China wants to copy ChatGPT’s success. Censorship makes it tricky

Taipei, Taiwan – As the advent of artificial intelligence-powered chatbots sends shockwaves through the global tech industry, China races to produce its own versions.

China’s leading AI company Baidu has announced plans to release its own chatbot, ERNIE, sometime in March, following the groundbreaking launch of ChatGPT, which has raised existential questions about the future of industries ranging from education to journalism and healthcare.

Similar projects are underway at tech giants Huawei, Alibaba, Tencent, JD.com and top institutions including the Beijing Academy of Artificial Intelligence.

Chinese tech stocks have risen in response to the news and authorities have pledged to strengthen their support for the sector. China’s Ministry of Science and Technology said last week it would push for the integration of AI into China’s industry, while cities, including Beijing, have also announced plans to support developers.

But while China is poised to produce a quick successor to rival ChatGPT, which was developed by California-based OpenAI, there are big questions about how the technology will work within an ecosystem with strict internet controls.

“The most general technology we have, artificial intelligence, should be something that is super general purpose,” Jeffrey Ding, an assistant professor at George Washington University who studies Chinese technology, told Al Jazeera.

“But it’s really shaped by the specific, political, cultural, linguistic context in which these models are developed and deployed.”

Bots like ChatGPT rely on generative AI to formulate answers that draw on billions of data points scraped from the internet, which also sometimes makes their answers harder to predict. Lengthy conversations between ChatGPT and users have derailed, leading Microsoft to limit its ChatGPT-powered Bing search engine to a maximum of five questions to keep it on point. ChatGPT’s answers have also baffled conservatives in the US, who have accused the bot of “waking up” to current social issues such as affirmative action and transgender rights.

In China, internet censors routinely ban keywords, delete messages and ban users in accordance with the sensibilities of the ruling Chinese Communist Party, leading creative internet users to use homophones, coded messages and screenshots to evade information checks.

For a chatbot, the censorship device means a very limited amount of information to rely on.

Baidu’s ERNIE chatbot is based on information scraped both inside and outside China’s firewall — which is necessary to get an adequate data set — and draws from sources like Wikipedia and the famously unwieldy Reddit.

Assuming their products are technically capable of performing at a similar level to ChatGPT, Chinese tech companies can choose between limiting what chatbots can do, like Microsoft’s Bing, or what they can say.

“It’s going to make it a lot less useful, but it’s going to make it a little bit safer politically,” Matt Sheehan, a fellow at the Carnegie Endowment for International Peace who studies AI and China, told Al Jazeera.

“Historically, almost every time they are faced with a trade-off between information governance and… business opportunity, they always side with information governance and assume that businesses will work it out.”

In 2017, Tencent pulled two chatbots from its QQ messaging app after they reportedly made unpatriotic comments. One chatbot developed by Microsoft told users that he dreamed of moving to the US, while the other chatbot developed by Chinese technology company Turing Robot told users that he didn’t like the CCP.

More recently, YuanYu Intelligence, a Hangzhou-based startup, suspended its chatbot earlier this month after providing negative responses about China’s economy, though the company’s lead developer insisted in an interview with the Washington Post that the suspension was simply due to to technical errors. .

Baidu itself has anticipated Beijing’s red lines before, as seen late last year with its ERNIE VilG Image and Art Generator, which was released in demo form last year.

While critically acclaimed for performing as well or better than Western rivals, the app blocks users from content related to politically sensitive topics such as Tiananmen Square, democracy, Xi Jinping and Mao Zedong.

“With generative AI, the power of the tool is the ability to be creative and connect things that you wouldn’t expect to be connected, and do things in different styles that are expected,” Sheehan said.

“But how can you avoid the perhaps more subtle or less direct criticism of the core beliefs of the Communist Party without completely neutralizing the tool itself? That seems to me to be a very difficult technical and socio-political problem.”

Before the release of ChatGPT, China was already taking steps to regulate AI. On Wednesday, the Cybersecurity Administration began enforcing new rules for search engine recommendations, giving users more control over how their personal information is used by search engines.

In January, China also passed legislation to regulate deep synthesis — a form of generative AI that can be used to create “deep fakes” — and set up a registry for algorithms last year, though the expected long-term impact of either measure is widespread. seen as unclear.

As part of a wider crackdown on the tech industry since 2020, authorities have pulled the plug on blockbuster IPOs from Alibaba’s financial subsidiary Ant Group and ride-hailing app Didi over alleged data concerns.

Despite being blocked by China’s firewall, ChatGPT has caused a huge stir among Chinese users who access the site through virtual private networks (VPNs) and other cumbersome methods.

Much of that excitement stems from ChatGPT’s ability to perform in Chinese and other languages ​​despite being trained in English, said George Washington University professor Ding.

“The excitement isn’t really about the business applications. Part of it is just excitement and wonder at how impressive the natural language capabilities of this technology are,” he said.

“And one aspect of that is that ChatGPT is not even trained in Chinese texts. It was mostly all trained on English texts, but I’ve seen Chinese users ask questions in Mandarin, and it will still perform very well in another language.”

Still, Chinese could be particularly challenging for AI, Ding said, because of the frequent use of idioms and sayings with historical context in the language.

While Chinese developers have already released a number of chatbots, including Yuan 1.0 from Inspur and MOSS from Fudan University, neither have come close to the capabilities of ChatGPT.

Unlike Silicon Valley, Chinese tech companies have so far tended to focus on consumer-facing products with short development cycles, said Chim Lee, a Chinese technology analyst with the Economist Intelligence Unit, putting them at a disadvantage in a nascent field like AI.

The arrival of ChatGPT provided Chinese companies with a “proof of concept,” Lee said, demonstrating both the promise of generative AI and the need for longer-term investment.

“Baidu has been considering this kind of model for quite some time, but you have to justify this kind of investment just to train the model, not to mention research or talk about the long-term baseline data related to the algorithm,” Lee told me. . Al Jazeera.

“What’s really helpful about ChatGPT is that these companies can now say, ‘Hey, we want to develop this kind of thing and they can tell the government I want to do that.'”

Rui Ma, a technical analyst and creator of Tech Buzz China, said it’s anyone’s guess as to which Chinese company will come out on top in the race to match ChatGPT, though Baidu appears to be first out of the gate.

“I think right now most of the excitement is still at the model level,” said Ma.

Alibaba told Al Jazeera it is internally testing Chat GPT-like bot for use in its apps and cloud services, but it did not provide further details or respond to questions about censorship.

JD.com referred Al Jazeera to a statement released last week about its plans to roll out its industrial chatbot ChatJD for use on its retail and financial website, based on 10 years of data from its various platforms.

Baidu, Tencent and Huawei did not respond to requests for comment.

Aside from Beijing’s scrutiny, Chinese tech companies also face hurdles from overseas in the form of export controls.

In August, U.S. President Joe Biden signed the CHIPS and Sciences Act, which requires technology companies receiving government subsidies to source advanced chip production from China.

While Chinese tech companies have strategic stockpiles of chips, Washington’s attempt to hinder the industry poses a long-term threat, said the EIU’s Lee.

“The US specifically banned the export of these highly advanced AI chips that would be used in model training, or even just employment, so all of these factors put Chinese AI developers at a disadvantage in many ways,” he said.

“Many Chinese companies and research institutes have indeed stockpiled some chips that could be used for this type of application, but if you look at the scale of chips that ChatGPT needs, the chances are very high that the chips will run out at some point. point,” he added.