Baidu, China’s largest search engine provider and robotaxi developer, is apparently working on its own ChatGPT counterpart. The news, first reported by Bloomberg and The Wall Street Journalcaused Baidu’s share price to rise on Monday, reaching its highest since September.
A Baidu spokesperson declined to comment on the reports. But it wouldn’t be surprising if Baidu, which bills itself as China’s pioneer of artificial intelligence, goes a step further to build the Chinese equivalent of the most powerful chatbot to date. The question is how big the difference can be with the tool and where the limitations lie.
A driving force behind China’s technological development in recent years has been the rise of digital sovereignty, which refers to a country’s ability to create its own “digital destinationand can include autonomy in critical software and hardware in the AI supply chain. Episodes of US export bans on China have prompted Beijing to further push for tech independence in areas ranging from semiconductors to fundamental research to AI.
As OpenAI’s ChatGPT demonstrates the potential to disrupt industries from education and news to the service industry, Chinese technology leaders and policy makers are likely wondering how AI can also be used to boost productivity at home. Of course, China wants its own ChatGPTs, not only to gain control over how data flows through such tools, but also to create AI products that better understand local culture and politics.
According to The Wall Street Journal, Baidu’s conversational robot will make its debut in March and will be integrated into the company’s search engine first. That suggests that the chatbot will mainly generate results in Chinese. Nevertheless, the deep learning model has been trained on both Chinese and English data sources, including collected information Outside the Great Firewall, the country’s extensive internet censorship infrastructure.
That’s where things get interesting. Like all other information channels in China, the Baidu chatbot will undoubtedly be subject to local regulations and censorship rules. As we wrote earlier, the company’s text-to-image application, ERNIE-VilG, already rejects politically sensitive requests. But conversational AI deals with much more complex questions than image generators – how will Baidu walk the line between censorship restriction and leaving enough freedom and creativity to its bone?
Also important to machine learning performance are the underlying algorithms. According to The Wall Street Journal, Baidu has adapted a “core breakthrough” that Google developed and open sourced in 2017, an algorithm that also powered ChatGPT. Most likely, however, there are other important bits of proprietary algorithms that Baidu has acquired or developed to form the backbone of its chatbot.
Hardware plays another important role in training large-scale neural networks. US chip sanctions against China threaten China’s AI industry as companies lose access to advanced semiconductors that power supercomputers and large data centers.
However, Baidu believes the chip ban has a “limited” effect on its AI business, as we reported. In the short term, the company “already has enough stock [chips] in hand.” Looking ahead, Baidu is counting on its Kunlun AI chip developed in-house to drive high-performance computing. Alternatively, it could work on increasing the efficiency of its algorithms to take some work off the hardware.