China’s future as an AI superpower is constrained by the Chinese censorship which aims to control all aspects of expression — newspapers, film, literature, media and education. There are strict political guidelines. There is aggressive editing of information. All this affects the scale of searchable information. There are new draft rules on chatbots. Generative AI content must embody socialist values. The content must not subvert state power. There was pre-existing censorship. In addition, there are new efforts at AI oversight. Traditionally there was key word filtering to block unacceptable information. Of late there is Whac-a-Mole approach to contain generative processing of such information.

Baidu’s Ernie Bot, a large language model from China, has been criticised for operating in a ‘firewalled internet rules by government censorship’. Other language processing models, such as Robot, Lily, Alibaba’s Tongy Qianwen (truth from thousand questions) have shown disappointing results.

OpenAI’s ChatGPT 3.5 provides false or hallucinogenic information in Chinese.

Information is the raw fuel for large language AI models. China’s AI models due to state censorship have been small language models. This is the key difference, and it will tell upon China’s ability to be a global power.