OpenAI’s ChatGPT has kicked off an artificial-intelligence arms race — and this race is global. In January, AI company Anthropic began quietly testing its ChatGPT rival named Claude. In February, Google unveiled its own AI-powered chatbot Bard. Although ChatGPT itself is blocked in China, its knockoffs have popped up all over Chinese social media platform WeChat. Several Chinese internet companies — from Alibaba to JD to iFlytek — have announced their own AI chatbot plans.
The most high-profile AI chatbot from China is internet company Baidu’s Ernie Bot, which it plans to integrate into its search engine, the same way that Microsoft has integrated ChatGPT into Bing. Most reporting, from CNBC to Barron’s, has framed Ernie as another Chinese “copycat” or “clone,” and a short-term tactic to boost its company’s search engine. Like so many other tech companies, Baidu appears to be glomming onto the ChatGPT-induced frenzy. Given that it has become lackluster in recent years, and less relevant to Chinese users than the more popular short-video platforms such as Douyin and Kuaishou, Baidu is in particular need of such a media blitz.
But this perspective — that Ernie is a copycat of ChatGPT — overlooks Baidu’s decade-long journey in building its own AI capabilities. Ernie did not pop up out of nowhere; it was born on PaddlePaddle, the homegrown deep-learning framework that Baidu had begun developing in 2013. It started around the same time as similar frameworks were being realized by its American counterparts, Google’s TensorFlow and Facebook’s PyTorch. Ernie relies on PaddlePaddle to train its large language model (LLM) — a necessary building block for any LLM-based AI application, like a chatbot. Since then, Baidu has kept pace with OpenAI: Ernie 3.0 was announced in December 2021, just over a year after GPT-3, ChatGPT’s predecessor, was released.
This view of Baidu as a follower of AI trends is also limited in its understanding of how generative AI applications can generate business value. If Ernie is to bolster Baidu’s growth prospects and make the company relevant again, this transformation would not come from its search engine. After all, Baidu’s search engine, artificially insulated from international competition thanks to China’s walled-off internet, already holds the largest market share in the country. Instead, Ernie will enable growth in two of its other product lines: its cloud computing platform and self-driving cars.
Search as a user behavior has been in decline, as recommendation-driven content apps like Douyin and TikTok have become more popular. As the dominant search engine in China, Baidu Search does not have much potential to grow from Ernie. (Bing, on the other hand, is a small player in all search markets, so Microsoft has a lot to gain from a ChatGPT boost.) In contrast, Baidu’s AI Cloud, not even in the top three in China’s cloud sector, would profit greatly — not only from those who want to use Ernie directly, but also from customers who want to use Baidu’s technology to build their own chatbots. Recent news that regulators may prohibit Chinese tech companies from offering products built on OpenAI’s ChatGPT may benefit Ernie, since it will have less competition if Baidu’s AI technology remains the only accepted alternative.
Similarly, Ernie would enhance Apollo, Baidu’s operating system for self-driving cars, which powers Baidu’s own autonomous vehicles and other cars that use the Apollo system. If Ernie becomes popular among Chinese users as a search tool and AI assistant, integrating Ernie into Apollo would make its self-driving interface more “sticky,” or more familiar and dependable for passengers who already use the chatbot for other purposes, like vacation planning or food delivery. This will help Baidu’s recently piloted robotaxis service generate more revenue, while also making Apollo more appealing to other carmakers. The more the Apollo software is used, the more services and applications — such as real-time navigation and software upgrades — will be run on Baidu AI Cloud, further benefiting the cloud platform.
Looking beyond Baidu, chat-based AI products could evolve differently in China’s tech landscape in other ways, too. It is possible, for example, that chatbots will be used more via voice than via text. Since WeChat popularized communicating through voice recordings, voice interactions in a messaging app have become much more common in China than in the U.S. Currently, most user interactions with ChatGPT are text-based. But in a more voice-driven Chinese tech ecosystem, chatbots could produce an entirely different product experience and technical capabilities (such as detecting regional accents). Voice-driven AI interactions could also be particularly senior-friendly, which would appeal to China’s rapidly aging population.
Baidu has all the pieces, and is as well-positioned as any company to capitalize on the ChatGPT frenzy. While regulations around generative AI remain uncertain and success is not guaranteed, Baidu has more than earned this opportunity.
