China’s chatbots will have to toe the Party line

Last week, Baidu unveiled China’s latest rival to OpenAI’s ChatGPT – ERNIE Bot or short for “Enhanced Representation through kNowledge IntEgration.” The “multi-modal” AI-powered chatbot can generate text, images, audio, and video from a text prompt.

But ERNIE was poorly received by the public. Baidu’s Hong Kong-listed shares fell by 10% during the media conference, and the beta test is only open to a group of organizations approved by the company.

ERNIE Bot will not be a Chinese substitute for ChatGPT, but that might be how the Chinese state wants it. As earlier efforts to make AI chatbots have shown, the ruling Communist Party prefers to maintain strict censorship rules and government steering of research – even at the cost of innovation.

ChatGPT is not directly accessible in China due to Beijing’s protectionist approach to digital sovereignty. Chinese data are confined within the country, and information that conflicts with government propaganda is censored.

Chinese tech companies including Baidu and Tencent prohibit third-party developers from plugging ChatGPT into their services.

Still, the prominence of ChatGPT created a booming illicit market. Until a crackdown, logins for the chatbot were sold on the e-commerce platform Taobao, and video tutorials were published on Chinese social media to demonstrate its abilities.

Censorship rules

Baidu isn’t the first or only tech company in China to trial a generative AI chatbot. In March 2017, Tencent launched two social chatbots, called XiaoIce and BabyQ, on the WeChat and QQ messaging apps respectively.

XiaoIce was developed by Microsoft, while BabyQ was created by a Beijing-based AI company called Turing Robot. Within months, the two chatbots were taken down to be attuned according to China’s censorship rules.

BabyQ never came back, but Microsoft’s XiaoIce returned and has been providing AI companionship services to millions of users on major platforms including WeChat, QQ, and Weibo.

Beijing, of course, would be on the defensive if China adopted only AI chatbots developed overseas. As they run on human feedback, it would be impossible to prevent transnational flows of data and the political interests of the Chinese Communist Party may be threatened.

Since 2015, during the administration of former premier Li Keqiang, the Made in China 2025 policy has endeavored to bolster the country’s technological capacities. AI is a major focus.

Chinese tech companies across AI, food delivery, e-commerce and gaming have scrambled to catch up with OpenAI since February and provide their own ChatGPT-style products.

Beijing’s Municipal Bureau of Economy and Information Technology is supporting this ambition, but only for some leading tech companies.

We can expect China will witness the short-term proliferation of versions of ChatGPT services, many of which will vanish or be acquired by big tech companies.

Smaller firms, with little support from the government, are unlikely to be able to afford the costs of censorship.

Political questions

A small startup called YuanYu launched China’s first ChatGPT-style bot in January. Dubbed ChatYuan, the bot ran as a “mini-program” inside WeChat. It was suspended within weeks after users posted screengrabs of its answers to political questions online.

Even so, Chinese users are still interested in large language models based on the Han Chinese linguistic system.

ERNIE Bot, for example, claims to be more culturally savvy than ChatGPT, with a better understanding of Chinese history, classical literature, and dialects.

Beijing has tightened its governance of the tech industry since a crackdown in 2021.

One upside for the industry is a secure flow of funding and talent support. The flip side is that resources are steered towards technologies that serve Beijing’s immediate interests in domestic governance and military defense.

China’s ChatGPT imitators are more likely to be designed to benefit enterprises than individuals. For tech giants, the objective is to form a “full AI stack” by integrating generative AI products into every level of their business, from search engines and apps to industrial processes, digital devices, urban infrastructure, and cloud computing.

But AI-driven chatbots can also lead to adverse outcomes. Alongside the universal concerns around job security, copyright, and academic integrity, in China there are also extra risks of emotional surveillance and disinformation.

Chatbots can identify users’ emotional status through conversations. This emotion-reading ability extends the power of big data and AI to invade people’s privacy.

Controversial events

In China, such emotional surveillance could further establish “authoritarianism”. Any sentiments that could threaten the leadership of the Communist Party, even if not directly stated, have the potential to attract punishment for the user.

AI-powered chatbots and search engines are also likely to legitimize Chinese state-organized propaganda and disinformation. Users will come to trust and depend on these services, but their inputs, outputs, and internal processes will be heavily censored.

Chinese politics and leadership will not be up for discussion. When it comes to controversial events or history, only the perspectives of the Communist Party will be presented.

Fan Yang is a Research Associate at RMIT and Alfred Deakin Institute, Deakin University.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

The views and opinions expressed in this article are those of the author and do not necessarily reflect the official policy of China Factor.