China’s rush to dominate AI has a twist: It depends on American technology

In November, a year after ChatGPT launched, a relatively unknown Chinese startup jumped to the top of a ranking judging the capabilities of open source artificial intelligence systems.

The Chinese company, 01.AI, was only eight months old, but it had deep-pocketed backers and a billion-dollar valuation and was founded by well-known investor and technologist, Kai-Fu Lee. In interviews, Lee presented his AI system as an alternative to options like Meta’s generative AI model, called LLaMA.

There was only one twist: part of the 01.AI system technology came of flame. Mr. Lee’s new company then built on Meta’s technology and trained his system with new data to make it more powerful.

The situation is emblematic of a reality that many in China openly admit. Even as the country races to develop generative AI, Chinese companies rely almost entirely on underlying systems from the United States. China is now behind the United States in generative AI by at least a year and may be falling even further behind, according to more than a dozen leading tech industry experts and engineers, setting the stage for a new phase in technology competition. ruthless between the two nations. which some have compared to a cold war.

“Chinese companies are under tremendous pressure to keep up with American innovations,” said Chris Nicholson, an investor at venture capital firm Page One Ventures who focuses on artificial intelligence technologies. The launch of ChatGPT was “yet another Sputnik moment that China felt it had to respond to.”

Jenny Xiao, a partner at Leonis Capital, an investment firm that focuses on AI-powered companies, said the AI ​​models that Chinese companies build from scratch are “not very good,” leading many Chinese companies to use often “tuned versions”. of Western models. She estimated that China was two or three years behind the United States in generative AI developments.

The fight for AI primacy has enormous implications. Advances in generative AI could tip the global technological balance of power, increasing people’s productivity, helping industries and leading to future innovations, even as nations struggle with the risks of the technology.

As Chinese companies play catch-up by turning to open-source artificial intelligence models from the United States, Washington finds itself in a difficult situation. Although the United States has tried to curb China’s advances by limiting the sale of microchips and curbing investment, it has not stopped the practice of openly releasing software to encourage its adoption.

For China, the new dependence on US artificial intelligence systems (mainly Meta’s LLaMA) has raised deeper questions about the country’s innovation model, which in recent decades surprised many by producing world-leading companies such as Alibaba and ByteDance despite Beijing’s authoritarian controls.

“When Chinese companies leverage American open source technologies to play catch-up, the questions become very complicated, wrapped up in questions of national security and geopolitics,” said Oren Etzioni, a University of Washington professor specializing in artificial intelligence and founder . from TrueMedia.org, a nonprofit organization that works to identify online misinformation in political campaigns.

In an emailed statement, Lee, founder of 01.AI, said his startup’s AI model was built at LLaMA “like most other AI companies,” adding that its use of hard-coded technologies open is standard practice. He said his company had trained its AI model from scratch, using its own data and algorithms. Those were “the main determinants” of the 01.AI model’s “excellent performance,” Lee said.

Meta pointed to comments from Nick Clegg, who leads global affairs, in which he said that openly sharing the company’s AI models helped spread its values ​​and standards and, in turn, helped secure American leadership.

(The New York Times has sued ChatGPT maker OpenAI and its partner, Microsoft, for copyright infringement of news content related to artificial intelligence systems.)

AI has long been a priority in China. After AI tool AlphaGo defeated two major players in the board game Go in 2016 and 2017, Chinese policymakers set out an ambitious plan to lead the world in technology by 2030. The government pledged billions to researchers and companies focused on artificial intelligence.

When OpenAI launched ChatGPT in November 2022, many Chinese companies were hamstrung by a regulatory crackdown from Beijing that discouraged experimentation without government approval. Chinese tech companies have also been burdened by censorship rules designed to manage public opinion and silence significant opposition to the Chinese Communist Party.

Chinese companies with the resources to build a generative AI model faced a dilemma. If they created a chatbot that said the wrong thing, its creators would pay the price. And no one could be sure what might come out of a chatbot’s digital mouth.

“It’s simply not possible to get rid of all the problematic ways these systems can express themselves,” said Andrew Ng, a computer science professor at Stanford and former executive at Baidu, the Chinese search giant.

Chinese tech giants were also grappling with new regulations dictating how AI models can be trained. The rules limited the data sets that could be used to train AI models and the applications that were acceptable, and also set requirements for registering AI models with the government.

“It’s harder and riskier to innovate in generative AI under the current regulatory regime, which remains a moving target,” said Kevin Xu, founder of US-based Interconnected Capital, a hedge fund that invests in AI companies. .

Technology investors in China have also pushed for quick results from AI, which has meant money has flowed toward easy-to-execute applications rather than more ambitious goals focused on fundamental research, said Yiran Chen, distinguished professor of electrical and computer engineering John Cocke. at Duke University. Up to 50 percent of China’s AI investment has gone into computer vision technology, needed for surveillance, rather than building basic models for generative AI, he said.

Now Baidu, Alibaba, dairy company Mengniu and tutoring company TAL Education have joined the generative AI race in China, prompting Chinese media to coin the phrase “the battle of the 100 models” to describe the frenzy.

Some have criticized wrestling as publicity stunts that add unnecessary competition. In a roundtable last year, Robin Li, CEO of Baidu, described having hundreds of basic AI models as wasteful.

“More resources should be allocated to applications in various industries, especially considering the limitations of our computing power,” he said.

Success has been difficult to achieve. When Baidu unveiled its chatbot, Ernie, in March, it was revealed that the “live” demo was pre-recorded. Baidu shares plunged 10 percent that day.

Despite the setback, Baidu remains one of China’s few major efforts to build a basic AI model from scratch. Others are led by Alibaba and Tencent, China’s tech giants, as well as a startup linked to Tsinghua University.

A Baidu spokesman declined to comment.

US restrictions on the sale of AI chips to China pose additional challenges, as many of those chips are needed to train generative AI models. Baidu and 01.AI, among others, have said they have stockpiled enough chips to sustain their operations for the foreseeable future.

There are some positives for China with AI, including in fields such as computer vision and autonomous vehicles. Some Chinese entrepreneurs are also looking to surpass the United States with advances in other parts of generative AI.

Wang Changhu, former director of ByteDance’s AI lab, founded a company called AIsphere in Beijing last year to spearhead what he saw as the next big frontier in technology: video generation. In November, the startup launched PixVerse, an AI-powered generator that can create videos from a text description.

“We are moving forward, building our models from scratch,” Wang said. “This gives us a significant advantage as true pioneers in the field of video generation.”

That advantage may have lasted only a few months. Last week, OpenAI introduced Sora, an artificial intelligence tool that turns a simple text message into videos that look like something out of a Hollywood movie. Sora instantly went viral.