China is quickly closing the once formidable lead the U.S. maintained on AI research. Chinese researchers now publish more papers on AI and secure more patents than U.S. researchers do. The country seems poised to become a leader in AI-empowered businesses, such as speech and image recognition applications. But while China has caught up with impressive speed, the conditions that have allowed it to do so — the open science nature of AI and the nature of the Chinese market, for instance — will likely also prevent it from taking a meaningful lead and leaving the U.S. in the dust.
Twenty years ago, there was a huge gulf between China and the United States on AI research. While the U.S. was witnessing sustained growth in research efforts by both public institutions and private sectors, China was still conducting low-value-added activities in global manufacturing. But in the intervening years, China has surged to rapidly catch up. From a research perspective, China has become a world leader in AI publications and patents. This trend suggests that China is also poised to become a leader in AI-empowered businesses, such as speech and image recognition applications.
China’s feat is dramatic. According to earlier research — the China AI Development Report 2018 project, which one of us (Li) helped spearhead — as well as an ongoing study of the economic and social impacts of AI technologies, the country’s progress is stunning. China’s global share of research papers in the field of AI has vaulted from 4.26% (1,086) in 1997 to 27.68% in 2017 (37,343), surpassing any other country in the world, including the U.S. — a position it continues to hold. China also consistently files more AI patents than any other country. As of March 2019, the number of Chinese AI firms has reached 1,189, second only to the U.S., which has more than 2,000 active AI firms. These firms focus more on speech (e.g., speech recognition, speech synthesis) and vision (e.g., image recognition, video recognition) than their overseas counterparts.
Impressive as this may be, however, there’s no guarantee it will translate into a robust advantage in AI innovation and global leadership moving ahead. Paradoxically, the conditions that helped China catch up might also pose a challenge to its future development in AI as the country reaches the innovation frontier. To explain why — and build on earlier research — we conducted field interviews with 15 AI related organizations of different types (including firms, universities, research institutes, and government agencies) and used the idea of catch-up cycles, a theoretical framework developed to explain countries’ successive changes in industrial leadership.
How China Caught Up
How was China able to leapfrog countries that had been working on this technology for much longer to build a world-leading AI research infrastructure in just 20 years?
Here, the concept of “catch-up cycles” can help us understand. In essence, the catch-up cycle framework suggests that, in certain circumstances, changes in technology, market conditions, and policy environments can put latecomers and forerunners more or less on an equal footing. According to the framework, these changes can open windows of opportunity for latecomers by quickly reducing the advantage of incumbents — for example, the emergence of Android smartphones was a technological change that flattened market leader Nokia’s advantage and allowed fast-movers like Samsung and Huawei to displace it. The framework also helps us understand when — and why — newcomers will displace incumbents.
In the story of how China managed to catch up, this framework highlights a few important factors: how the nature of AI research means that leaders’ technological advantages aren’t particularly robust; how China’s huge market is particularly conducive to improving AI; and how the country’s friendly regulatory environment is especially encouraging to AI investment and adoption.
In AI, research doesn’t provide a durable advantage.
AI is different from other technologies in a few significant ways. While research propels the field forward, that research is often shared openly, the patents research yields matter less, and improvements often come from the virtuous cycle of users generating data and firms refining their product based on what they learn from that data.
Unlike computer hardware or drug development, AI is open science. In terms of knowledge and technologies, many of the essential algorithms in the field of AI have become public knowledge, accessible from published papers and conference proceedings. “Currently, everyone is proud of publishing AI research results,” one manager of NISE Intelligent Technology, a startup specializing in AI algorithms and AI chips, told us. “Generally speaking, if you publish the paper, in this profession it is not too difficult for others to figure out the code and implement it.”
The open science nature of AI is important for latecomers’ catching-up with respect to forerunners, because it allows the former to close the knowledge gap with the latter in a short period of time.
The second way that AI differs from traditional sectors is where innovation creates profit. Put simply, data and talent trump patents in AI research. In traditional sectors such as pharmaceutical or mobile communications, patents play a critical role in securing firms’ positions and protecting profit streams. The open science nature of AI means that firms’ competitive advantages often stem from the extent to which they can assemble a large database — and develop domain-specific knowledge or applications around the database — faster than anyone else.
This means that there are two critical assets in the AI era: data and computer science and engineering talent. China happens to be quite abundant with both. Its large population gives it advantages in generating and utilizing big data, and its decades-long effort in promoting technology and engineering gives it a rich supply of high-quality computer scientists and engineers.
Finally, the “weak AI” we are developing today — AI that solves narrowly defined problems — requires domain-specific knowledge and user-generated data to improve. For example, AI often needs to be customized to specific business scenarios. You first make a product (e.g., voice recognition). Then, you attract many users and these users generate data. Finally, you use machine learning to improve products with data. Improvements occur through this virtuous cycle.
China has a vibrant market that is receptive to these new AI-based products, and Chinese firms are relatively fast in bringing AI products and services to the market. Chinese consumers are also fast in adopting such products and services. As such, the environment supports rapid refinement of AI technologies and AI-powered products.
China’s market is conducive to the adoption and improvement of AI.
Given how important large data sets are to innovation in AI, it’s east to see how China’s gigantic market size helps explain how its rapid catching-up in AI. The volume of this market offers Chinese firms a unique opportunity to assemble big databases. Consider Didi, China’s counterpart of Uber and the largest ride-sharing company in the world today. According to its CEO Liu Qing, each day, Didi processes more than 70TB of data, with 9 billion routes being planned a day and 1,000 car requests a second.
China’s huge market not only provides advantages in big data, but also offers firms strong economic incentives to tackle technological challenges. For example, although chipsets have long been a weak part of China’s information and communication technology (ICT) industry, Chinese firms recently are making big strides in narrowing the gap in AI chipsets. A senior manager from ZTE, one of the world’s largest ICT companies, told us, “China’s development of AI chips is relatively fast. … Once there is a market, firms are motivated to [develop AI chips].” China’s huge market brings large economies of scale to the ICT industry, meaning investments that push the technology pay off quickly.
In addition to its sheer size, the Chinese market also shows large variety and is fast changing. This creates a dynamic range of opportunities for startups and established firms alike to explore different AI applications in different market segments at a fast pace. As earlier research suggests, these kinds of market dynamics often helps latecomers catch up, leading to the emergence of new products and new ventures.
China has strong AI promoting policies and weak privacy regulations.
The final pillar relates to the policy environment. China has in recent years passed a number of policies to promote the development of AI. Such policies include, but are not limited to, “Made in China 2025,” “Action Outline for Promoting the Development of Big Data,” “Next Generation Artificial Intelligence Development Plan,” and so on. These policies send a clear signal to different AI stakeholders, including entrepreneurs, investors, and even researchers, that AI is a field that is being backed by the government and is worth investing.
China’s lack of clear policies and regulations in areas such as privacy can explain how it caught up so rapidly in certain AI application fields. For example, the ubiquity of surveillance cameras in China creates a big market for AI firms specializing in visual and facial recognition. This market would not have grown so fast in many other countries with tighter regulations on privacy. As a project leader, also from NISE Intelligent Technology, told us, the loose privacy regulations in China are a critical advantage for some domestic AI companies.
Challenges and Future Prospects
By many indicators, China is now on the global frontier of AI in terms of technological development and market applications. The unique technological, market, policy environments that Chinese firms face in the emerging AI sector have given them a window of opportunity to catch up with global leaders rapidly.
But, paradoxically, while China may have caught up in record time, the conditions that have allowed it to do so may impede its further development in AI.
For example, given the open science nature of AI and the advantages of being quick followers, Chinese firms often lack strong incentives to invest in developing core AI technologies. Unlike in Western developed economies where companies are the primary holders of AI patents, in China, the majority of AI patents are filed by universities and research institutes, most of which are government owned or sponsored. However, university-industry linkages in China are relatively weak, and technology transfer remains rather limited. Overall, although aggregate AI research outputs (e.g., scientific publications, patents) are rising rapidly in China, truly original ideas and breakthrough technologies are lacking.
Further, the uncertain business environment in China, coupled with the huge market for AI products and Chinese consumers’ enthusiasm to adopt them, leads companies and investors to favor applied AI research that can bring quick money instead of more basic research that promises to have long-lasting impacts. At a more fundamental level, the research culture in China needs a great deal of improvement, as many researchers have highlighted.
On the policy front, the relaxed regulatory environment has proved to be a double-edged sword. While some firms are bold enough to take advantage of such environment by aggressively pushing different AI applications to markets, others feel frustrated as they don’t know what is allowed due to such policy uncertainty. The chairman of Suzhou Blue Amber Medi-Tech, a medical device company, lamented that this uncertainty has led his company to decide to not touch any data that might fall into certain gray areas (e.g., use of personal health data for other commercial purposes). “Our current thinking is that if we don’t need to touch the data, we will not touch it. … But, if we do not touch the data, a significant part of the value [of the data] is not realized. So, from our company’s point of view, we do hope that the government will make the regulations clear sooner.”
Today, the global business and technology environments face a set of political uncertainties. These include the U.S.-China trade war and heightened conflicts over intellectual property rights, the deglobalization movement, increasing protectionism, and so forth. These challenges will have an immediate impact on China’s further catching-up in AI, but their long-term influences on the rate and direction of China’s AI innovation remain to be seen. In the meantime, regardless of such uncertainties, the coopetition between the U.S. and China in the AI space will continue for many years to come.