The essence of evolution is large-scale data training—and in the digital era, AI can "run faster than nature" by leveraging massive datasets, Fei-Fei Li, co-founder of World Labs and a member of the U.S. National Academy of Engineering, said on Sunday.
She shared her insights at a discussion themed "On the Eve of the Explosion of Spatial Intelligence" during the 2025 T-EDGE Conference with Jany Hejuan Zhao, founder and chief executive of NextFin.AI and chair and chief executive of TMTPost Group.
Li emphasized that the essence of evolution is comparable to large-scale data training. "The long process of natural evolution is, in effect, a period of big data training," she said. "In the digital era, we no longer need to wait billions of years to collect data. We can gather it on a massive scale, and the principle is the same—it's just executed differently than in nature."
"When you process data at scale, it's as if you're seeing the equivalent of millions of years of evolution in one go. It's not a comparison you can make directly," Li added.
Li recalled that while ImageNet may seem modest today, at the time it was the largest dataset of its kind and a critical step forward for computer vision.
Reflecting on her own research approach, Li pointed out the importance of hypothesis-driven work. "Believing in your hypothesis isn't unusual in science," she said. "You think deeply about your assumptions, test them, and learn. Some will be wrong—but the process is what matters."
She described AI as fundamentally about generalization, achieved through the combination of algorithms and data. "If the algorithm is too complex and data is limited, you overfit. If data is abundant but the algorithm is inadequate, you still overfit. There's a precise mathematical relationship between the two."
Li traced the evolution of AI development from ImageNet's vast visual datasets to the rise of natural language models powered by massive internet data. Video modeling and autonomous driving advancements, she noted, are similarly driven by large-scale datasets. "It's never been about leaving algorithms behind," she said. "Data and algorithms have always been inseparable."
Li also observed that while algorithms often receive more attention in public discourse, professionals in AI research and industry recognize that data is equally important. "Data is a science in itself," she said.
