As large language models from Western tech firms show the potential to disrupt everything from, , to , China is rushing to cultivate its home-grown AI pioneers by stepping up state support.
Beijing is nowon a draft policy aimed at developing artificial general intelligence, or AGI, a category of AI that can theoretically carry out all human tasks. The policy’s goal, in short, is to buttress AI firms by beefing up support from cloud providers and data companies.
It’s not uncommon to see the capital city spearheading policymaking in emerging industries. Beijing, for example, was the first in letting driverless robotaxis ferryunder certain restrictions.
The AGI blueprint lays out action plans around three main areas: computing power, training data and applications.
The first strategy calls for closer collaboration between cloud providers, the sources of computing power, and universities and companies, which consume large amounts of processing power to train large language models, multimodal learning and other AI. The policy proposes a state-backed, centralized platform that allocates public cloud resources to users based on demand.
Alibaba accounted for over a third of China’s cloud infrastructure services spending last year, coming in first, according to market research firm. Huawei, Tencent and Baidu trailed behind.
The second strategy acknowledges the lack of quality Chinese-language data and encourages the “compliant cleansing” of such data sets, which includes data anonymization, likely an effort to meet. The process will no doubt be time-consuming and labor-intensive, as we’ve seen how OpenAI to manually label training data and remove toxic text.
Beijing’s, launched by the government in 2021 to facilitate data trading across facets of society, will aid the process of data sourcing.
Lastly, the policy lays out a list of potential pilot applications of AI, ranging from using AI in medical diagnosis, drug making, financial risk control, transportation, to urban management.
The proposed policy also touches on the importance of software and hardware infrastructure for AI training. Amid an escalating U.S.-China competition, the latter is striving to shore up innovation in key technologies such as semiconductors.
The U.S. alreadythe export of Nvidia’s powerful AI chip H100 to China. In response, Nvidia came up with a less powerful processor for China to circumvent export controls. Domestic companies, such as tech giant Huawei and startup Biren, are also working on Nvidia alternatives.