(Edit Notes: China’s aggressive policy to develop its artificial intelligence industry has been a subject of much debate. In today’s piece, Manoj Kewalramani, Associate Fellow-China Studies at The Takshashila Institution breaks down the policy and brings you a detailed look at what it means. This is an edited excerpt from his report published earlier this month titled China’s Quest for AI Leadership: Prospects and Challenges for The Takshashila Institution. Also see: FactorDaily’s China Coverage.)
China unveiled a plan to develop the country into the world’s primary innovation centre for artificial intelligence in 2017. It identified AI as a strategic industry, crucial for enhancing economic development, national security and governance.
The Chinese government’s command innovation approach towards AI development is crafting a political economy that tolerates sub-optimal and even wasteful outcomes in the quest for expanding the scale of the industry. Consequently, the industry is likely to be plagued by concerns about overinvestment, overcapacity, quality of products and global competitiveness.
In addition, increasing friction over trade with other states and President Xi Jinping’s turn towards techno-nationalism along with tightening political control could further undermine China’s AI industry. Before we dive into the challenges, here’s some background.
The origins of China’s AI ambition
In October 2017, at the 19th Party Congress, President Xi Jinping outlined (pdf) the objective of developing China into a “country of innovators,” which aims for “the frontiers of science and technology.”
Developing AI technologies, which found a mention in Xi’s speech, is critical for achieving that objective. A few months earlier, in July 2017, the State Council (the Chinese equivalent of the Cabinet of Ministers), had issued a landmark plan on the development of AI technologies in the country.
The document, A Next Generation Artificial Intelligence Development Plan, built upon previous technological and industrial development plans to provide an overarching vision for the development of AI in China. It categorically states the goal of developing China into a world leader in AI innovation by 2030.
The plan describes AI as the “new engine of economic development” and “a core driving force for a new round of industrial transformation.”
It does not view AI as a particular technology or a specific industry. Instead, it identifies AI as the next frontier of technological evolution, i.e., a set of technologies that enable the shift from ‘digitization’, ‘networkization’ to ‘intelligentization.’
The State Council, however, draws a distinction between core AI and AI-related sectors. Core AI essentially covers specific, industry-agnostic technologies, while AI-related sectors cover applications that rely on core technologies and are relevant to specific industries.
Further, the plan identifies the utility of AI technologies from three broad angles – economic development, preservation of national security and enhancing social construction.
From the point of view of the economy, the government anticipates AI to “inject new kinetic energy” in China’s economic development, enabling manufacturing to move up the value chain and facilitating leaps in factor productivity.
In terms of national security, the State Council’s plan takes a macro view. The focus is on AI’s role in supporting command systems and decision-making along with enhancing defence equipment and cybersecurity by boosting civil-military integration. To be sure, deepening civil-military integration is not an AI-specific policy as civil-military integration was upgraded to national strategy status in March 2014.
The State Council’s plan also identifies three phases of development for China’s AI’s industry till 2030, with specific quantitative and qualitative targets. The plan aims that by 2020 core AI industry must exceed RMB 150 Billion, with the scale of related industries exceeding RMB 1 trillion. These figures are projected to touch RMB 1 trillion and RMB 10 trillion, respectively, by 2030.
Independent assessments and even those published by Chinese think tanks and state media indicate that these targets are nothing short of extraordinary considering the current scale of China’s AI industry. For instance, the Ministry of Industry and Information Technology estimates that China’s AI industry output in 2017 was RMB 18 billion ($2.63 billion).
The State Council plan also indicates what analysts like Elsa Kania term “an integrated, whole-of-nation approach” to AI development.26 The plan is, in effect, a guiding document or a call to action for local governments, businesses, academia, individuals and organisations working across sectors to “firmly seize the major historic opportunity for the development of AI.”
Taking a cue from the State Council, as of May 2018, at least 18 provinces, regions and municipalities have announced specific AI plans. A report published by the Qianzhan Chanye Research Institute in March this year evaluated 12 of these plans to find that the sum of their targets for scale of core AI industry by 2020 was approximately RMB 429 billion ($62.5 billion), which is nearly three times the national target set by the central government. This is indicative of the budding competition among local governments for Beijing’s approval and largesse along with private capital, which deepens local protectionism and distorts the market. Considering this, a report by the Mercator Institute for China Studies argues that while such “enthusiasm of local governments will accelerate China’s AI development considerably, it also carries the risk of creating overcapacities.”
Beyond the scale of industry targets, there are certain other features of these local plans that are noteworthy. First, a number of them call for the establishment of dedicated AI industry parks or hubs. For instance, Shanghai aims to build AI industrial clusters across the city with different focuses such as intelligent driving, intelligent robots and intelligent software and hardware, while Beijing is planning to build an RMB 13.8 billion ($2.01 billion) AI intelligence development park.
Second, most local governments are looking to leverage their distinctive advantages while devising their AI strategies. So Anhui province is seeking to build on the speech recognition expertise available in the capital city, Hefei. Hebei province wants to focus on intelligent equipment and manufacturing industries. And the Hubei provincial government is banking on leveraging the clout of the East Lake High-tech Development Zone (Optics Valley) in Wuhan to develop an AI industrial cluster of global influence.
Four factors lie at the heart of China’s vision for global AI leadership -technology and talent, research and rules, data management, and a commercial ecosystem.
Technology and Talent
Despite being the world’s biggest semiconductor market, data shows that China manufactures only around 16% of the semiconductors it uses domestically. Estimates suggest that as of 2015, China accounted for only 4% of the global semiconductor production market share. The country reportedly imported $227 billion worth of integrated circuits in 2016, an amount greater than its combined imports of crude oil, iron ore and primary plastics. This is the scenario despite massive capacity upgrades in the past decade. Therefore, starting with the Made in China 2025 policy, the Chinese government has set out clear self-sufficiency targets – 40% by 2020 and 70% by 2025 – with regard to integrated circuits. The roadmap to achieve this is through increased government spending aided by the China Integrated Circuit Industry Investment Fund, establishing focused industry clusters, allowing local private equity firms to allocate public funds, supporting consolidation in the domestic market and encouraging overseas partnerships, mergers and acquisitions. Given China’s dependence on imports, M&As and global partnerships, the success of this strategy is heavily contingent on geopolitical factors. Growing friction with the US and Europe over trade and market access, intellectual property rights and allegations of forced technology transfers pose a clear and present threat. In response, the Chinese government appears to be tilting towards a nationalistic narrative of self-reliance. In May 2018, Xi Jinping outlined a new approach to develop China into a science and technology leader, calling for core technologies to be “self-developed and controllable” with the initiatives of innovation and development being “securely kept in our own hands.” While such an approach is likely to expand state support, it could potentially adversely impact the pace and quality of core technology industry development in China, along with deepening frictions with international actors.
In addition, adopting a techno-nationalist approach could hinder academic, business and research partnerships. It could also immediately impact the attractiveness of China as a destination for international AI talent. Estimates of the global AI talent pool vary widely. However, the common consensus is that there is an acute shortage of talented people. According to the People’s Daily, China faces a severe AI talent shortage with over 5 million professionals urgently needed. The Tencent Research Institute’s 2017 Global AI Talent White Paper (pdf) estimates that there are around 300,000 AI researchers and practitioners around the world, with 200,000 of them already employed in various industries. The State Council’s plan calls for a training and gathering approach. There has been a concerted effort to offer extremely competitive salaries, easier visa processes, financial and non-financial incentives and specific subsidies to enhance living standards. Data on overall employment suggest that such policies have resulted in some gains in attracting talent, with China jumping 11 places to the 43rd position in The Business School for the World (INSEAD) Global Talent Competitiveness Index from 2017 to 2018. Moreover, improving career prospects, particularly in the technology sector, have meant that increasingly Chinese international students are returning home after their education overseas. Despite improvements, China continues to lag behind in the US and major European states in terms of talent competitiveness. The socio-political and cultural peculiarities of China, with growing censorship and a walled internet, can also potentially hamper its attraction. Western states would likely face a similar problem if populist, anti-immigrant and xenophobic movements were to continue to grow in scope and intensity.
To train more talent, an increasing number of Chinese universities have been introducing AI courses. In April 2018, the Ministry of Education released its first AI Innovation Action Plan for Colleges and Universities which aims to establish China’s universities as hotbeds for AI talent by 2030. The first step towards this is to bring out 50 world class AI textbooks; set up 50 national-level high quality online AI courses and establish 50 AI research centres by 2020. Along with this, the Ministry of Education also wants to train 5,000 students and 500 teachers in AI within five years. In June 2018, SenseTime and the East China Normal University published the first Chinese high-school AI textbook.
Research and Rules
Assessment of research capacities involves quantitative and qualitative measures. From a purely quantitative viewpoint, Elsevier data from 2011 to 2015 show that China has long surpassed the US, its closest competitor, in terms of the number of papers published on AI. In its 2018 China Artificial Intelligence Development Report, Tsinghua University estimates that China’s share of AI research papers published globally has grown from 4.26% in 1997 to 27.68%. The report also claims that China holds the most AI patents, inching ahead of the US and Japan. However, given that China is in a catch-up phase vis-a-vis international competitors in terms of its technological base, the cost it incurs to generate a patent is significantly higher than countries like Japan and Germany. Moreover, data regarding numbers of patents simply associated with the term AI are an insufficient gauge of quality.
Data on research citations and their weighted impact offer a far more comprehensive qualitative perspective.
Data on research citations and their weighted impact offer a far more comprehensive qualitative perspective. In this regard, Elsevier data from 2011 to 2015 placed China at the 34th position in terms of field-weighted citation impact. The US, in contrast, was ranked 3rd. However, that trend appears to be changing. SCImago Journal & Country Rank, which is based on citation data derived from the Scopus Database, evaluates China’s 2017 H-Index value in the AI subject area at 213, i.e., the third highest after the US’s 479 and the UK’s 228. In addition, there has been a sharp spike in Chinese authors’ clout at international AI conferences, indicating qualitative improvements. For instance, an assessment of the papers presented at the annual conference of the Association for the Advancement of Artificial Intelligence revealed that 23% of the authors who presented papers in 2017 were Chinese, compared to just 10% in 2012. The share of US authors, meanwhile, fell from 41% to 34% during the same period.
As impressive as the above-mentioned trend appears, certain structural impediments are likely to persist. Over the years, a flawed incentives structure, such as high stress on meeting numerical targets and making breakthrough achievements, has led to the emergence of a perverse black economy of research in China. The Chinese Party-state’s attempts to intensify political control are only likely to lead to the persistence of this trend. The challenge for Beijing, therefore, is to foster a regulatory environment that enables collaborative, authoritative and ethically responsible research while meeting political objectives of control and stability. How the Chinese government strikes this balance will also be one of the key determinants of its ability to influence the debate around global norms, ethics and standards governing AI.
China often is seen as having an edge when it comes to the availability of data, which fuels AI. Over 57% of the Chinese population, which is roughly over 800 million people, now have access to the Internet. Lack of effective privacy protections ensures that there are fewer constraints with regard to accessing data. Moreover, the Chinese public appears to have a far more relaxed view with regard to privacy protections, particularly in the context of the state having greater access to private data. This is evident in the popularity of the rapidly developing Social Credit System.
It is important to keep in mind that this does not imply that privacy is not a matter of concern for Chinese consumers.
However, it is important to keep in mind that this does not imply that privacy is not a matter of concern for Chinese consumers. Data from the Internet Society of China also shows that 54% of Chinese Internet users hold the view that privacy breaches are a severe problem, particularly given the rising cases of fraud and theft. In addition, social media outrage against Baidu’s Robin Li for arguing that Chinese people tend to disregard privacy for convenience and official criticism of Ant Financial for violating personal information security standards also indicate shifting attitudes. One can locate the Chinese government’s increasing efforts at framing data protection and usage laws within this changing environment. Therefore, new Chinese laws are increasingly aiming to protect against the misuse of data by companies, without placing such restrictions on the state. This leads us to another factor i.e., as a consequence of the above, increasingly Chinese private companies are building their business models around the needs of the state. As a result, the state wields strong influence over tech companies, which in turn enables resource sharing and coordinated action.
The mere availability of large quantities of data, however, does not imply a winning advantage in terms of AI development. The qualitative component of training data for AI systems is crucial, particularly if these systems are to serve as standard bearers internationally. Among other things, quality is primarily a factor of diversity, technology and manpower, and China’s challenges in these contexts have been outlined above. In addition, laws implemented by other states that restrict cross-border transfer and mandate data localisation, such as the Chinese Cybersecurity Law, which went into effect in June 2017, could limit data flow to China, impeding its AI development.
The Chinese Party-state is intimately involved in the development of the country’s AI sector. To foster growth and meet the targets that it has laid out, the Party-state broadly performs four distinct functions. It is a regulator, investor, partner, and consumer.
In November 2017, the Chinese government announced a major partnership with private enterprises to boost AI development. The Ministry of Science and Technology identified tech giants Baidu, Alibaba Group, Tencent Holdings and iFlyTek as national champions that would spur advancements in AI in the country. Each enterprise was tasked with a specific focus area, with commitments of government support. This underscores the deep linkages between private enterprises and the state in terms of China’s approach to technology development, and while such an approach might encourage focused development in certain areas, it also results in market distortion. Assured state support for certain firms undercuts competition and can encourage moral hazards, resulting in a greater prevalence of and tolerance for suboptimal outcomes. An excellent example of this is the recent iFlytek controversy, with the company being accused of hiring humans to fake its simultaneous interpretation tools, which are said to be powered by AI. Along with engendering problems as this, such a policy raises entry barriers for new innovators, undermining new ideas and enterprises. Following this announcement, the December 2017 three-year AI action plan issued by the Ministry of Industry and Information Technology emphasised that technological developments must be driven by market demand.
The plan identified industry, healthcare, transportation, agriculture, finance, logistics, education, culture, and tourism as key sectors for the development of smart products. Such signalling by central authorities acts as an important guide for local governments, state-owned and private enterprises to direct their attention, with state support likely to be concentrated for activities in these sectors. This, in turn, also tends to attract the attention of central and local guidance funds, venture capitalists and foreign investors.
Already, this has had a significant impact in terms of the number of players and size of the AI industry in China. But, while the numbers have swelled, there are serious questions about the quality of Chinese AI startups. SenseTime’s Yang Fan and investors like Everbright’s Ai Yu have argued that the fundamental challenge for Chinese AI startups is the commercialisation of new technologies, which need to be focused on real demand. This, they argue, is likely to come into greater focus as the initial flow of funding begins to slow down. Such assessments indicate that already there are concerns about overcapacity, quality and inefficiency and the emergence of a bubble. Addressing this is likely to require greater and continued government intervention in the AI sector, which will further have implications for quality and capacity for innovation.
Subscribe to FactorDaily
Our daily brief keeps thousands of readers ahead of the curve. More signals, less noise.
Disclosure: FactorDaily is owned by SourceCode Media, which counts Accel Partners, Blume Ventures, Vijay Shekhar Sharma, Jay Vijayan and Girish Mathrubootham among its investors. Accel Partners and Blume Ventures are venture capital firms with investments in several companies. Vijay Shekhar Sharma is the founder of Paytm. Jay Vijayan and Girish Mathrubootham are entrepreneurs and angel investors. None of FactorDaily’s investors has any influence on its reporting about India’s technology and startup ecosystem.