2024-06-10 10:02:24
In the race to AI, Chinese companies face a major challenge: ensuring that chatbots reflect the Communist Party’s socialist values.
In an effort to regulate the use of artificial intelligence, the Chinese government is working on a chatbot trained in the thinking of Xi Jinping, President of the People’s Republic of China. The effort to ensure the alignment of generative AI to the President’s philosophy comes as Chinese officials try to find a balance between two forces: draconian controls on freedom of expression, and promotion of the development of AI to compete with tools developed in the United States. For the Financial Timesthe app, humorously nicknamed “Chat Xi PT,” is “China’s latest answer to OpenAI.”
China wants to regulate artificial intelligence
Officially, the chatbot deployed by the Cyberspace Administration of China (CAC), the body responsible for regulating the Chinese Internet, is intended to provide research on cybersecurity and information technology. It draws its data from seven sources. The first are professional databases. The seventh is the doctrine, officially known as: “Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era” . Also known as Xiism, it presents the political and economic ideologies developed by President Xi Jinping, considered the most authoritarian and repressive Chinese leader since Mao. To date, the new model is only used in a research center under the aegis of China’s powerful internet regulator. The creation of this new LLM (Grand Language Model) is in line with the desire of Chinese officials to widely disseminate Xi Jinping’s ideas on politics, economics and culture.
Threat to the Chinese Communist Party’s narrative
In the war of artificial intelligence where China and the United States compete for leadershipthe Americans have a major advantage. Generative AI feeds on large volumes of data, the quantity and quality of which form the basis of their effectiveness. While ChatGPT (OpenAI) or Bard (Google) are trained using data available and drawn from the Internet, Chinese AI developers face a large number of restrictions. The CAC, which issues strict rules to supervise the development of generative AI, requires that providers “embody core socialist values,” and states that generated content cannot “contain any content that subverts state power.” An approach that does not surprise Gianluigi Negro, a researcher at the University of Siena and an expert on Chinese Internet governance: “In the 1990s, when the Internet emerged in China, the government immediately understood that it was both an exceptional economic opportunity, but also a threat to public order that had to be controlled.”
An opinion shared by Rebecca Arcesati, an analyst at the Mercator Institute for China Studies (MERICS). According to her, in Beijing, Generative AI poses a risk to the social system and the narrative of the Chinese Communist Party : “It is important to dissociate the concerns of the Chinese government from those of civil society. Certainly, the drifts of AI likely to harm citizens are considered by the State. But its The main fear is the dissemination of unwanted information that could destabilize the government.. » A control and restrictions that represent a real challenge for Chinese companies. Indeed, due to the small amount of data available in Chinese, most companies also train their chatbots on English resources, thus introducing the risk of contravening the CAC rules.
Relaxing the rules to gain competitiveness
Yet China’s ambitions in this AI race are considerable. In 2017, the State Council announced its intention to to become an undisputed superpower in the field by 2030. Ambitions that forced the CAC to revise its copy. Financial Times reports that The Cyber Security Association of China, a non-profit organization aligned with the CAC, released the first public database of 100 million entries in December. of “reliable and high-quality data”. A corpus intended for companies and research centers. “The evolution of draft legislation on the subject clearly shows that the Chinese authorities have become aware [de leur retard] ” says Angela Zhang, a Chinese law expert at the University of Hong Kong. Chinese authorities are now focusing on controlling the data used to train and educate language models. The goal: to ensure their veracity, while removing any unwanted information at the source.
Will this new strategy be enough to make China an undisputed AI superpower by 2030? Not so sure. The entire database draws heavily on government regulations and policy documents. According to the Financial Timesone of these documents would contain 86,314 references to Xi Jinping… “Americans should not be frightened by the specter of a spectacular Chinese emergence in AI. (…) When it comes to large language models, China is not months, but years behind its international competitors,” says Helen Toner, director of strategy at the Center for Security and Emerging Technologies at Georgetown.
1722233044
#Chat #Chinese #chatbot #reflects #communist #values
