Artificial Intelligence in China - Myth and Reality

There's been a great deal of discussion surrounding the advancement of AI in China in recent years. While some hail the strides Chinese corporations have made in AI research and its applications, serving the benefits for humankind, others warn of doomsday-like scenarios where China, as the global frontrunner in AI, could misuse its power for evil purposes. It seems that now is a good time to separate the myths from the realities of China's AI roadmap and the current state of affairs within the country.

Drivers of AI in China

China houses 28% of the world's manufacturing capacity. With the advent of AI, commencing with Industry 4.0, it started integrating into these manufacturing processes. Factory shop floors became capable of self-configuration, while robots developed the ability to identify errors, alter procedures, "talk" to fellow robots, and accomplish single-lot production within a mass-production environment.

The China 2025 plan, unveiled nearly a decade ago, distinctly emphasized the government's ambition to achieve global leadership in many areas, including integrating AI into business applications. This initiative was swiftly followed by actions that produced striking results. Today, Chinese researchers contribute roughly 40% of all AI-related articles published in international journals. In the last six months alone, China has released 40 large language models (LLMs) for public use, even though most global rankings are still dominated by US-based LLMs. Public acceptance of AI in China is among the highest globally. In many cities in China, you can drive in autonomous taxis; in many airports, boarding works purely operated by facial recognition; and in many hospitals, doctors use AI to analyze X-rays and formulate treatment plans. Overall, there's a prevailing sentiment that AI makes life easier.

Which industry sectors are impacted most by AI?

AI is increasingly important in many aspects of life in China and most industries. However, it can be challenging for an outside observer to understand this change in context. Therefore, I will describe only two areas here that exemplify both the advances and challenges that AI applications are facing.

• AI in industry and production

China, home to an immense number of large-scale factories, is now recognized by the European Centre for Economic Policy Research (CEPR) as the world's unrivaled manufacturing powerhouse.  Some decades ago, during the initial phase of China's opening and reform process, most factories had hundreds or thousands of workers in blue overalls. They were stationed along production lines, diligently manufacturing textiles, toys, and other relatively simple products. However, China's approach has since evolved. The country is now fostering the production of high-tech goods such as advanced printing machines, lasers, sophisticated drones, and premium electronics.

Indeed, I have experienced a factory operating in total darkness, populated only by silently functioning robots. Humans were summoned—by the robots themselves—only when an issue arose that required human intervention. In such instances, the closest engineer possessing the right skills would be called upon, identified via an RFID chip worn by everyone on site. While not always suitable, this extreme AI manufacturing example showcases automation potential.

China continues to push the robotization of its manufacturing industry in high speed and along with it the application of AI on the shopfloor. In 2021 China had 18% more robots installed per worker than the US, despite its much lower labor costs. The rate will continue to grow, probably tale over the robot density of Japan and Germany too.  Of course, one will find super modern factories like Tesla’s Gigafactories in US and Germany or Siemens’ digital factory and industrial research center in Erlangen, Germany. As many German manufacturers in China tell me – the key difference to he rest of world in China is the speed of creating new factories, the high willingness to try and apply new technology, and the general openness to AI.  

• AI in mobility

Most people immediately think of autonomous driving when AI is mentioned alongside traffic and mobility. However, there are myriad additional intelligent traffic solutions. Many cities in China already use large-scale, interconnected AI to maintain safe traffic. If I order a “Didi” Car through my App it will not just give an estimate of its arrival, but, for example, also tell me, that the car is standing just one traffic light away, and the waiting time there for the next green light is just 40 seconds.

 Although fully autonomous driving progress is surprisingly slow, recognized test areas exist in Guangzhou, Shenzhen, Shanghai (Anting), and Beijing, hosting several hundred autonomous taxis. However, most Chinese car manufacturers are focusing their strategies on AI-assisted driving for time being —where a driver remains necessary to assume control—as opposed to fully autonomous vehicles. These systems rely heavily on AI algorithms, such as Convolutional Neural Networks, Recurrent Neural Networks, and Reinforcement Learning. Western companies, like Tesla, also adopt this model, leaving full autonomy to startups like Waymo in San Francisco, a ride-hailing service that offers fully autonomous rides. In Europe, fully autonomous taxi rides seem to be still years away, stuck in bureaucratic and legal processes hindering the automotive industry to bring pilot vehicles on the road. The traffic laws are old and complex. European manufactures rather chose the US and Asia as test grounds, so Europe loses in my opinion out on an opportunity to massively reduce carbon footprint, leave alone on the expertise.

Generative AI in China is fast growing.

Like everywhere else, China experienced a breakthrough in generative AI in 2023. Throughout the year, several large language models (LLMs) such as Baidu's ERNIE, Huawei's Cloud Pangu, Tencent's Hunyuan, and Alibaba's Tongyi Qianwen were introduced to the public These Chinese LLMs have sparked considerable discussion in the West. Some observers argue that China is trailing behind tech giants like OpenAI, Alphabet, and AWS, while others contend that Chinese LLMs are superior, offering more detailed responses and data that are more rooted in industrial contexts, thus making them more user-friendly in B2B scenarios. My personal experience with questions about business context has been good. However, the reviews are quite diverse; Forbes comes to a different assessment than the China Daily.  Furthermore, ERNIE’s capabilities in Chinese are reportedly much better than what it can do in English.

Generative AI can be a headache for regulators.

Chinese regulators have been particularly keen on encouraging the use and application of LLMs among the public and industry. The utilization of LLMs from Chinese tech powerhouses, including those previously mentioned, received rapid approval in 2023, with almost two approvals granted each week. At the same time, one of the key features of generative AI is doing exactly what its name suggests—generating text, pictures, and now video—yet not all such generated content would pass the Chinese censors if it were posted by humans, so the Gen AI systems are also restricted. Since Gen AI systems cannot (yet) be held legally accountable for their actions, the responsibilities fall on the publishers of the LLMs. They must exercise caution, using the "right" algorithms and rigorous training of the models to ensure that the LLMs only generate content that complies with the law. This creates a challenge for Chinese regulators - how to maintain order and enforce compliance without hindering innovation. China was the first country worldwide to issue regulations on generative AI and is expected to continue to issue AI laws and regulations that achieve both objectives above.

How do the key regulations for AI in China impact its progress?

Since enacting its Cyber Security Law in January 2017, China has repeatedly published regulations for the governance of cyberspace, with a focus on both national security and data privacy. Many rules that apply to, for example, cross-border data transfers and data localization have these two areas in mind. Firstly, they consider whether data, if controlled by foreign entities, could threaten China's national security. Secondly, they examine whether the citizen's right to determine the location of their data storage is safeguarded. Notably, in terms of data localization, the EU has similar regulations as part of the GDPR.

Certain data inevitably needs to be stored within China. This can include information gathered from Tesla or VW vehicles navigating the streets of Shanghai and Beijing or a large-scale social media advertising campaign by Nike targeting millions of recipients. Non-Chinese LLMs that include data from China face a significant issue. By collecting data from global sources, their usage within China could be limited by Chinese regulators, among others. As a result, many of these LLMs opt to exclude Chinese users entirely.

In the EU, meanwhile, the EU Commission took a very different approach to regulating AI in its so-called AI Act starting from a risk classification of AI applications. Similar to China, Europe struggles with the challenge of regulating AI without killing innovation. The European Computer and Industry Association has called the impact of the EU AI Act on Innovation disastrous.  Other observers don’t see it that threatening to innovation and do stress to its aim of providing a technology-neutral definition of AI systems and establish harmonised horizontal regulation.

The worldwide AI race and global regulations

There can be little doubt that the global race for AI supremacy is underway. Only a few nations have the financial and technological capabilities to compete, and China is ranked by most observers and researchers as second only to the US—depending on what criteria are used. Despite a promising beginning at the Bletchley Park Summit, which essentially all nations with AI research capabilities attended, binding global regulations for AI and its applications are probably years away. Therefore, national governments (or the EU) are responsible for issuing regulations that strike the right balance between security concerns and innovation; this is no different in China than anywhere else.

Undoubtedly, all stakeholders—governments, civil society, industry, and researchers—must collaboratively establish worldwide guidelines for AI. Two domains frequently highlighted at summits, in AI discussion groups, and during conferences are the need for governance in AI responsibility—encompassing ethics, truth, and good faith—and AI safety, which involves controlling and securing AI to prevent it from posing a potential threat to society or even humanity itself. Considering the worldwide diversity in beliefs, values, and principles, considerable effort and determination will be required to reach a consensus on enforceable regulations in these areas.

Conclusion

China continues to forge ahead in AI development. While global attention is often centered on the US—thanks to ChatGPT and others—the goings-on in China tend to be less understood beyond its borders. Yet, the extent to which AI has been integrated into everyday life in China exceeds what I've observed nearly anywhere else. However, the degree to which this will have a global impact remains to be seen.

Geopolitics, for the time being, remains the "great unknown" in the equation. Despite the clouds on the horizon, I maintain an optimistic outlook. As with other global issues that affect all of humanity—such as climate change or genomics—I believe that AI will eventually unite all stakeholders to establish essential guidelines. This, I hope, will foster more international cooperation, encouraging the peaceful use of this technology for the benefit of everyone, everywhere.

(Disclaimer: The ideas, views, and opinions expressed in my LinkedIn posts, articles, videos, and profiles represent my own views and not those of my current or previous employers or any organizations with which I am associated. In addition, any and all comments on my posts from respondents/commenters to my posts belong only to the responder posting the comment.)