‘I have to do it’: Leading AI scientist Song-Chun Zhu returns to China after decades in US

Born in 1968 in Ezhou, Hubei Province, Zhu grew up in his father’s small grocery store surrounded by stories of poverty and struggle. "That was really tough," he told The Guardian. "People were so poor."

When he once asked why his family genealogy recorded only dates of birth and death but not life stories, the keeper told him there was "nothing worth recording" because his ancestors were peasants. This response deeply disturbed Zhu and instilled in him a resolve to shape his destiny differently.

He excelled as a top student at his local high school. His academic success led him to the University of Science and Technology of China in Hefei, where he majored in computer science. In the mid-1980s, as academic exchanges between China and the U.S. expanded, some of his professors traveled to the U.S. and returned with scientific texts for translation. "At the time we saw America as a beacon, a cathedral of science," he said.

Among those books was "Vision" by British neuroscientist David Marr, who reimagined human sight as a mathematical process. Zhu was captivated by the idea that machines might one day "see" the world as humans do, sparking his dream of mapping intelligence, how humans think, reason, and exercise moral judgment, with the precision of physics.

Song-Chun Zhu, Founding Director of the Beijing Institute for General Artificial Intelligence. Photo courtesy of Song-Chun Zhu's website

In 1991 Zhu graduated but could not afford the application fees for U.S. universities. On his professor’s advice, he applied again the following year to Ivy League institutions that might overlook the unpaid fees. A few months later he was astonished to receive a letter from Harvard University, offering him a full PhD fellowship in computer science. "It changed my life," Zhu said.

After earning his doctorate in 1996 he joined Stanford University and later the University of California, Los Angeles (UCLA), in 2002, following a brief stint at Ohio State University, according to the South China Morning Post newspaper.

While lecturing at Stanford, Zhu witnessed the San Francisco Bay Area at the peak of the dot-com boom. Yahoo had just gone public, and venture capitalists swarmed the campus. At around the same time, two PhD students, Larry Page and Sergey Brin, created a new search engine called Google. As students flocked to web development courses, Zhu’s theoretical classes on pattern recognition struggled to attract attention.

Zhu spent 18 years at UCLA, becoming a leading figure in computer vision and statistical learning and winning the Marr Prize in 2003. However, as machine learning shifted toward data-heavy models, he grew increasingly skeptical of the field’s direction.

In 2004 he co-founded the Lotus Hill Institute in his hometown in China to construct the large-scale datasets needed to improve and test AI systems. But by 2010 he had shut it down, convinced that massive datasets and machine learning alone could not create true intelligence. "One of the major Chinese philosophical schools, the Yangmingism or the ‘Teachings of the Heart’, argues ‘the reality we see comes from how our minds perceive’,’" Zhu once told Science magazine. He began developing agents with "cognitive architectures" that could reason, plan and evolve using minimal data.

His timing coincided with the rise of ImageNet, the large-scale dataset developed by Fei-Fei Li, who attended a Lotus Hill workshop and later called Zhu one of her influences. In 2012 Geoffrey Hinton’s neural network model triumphed in the ImageNet competition, triggering the deep learning revolution. "Just as I turned my back to big data, it exploded," Zhu wrote to his mentor David Mumford.

Zhu maintained that true intelligence lies in the ability to reason toward a goal with minimal inputs, a philosophy he called "small data, big task." He said today's large language models like ChatGPT follow a "big data, small task" approach. "The difference between AGI and current LLM-based AI is just like the difference between a crow and a parrot." While parrots can mimic many words, crows can achieve their goals autonomously in the real world, he said.

By 2020 Zhu felt increasingly alienated in western academia as tensions between the U.S. and China escalated. According to a 2020 study by the U.S. think tank MarcoPolo, China was the largest global source of top-tier AI talent, but 88% of Chinese AI researchers working in the U.S. chose to stay there. As the U.S.–China tech war intensified, Washington increased scrutiny of Chinese researchers and tightened visa restrictions, casting uncertainty on global talent flows.

Zhu also faced growing challenges securing funding after his past Huawei grant drew suspicion. U.S. magazine Newsweek later reported that U.S. federal agencies had provided at least US$30 million to support Zhu’s research, including defense-oriented grants in 2021. A personal factor also influenced his decision to move to China: his daughter, Zhu Yi, a figure skater, had been selected to represent that country at the 2022 Beijing Winter Olympics.

In August 2020, after nearly 28 years in the U.S., Zhu boarded a one-way flight to Beijing. His long-time friend and Harvard classmate Mark Nitzberg recalled asking: "Are you sure you want to do this?"

Zhu replied, "They are giving me resources that I could never get in the United States. If I want to make this system that I have in my mind, then this is a once in a lifetime opportunity. I have to do it."

TongTong, the world's first artificial general intelligence agent developed by Beijing Institute for General Artificial Intelligence. Photo from Facebook

Upon returning, he accepted professorships at Peking University and Tsinghua University, and Beijing’s municipal government agreed to fund his Beijing Institute for General Artificial Intelligence. In a proposal submitted to the Chinese People’s Political Consultative Conference, Zhu called advanced AI "the strategic commanding heights of international scientific and technological competition in the next 10 to 20 years," comparing its impact to "the atomic bomb in the information technology field," Newsweek reported.

Chinese media hailed Zhu as a patriotic returnee contributing to the nation’s AI ambitions. While some collaborators feared his research would lose independence in China, Zhu says he feels more focused and supported in Beijing.

Sources close to him told The Guardian that in the five years since his move he has received several hundred million dollars in research funding. When asked whether it matters which country wins the AI race, Zhu says: "Do I want the Silicon Valley people to win? Probably not." He hopes it is the most ethical version of AI that will win.

"Over the last 30 years I’ve been focused on one thing. It’s the unified theory of AI. To build understanding. That’s my only drive."