This article profiles Song-Chun Zhu, a prominent artificial intelligence scientist, and details his unexpected return to China in 2020 after nearly three decades in the United States. It explores the intellectual and geopolitical factors driving his move, including his growing disillusionment with the "big data, small task" paradigm of large language models prevalent in Silicon Valley, and the increasingly hostile environment for Chinese-born scientists in the US under the "China Initiative." Now leading China's Beijing Institute for General Artificial Intelligence (BigAI), Zhu advocates for a "small data, big task" approach focused on cognitive architectures and commonsense reasoning for achieving Artificial General Intelligence (AGI), positioning him as a key figure in China's AI strategy. His journey highlights the broader geopolitical contest for AI supremacy and the shifting global landscape for scientific talent.
Summarizing "I have to do it’: Why one of the world’s most brilliant AI scientists left the US for China"
0:00 Early Life and Driving Ambition: Song-Chun Zhu grew up in rural China during the Cultural Revolution, witnessing poverty and death, which instilled in him a resolve to make his life count and avoid an unremarkable fate.
1:00 Distinguished US Academic Career: After pursuing a PhD in computer science at Harvard in 1992, Zhu became a leading figure at UCLA, directing one of the world's most prolific AI research centers, winning major awards, and securing prestigious grants for his pioneering work in pattern recognition.
3:00 Intellectual Divergence from Mainstream AI: From around 2009, Zhu became increasingly disillusioned with the data-driven "big data, small task" approach of neural networks (like ChatGPT), which he believed lacked deeper cognition and reasoning necessary for true intelligence. He advocated for a "small data, big task" approach emphasizing cognitive architecture.
6:00 US-China Tensions and Hostile Environment: In the late 2010s, the US "China Initiative" intensified scrutiny on Chinese-born scientists, leading to funding difficulties and pressure for Zhu and his students, contributing to a feeling of being "caught in the middle."
8:00 Decision to Return to China (August 2020): After 28 years, Zhu left the US, citing his intellectual estrangement from the dominant AI paradigm, the "hothouse" of American politics, and his younger daughter Zhu Yi's move to China to compete in the 2022 Winter Olympics.
10:00 Key Role in China's AI Strategy: Upon returning, Zhu secured full professorships at top Beijing universities and became director of the state-sponsored Beijing Institute for General Artificial Intelligence (BigAI). He also joined China's top political advisory body, proposing AI be treated with the same strategic urgency as a nuclear weapons program.
12:00 BigAI's AGI Philosophy and "TongTong": Zhu's institute focuses on achieving AGI through qualities like resourcefulness, social/physical intuition, and cause/effect understanding, rather than purely data-driven methods. They debuted "TongTong," a virtual humanoid child designed to mimic cognitive and emotional capacities, demonstrating commonsense problem-solving.
14:00 Critique of Silicon Valley's Approach: Zhu explicitly contrasts his philosophy with the "Silicon Valley narrative" that AGI can be achieved simply with more data and computing power, asserting that this misguided approach holds back true AI progress.
16:00 Geopolitical Implications and Talent Exodus: Zhu's story exemplifies a broader trend of scientific talent potentially leaving the US due to restrictive policies, with Chinese universities actively recruiting. He notes that his PhD students in Beijing are "on a par with those at MIT and Stanford."
17:00 Underlying Motivation: Despite adopting the Communist party's vernacular to secure resources, Zhu maintains his primary motivation is a lifelong intellectual quest for a unified theory of AI, aiming to carry on the legacy of the "Harvard and Brown school" of computer science.