Fei-Fei Li advises leaders to focus on science in AI debate

Fei-Fei Li has a message for the world’s most powerful decision-makers: put down the sci-fi novels and pick up a research paper.

The Stanford professor and co-director of the Stanford Institute for Human-Centered AI (HAI) used the lead-up to the AI Action Summit in Paris to make a pointed case for evidence-based AI governance. In a landscape where policy conversations frequently veer into existential dread about sentient machines, Li argued that effective regulation needs to be anchored in what AI actually does today, not what a screenwriter imagines it might do in 2045.

Three principles for AI policy that doesn’t embarrass itself

Li laid out a clear framework built on three pillars. First, AI regulation should be grounded in scientific evidence rather than fictional projections about machine consciousness. Second, rules need to be pragmatic enough to reduce unintended consequences while still leaving room for innovation. Third, policy should support the entire AI ecosystem, from well-funded corporate labs down to open-source projects and resource-constrained academic researchers.

Her concern about science fiction framing isn’t abstract. When lawmakers spend hearings asking whether an AI might “wake up” or develop feelings, they’re burning time that could be spent on the far less cinematic but far more urgent questions of bias in hiring algorithms, misinformation generated by language models, and the concentration of computational resources among a few corporations.

The woman behind the framework

Li’s authority on this subject is hard to overstate. Known widely as the “Godmother of AI,” she created the ImageNet dataset, a massive collection of labeled images that became the foundation for modern computer vision breakthroughs.

Li has also moved into the entrepreneurial arena. She co-founded World Labs, a company focused on advancing spatial intelligence, the ability of AI systems to understand and interact with three-dimensional environments. The company launched with $230 million in funding.

Why the science fiction problem is a real problem

Chatbots don’t have desires. They predict the next word in a sequence based on statistical patterns in training data. That’s impressive engineering, but it’s not sentience, and pretending otherwise leads to regulations that solve imaginary problems while ignoring real ones.

The open-source dimension of her argument is particularly timely. Li’s position suggests that locking down the technology too aggressively could stifle the academic research that drives genuine understanding of AI’s capabilities and limitations.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.