Much has been speculated on the Arms Race in AI between the United States and China and more specifically on which nation will be the first to claim AI supremacy. Many reports compare these countries on venture funding and policy choices while largely ignoring the fact that talent is the scarcest resource in the AI industry. Interestingly, the AI Industry is fueled by Chinese talent both in the East and in the West, with undergrads of Chinese Universities making up 27% of the AI Research workforce of US Academia and Industry. Successfully attracting or keeping this Chinese talent will play a large part in determining the outcome of the US-China race towards becoming the leading AI nation.
Recent “western” breakthroughs in Deep Learning are often fueled by Chinese talent.
Deep Learning has fueled some incredible advancements over recent years, allowing machines to bridge the gap with humans in extremely challenging tasks that have unlocked entirely new industries. As such, Microsoft Research was the first to outperform humans in image classification  and image captioning  with Machine Learning. Google Research, on the other hand, demonstrated the first model to perform on par with humans in language translation . One common feature these recent developments share is that they happened under the hood of western companies. This property is often highlighted and employed to illustrate the supposed AI supremacy of western research institutes over Chinese ones. However, this is a far too simplistic take. When we dig deeper, we find that there is an even more profound characteristic that these breakthroughs share. Kaiming He, Hao Fang and Yonghui Wu, the respective first authors of these papers and breakthroughs, all received their undergraduate degree at a Chinese university.
Policy changes and capital infusions are important, but the real race for AI supremacy is a race for talent.
As illustrated above, China clearly did not lack AI talent as a raw resource, and thus the question remains how it came to be that Western companies were the ones to lead the charge in these discoveries. A recurring argument is that China did not have the resources on a policy and money level. However, much has changed in the short time between these scientific breakthroughs (2014-2016) and now. In 2017, the “New Generation Artificial Intelligence Development Plan” (新一代人工智能发展规划) was released, outlining China’s strategy to become the leading AI power by 2030. This policy plan across R&D, product applications and industrial cultivation officially marked the AI sector as a national priority and was included in President Xi Jinping’s grand vision for China. The same year, Chinese venture-capital investors surpassed their United States counterparts by accounting for 48% of worldwide AI venture funding as seen in the figure below. As a direct result of this rise in funding, we have witnessed the birth and rise of AI giants in China such as SenseTime, Megvii and many more.
Academia, however, preceded both industry and policy in the “China catching up to AI”- trend. The Institute of Interdisciplinary Information Sciences (IIIS) and its famous Yao Class were founded in 2011 and bred startups such as Megvii (US$4.5 billion valuation) and Pony.ai (US$5.3 billion valuation) while building a solid basis of world class interdisciplinary researchers. At the same time, we saw the rise of both the Tsinghua Statistical Artificial Intelligence & Learning (TSAIL) lab and Natural Language Processing Lab at Tsinghua University (THUNLP) at the Tsinghua Computer Science faculty to the absolute world top of AI research labs. Although often overlooked in analyses like these, the role of academia and academic institutions in this industry is hard to be overstated. The reason is as simple as “incontournable”: AI is just really hard and thus needs to be taught. None of the amazing strides that companies such as Microsoft, Google, SenseTime and many more have made recently could have been possible without academia. Simply put, the people that fuel the discoveries of these companies had to be schooled before. This extensive need for schooling is a serious bottleneck to the incredible growth of the AI industry. Incredible growth is not an exaggeration: in 2017 Tencent released a report stating there were just 300,000 AI researchers and practitioners worldwide, but the market demand was for millions of roles.
AI Talent is becoming increasingly Chinese and that includes the absolute top talent.
Having established talent as the bottleneck of the AI industry and looking at the impressive growth of quality researchers in Chinese universities between 2007 and 2011 (Figure 1), we could have therefore predicted the catch-up of China in the US-China AI Arms race from 2017 onwards. As the graph clearly illustrates, the future AI workforce will be proportionally even more Chinese than it is today and that presents both a huge challenge for the US and a tremendous opportunity for China. As such, our earlier “breakthrough research”- examples are just a symptom of a world-scale phenomenon. Now already, 27% of AI researchers working for US institutions (Academia and Industry) received their Bachelors degree at a Chinese University.
It is important to establish that Figure 1 does not only illustrate a quantitative leap in AI research but also a qualitative one. China is still labeled by many as a country that copies and produces lower quality versions en masse. Given that our Figure 1 only includes faculty publishing at the very top of AI conferences, this clearly does not hold for AI research. China is becoming a larger and larger contributor of the absolute best AI research and as such holds a larger part of the top quality researchers. With the growth of quality research not showing any sign of slacking in recent years, it is not difficult to foresee that the Chinese graduate programs will grow to be a very attractive alternative to their US counterparts in the future.
US universities are still world-class in nurturing raw talent to great researchers, yet China is catching up fast which leaves the US with sorrows.
For now, the United States is still the absolute world top at nurturing AI talent, from China and everywhere else - leaving only the very top of China institutions to compete on faculty-level. Furthermore, the percentage of Chinese undergraduate students that leave for the United States has not shown any sign of going down either. However, judging by the trends we laid out earlier, a stagnation of the emigration of Chinese undergraduates for the US is not good enough to keep fueling its AI advancements. To meet the demands of the AI industry growth, the US will need to find new ways to attract more raw Chinese AI talent. This might prove especially challenging in a time where US-China relations are at best shaky and where Chinese universities keep bringing out top-tier research as both these trends could severely harm the attractiveness of US institutions for Chinese undergraduates.
To conclude, many of the biggest “western” breakthroughs spurred by Deep Learning were often a collaboration of raw Chinese talent and successful Western nurture. While China is quickly catching up on AI-friendly policy and AI-targeted capital, the US grows more and more dependent on Chinese talent. Chinese AI research is not only growing quantitatively but also qualitatively making staying in China more and more attractive to Chinese undergraduates. The US will have to either tap into another source of human capital or be prepared for fiercer competition in attracting Chinese AI undergraduates in the future. Whoever will become the leading AI power remains to be seen, but one thing is for sure: Chinese undergraduates and academic institutions have a large role to play in the outcome.
All opinions expressed in this essay represent my personal views only.