and China cut off trade with each other, both countries would still have to worry about security risks from components, since risks along the supply chain exist everywhere.”⁸ There is far more interconnection and interdependence than may otherwise be apparent in an “arms race” framing.

American firms rely heavily on Chinese manufacturing technologies (e.g., in Shenzhen),⁹ and that reliance is likely to grow as artificial intelligence applications made in the United States are increasingly deployed in drones, robots, and the like which may depend upon Chinese-made hardware. Trade and supply-chain links aid both countries in further developing artificial intelligence.¹⁰ China is a major market for U.S. AI hardware, and researchers from around the world—including between China and the United States—might work on similar AI problems, share data used to develop AI systems, or coauthor research papers.¹¹ All of these factors further bolster the interconnections between the two countries’ AI sectors.

As is perhaps expected, money, too, has an impact. In 2013, Chinese investment in Silicon Valley was at $1.17 billion. By 2015, it was $11.52 billion.¹² From 2012 to 2016, American firms had invested $2.6 billion in the other direction.¹³ While Chinese venture capitalists have invested heavily in U.S. tech companies like Uber and Airbnb,¹⁴ Tencent and Alibaba—both tech giants in China—are themselves multinational public corporations with significant ownership by international stakeholders.¹⁵ Worth noting is that both companies have extensive ties to the Communist Party of China (CCP) and the Chinese government, and are currently subject to Chinese laws mandating information-sharing with the government (e.g., via the 2017 Cybersecurity Law).¹⁶ But even this considered, there is notable economic interconnection and interdependence between AI development in the two states.

Interconnections and interdependencies between AI development in the United States and AI development in China don’t end there. China’s Tsinghua University opened an Institute for Artificial Intelligence in June 2018, for which Google’s AI Chief Jeff Dean is an advisor.¹⁷ Alibaba, another Chinese tech giant, has multiple research labs located in the United States as part of its global AI research efforts.¹⁸ China’s largest retailer has a research partnership with Stanford University’s Artificial Intelligence Lab to fund such research areas as computer vision, robotics, machine learning, forecasting, and natural language processing.¹⁹ Kai-Fu Lee, former head of Google’s operations in China, runs an AI training institution in the country that leverages the expertise of Chinese government personnel and some of North America’s “leading” computer scientists.²⁰ And Baidu, the Chinese search company, belongs to the U.S.-based Partnership for AI, which aims to develop best practices for AI technology.²¹ It is worth noting potential risks associated with these kinds of relationships, including the Chinese government’s intentions to bring Western technology talent from the United States into its own country²² and previous instances of intellectual property theft. But that still doesn’t change the existence of collaborative relationships and the value garnered from at least some of them. The United States and China have many interconnections and interdependencies between their AI sectors—and so the key is in managing the risks that result therefrom, not denying the interconnectivity and interdependence in the first place.

The “AI arms race” framing implies isolated competition between two global powers, which is clearly inaccurate when it comes to the United States and China; their AI development is anything but isolated from one another. Ideally, the “openness that is so integral to American innovation should be sustained and safeguarded”²³ rather than building walls between American and Chinese AI development.²⁴ But as a result of this winner-takes-all framing, a growing emphasis on a winner-takes-all race threatens to push the two countries’ tech sectors apart, potentially damaging AI development in ways we can’t foresee. It doesn’t matter what exactly initialized this way of thinking—for instance, some would argue this is a reaction to China’s aggressive push for AI supremacy in many areas²⁵—because this is the reality.

This bad fit could result in, among other undesirable outcomes, damage to AI development, missed opportunities for AI development, and mishandling of real AI risks.

As a result of this winner-takes-all view of AI development, mutually damaging U.S. policies towards China—like reckless trade policies or trying to hamper all collaboration on AI whatsoever—may very well hurt American AI development,²⁶ not to mention play into Beijing’s vision of “science as a tool of national greatness and scientists as servants to the state.”²⁷ The Trump administration’s recent talk of limiting the “export of artificial intelligence”²⁸ is just one recent example. This winner-takes-all, race-to-the-bottom approach, which Remco Zwetsloot, Helen Toner, and Jeffrey Ding argue is guided heavily by fear and speculation,²⁹ could further compromise the likelihood of developing sound policies that can advance mutual interests—while not lending too much advantage to Chinese AI capabilities.³⁰

In addition to crafting hurtful policies and missing opportunities to advance mutual interests, United States policymakers, as a result of the AI arms race framing, may mishandle AI risks. An increasingly common refrain, for instance, is that any privacy regulation in the United States is going to doom AI development because Chinese competitors face no restrictions on their data collection. This argument risks exploiting individuals’ information, by which American data privacy legislation—which is desperately needed—is guided by what the Chinese government has or has not done in the same vein.³¹ American companies can still remain competitive in the AI sector while working under the guidance of some form of national privacy regulation.

Even generally, total disengagement with China on issues of AI ethics is not preferable either. It is well known and oft-discussed that Chinese society has different views than American society on such issues as data privacy.³² According to the executive director at Partnership for AI, “we cannot have a comprehensive and global conversation on AI development unless China has a seat at the table.”³³ But with the AI arms race framing, U.S. policymakers may very well damage American AI development, miss opportunities for bolstering it, and dismiss or ignore ethical issues that need addressing under the belief of winner-takes-all AI competition.

Relatedly, the arms race framing often prompts discussion of artificial intelligence as a single technology—but this, too, is inaccurate and could lead to bad policymaking. There is no true consensus definition of artificial intelligence, even among experts, but it’s clear that AI is not one technology. Instead, artificial intelligence is “a catch-call concept alluding to a range of techniques with varied applications in enabling new capabilities,”³⁴ from image recognition to disease prediction. However, thinking of AI as a single thing threatens to guide policymakers to mishandle AI risks and miss out on upsides.

To provide context on the topic, much of what today’s commentators refer to as “AI” is just machine learning. While the term is often used without definition, the premise is relatively simple: Computers can identify patterns from data and use that pattern analysis to make decisions on their own.³⁵ For instance, a computer that receives labeled images of cats and dogs can, with the right algorithm, learn to distinguish between the two classes of images. Researchers would feed labeled images of the two animals, perhaps hundreds or thousands at a time, to the model. As this occurs, the model begins to make observations about what characterizes each type of image, as well as what distinguishes one type from the other (e.g., through various statistical techniques often unknown to the programmer). Eventually, this so-called training process will be complete, at which point the human programmers would feed the machine unlabeled photos of cats and dogs, hoping it can now identify which is which. The computer’s skills on these tests provide success benchmarks from which further improvements can be made.³⁶

Thinking of AI as a single thing threatens to guide policymakers to mishandle AI risks and miss out on upsides.

This is precisely why AI should not be used in reference to one technology: The process for teaching a computer to identify pictures of household pets is different than the process for teaching it to comprehend human language. Similarly, machine learning models for facial recognition are designed differently than machine learning models used to generate risk scores on convicts or estimate someone’s likelihood of defaulting on a loan. Depending on the task, the code itself—the machine learning model and/or its specific properties—will differ between AI implementations. The same goes for the dataset, which is often specifically tailored to single use cases. Even within a single application area of AI like image recognition, detecting enemy combatants’ faces would require a notably different dataset than the one used for a cat-versus-dog image classifier.

Discussing a single arms race, though, makes the development of AI sound as if it’s focused on one technology.³⁷ Commentators, in turn, talk of China “beating” the United States in AI³⁸ without any particular understanding of what that means—the winning, or what it means for China to do it. Will Chinese tech giant Tencent develop a more accurate facial recognition system than the FBI? Will Chinese company Baidu minimize the bias³⁹ of its AI systems compared to bias in systems built by Amazon? Will China’s military drones be able to fly faster than ours, or its intelligence service’s natural language processors better spy on phone calls than an American private security firm?

In reality, it’s difficult to decipher the answers to these questions since the underlying logic behind a single AI arms race, with a clear winner and clear loser, is flawed.⁴⁰ Further, referring to China as a single entity—albeit considering the government’s relative control of industry compared to the United States—still loses much important nuance. This could have many potentially damaging effects. As the U.S. government still needs to develop a cohesive national AI strategy, putting many forms of AI in a single bucket may yield disastrous risk management. Skin cancer predictors yield different economic, legal, social, political, and ethical risks than do facial recognition systems deployed in poor urban centers. Relatedly, some forms of artificial intelligence—like a system that’s world-class at playing games like Dota 2 or Go—could make for grabby news headlines, but likely have much less strategic value in U.S.-China great power competition than, say, an AI system in a lethal autonomous weapon.

Artificial intelligence is not a single technology, and policymakers must recognize that AI is instead a catch-all term that refers to many unique technologies, many of which that have their own distinctive development methods and timelines. Different applications of AI will develop at different speeds and with different levels of accuracy and effectiveness. Other factors, such as the computing power needed to run particular functionalities or the data used to test a given system, will also vary. For policymakers to better invest government resources in AI development—and to better coordinate non-governmental efforts—it’s essential that artificial intelligence’s many forms are not thrown into a single bucket and treated in the same fashion. As just iterated, doing so will almost certainly result in U.S. policymakers mishandling AI risks while missing out on critical AI upsides. Treating AI as one thing greatly oversimplifies AI development at the peril of the United States’ economic, technological, and strategic leadership.

Competing AI development in the United States and China needs to be reframed from the AI arms race rhetoric, but that doesn’t mean AI development itself doesn’t matter. In fact, the opposite is true. We are in an era of great power competition, and U.S. policymakers must pay greater attention to artificial intelligence development domestically and in China, primarily for two reasons. First, artificial intelligence will have a profound impact on state power, mainly through economic growth and enhanced military capability. Second, global leaders in AI will set norms around its use—and around the use of technology in society writ large—which will have important influence on other “undecided” states and the future international order. This is why American policymakers should focus on engaging with China on AI projects without giving up critical expertise or technologies that could potentially enhance harmful applications of artificial intelligence, whether they are in governance, business, or the military.

We are entering an era of great power competition, characterized by “struggle, change, competition, the use of force, and the organization of national resources to enhance state power.” Industrial productivity, science, and technology are critical in this struggle as well, notes international relations scholar Paul Kennedy.⁴¹ Indeed, that seems to be the case. Amidst a heap of articles on the death of the liberal world order,⁴² there is also much concern in the American national security arena around China, Russia, Iran, and North Korea—what political scientist Amy Zegart refers to as the “big four” due to their various elements of state power and the threat vectors they present (nuclear risks, cyber risks, territorial aggression against U.S. allies, and disruption of the international order).⁴³ The great power competition that Paul Kennedy identified from the twentieth century is back in full force.⁴⁴

“Great power competition returned,” declared President Trump’s National Security Strategy in 2017.⁴⁵ While the strategy isn’t without its flaws—one historian notes how “the grim worldview at [the strategy’s] core threatens to undermine the strategies that have long made U.S. global leadership work”⁴⁶—the document isn’t wrong in its premise: We are in an era of great power competition. The U.S. defense establishment similarly recognized this fact in its 2018 National Defense Strategy, asserting that

“The central challenge to U.S. prosperity and security is the reemergence of long-term, strategic competition by what the National Security Strategy classifies as revisionist powers. It is increasingly clear that China and Russia want to shape a world consistent with their authoritarian model—gaining veto authority over other nations’ economic, diplomatic, and security decisions.”

Alongside the ongoing weakening of the post-World War II international order, the strategy notes, “China is leveraging military modernization, influence operations, and predatory economics to coerce neighboring countries to reorder the Indo-Pacific region to their advantage.” Further, “As China continues its economic and military ascendance, asserting power through an all-of-nation long-term strategy, it will continue to pursue a military modernization program that seeks Indo-Pacific regional hegemony in the near-term and displacement of the United States to achieve global preeminence in the future.”⁴⁷ This is undoubtedly great power competition between China and the United States, as then-Secretary of Defense Jim Mattis further emphasized when presenting the strategy document.⁴⁸

In this great power competition, artificial intelligence is of vital importance and will become even more important for state power in the coming decades, particularly as AI accelerates economic growth and enhances military capabilities. As China enhances its state power through both of these dimensions, the United States must pay close attention and work to maximize its own gains in these respects.

Artificial intelligence is poised to contribute greatly to bolstering a developed nation’s economy. Accenture Research and Frontier Economics predict, based on research in 12 developed countries, that AI could “double annual economic growth rates” in 2035 while also increasing labor productivity by up to 40 percent.⁴⁹ McKinsey Global Institute predicts AI may deliver $13 trillion in global economic activity by 2030.⁵⁰ PricewaterhouseCoopers puts that figure even higher at up to $15.7 trillion in global GDP growth by 2030, much of which will be due to productivity increases.⁵¹

These estimates are varied, but they all rightfully predict enormous economic growth due to an explosion in AI uses worldwide.⁵² However, these gains will not be evenly spread. As research from McKinsey Global Institute articulates, “leaders of AI adoption (mostly in developed countries) could increase their lead over developing countries,” and “leading AI countries could capture an additional 20 to 25 percent in net economic benefits, compared with today, while developing countries might capture only about 5 to 15 percent.”⁵³ With the United States and China already representing the largest economies in the world, maximizing uses of AI within either nation could lead to massive gains in state power and influence on the global stage. “After all,” writes political scientist Michael Horowitz, “countries cannot maintain military superiority over the medium to long term without an underlying economic basis for that power.”⁵⁴ Further, there is in part a question of pure economic power: If Chinese companies don’t just develop better but also use that AI more profitably than American firms, China benefits economically and by extension has more resources to build state power generally. That the United States currently has significant AI talent does not mean an American edge in AI development is decisive and everlasting.

Militarily speaking, artificial intelligence is also revolutionary for state military power. The People’s Liberation Army (PLA) in China views AI as a revolutionary factor in military power and civil-military fusion,⁵⁵ just as the U.S. Department of Defense has similarly recognized how advances in artificial intelligence “will change society and, ultimately, the character of war.”⁵⁶ China is investing in this future. The PLA has already funded a number of AI military projects as part of its 13th Five-Year Plan, spanning command decision-making, equipment systems, robotics, autonomous operating guidance and control systems, advanced computing, and intelligent unmanned weapon systems.⁵⁷ In 2017, President Xi Jinping called for the military to accelerate AI research in preparation for the future of war.⁵⁸ There has even been a report of the Beijing Institute of Technology recruiting high-talent teenagers for a new AI weapons development program.⁵⁹ The Chinese government is undoubtedly preparing to maximize its AI development in the service of maximizing its military power.

That the United States currently has significant AI talent does not mean an American edge in AI development is decisive and everlasting.

The United States has started to do the same, in some respects: It has established a Defense Innovation Board for ethics of AI in war,⁶⁰ as well as a Joint Artificial Intelligence Center to develop “standards….tools, shared data, reusable technology, processes, and expertise” in coordination with industry, academia, and American allies.⁶¹ DARPA, the Defense Advanced Research Projects Agency, currently has 25 programs in place focused on artificial intelligence research, and in September 2018, its director announced a plan to spend up to $2 billion over the next five years on more AI work.⁶² But there is still much to be done, as I’ll address in the last section.

Even within the U.S. military’s approaches to artificial intelligence, as one West Point scholar notes, “the military is facing some hard questions about how it will adapt its culture and institutions to exploit new technologies—and civilians face a tough job ensuring they answer them effectively.”⁶³ There are certainly military leaders aware of this fact—in announcing the $2 billion in AI funding, DARPA’s director depicted it “as a new effort to make such systems more trusted and accepted by military commanders”⁶⁴—yet the road ahead will have its challenges. In general, the U.S. defense apparatus’ willingness to engage in cultural and operational shifts will greatly influence how successfully AI is integrated into the United States military.

It’s also important to note that China’s government and its private companies will likely be less constrained by ethical and legal norms when developing AI than will their American counterparts.⁶⁵ Faster deployment of and greater experimentation with AI may result, even though this may lead to perhaps chaotic or more unpredictable deployments of artificial intelligence—or, perhaps, plainly unethical uses of AI. This leads into the second main reason why U.S.-China AI competition still matters.

Artificial intelligence is increasingly enabling authoritarian governance around the world. Many commentators have referred to this as “digital authoritarianism,”⁶⁶ by which technologies like AI—deployed at scale to, say, bolster citywide facial recognition—enable or enhance authoritarian principles of state governance. Chinese companies have been complicit—and at times instrumental—in the diffusion of the technology and practices that enable this authoritarian governance. While Western companies sometimes export surveillance technology to dictators as well, most democratic governments take steps to prevent it through export controls, such as those established in the Wassenaar Arrangement.⁶⁷

Artificial intelligence is increasingly enabling authoritarian governance around the world.

Chinese firms have exported facial recognition systems to governments in Singapore, the United Arab Emirates, Zimbabwe, and Malaysia. Broader categories of surveillance technology—from national identity cards to biometric sensors, internet monitoring software, and more—are also exported by Chinese companies to governments in Ethiopia, Ecuador, South Africa, Bolivia, Egypt, Rwanda, and Saudi Arabia.⁶⁸ In addition to enabling insidious social control and large-scale human rights abuses, such spread of surveillance technology—which includes AI—consolidates power in the hands of governments that are hostile, or typically align with powers hostile, to American interests.⁶⁹ Democracy is under attack around the world,⁷⁰ and authoritarian uses of AI further harm democratic norms and the state of democracy globally. This is especially true for the many countries who remain undecided in their use and regulation of technologies like artificial intelligence.

This is an obvious strategic problem for U.S. policymakers, especially as Beijing has already sought to become a global leader in technology norms through an assortment of proposals at the likes of the United Nations⁷¹ and through channels such as IEEE.⁷² Unsurprisingly, this interest has already been reflected with respect to artificial intelligence. The China Electronics Standardization Institute, involved in standard-setting under China’s Ministry of Industry and Information Technology, released a white paper in January 2018 on a framework for AI standardization.⁷³ In April 2018, Chinese organizations hosted a major international AI standards meeting in Beijing.⁷⁴ Similar to how China is proactive in trying to set global norms around the internet, it is also intent on setting global standards and norms around artificial intelligence and its use in society.⁷⁵

As Jeffrey Ding, Paul Triolo, and Samm Sacks explain, “realizing that China’s many large companies are increasingly global players, and Chinese-developed AI algorithms will have effects on users outside of China, China’s government aims to advance global efforts to set standards around ethical and social issues related to AI algorithm deployment. Should Chinese officials and experts succeed in influencing such standards and related AI governance discussions, the policy landscape may skew toward the interests of government-driven technical organizations, attenuating the voices of independent civil society actors that inform the debate in North America and Europe.”⁷⁶ This may very well become a greater trend—as China’s government and private industry develop increasingly sophisticated AI applications—that lends even further global influence to China alongside bigger economic advantage.

It’s not just through diplomats, trade negotiators, and military officers that the Chinese or American government will exercise influence in areas of competition between the two countries (most obviously Southeast Asia, Africa, and Latin America). Without a clear, viable model of digital governance to oppose digital authoritarianism, the United States stands to lose political influence over a number of countries in these areas. This, too, will significantly impact great power competition, potentially in ways that diminish the relative importance of military power.

The Chinese government, in this larger process of global norm-setting on uses of contemporary technologies, is attempting to remake the world in its image. This is why U.S. policymakers desperately need to reframe their thinking on AI competition.

We are in an era of great power competition, and the United States and China are undoubtedly in competition with respect to artificial intelligence. AI, which is a catch-all term for a number of technologies, will impact state power—primarily via economic growth and military capability—and allow global norm-setting on AI and technology writ large in fashions that impact the future world order. In short, artificial intelligence is a vital element of U.S.-China great power competition. But the winner-takes-all arms race view of this competition is wrong and dangerous for American policymaking, which is why it must be reframed.

First, U.S. policymakers need to understand the interdependence and interconnection of AI development between the United States and China. Competition is a fine way to put it, but an arms race sounds as if AI development is siloed within each country. This leads to impractical statements about export controls on artificial intelligence writ large—which anyone aware of AI research’s open source nature would certainly dismiss as impractical.

Some experts have agreed on the value of paying attention to China potentially exploiting U.S. policy gaps to undermine American technological advantages, which the export controls in some sense address. But others point out that top universities and businesses alike are concerned about “possibly throttling a vital source of research” due to proposed blocks on industry collaboration.⁷⁷ And it isn’t just about research collaboration. Important funding and resources for American AI research could also be blocked as a result. The Center for Data Innovation, for instance, writes that export restrictions on AI technologies could “substantially reduce” opportunities for American firms to sell their AI products and services, thereby harming U.S. AI competitiveness.⁷⁸ Jack Clark, head of policy at OpenAI, holds that “the number of cases where exports can be sufficiently controlled are very, very, very small, and the chance of making an error is quite large.” Further, MIT’s R. David Edelman notes, trying to distinguish in export control policies between what is military versus civilian use of AI “may be impossible.”⁷⁹ In short, overlooking U.S.-China interconnection and interdependence in AI development may very well lead to policies that try to sever connections between AI sectors and thereby harm AI development in the process.

Rather than try to blindly and broadly apply export controls to AI under the myth of winner-takes-all AI development, American policymakers should focus on ways in which China is actually aided in a winner-takes-all fashion by American ideas or resources—like through its theft of American intellectual property.⁸⁰

In September 2015, President Obama and President Xi announced at a White House press conference: “We’ve agreed that neither the U.S. or the Chinese government will conduct or knowingly support cyber-enabled theft of intellectual property, including trade secrets or other confidential business information for commercial advantage.”⁸¹ This seemed to work initially, and Chinese hacking of American companies notably dropped for a brief period, though this didn’t directly translate to increased U.S.-Chinese cooperation on other cyberspace issues.⁸² (For evidence of this since 2015, see the aforementioned work my colleague and I have done to document China’s proposals for cyber codes of conduct in the UN and other international bodies, which the United States and its allies have resisted due to fundamental disagreements over issues of internet governance and so-called cyber sovereignty.)

American policymakers should focus on ways in which China is actually aided in a winner-takes-all fashion by American ideas or resources—like through its theft of American intellectual property.

It’s clear, however, that this no-IP-theft agreement has fallen apart since President Trump took office,⁸³ despite initial proclamations that the Trump administration would stick with the agreement.⁸⁴ The volume of hacking in this vein is back up. By the estimation of one independent commission, China now accounts for 50 to 80 percent of the annual $300 billion in American economic losses from thefts of intellectual property.⁸⁵ A March 2018 report from the Office of the U.S. Trade Representative reached similar conclusions: “Beijing’s cyber espionage against U.S. companies persists and continues to evolve,” as “Chinese state-sponsored cyber operators continue to support Beijing’s strategic development goals, including its S&T advancement, military modernization, and economic development.”⁸⁶ Without getting too much into the nuance of the Obama-Xi agreement⁸⁷ and the changes in Chinese hacking that resulted therefrom, the point is that American AI development is necessarily harmed by China’s industrial espionage, both online and offline.

The same goes for the potential security risks of Chinese investments in American AI companies; a January 2017 report to the president warned, for instance, about China’s challenge to American leadership in semiconductors—which make the microchips in many advanced technologies—via investment in U.S. firms.⁸⁸ But between these problems and broad, sweeping plans for limiting industry collaboration on AI, the U.S. government cannot effectively combat China’s technological rise without recognition of the interconnection and interdependence of American and Chinese AI development.

Second, U.S. policymakers cannot approach artificial intelligence as if it’s one technology. Doing so treats all AI implementations as the same, which is wrong—and leads to narrow thinking about how to bolster AI development. To see this in practice, look no further than U.S. policymakers’ intense focus on AI’s military applications at the cost of neglecting its non-military ones (perhaps another result of calling it an arms race).

Particularly for national security professionals who speak vaguely (and widely speculatively) of a world with automated fighting and a changed character of battle and war, military applications of AI—such as autonomous surveillance drones or intelligently automated command and control systems—are a stereotypical answer for how AI will impact state power. This answer is not wrong. China, as already discussed, well recognizes this fact, as does the United States and other countries like Russia⁸⁹ who have invested in defense-focused AI applications.

But many other non-military applications of AI are particularly important for state power in the ways they could boost the economy, and they too must be a part of policymakers’ thinking on the technologies captured in the term artificial intelligence. Healthcare, for instance, is a particularly promising area for AI’s impact on economic growth. Cancer detection, eye health, coma treatments, and depression prediction are just some of the varied ways in which AI implementations are already revolutionizing medicine around the world.⁹⁰ Disease prediction in particular has received much research attention, insofar as AI systems may combat such issues as doctors’ decision fatigue.

This is not to overstate the ability of modern machine learning algorithms to identify cancers or predict epidemics; many legal and ethical issues plague AI in healthcare (e.g., data privacy, AI bias) and other challenges such as data labeling, data sampling, and clinical integration will put additional limits on how, and how quickly, AI implementations will impact the American and global healthcare systems.⁹¹ But this is to say that the U.S. government should not only focus on military applications of AI. This entirely ignores potential AI application areas with promise to improve quality of life and greatly boost American economic power in the process.

Moreover, in either case, AI is still not one technology: Facial recognition systems deployed in military drones are different than natural language processors used to spy on phone calls, and image recognition algorithms to detect brain tumors are different than intelligent systems that manage hospital supply chains. Yet all could fall under the banner of AI, and all can have important impacts on state power via the military and the economy.

U.S. policymakers must therefore prioritize investments in AI, and policies towards AI development, that attempt to maximize state power in both military and economic dimensions—all while understanding that Congress and other bodies must address the legal and ethical issues raised by the various forms these AI implementations may take.

Without a doubt, the U.S. government needs to invest more in developing artificial intelligence within its borders. Congress must work to advance standard development for safe artificial intelligence⁹² and explore regulating certain public uses of AI that disproportionately harm minorities and other already disadvantaged groups, such as racially biased facial recognition in urban centers.⁹³ Both of these policy actions would help integrate AI into American society in ways better aligned with democratic principles. This would further help to promote global norms around democratic uses of AI.⁹⁴

More broadly, the United States needs a whole-of-government approach to artificial intelligence that its Chinese counterparts have already begun to execute. The United States must focus on deciphering the investments and policies most important to maximizing military and economic gains, while still working carefully to promote and ensure democratic uses of AI domestically. For instance, Congress should work to support collaborative health research related to AI while still working to legally protect the privacy of patient information. The National Institute of Standards and Technology (NIST) should work with industry trade organizations like the Institute of Electrical and Electronics Engineers to help set standards for ethical uses of AI facial recognition, while the American defense apparatus should do similar, parallel work in the military vein. A multi-pronged and multi-stakeholder approach is needed for AI development. This is especially true given China’s multi-pronged and multi-stakeholder investment in AI.

The United States needs a whole-of-government approach to artificial intelligence that its Chinese counterparts have already begun to execute.

Chinese business and government entities have poured billions of dollars (USD) into artificial intelligence development over the last decade.⁹⁵ The Chinese government has also released a variety of plans and held a number of dialogues on further developing AI, which spans the founding of AI-specific educational institutes, creating AI majors at universities, and communicating and coordinating AI research among research institutes, universities, enterprises, and military industry.⁹⁶ It remains to be seen how well these plans will be executed upon,⁹⁷ but their construction nonetheless reflects government effort to bolster AI development within Chinese borders—clearly, acknowledging AI’s many forms.

In sum, the Chinese government’s actions are “a clear indication of governmental commitment to this agenda at the highest levels,”⁹⁸ while the United States, meanwhile, has yet to implement a cohesive, national AI strategy.⁹⁹ The U.S. Treasury Secretary said back in 2017 that AI worker displacement was “not even on [their] radar screen.”¹⁰⁰ (This is only further evidence of U.S. policymakers focusing, at the highest levels, too much on AI’s military applications and not enough on its potential applications in, say, healthcare.) While the United States has strong advantages in developing various forms of artificial intelligence—such as a talented workforce¹⁰¹ and highly influential research coming from its scholars and practitioners¹⁰²—the country still needs whole-of-government investment in developing artificial intelligence. The likes of a Defense Innovation Board for ethics of AI in war,¹⁰³ a Joint Artificial Intelligence Center,¹⁰⁴ and a new Congressional AI commission,¹⁰⁵ while valuable steps, are not enough.¹⁰⁶

U.S. federal agencies should all be strategizing about the research, development, and implementation of AI in their organizations, and this should be happening with top-down direction from the White House. Congress should simultaneously be exploring regulatory data privacy frameworks that seek to maintain AI competitiveness in the military, government, and industry while still protecting consumers’ and citizens’ information. But these lofty goals must start with a few tangible policy steps.

• Stop with the AI arms race rhetoric. American policymakers must acknowledge that AI development is not winner-takes-all and that AI is not a single technology—and then ditch the arms race framing. Journalists, too, should take greater care in reporting on AI development in ways that don’t imply a winner-takes-all competition. Alongside this, the national security establishment—spanning think tanks, academia, and high-level U.S. policy offices—should, alongside writing and strategizing about AI’s impact on military capability, take care to put similar focus on AI’s impact on economic power. The U.S. government in particular should explicitly include the influence of AI on economic power in national security and defense strategies, paying far more attention to non-military AI applications.
• Develop a national AI strategy. The White House needs to develop a cohesive, national AI strategy document highlighting the importance of AI for bolstering economic and military power, as well as the importance of working to promote democratic uses of AI domestically and around the world. From China to France, other countries have done so—yet the United States has not, despite vague term-dropping of “artificial intelligence” in such documents as the 2017 National Security Strategy¹⁰⁷ or the 2019 National Intelligence Strategy.¹⁰⁸ Individual agency strategies or reports on artificial intelligence, such as that from the Director of National Intelligence,¹⁰⁹ are not enough either, and the February 11 Executive Order on maintaining American leadership in AI¹¹⁰ is still not a cohesive, national strategy that compares to what China has developed. In spite of the Trump administration’s unprecedented vacancies in the White House Office of Science & Technology Policy,¹¹¹ the White House should also hire, and consult with, artificial intelligence experts in the design of this strategy.
• Bolster diplomatic cyber capacity. Especially after the closing of the Office of Coordinator for Cyber Issues at the U.S. State Department—amid broader State Department cutbacks and the retraction of American diplomatic arms—the U.S. federal government needs to devote more diplomatic capacity¹¹² to fighting the model of digital authoritarianism that China currently champions. This involves such policy actions as helping smaller countries build diplomatic cyber capacity; building international norms that champion the value of a global and open internet and ethical uses of ethical AI; and emphasizing the value of democratic uses of AI for economic growth. Major agreements and dialogues around cyberspace and AI are occurring in international forums, yet the United States isn’t nearly active enough in delivering a clear, cohesive message that opposes digital authoritarianism. The State Department has announced the creation of a new cybersecurity bureau,¹¹³ but this should be further supplemented by greater diplomatic focus on AI—on formal agreements, standard-setting, and global norms and practices around AI’s use and regulation in society.
• Tighten controls on selling AI surveillance products to dictators. American policymakers must also evaluate how some of its own private companies slip through the cracks of existing export controls and sell surveillance technology to authoritarians like Saudi Arabia.¹¹⁴ Such practices make the United States look hypocritical and serve to further justify China’s narrative around undemocratic uses of AI for social control and “national security.” From the 2013 multilateral arms-control Wassenaar Arrangement, for instance, the United States has not implemented the “IP network communications surveillance systems” control, unlike the entire E.U. bloc and most other Wassenaar participants.
• Address Chinese intellectual property theft. While many research collaborations between American and Chinese AI sectors undoubtedly hold benefits for both countries, the United States is losing massive technological advantages in a number of sectors due to China’s widespread IP theft; and this certainly includes AI. It’s likely that the widely publicized 2015 U.S. indictments of Chinese hackers influenced the subsequent Obama-Xi agreement, so the Department of Justice should make it a clear priority today to indict Chinese hackers for stealing American IP. Among other policy changes, this would be helped by bolstering incentives for the FBI to prosecute cybercrime cases, increasing interagency cooperation, and bolstering diplomatic cyber capacity.¹¹⁵ Borrowing from Lorand Laskai and Adam Segal at the Council on Foreign Relations, the U.S. government should also combine indictments for Chinese IP theft with targeted sanctions against the entities from which thefts occur, and work with potential American targets to strengthen their cybersecurity.¹¹⁶
• Enact national data privacy legislation. The United States must develop a stance on data governance that contrasts with China’s model of pervasive government surveillance, while still upholding democratic norms around consumer protection. To think that any and all privacy laws will massively hinder American AI development is not only quite speculative, but follows a dangerous narrative whereby ethical considerations around AI should be cast aside in the service of trying to bolster national AI power. Not only does this narrative seek to serve major American technology companies which desire minimal regulation,¹¹⁷ but it also ignores the importance of the United States and its allies upholding democratic norms around AI—which includes addressing such issues as AI bias and data privacy—in order to promote a less authoritarian global order.

While the United States needs to worry about China’s AI ambitions, an arms race framing is not the right approach. Before any true policy changes can be made to aid the United States in this great power competition with China, American policymakers at the highest levels—as well as American academics, journalists, and national security analysts writ large—must ditch the AI arms race metaphor.