US-China AI race. Who is winning?

I have been closely following the development of technology in China for some two years now. Why am I so interested in it?

My interest started with Kai-Fu Lee's book "AI Superpowers." Lee is the former head of Google China and a VC investor in artificial intelligence. What piqued my interest were the opinions and facts cited in the book that artificial intelligence in China is developed based on a different model than in the West.

First, Chinese AI start-ups use much larger data sets and rely much more on training models based on big data than on fine-tuning algorithms. The size of the Chinese market makes this possible and gives them an advantage over Western start-ups.

Second, Chinese AI-based businesses are often based on an online merges offline (OMO) model.Explaining this with the example of a ride hailing business: the Chinese business has its own fleet of vehicles and its own employees. It is not like Uber, for example, a platform that manages a fleet of independent contractors and their vehicles. This allows it to collect much more data about its customers. Which allows it to develop new ideas and business lines using AI models. It gives an advantage over platform-type businesses.

Third, China has officially declared AI a key strategic industry. The central authority in Beijing is delegating huge resources to the authorities of 32 Chinese provinces to create ecosystems for AI development: grants, VC investment, subsidized offices, tax breaks, etc. The provinces are supposed to compete with each other in the area of AI. In addition, China has several times more AI students than the US.

In the times of ChatGPT it is interesting whether this all gives China competitive edge in AI development.

Kai-Fu Lee's book aside, of course we are all following the competitive race between the US and China in the technology area. This race is turning into another "Cold War" similar to the one between the US and the Soviet Union. We who live in the Western world are concerned about the progress of an authoritarian country in the technological race. In addition to AI, US sanctions on chip exports and the race between China and the West in chip production have also been in the news in recent months. Overlaid on top of this are Chinese territorial claims to Taiwan, which is a major producer of advanced chips globally, and a potential Chinese invasion of Taiwan would spell complete disaster for the global electronics supply chain and consequently the entire global tech industry.

Imagine that China wins the race in AI or high-tech chip manufacturing. What happens?

All breakthrough new drugs are produced in China and made available to the Chinese public first. The rest of the world has to wait.

China is gaining an advantage in the production of autonomous weapons, which give it a huge military edge.

Chinese companies, with a technological edge, are becoming leading global investors, just as WeChat owner Tencent, which owns stakes in Spotify, Epic Games and Discord, among others, is already becoming one. Authoritarian China gains access to tens of millions of data on Western citizens.

Is this definitely going to happen? And is China really winning the race for technological dominance? Not necessarily.

The following are my opinions based on books I've read (sources below) and research on the Internet. I am aware that this is quite subjective, as we have few hard sources on what is happening beyond the "Great Wall." Nevertheless:

First, the chips. China does not have developed production of the most advanced chips with the smallest gates available today. In addition, most Chinese chip manufacturers are fabless companies. They design chips, but do not manufacture them. Despite China's non-recognition of Taiwan, much of China's chips are produced by Taiwan's TSMC-the world's largest manufacturer of advanced chips. China's history of producing advanced chips is almost 40 years old, and so far none of China's flagship projects have worked out. Neither a joint venture between Shanghai Huahong and NEC, nor an attempt to create SMIC-a rival to TSMC. Now China is talking a lot about producing "AI chips" with which it can "get ahead of the curve" rival in the race. This does not seem likely-AI chips are some of the most advanced designs. Both GPUs and chips that consume minimal power and emit minimal heat are expected to find applications in medical devices in the near future.

And the invasion of Taiwan? As of today, it fortunately seems highly unlikely. TSMC's factories could easily be destroyed in the event of an invasion, lest they fall into Chinese hands, which defeats justification of invasion. In addition, Chinese companies are a major electronics manufacturer for Apple, for example, and use TSMC as a supplier. An invasion doesn't pay off for anyone. What is likely, however, are Cold War incidents in the South China Sea involving the China and US navies. China will try to show its people that it is capable of militarily blockading Taiwan. The question is, how will this affect supply chain liquidity and the global IT industry?

Secondly- the development of AI. According to most authors and commentators, including the aforementioned Kai Fu Lee, as of today the US is ahead of China in AI development. However, it is worth mentioning in this context that, for example, Forbes says China and US achieved parity, but China is to edge ahead this year (2023). Nevertheless, as Kai Fu Lee stated in an interview: All 16 AI Turing prize recipients are US or Canadian, top 1% papers published are predominantly American, overall more papers are Chinese. At the same time: China has more data and is better with commercialization of AI.

The only area where China is ahead of the US is in image recognition. Certainly the government's citizen monitoring programs and the hundreds of millions (population 1.4 billion people) of images of citizens collected in this way contribute to this. China also leads in the number of CCTV cameras per capita. The leader in face recognition solutions is Megvii and its Face++ platform, which creates APIs for various person recognition solutions. Of course, face recognition is used for citizen surveillance.

As for the big language models, it seems that we won't soon see a Chinese competitor to ChatGPT. Although Baidu has announced its chatbot, and the Chinese Academy of Sciences has also announced that its work on LLM is advanced.?

Chinese LLMs are based on data available in Mandarin, which is far (several dozen times?) less than data available in English, which reduces tremendously the possibilities for LLM training. Additionally, the largest cloud providers are American big tech.

Interestingly, despite the "Great Firewall" that seals off the Chinese Internet from the rest of the world, models are also trained on data available outside the Chinese Internet, but are censored. An attempt by a Western journalist to ask one of the chatbots whether Xi Jinping is a good leader was simply met with a refusal.

China's advantage is certainly its physical infrastructure. Given the authoritarian political system, it's much easier to build 5G networks, or sensor-equipped highways that will allow autonomous trucks to move (paradoxically, it's hard to train autonomous cars in China's crowded metropolises with chaotic traffic).

5G infrastructure could give China an edge in building medical devices backed by AI.

Finally, it is worth stating that the development of AI is not just a matter of a race for world domination for China. It is also, and perhaps most importantly, a matter of mastering the country's own social development and rapid urbanization. From an eminently agricultural country, China has achieved 60% urbanization of society in the past 30 years and expects this to rise to 80% within a few years. The 30,000-strong fishing town of Shenzhen, which served as a border crossing with Hong Kong, has grown into an officially 13-million (unofficially 20 million) agglomeration and global hardware design capital. Without AI, managing transportation or countering crime in such large and dense human concentrations seems much more difficult in the future.

Books:

Videos: