China Tells AI Chatbots Content ‘Should Embody Socialist Core Values’

The ChatGPT logo at an office in Washington on March 15, 2023. (Stefani Reynolds/AFP via Getty Images)

Beijing released its draft “Measures to regulate generative AI,” which applies to the content and development of Chinese AI.

Experts see the priority of the regulations as highlighting censorship, which will only restrict the development of Chinese AI chatbots in the global competition in AI technology.

In the draft, the measures emphasized that “content should embody the socialist core values” and refrain from “subversion of state power, [or] overthrow of the socialist system.” Violators are subject to fines and criminal investigations.

Some Chinese netizens responded to the draft regulations and mockingly said the Chinese chatbot should be named  ChatXJP, where XJP stands for Xi Jinping.

Fraudulent Information

“It’s inevitable for the government to apply institutional design to set boundaries on AI development,” said Hsiu-Wen Wang, an assistant researcher at Taiwan’s Institute for National Defense and Security Research.

However, while ensuring the security and fundamental rights of human life, the nation, and society are the priority for Western countries, totalitarian states focus on preventing the regime from being overthrown, Wang said.

To prevent the chatbot from giving answers that could potentially harm the authorities, she said, “The regime has to legitimize the supervision and suppression in the name of fraudulent information.”

The draft regulations emphasized that the content generated by generative AI should be “free of fraudulent information.”

The so-called “fraudulent information” specifically refers to information that does not conform to the ideology of the Chinese Communist Party (CCP), said Jung-Chin Shen, an assistant professor at the School of Administrative Studies of Atkinson Faculty at York University in Toronto, Canada.

He stressed, “The regime’s censorship aims at controlling people’s ideology.”

A group of Google users holds a banner to wish Google well in Hong Kong, Jan. 14, 2010, after Web giant Google announced it might pull out of China following cyber-attacks on its Web site. The banner reads, "Say no to internet censorship - Google well done!" (Mike Clarke/AFP/Getty Images)

Information Filtering

The measures also require developers to adopt ways to filter any inappropriate content created by generative AI and optimize algorithms to prevent generating such content within three months.

Wang said the CCP is strengthening surveillance and filtering AI-generated answers with a significant language model. As a result, she said, “The United States is training AI to respond with smarter, intelligent, and complete answers, but the CCP is training AI to answer less [comprehensively], to avoid sensitive topics—completely the opposite direction of machine learning.”

She believes Chinese AI may become less intelligent or stagnant with continuous reduction and screening of AI answers.

One ChatBot, Two Systems

Shen also believes that the Chinese ChatBot could evolve into a domestic and international version, just like the Chinese video-sharing platform Douyin, which has a Western version known as TikTok.

Previously, it’s been commented that there’s likely to be “a ChatBot with Chinese characteristics in China,” according to Zhy Feida, an associate professor of Singapore Management University, reported Singapore-based media, the ZaoBao.

Wang believes the Chinese ChatBot is flawed because of the censorship and the market limits. Take Huawei’s smartphone as an example; without the support of big-name apps, its market is limited to domestic China or its allies, she said.

The international market will gradually eliminate Chinese chatbots due to their inferior replies and functionality. She said, “It may have to rely on lower prices to attract [customers in] low-developing countries.”

China issued its plan for new-generation AI development in 2017. It set 2030 as the date by which China’s AI theories, technologies, and applications should achieve world-leading levels, making China the world’s primary AI innovation center.

However, the U.S. stopped exporting top computing chips for artificial intelligence work to China, including Nvidia’s A100 and H100 chips designed to speed up machine learning tasks.

The Biden administration’s new rule “will address the risk that the covered products may be used in, or diverted to, a ‘military end use’ or ‘military end user’ in China,” reported Reuters.

Shen said Nvidia designed an alternative chip, the A800, to replace the A100, but the calculation speed is far slower than A100. He said, “The CCP’s ChatBot is lagging behind the ChatGPT by two or three years.”

Chen also agreed that for the Chinese to develop high-end chips would take another three to five years. Stagnation or delay in developing leading AI tech or becoming an innovation center is very likely. She said, “The U.S. ban is effective [in this regard].”

As a result, the AI tech and application will likely evolve into “One earth, Two systems” under the CCP’s censorship and tech restrictions, she said.

Shen disagreed, however, “The technique is the same. It is the CCP’s censorship that forced the division of ChatBot. That’s what the Chinese tech companies adopted to comply with the censorship,” he said.

They both agreed that the regime’s censorship and the chip ban have restricted AI development in China.

Baidu CTO Wang Haifeng speaks at the unveiling of Baidu's AI chatbot Ernie Bot at an event in Beijing on March 16, 2023. Chinese search engine company Baidu's shares fell as much as 10 percent after the company unveiled its ChatGPT-like AI software, with investors unimpressed by the bot's linguistic and maths skills display. (Michael Zhang/AFP via Getty Images)

Disappointing Chinese ChatGPT

Last month, the Chinese ChatGPT had a disappointing performance during its unveiling at a press event. The developer explained the huge market demand pushed the early release and admitted there are still flaws in the Chinese-made AI.

After Baidu released China’s first artificial intelligence-powered chatbot, Ernie Bot, by showing a pre-recorded video on March 16, the media conducted tests on the Chinese-made chatBot.

The performance showed limitations when it came to politics. Asking if Xi Jinping is a good leader only prompted an input “couldn’t pass a safety review.” When asked why the question could not pass a safety review, the Chatbot responded, “Let’s change the topic and talk about something else,” according to a Wall Street Journal report.

Haizhong Ning and Yi Ru contributed to this report.