China’s Bid to Lead the World in AI 

Recent Features

Interviews | Security | East Asia

China’s Bid to Lead the World in AI 

Insights from Huey-Meei Chang.

China’s Bid to Lead the World in AI 
Credit: Depositphotos

The Diplomat author Mercy Kuo regularly engages subject-matter experts, policy practitioners, and strategic thinkers across the globe for their diverse insights into U.S. Asia policy.  This conversation with Huey-Meei Chang – Senior China Science and Technology Specialist at Georgetown University’s Center for Security and Emerging Technology (CSET) and co-editor with William C. Hannas of “Chinese Power and Artificial Intelligence: Perspectives and Challenges” (Routledge, 2023) – is the 422nd in “The Trans-Pacific View Insight Series.” 

Evaluate the viability of China’s plan to lead the world in artificial intelligence by 2030. 

It is plausible, even probable. The usual predictors are talent, computing strength, and data availability. Let’s look at them in reverse order.

China is said to have an edge on data, needed to feed the LLMs [large language models] that many believe will dominate AI through the end of the decade. Chinese data quality is a problem, but the data can be cleaned. In any event, the entire world is running out of training data, so call this even.

Computing strength turns on the availability of high-end chips, where China lags. Efforts to restrict China’s access to graphic processing units (GPUs) have succeeded temporarily, but China has workarounds, such as substituting quantity, third-party purchases, writing efficient algorithms, outsourcing the training, investing in R&D, and, hypothetically, stealing. If the fate of a reborn Huawei is illustrative, we may erase our advantage. But for now, the U.S. has an edge.

For AI talent, China is the winner as measured by published papers, conference presentations, number of graduates, university ranking, and number of patents. More than a third of U.S. AI researchers hail from China, which has its own slippery dynamic. 

In sum, the race to “lead” in AI is a draw with the momentum in China’s favor.

Examine the core elements of China’s 2017 “New Generation AI Development Plan.” 

The PRC has a history of publicizing its plans and the 2017 document does not disappoint. While aspirational, these ambitions guide funding and should be taken seriously.

All the strengths and weaknesses of industrial policy are on display, including support for AI research and its needs in education and talent. Six broad areas get attention: fostering innovation, infusing AI into the economy, integration into society, military-civil fusion, building a “safe” AI infrastructure, and planning for next-generation AI, i.e., general AI and associated megaprojects.

The terms “brain” and “brain-inspired” appear in multiple places, reflecting China’s view of its importance in AI research. Also, references to foreign dependencies building links, “going global” (走出去), attracting foreign support via “Thousand Talents” and other venues occupy much of the narrative. 

Three other points are worrisome: state-backing for world AI leadership, a commitment mostly absent in the U.S.; the call to “merge” (混合) human and artificial intelligence, which carries its own special risks; and China’s goal of achieving a “first mover advantage” (先发优势) which in an AI context might be irreversible.

Identify China’s three research areas leading to advanced general intelligence. 

We omit the convoluted discussion about what these terms mean separately and together. Roughly, advanced “general” AI is human-level AI (artificial general intelligence) or AI that exceeds human levels at most tasks (artificial superintelligence).

There are multiple (theoretical) paths to AGI/ASI, and we won’t pretend to know everything happening inside China. However, three areas show potential based on statements by credible practitioners, the quality of their work, and visible investment and infrastructure.

The first is China’s work in symbolic and sub-symbolic AI to replicate cognition. Literature surveys of Chinese research show most of the “hard” problems of brain emulation – planning, continual learning, creativity, intuition, sensemaking – are being studied, efforts that are brain-inspired in a derivative sense.

The second approach is making mathematical models of the physical processes that produce these elements of “mind” in lay terms, brain modeling. Pu Muming’s Mesoscopic Connectome Project and the HUST [Huazhong University of Science and Technology] – Suzhou Institute for Brainmatics are examples.

Finally, a major part of China’s BCI [brain-computer interface] research aims at cognitive enhancement of healthy persons. Several such institutes acknowledge AGI as a goal, while BCI pioneer Gao Xiaorong (Tsinghua) sees BCI as a link to superintelligence.

Describe China’s generative approaches to advanced AI. 

We emphasized China’s alternative paths to advanced AI at the risk of slighting its significant mainstream research in machine learning and its generative offspring: “large language models.” LLMs have taken the world by storm due to their success as chatbots, at writing code, designing proteins, doing translation, and the popularity of ChatGPT and other such products that process language at the “Turing Test” level. 

China has been quick to follow. Some 100 LLM models were released in 2023 alone, with the count now at around 300. Originality is a moot point (Kai-Fu Lee’s 01.AI release is modeled on Meta’s open-source architecture) – what matters is that the performance is roughly comparable. Problems involving tokenization, data quality, and GPUs are not dealbreakers.

Whether LLMs are a path to general AI is hotly debated in China, perhaps more so than in the West. Some top AI scientists (Zeng Yi, Chinese Academy of Science’s Institute of Automation, and Zhu Songchun, Beijing Institute for General Artificial Intelligence) are working on brain-inspired “small data” models, which at some point may marry up with LLMs whereupon genuine AGI becomes a real possibility.

Assess China’s emerging advanced AI potential vis-à-vis U.S. AI advancements. 

China is on a par with or ahead of the U.S. on many predictors of AI performance and has the will, wherewithal, savvy, and support to push ahead. Notions about free markets, globalization, and the role of political freedom in innovation will have little impact on what happens.

The one area, besides chips, where China acknowledges a deficit is basic science, which if true is offset by China’s ability to tap foreign sources, which is nearly impossible to control. 

Western universities and technology companies for the most part understand the need to protect their IP but the problem is huge and solutions conflict with our tradition of openness. The PRC’s ability to exploit these vulnerabilities dates from 1956, when then-Premier Zhou Enlai instructed China’s S&T managers to build an “intelligence” apparatus for foreign transfers. That enterprise is still blossoming.

Another challenge for the West, beyond shedding its hubris, is its attachment to the one-horse paradigm of LLMs, which may end up at a dead end. China lacks this encumbrance.