What will China’s ChatGPT look like?

When the Microsoft-backed start-up OpenAI launched ChatGPT at the end of November, there was a swift outpouring of excitement among users, even in China, where the product was not officially available.

But the response in China was also accompanied by an undercurrent of frustration, as people began to wonder why China did not beat the US to making a chatbot capable of such feats as generating essays and creative writing in various styles, writing code, answering research questions and summarising content. Some were quick to blame a risk-averse environment that prioritises monetisation above innovation.

In the following weeks, though, people would see first-hand the kinds of challenges that a product like ChatGPT faces in China.

When third-party applications on WeChat allowed people in China to experience ChatGPT without circumvention tools like virtual private networks, the country’s largest social media network sprang into action – by banning the services.

Authorities have also warned about how the technology can be used to commit crimes and spread rumours. Minister of Science and Technology Wang Zhigang went as far as suggesting ChatGPT presents an “ethical” problem for China, noting that it can give “an answer based on values, not facts”.

That ChatGPT is designed to riff on material it has absorbed from other sources and often spits out fallacious information has concerned teachers and researchers as much as politicians. Some schools, like University of Hong Kong and Hong Kong Baptist University, have banned students from using the tool for coursework.

But whereas professionals can use a tool like ChatGPT while appreciating its limitations, for Beijing it is a direct challenge to its tightly controlled internet environment. Fang Bingxing, known as the father of the Great Firewall, warned of ChatGPT’s ability to create an “information cocoon” in which “people’s perspectives can be manipulated”.

The ideal scenario for the ruling Communist Party is for China to have its own home-grown version of ChatGPT. It has been working hard towards that end, fielding ideas at the ‘two sessions’ parliamentary gathering last month and mustering the might of its tech giants.

The country’s best hope at catching up to ChatGPT so far appears to be Baidu’s Ernie Bot. After a bumpy launch with no live preview, Ernie was found by early users to have better English-to-Chinese translations than ChatGPT, although the Baidu bot struggled more in other areas.

Baidu so far appears to be moving cautiously in pushing Ernie adoption. After initially announcing that the bot would be integrated with its cloud services to help enterprises, it later cancelled a related launch event.

Baidu is hardly the only game in town, though. Since ChatGPT created a torrent of publicity, every big name in tech has been eager to tout its AI bona fides.

Alibaba has been testing its own ChatGPT rival. WeChat owner Tencent has also said it is exploring large language models, the technology behind ChatGPT.

Tencent rival NetEase and game developer miHoYo are exploring the use of conversational AI in video games. The CEO of on-demand delivery firm Meituan has also invested in an AI start-up since hype around ChatGPT exploded.

Even companies sanctioned by the US, which would seem to be at a disadvantage, are joining the generative AI race. iFlyTek has been investing in large language models and the company’s founder said China needs a strategy for catching up with and overtaking the US.
SenseTime, another sanctioned Chinese AI firm, is expected to announce its own ChatGPT-like product next week.

Still, developing this kind of AI tech is “difficult to achieve”, as science and technology minister Wang put it. Some of the challenges China faces are external. In addition to companies being sanctioned by Washington, the US has been making it harder for Chinese companies to acquire the advanced graphics processing units used to train and power AI models.

But other challenges include content moderation policies that the Chinese government has itself put in place. Experts have noted the challenges that censorship poses to producing a ChatGPT rival.

China certainly has the talent base needed to be an AI powerhouse. It is a leader in some areas of AI and follows closely behind the US in others. But the machine learning used to train generative AI can lead to unpredictable results, which is anathema to Communist Party.
Moreover, training AI in censored media environments limits the inputs, potentially resulting in less useful outputs.

There is no putting the genie back in the bottle, though. Companies, schools and governments alike are all trying to come to grips with what it means to live with generative AI. While Beijing has pledged support for developing a ChatGPT alternative, it may take some time yet to understand what Chinese generative AI really looks like, and whether restrictions on content are permissive enough for such tech to be as transformative for China as some expect OpenAI’s tech to be elsewhere in the world. South China Morning Post