Visitors walk past the Baidu booth during the World Artificial Intelligence Conference (WAIC) in Shanghai
Beijing (AFP) - Tech giant Baidu launched China’s first public artificial intelligence chatbot on Thursday, with ERNIE trained to censor highly sensitive topics for the ruling Communist Party such as the Tiananmen crackdown.
Beijing issued new regulations this month for China’s AI developers that maintain the government’s tight control on information while allowing them to stay in the race with the likes of Microsoft and ChatGPT maker OpenAI.
Baidu led several companies in rolling out AI chatbots on Thursday: SenseTime opened its service for registration, and two other firms – Baichuan Intelligent Technology and Zhipu AI – said their bots were online and open to the public.
“We are thrilled to share that ERNIE Bot is now fully open to the general public starting August 31,” Baidu said in a statement on Thursday.
“In addition to ERNIE Bot, Baidu is set to launch a suite of new AI-native apps that allow users to fully experience the four core abilities of generative AI: understanding, generation, reasoning, and memory.”
The chatbot was first released in March but its availability was limited.
- ‘Let’s talk about something else’ -
Timeline of key developments in large language models and artificial intelligence chatbots.
Chinese generative AI apps must “adhere to the core values of socialism” and refrain from threatening national security, according to the guidelines published this month.
When tested by AFP on Thursday, ERNIE Bot easily answered mundane questions such as “What is the capital of China?” and “Do you have any hobbies?”
But on sensitive topics such as China’s bloody clampdown on pro-democracy protesters at Beijing’s Tiananmen Square in 1989, it said: “Let’s change the topic and start again.”
Tiananmen is a particularly sensitive subject in China and online information and discussion about it is strictly censored.
When asked about Taiwan, a self-ruling island that China claims as its territory, ERNIE Bot offered a longer answer.
“Taiwan is part of the sacred territory of the People’s Republic of China,” it responded. “China’s sovereignty and territorial integrity cannot be violated or divided.”
Then, it said: “Let’s talk about something else.”
- Global AI race -
Chinese generative AI apps must 'adhere to the core values of socialism', according to guidelines published this month
By making ERNIE widely available, Baidu will be able to gain “massive” human feedback to improve the app at a swift pace, CEO Robin Li was quoted as saying in the statement.
Generative AI apps, including chatbots such as ERNIE, are trained on vast amounts of data as well as their interactions with users so they can answer questions, even complex ones, in human-like language.
The rapid success of US-based OpenAI’s ChatGPT – which is banned in China – sparked an international race to develop rival apps, including image and video generators, but also widespread alarm about the potential for abuse and disinformation.
Under Chinese regulations, AI developers must conduct security assessments and submit filings on their algorithms to the authorities if their software is judged to have an impact on “public opinion”, according to the rules.
Baidu CEO Robin Li unveiled Ernie Bot in March but it was only made fully available to the public in China on August 31
They are also required to label AI-generated content.
Baidu is one of China’s biggest tech companies, but has faced competition from other firms such as Tencent in various sectors.
In addition to AI, it has also looked to grow its cloud computing business and develop autonomous driving technology.
Baidu shares were up 3.2 percent in Hong Kong at around 0430 GMT on Thursday.