If the US slows down AI research won’t China just get there first?

It is reasonable to consider the competition for AI supremacy between the US, China, and other countries. China has recently emerged quickly as an AI superpower, with a majority of publications in AI coming from China. However, concerns of China’s AI capabilities are often overblown. We will first explore some reasons why China can make fast advances in AI, and then we will explore reasons against China being a big threat to the US. Here are some reasons for China’s rapid progress on AI:

• China’s economy is large. This economy can be used to boost strategically important industries. For example, it has been less than 35 years since China opened its first nuclear power facility, but it is already the world’s second largest generator of nuclear power.

• 

• China’s large economy offers Chinese companies a lot of data to train their AI models with. This is vital for development of stronger AI models.

• Chinese AI progress is often the result of direct interaction between government, academia, and industry. This might be crucial if there are strong feedback loops between different actors.

• China has weak privacy regulations, which can explain why it has made rapid progress in some areas of AI. For example, the large number of surveillance cameras in China supports the development of visual and facial recognition.

• China already produces more AI publications than any other country. However, it is important to note that a lot of the most important advances in AI come from industry, and industry labs publish significantly less papers than academia.

However, there are several reasons to believe why the perception of China’s importance in the global AI race might be exaggerated:

• The semiconductor supply chain is largely controlled by the US and its allies. They have previously used this to restrict exportation of semiconductors to China.

• The top AI labs are currently based in the US and UK. These include OpenAI, Google DeepMind, Anthropic, and Meta AI.

• Important breakthroughs in modern AI research have largely come from the US and Canada. This includes deep convolutional networks, Generative Adversarial Nets, reinforcement learning, AlphaGo, transformers, reinforcement learning from human feedback, etc.

• The US produces most of the top AI publications, measured by the number of citations.

• The US has better access to top AI talent than China. This can be seen in the number of AI authors on arXiv (a platform for pre-publishing papers), which is dominated by US authors. The US also attracts foreign AI expertise much better than China does.

• 

• Censorship in China will hinder AI development, especially in the field of generative AI. Large language models (LLMs) are not allowed to be trained on datasets that include forbidden discourse, and if a model outputs content deemed harmful by the Chinese government, the company responsible for the model is liable.

• If the US slows down AI development, it might also slow down AI progress in China, as Chinese work on AI seems to significantly rely on research published abroad. The dependence of other countries’ advancements could mean that it would be difficult for China to establish a lead in AI.

• Many Chinese startups seem to be moving their operations overseas to escape strict regulations and economic slowdown in the country.

• China has a low capacity for diffusing technologies, meaning the adoption rate of technologies across businesses.

• Historically, the US tends to overvalue the technological capabilities of its competitors. During the Cold War, US officials made policy based on the illusory “bomber gap” and “missile gap”, often inflating the perceived threat of the USSR.

Overall, while China plays an important part in the global AI race, there are indications that its capabilities might be overstated. Furthermore, it is important to note that it does not matter which country gets to AGI first, if we don’t know how such a system can be built in a safe way, since this could possibly lead to an existential catastrophe.