We see the phrase everywhere — the United States and China are in an artificial intelligence “arms race.” It manifests in op-eds, news articles and television segments. It’s in books, think tank pieces and government documents. All this to capture the fear that another country might develop AI more powerful than our own.
But calling the AI competition an “arms race” is both wrong and dangerous. It suggests AI development is winner-take-all, in that two isolated national AI sectors struggle for total domination, leading to policies that cut off valuable interconnection. Simultaneously, it misrepresents AI research more generally by implying that this varied field is a single technology, almost inevitably focusing too heavily on AI’s military applications.
The premise that AI research is a zero-sum endeavor is especially easy to debunk. In reality, American firms invest billions of dollars in Chinese AI companies, and Chinese firms have invested tens of billions of dollars in the other direction. American firms also depend heavily on Chinese manufacturing, which will have an even greater impact on AI development, as artificial intelligence is increasingly deployed in hardware such as that of drones and robots.
The interconnections between U.S. and China AI development are also knowledge-related: To name just a few examples, China’s Tsinghua University opened in June an Institute for Artificial Intelligence, where Google’s AI Chief, Jeff Dean, is an adviser; Baidu, the Chinese search company, belongs to the U.S.-based Partnership for AI, which aims to develop best practices for AI technology; and China’s largest retailer has a research partnership with Stanford University’s Artificial Intelligence Lab to fund research areas such as computer vision, machine learning and forecasting. The open-source nature of some elements of AI research further contributes to a near-constant flow of information across borders.
To speak of AI as an arms race is also to ignore the many areas of AI development, such as the potential for improved public health outcomes, that may benefit both countries. Algorithms that better detect cancer, for instance, could notably reduce costs of care and increase the accuracy of early-stage cancer prediction. This could benefit the United States and China at once, not to mention other countries around the world. With a winner-takes-all “arms race” framing, though, U.S. policymakers may enact policies that hurt American AI development and foreclose opportunities by cutting off vital pipelines of funding, knowledge and other resources. Trump’s sweeping export controls on AI, for example, aim to limit the diffusion of certain knowledge and resources around AI to China. In the process, they might cut off beneficial relationships and exchanges and “substantially reduce” commercial opportunities for American companies.
The “arms race” metaphor is also misguided because it incorrectly treats “artificial intelligence” as a single technology. From recognizing a face to detecting skin cancer to assessing a convict’s likelihood of recidivism, different applications of AI have different properties and different sets of training data. These technologies also develop at different speeds, as they may require different data or computing power and may rely on different computer science techniques. Some (such as lethal autonomous weapons) may have wide-ranging effects on state power, while others (such as sophisticated chess programs) may function more as corporate showpieces. Equating these and other fields could easily lead us to prioritize the wrong things for the wrong reasons.
But with this “arms race” framing, policymakers and commentators talk of China “beating” the United States without understanding what “winning” means for either side. What happens if Chinese tech giant Alibaba develops better facial recognition systems than Google? Or what if China’s military drones autonomously fly faster than those developed by a San Francisco start-up? The end result for these and other scenarios is unclear, which means policymakers may not adequately invest in areas of AI development with the greatest strategic effect.
Additionally, an “arms race” framing may very well lead policymakers to mishandle the varied risks that some AI technologies present. The social, political, economic, legal and ethical challenges of a facial recognition algorithm deployed by a city’s police department are quite different from those of a racially biased skin cancer predictor or a “black box” missile-firing system. If we’re going to manage those dangers, we need to think about them carefully and discretely, which becomes more difficult when we’re just rushing to produce them first. At a time when the United States should be setting strong democratic norms around the design and use of AI — in opposition to the Chinese government’s digital authoritarianism — treating these technologies as if they were the same may yield disastrous risk management.
This doesn’t mean that the United States and China aren’t competing over AI — or that the competition is irrelevant. On the contrary, artificial intelligence will bolster national economies and enhance military capabilities, both of which are bound to have an effect on state power. As many countries around the world decide on the role of AI in society, their choices will inevitably affect the world order — influencing whether AI is used to bolster democratic or authoritarian forms of governance. That adds another worrisome complication to the “arms race” metaphor, which suggests that the United States and China are both coursing along the same track toward the same finish line. This premise could make it harder for the United States to pursue research according to more democratic norms, as it suggests that we’re just trying to snatch away whatever it is that China is grasping at before it can get to it.
The United States needs to design a cohesive national AI strategy — the recent executive order does not count, as it’s too vague and doesn’t adequately discuss a long-term American vision for AI — that addresses the many technologies at hand. China, on the other hand, does address AI’s many forms in its many documents that outline the government’s plans and ambitions for AI development in numerous domains. It’s a demonstration of commitment to AI development “at the highest levels,” from education to industrial transformation to driverless vehicles. An American strategy that approaches AI development as one “arms race” is going to fall short because it tells a story that is far too simple about technologies that are getting more complex every day.