Artificial intelligence will be a focal point of the scientific agenda at Beijing’s annual legislative meeting and coming five-year plan, as China and the US pursue divergent AI paths putting their tech ecosystems at odds.
Just over a year ago, Chinese start-up DeepSeek released an open-source AI model that changed how the world viewed the country’s AI capabilities and ambitions.
Chinese companies have embraced an open-source approach to AI development, which has rapidly scaled usage of their models worldwide and sped up AI adoption across domestic industries, including healthcare, energy and transport.
Meanwhile, US-based companies have largely followed a closed-off, paid approach that may limit how these models become integrated into daily life and slow the adoption of advanced AI-reliant technologies.
Last month, Anthropic accused DeepSeek, along with fellow Chinese AI companies Moonshot and MiniMax, of extracting the capabilities of its Claude model to improve their own models.
In a statement, the US-based AI company said the labs generated millions of exchanges with Claude using tens of thousands of accounts to execute “distillation” – a technique of training a less capable model by using the outputs of a more capable one.
While the accusation raised questions about the merits of China’s AI models, distillation itself is not illegal. And to many, it highlighted a hypocrisy present in the AI ecosystem.
To begin with, AI models are predominantly trained using massive amounts of data scraped from the internet that can include copyrighted material used without permission.
Chinese companies have embraced the open-source approach to AI model development, in which source codes are available for anyone to use, modify and distribute. This has contributed to a collaborative environment for AI development in the country.
Chinese President Xi Jinping has stated that the acceleration of innovation and consumption, particularly in AI, is an essential task for the country in the coming year, according to a speech excerpt published by the ruling Communist Party’s leading theoretical journal in February.
In recommendations for China’s 15th five-year plan released in October, Beijing said that “forward-looking plans should be put in place for industries of the future”, including embodied AI – the integration of AI into agents like robots that interact in the real world.
The full text of the five-year plan, which will serve as a blueprint for China’s socioeconomic development from 2026 to 2030, will be approved and unveiled during the “two sessions”.
China’s two sessions are annual parliamentary meetings where its two main political bodies – the National People’s Congress and the National Committee of the Chinese People’s Political Consultative Conference – reveal the country’s policy plans involving the economy, diplomacy, trade, the military, and science and technology.
The five-year plan recommendations also called for advancement of the AI Plus Initiative, a national strategy launched in 2024 aiming to integrate AI into all social and economic sectors.
The integration of AI would help China steer the transformation of scientific research, cultural advancement, industrial development and industrial applications, according to the recommendations.
Meanwhile, the US has sought to slow China’s tech development, including AI, mainly through blocking the export of advanced chips used to power AI.
According to a report to Congress by the US-China Economic and Security Review Commission in November, “Whoever leads in quantum (and artificial intelligence) will control the encryption of the digital economy; enable breakthroughs in materials, energy and medicine; and gain asymmetric and likely persistent advantage in intelligence and targeting.”
“Congress recognises that autonomous systems – including humanoid robots, industrial automation and unmanned systems – represent the physical embodiment of artificial intelligence and a critical domain where the People’s Republic of China is rapidly advancing,” the report added.
Despite US containment efforts, some of China’s open-source models have caught up to their US counterparts in both capabilities and adoption.
The future of AI development is facing several pressing issues, including limits in computational power, massive energy demand, a shortage of training data, and safety and reliability concerns.
China has made efforts to address these issues, including developing synthetic training data that can mimic real-world data.
The open-source research pipeline called SynthSmith, developed by researchers from Tsinghua University and Microsoft and debuted in January, could be used to address the growing issues of real-world data scarcity for improving AI models.
Researchers from Peking University and the Chinese Academy of Sciences also recently unveiled the world’s smallest and most energy-efficient transistor, which could help develop next-generation AI chips.
Nevertheless, there are still pressing issues to address, such as the reliability of AI systems, especially as they become increasingly integrated into real-world workflows.
Western leaders have framed the US-China AI competition as a race to build the biggest, smartest models, but this misses the “real determinant of power”, which is the ability to deploy AI at scale, according to a January report by the US think tank Stimson Centre.
China “may already be years ahead” on this front, the report said, while companies in the US and Europe have lagged behind on integrating AI into their workflows. -- SOUTH CHINA MORNING POST
