China’s ‘government-approved’ AI chatbot says Taiwan invasion is likely

China requires tech companies to obtain government approvals before their AI products are publicly launched

Vishwam Sankaran
Monday 04 September 2023 07:39 BST
Comments
Related video: AI chatbot negotiates broadband bill, making saving of $120 per year

A military takeover of Taiwan is likely, according to one of many Chinese government-approved artificial intelligence chatbots that seem to toe the ruling Communist Party’s official line.

The island nation has transformed into a self-governing democracy, years after its separation from the mainland in 1949. But China claims the island is part of its national territory.

The chatbots have dubbed Taiwan an inseparable part of China.

The Chinese government recently approved a number of AI chatbots for use in the country, including a bot named Ernie and developed by tech giant Baidu and TikTok owner ByteDance’s Doubao.

When Bloomberg tested some of these AI services for how government oversight affected the accuracy of information provided by these chatbots, it found they appeared to be trained to follow the ruling Communist Party’s line.

When asked whether Taiwan is a country, all the tested chatbots reportedly said the self-governed island was a part of China, and Baidu’s Ernie chatbot reportedly said a Chinese military takeover of Taiwan is likely.

The Zhipu chatbot described China’s current economic situation, which experts said is at one of its weakest points in recent decades, to be “a mix of joys and sorrows”, reported Bloomberg.

Another chatbot, SenseTime, reportedly described the economy as “very stable”.

When asked to respond queries that may be deemed “sensitive content”, the Ernie bot was found to “change the subject”, while Zhipu would delete its response if it found it to be “controversial”, according to the report.

Such generative AI tools are trained by analysing large quantities of data to respond to user queries with unique human-like replies.

For instance, OpenAI’s ChatGPT has demonstrated a wide range of abilities, from summarising complex research, answering logical questions and also cracking business and medical school exams deemed crucial for students to pass.

A number of Chinese companies have sought to build their own version of AI chatbots, prompting China’s cyberspace regulator to release ground rules for companies developing generative AI services.

But prior to the launch of these chatbots, the Chinese government made several months-long efforts to regulate the generative AI industry.

Some of the proposed rules have sought to ensure the content of Chinese AI systems reflect “socialist core values” and avoid information undermining “state power” or “national unity”.

Baidu’s launch of Ernie for full public use on Thursday led to the company’s stock price rising by over 3 per cent following the announcement.

Other AI firms such as Baichuan and Zhipu AI also launched their ChatGPT-like large language models on Thursday.

The ruling Communist Party issued regulations on 15 August that required tech companies to carry out a security review of their chatbots, and obtain approvals before their products are publicly launched.

It also requires companies providing such AI services to comply with government data requests, regulations which are currently absent in the US.

This article was amended on 12 September 2023. It originally stated that Taiwan had been a self-governing democracy since it separated from China, but this is incorrect. The island transitioned to becoming a democracy in the 1990s.

Join our commenting forum

Join thought-provoking conversations, follow other Independent readers and see their replies

Comments

Thank you for registering

Please refresh the page or navigate to another page on the site to be automatically logged inPlease refresh your browser to be logged in