Artificial intelligence should be hard to squeeze into the antiquated framework of nation-state competition. AI is not a territory, a scarce resource or a fiercely protected technology like nuclear weaponry; it’s theoretically borderless thanks to abundant open research; the physical raw materials required for its development — like silicon — are cheap and plentiful; and ultimately, the only major constraint to its spread is the ability of people to learn.
Yet the competition narrative, usually pitting the U.S. against China, is powerful and expressed in the terms of utmost urgency. “We must win the AI competition that is intensifying strategic competition with China,” the U.S. National Security Commission on Artificial Intelligence, chaired by former Google chief executive officer Eric Schmidt, wrote in its final report to Congress earlier this year.
Part of the problem is that more often than not, research into the state of AI reinforces the framework of its development as a competitive race between countries rather than the hodgepodge of interdependencies that globalization has made out of many markets. Consider a series of reports by Georgetown University’s Center for Security and Emerging Technology: CSET’s briefs on U.S. and non-U.S. AI hubs make it clear that Chinese entities have invested very little in Western AI start-ups. The disclosed investments by Chinese-based funders in U.S. AI companies from 2000 to 2020 add up to a mere $1.9 billion, 2.4% of total investment. In Europe, Canada, Australia, New Zealand and the U.K., Chinese investments in AI firms “never exceed 3%,” and in many countries covered by the study there has been “zero disclosed venture capital [AI] investment from entities headquartered in China.”
This absence isn’t explained by any dearth of Chinese resources ploughed into AI. The resources simply tend to stay at home. The size of China’s AI market was estimated at 71 billion renminbi ($10.8 billion at the current exchange rate) in 2020, and about 800 specialized local companies were active in that market. The investment appears to be geographically segregated, as if an invisible silicon curtain separated the U.S. and China. That means hardly any interdependence of the kind that would make it hard to split the semiconductor, telecom equipment and consumer electronics industries into a U.S. camp and a China-led one.
Apart from the investment divide, however, few data exist that would support the framing of AI as a competitive battlefield of nation states, mainly because the commonly used measures of leadership don’t seem to confer real-world advantages on a given country or bloc.
Take, for example, CSRankings, a comparison system that ranks universities by the number of papers produced in specific computer science fields. Tsinghua University in Beijing leads in AI papers published between 2015 and 2021, with Carnegie Mellon University in second place. Does this mean that Tsinghua outcompeted Carnegie Mellon in any real sense? The quality and commercial applications of the papers are not easy to quantify. The number of citations a paper receives, for example, is heavily influenced by the resources an institution puts into getting its research cited, the researchers’ first language and their international contacts. These factors contribute to the disparity on citations: U.S. AI papers receive 40% more citations than the global average, European ones 10% more and Chinese ones 20% less.
Is Israel’s Technion or ETH Zurich in Switzerland, both in CSRankings’ AI top 30, “better” or “worse” than the U.S. or Chinese institutions that sit above them? Are the seven German AI research institutions counted by CSET among the top 30 outside the U.S. and China — the most of any country — not doing a good job because they are not featured in the CSRankings global top 30?
The honest answer is, it’s impossible to know. Bright ideas can be born anywhere that good brains can access good hardware; one great paper can be worth 20 run-of-the-mill ones — and that paper could be written in German or Chinese and only noticed by a few people with an especially keen interest in its narrow subject.
Or consider the elaborate comparison system created by the think tank Center for Data Innovation to figure out “who is winning the AI race.” Apart from various metrics that concern academic research, it includes investment and spending measures — but, because of purchasing power, political, regulatory and cultural differences, the comparative efficiency of a dollar spent on R&D or invested in a start-up in Silicon Valley, Beijing and Berlin is hard to fathom. The spending and investment would have been failsafe indicators of a country’s competitive strength if they defined the viability of the resulting technologies. But in knowledge-based markets with a relatively low entry barrier, there is no such direct relationship. China is seen as breathing down the U.S.’s neck and outcompeting the European Union, though it only has 398 AI companies that have received $1 million or more in funding, compared with 890 in the EU and 2,130 in the U.S.
Money, of course, helps buy the best brains — but there are plenty of reasons why a Chinese, German, Singaporean or French AI researcher or developer may not want to live in Silicon Valley or work for a major U.S. company. CSET counted 70,000 workers with self-declared AI skills in Germany, about 0.7% of the country’s LinkedIn user base, and more than 5 million in the U.S., or 2.9% of the LinkedIn user base — but then some 53% of the U.S. population and only about 13% of the German one are on LinkedIn.
One other area the Center for Data Innovation has searched for clues on nation-state leadership is hardware production. On the one hand, this is a strong indicator of power. The U.S., as the global leader in AI chip design, can simply cut off Chinese companies from the supply of state-of-the art equipment, as Huawei found out the hard way. But even that kind of advantage can quickly melt in the fluid AI space. Take supercomputer power: In 2019, Japan accounted for less than 8% of the aggregate performance of such systems globally, but in 2020, its share was above 24%. When such quick changes are possible, any lasting competitive advantage is hard to pinpoint. And if such an advantage cannot be secured, what’s the point of competing rather than just trying to solve specific real-world problems with the help of AI?
The nation-state competition paradigm is great for selling politicians on the importance of government investment in AI and regulation favorable to the field. In the U.S., China’s threat to America’s global leadership is top of mind for the political establishment; in industrial-age powerhouses such as Germany and France, it’s all about not falling behind the U.S. and Asian upstarts. But in reality, AI, like any technology that relies on brainpower and knowledge more than on hard-to-get resources, is going to power democracies and autocracies alike, European as well as American and Asian retail chains and militaries, banks and social networks. It’s part of global progress, not something one country or group of countries can hoard and use to achieve supremacy. The very notion of overtaking others in this area is faintly ridiculous. Instead of hyperventilating over geopolitics, policymakers could better serve their citizens by focusing more on specific technological and societal goals that AI can help achieve.