ChatGPT shakes up the AI research landscape – but who is ahead?

China and the United States are engaged in a heated race to lead in artificial intelligence (AI) research. China has been pouring billions into AI research and development in its goal to become a global leader in AI by 2030. America is also stepping up funding while seeking to limit China’s gains by restricting exports of sensitive technologies, in particular semiconductor chips.

China seemed to be pulling ahead in AI research by some measures. However, the emergence of ChatGPT and other generative AI technologies from companies in the West has significantly altered the AI research landscape, challenging China’s perceived lead as it struggles to catch up.

In July 2017, China’s State Council released a strategic plan for AI to ensure that the country catches up with the US in AI technologies and applications by 2020 and becomes the world’s primary centre for AI innovation by 2030.

This sparked a rapid acceleration in research. From about 25,000 published papers on AI in reputed journals, its research output rocketed to some 135,000 papers in 2021, according to a study released in January 2023 by Nikkei Asia, a Japanese media organisation.

Some reports, such as Nikkei’s, suggest that China has overtaken the US, not just in the quantity of AI papers but the quality as well, based on how many papers were in the top 10% of citations by other papers. This caused considerable waves among Washington policy-makers, adding to the pressure to ‘decouple’ from Chinese research.

According to Nikkei, in 2012 the US led with 629 of the most cited papers, with China in second place at 425. China “made dramatic progress” overtaking the US in 2019. In 2021, China accounted for 7,401 of the most cited papers, topping the American tally by around 70%, the Nikkei study revealed.

“Who is number one or number two in the world is less relevant, it is the trajectory of the Chinese effort that is impressive – the growth over the last five to seven years has been remarkable,” said Denis Simon, a China science and technology policy expert at the University of North Carolina’s Kenan-Flagler Business School at Chapel Hill in the US, and former executive vice-chancellor of Duke Kunshan University in China.

“It’s not that the US has stopped or the US is lagging in AI because we’re not putting in the effort. It’s that the Chinese are so committed,” Simon told University World News.

Steep rise in publications and citations

Caroline Wagner, professor at Ohio State University’s John Glenn College of Public Affairs in the US, specialising in China science and technology, has been conducting her own, as yet unpublished, research into China’s AI research prowess compared to the US. She acknowledges China’s phenomenal growth.

“The US was probably producing at close to maximum output per person and the US is highly developed, whereas China has had room to grow so they’re producing a lot more work,” said Wagner. “China makes policy statements and was offering money supporting AI research, so we see a change in output, which is what you would expect to some extent as people respond to these things.”

Wagner’s research team combed publications databases, looking at papers just with Chinese names, those that included names from other countries outside the US, and those with Chinese names within the US – though she cautioned that some of these may be Chinese-American co-authors.

Looking at the five years preceding 2022, China-only AI research has indeed caught up with US-only AI research, she discovered.

“We were surprised to find that in 2019, Chinese authors published a greater percentage of the most influential papers,” Wagner told University World News. China claimed 8,422 articles in the top category of the 1% most cited, while the US had 7,959 papers and the European Union had 6,074 papers.

Wagner and her co-researchers found that in 2022 Chinese researchers published three times as many papers on AI as US researchers: in the top 1% most-cited AI research Chinese papers outnumbered US papers by a two-to-one ratio.

According to her findings, Chinese research does not copy the West, as might have been the case in earlier years, but has emerged as novel and creative in its own right.

Nonetheless, she did not necessarily agree with the Nikkei study. “If you are just looking at numbers of AI articles then it would be the case that China is ahead, but it is certainly not in terms of the overall quality and impact of the work in those cases. US articles are still getting more citations and more attention from the [scientific] community,” she noted.

However, Wagner stressed that collaborations between US and Chinese scientists “are even more highly cited”.

China-US collaborations have dropped since the COVID-19 pandemic and the stepping up of US technology restrictions directed specifically at China. However, China collaboration with Europe on AI “has remained pretty strong, even while some of the China-US work has dropped”, Wagner noted.

Other commentators point out that academic publications and citations are just one metric. US defence-related research, classified US government research and industry research are not always made publicly available. Notably, ChatGPT – which took China completely by surprise – came from a private company, OpenAI.

China has some significant AI research and development advantages, according to Wagner. For example, it does not have the same privacy concerns about the use of data as countries in the West.

“They have massive data, probably more data than the United States or Europe has, and they have fewer regulations on that data. They have a data advantage while the US and the West may have a chips advantage,” she said.

Talent and talent development is another important indicator for assessing AI development, Wagner added, noting the ability of the US to attract some of the world’s best talent, including people from China.

Strategic importance of AI

Which country is on top is not just a matter of national pride.

“AI and emerging technologies, in general, are economically and strategically important to both the United States and the Chinese government, as well as others,” said Hanna Dohmen, senior researcher at the Center for Security and Emerging Technology at Georgetown University in the US.

“So assessing and understanding China’s AI capabilities and strengths, comparative to our own, is very important. AI capabilities will have consequences for many different industries in the future. And all of this factors into the broader US-China tech competition.”

US National Security Advisor Jake Sullivan, in a speech at a Global Emerging Technologies summit in September 2022, said: “Fundamentally, we believe that a select few technologies are set to play an outsized importance over the coming decade.”

Sullivan named AI and information systems as being among them, and described leadership in these, among other key technologies, as “a national security imperative”.

Their competition has driven AI up the national agenda in both China and the US. For instance, China has staked its digital economy ambitions on AI. “It will need advanced AI capability in order to support the entire onset of big digital driven infrastructures. The juice that makes them work is going to be AI,” said Simon.

This week China’s Ministry of Science and Technology and the National Natural Science Foundation jointly launched the Artificial Intelligence for Science programme to accelerate the use of AI in all science, technology, engineering and mathematics research and in finding solutions to major challenges such as climate change.

US intent on widening the research gap

“Even outside of generative AI, we’ve seen a shift in policies in the United States approach to tech competition and that will have consequences for AI competition,” said Dohmen, pointing out that the US is now intent on widening the research gap with China as much as possible.

US export controls that sought to restrict key US technology exports to China as part of the strategy to maintain relative advantages over competitors, and ensure the US stayed “a couple of generations ahead”, according to Sullivan, now have an even more ambitious goal.

A lead of “a couple of generations” of semiconductor technology is no longer enough. “That is not the strategic environment we are in today. Given the foundational nature of certain technologies, such as advanced logic and memory chips, we must maintain as large a lead as possible,” Sullivan said.

Hence, more aggressive restrictions have followed. “We have adapted our technology protection tools to new geopolitical realities,” Sullivan added.

The US ban on semiconductor chip exports to China and Chinese companies, announced in October 2022, prevents leading US computer semiconductor companies like Nvidia and AMD from selling high-end chips that are used for AI, and also supercomputing in China.

The US is now leaning on allies like Japan and the Netherlands to follow with their own restrictions on chip production tools and equipment to China.

This comes in addition to restrictions that apply to the US Entity List. Inclusion on the Entity List requires US companies and universities to seek permissions for certain types of research collaboration with China, as it applies to controls on ‘US-origin content’.

US export controls

Even those who accept that China may have overtaken the US on some research metrics question whether it can maintain its slim lead. US export controls have dramatically changed the timelines for China to achieve its AI goals.

Controls also substantially increase the cost of AI research and innovation in China as companies seek alternatives, including less powerful chips that are not subject to US export control restrictions.

Experts predict that this could hamper Chinese research and commercialisation of ChatGPT-like tools and other generative AI applications, which require huge computing power to speedily trawl through massive data sets.

Simon sees US export controls as curbing China’s AI progress, with a large number of Chinese universities and companies added to the US Entity List in recent years.

“The ability to rapidly and comprehensively process data – this is where the Chinese are now facing some constraints,” Simon said.

“China may build a one-of-a-kind, high-end supercomputer or quantum computer but they don’t have the ability to produce these in any kind of large quantity. So, AI can go forward to a certain degree in China, but when it gets to those very sophisticated levels, it’s going to hit a brick wall,” said Simon.

While China is stepping up its self-sufficiency drive in these critical areas, Simon noted that it “won't have the ability to manufacture the kinds of chips that are going to be needed to support some of this processing to support AI”.

Dohmen said: “The effect of the export controls on the hardware, which is necessary for the development of generative AI, could have short to medium term consequences in China and in their AI development.”

The future of research collaboration

However, Dohmen and others cautioned against seeing research and AI and AI development as a ‘black and white’ China-versus-the-US race.

“There is still a lot of US-China cooperation in joint publications and at conferences and in research labs,” Dohmen said.

“This may not be as robust as in the past,” she acknowledged, and US and other researchers cooperating with Chinese researchers will “need to take a more risk-based approach” to decide on whether or how to continue research collaboration with Chinese researchers.

She warned: “If you restrict the level of cooperation too much, it can potentially lead Chinese researchers to increasingly do joint research with third countries.”

This is already happening. Simon pointed to the acceleration of new joint research projects between China and universities in the United Arab Emirates and Saudi Arabia.

Wagner believes that research collaboration, including that between the US and China, will continue to underpin AI advances.

“What does it mean for a country to be ahead in a globalised knowledge system?” she asked rhetorically. “The political language [of competition] and concerns of who is ahead completely ignores the fact that the most advanced work is being done cooperatively between the two countries.

“I don’t see that retreating – at least not in the number of joint articles being published. These are pretty much staying the same or rising.”

Wagner acknowledged that given the climate in the US, there may be more caution around collaborations with China. But that did not mean they would stop altogether. “There are a lot of people who will keep going, just as they have always done.”

However, Simon and others doubt that research collaboration between China and the US can return to pre-COVID levels – particularly in AI, which is bound up with national security concerns. The geopolitical climate has altered too much for that.