Chinese AI researchers were already aware of large language models, but the release of chat GPT by OpenAI motivated them to catch up and develop their own models.
Chinese AI companies have primarily focused on practical applications of AI, but generative AI introduces new business models and expands the potential of AI in China.
Chinese regulators have introduced interim regulations on generative AI, which are less stringent than the initial draft regulations and primarily apply to public-facing platforms.
Deep dives
China's AI researchers react to OpenAI's chat GPT
Chinese AI researchers were not surprised by the unveiling of chat GPT by OpenAI, as they were already aware of the existence of transformers and large language models. However, they were surprised that OpenAI released it to the public as an experiment. Chinese companies and researchers were motivated to catch up and develop their own large language models. The release of chat GPT compelled Chinese regulators to respond and consider regulations on generative AI.
Focus on practical applications of AI in China
Chinese AI companies have been primarily focused on developing practical applications of AI, such as perception AI, internet AI, recommendation algorithms, and logistics. Generative AI introduces new business models and applications, expanding the potential of AI in China. However, the focus remains on practical usage, and the debate on the existential risks of AI has not been as prominent in China compared to the West.
Regulations and government support in China
Chinese regulators have introduced interim regulations on generative AI, which are less stringent than the initial draft regulations. These regulations primarily apply to public-facing platforms. Chinese companies provided feedback during the drafting process, and the regulators considered industry concerns. Additionally, the Chinese government has signaled support for the AI industry, encouraging Chinese companies to lead in AI development. There are also efforts to establish data centers and provide advanced compute capabilities to fuel the industry's growth.
US Controls on AI Hardware and Workarounds
The US government is implementing controls on AI hardware with the goal of preventing the emergence of AI in China as a military force. These controls aim to restrict China's access to advanced AI capability, which the US perceives as a potential game changer. The focus is on preventing China and other countries from developing an advanced AI capability that could alter the military balance. However, it is challenging to define what exactly constitutes a game-changing capability. While GPUs and advanced compute capabilities have tangible applications in high-performance computing and weapons systems development, the majority of generative AI applications are civilian-oriented, such as drug discovery and data mapping. The US is also considering investment controls in critical technology sectors to limit venture capital and private equity into Chinese companies. AI, being software-based, poses specific challenges in terms of control and application restriction, as it is difficult to define and target specific AI applications that are considered significant for national security. Overall, the US is navigating the complex landscape of controlling AI hardware and managing the potential implications for national security and global collaboration.
Workarounds and the Future of AI Development in China
Chinese companies are exploring workarounds to bypass the US controls on AI hardware. One strategy is to stock up on existing hardware like modified versions of NVIDIA GPUs before the performance thresholds change. Acquiring restricted semiconductors through various channels and using domestically developed capabilities like Huawei's ascend processors are also part of the workarounds. Chinese companies are focusing on optimizing large language models for specific industries like weather forecasting, energy optimization, and drug discovery. They are developing proprietary models that can process vast amounts of data effectively and efficiently. Additionally, companies like Alibaba and Baidu are providing access to models like Meta's LAMA through their cloud services. However, licensing and regulatory issues may arise from these workarounds. As a result of workarounds and the advancements in talent and education, China has become a major player in the AI space globally. The concerns regarding the future of AI development lie in the potential for decoupling and a lack of collaboration and information exchange between countries, which could lead to surprises and challenges in regulating the industry. It is vital to include Chinese companies and regulatory authorities in the global dialogue to avoid a fragmented and suboptimal future for AI development and regulation.
This week on Sinica, Paul Triolo returns to the show to give us a rundown on what’s happening in the exciting arena of generative AI in China. The veteran China tech watcher, who is now Senior VP for China and Technology Policy Lead at Dentons Global Advisors ASG, is Just back from a trip to China during which he spoke with numerous companies working in the space, Paul offers a great overview of what various companies are doing, and how they’re responding to U.S. restrictions on the export of key hardware needed for large AI training and modeling.
03:38 – The Chinese AI community’s reaction to the unveiling of ChatGPT by OpenAI
08:14 – What drives China’s National AI Development strategy?
14:31 – Chinese AI researchers and their perspectives on regulation
21:28 – Is there a lot of investor money going into Generative AI startups?
24:25 – U.S. policy on China’s AI development
35:53 – What will China’s Generative AI look like?
44:14– Companies involved in Chinese AI
51:31 – The changing availability of innovative AI scientists in China
55:10 – How will decoupling effect AI competition?
