Using AI to Increase Your Intelligence & Enrich Humanity | Dr. Fei-Fei Li

Full analysis

Stanford AI pioneer Dr. Fei-Fei Li joins Andrew Huberman for a wide-ranging conversation about what AI can and cannot do compared to human cognition, how AI is transforming healthcare and scientific discovery, and why the public conversation about AI is currently failing teachers, parents, and students. Optimistic but clear-eyed, Li argues AI should augment human agency — not replace it.

What was covered

• The origins of modern AI in vision science. Li describes how her ImageNet project (15 million labeled images, launched around 2006 at Princeton) provided the large-scale training data that, combined with neural network algorithms and GPU computing, triggered the 2012 inflection point now recognized as the start of the modern AI era. By roughly 2016, machines surpassed humans at naming one of a thousand object categories, with humans registering about a 4% error rate on that benchmark.

• How AI learns versus how children learn. Today's large language and vision models learn from enormous internet-sourced datasets; a child learns a cat from perhaps 3–10 real encounters but generalizes through a different, still poorly understood biological pathway. Li is direct: these are fundamentally different processes, and conflating them misleads the public.

• What AI still cannot access. Highly personalized, internal human states — a memory triggered by a cup, a felt but unspoken intuition, a creative insight that never got typed or photographed — have never been uploaded to the internet and therefore cannot be learned by current AI. Li calls this the irreducible gap.

• AI in healthcare and scientific discovery. Li describes her father's liver surgery performed by a Da Vinci robot guided by a human surgeon (he lost roughly 10 times less blood than typical). She explains why fully autonomous AI surgery for a complex, highly variable organ like the liver remains premature — insufficient training data. By contrast, common clinical scenarios like distinguishing vertigo from low blood pressure are well-represented in the data, which is why Huberman found AI diagnostically useful in his own case.

• Children, teachers, and the forgotten population. Li says policymakers, investors, and tech builders talk past teachers and parents entirely. Within days of ChatGPT's November 2022 launch, she emailed her child's elementary school principal to offer a guest lecture. Her concern: doom-or-utopia rhetoric is useless to a classroom teacher trying to decide how to handle AI-assisted homework.

• World Labs, her startup. Co-founded early 2024, World Labs focuses on "spatial intelligence" — building AI models that generate and manipulate 3D and 4D environments, useful for robotics training, architecture, healthcare simulation, and entertainment. She describes it as her life's work connecting computer vision to the physical world.

Notable claims & predictions

• Li on AI's hard limit: "When you cannot even access that [internal state], neither a different human nor a machine can do anything about it because there's no access to that." The implication: AI creativity and empathy are pattern-matching on available data, not felt experience.

• Li on robots and elder care: "I'm a single grown-up child taking care of two very advanced-aged and very sick parents... The amount of work I do is incredible. I would love to have help. It doesn't take away family's responsibility. It doesn't take away love." She sees domestic and caregiving robots as one of the most urgent near-term applications — but places full deployment on a 30-year horizon, not 12 months.

• Li on AI replacing doctors: "I won't say AI is better than all doctors, but AI was able to disambiguate vertigo from low blood pressure for me [Huberman] a few months back. And one of the people who got it wrong is an ENT who works on the vestibular system." She frames this not as a replacement argument but as a case for augmentation — especially for people without immediate access to a physician.

• Li on the student agency risk: "The absolute bad outcome is that our young generation, their agency and human-level motivation of learning and living is taken away by tools... But the other kind of danger is in the name of agency and motivation, the tools are denied to our students." She explicitly rejects both extremes.

• Li on prompting as a core skill: "Who is humanity's best prompter? Socrates — because that is the method of prompting. We should go back to teaching kids that." She advocates K–12 curricula that teach prompting as a foundational literacy.

Fact check

• Li's claim that ImageNet contains 15 million images. This figure is consistent with how ImageNet has been publicly described; the project's training data is well-documented. No issue.

• Li's claim that the transformer paper was published "around 2016, 2017." The landmark "Attention Is All You Need" paper (Vaswani et al.) was published in 2017. Accurate.

• Li's claim that ChatGPT launched in November 2022. Correct — OpenAI released ChatGPT publicly in late November 2022.

• Huberman's claim about the action potential shape paper in Nature. He describes a Nature paper showing action potential waveforms can vary substantially within a single neuron, published roughly 12 years before the recording. This is consistent with real published research on axonal action potential modulation, though Huberman presents it somewhat loosely ("everyone saw it and then no one wanted to deal with it"). The underlying science is real; his framing of scientific community dismissal is an anecdote, not a documented fact.

• Li's description of the Da Vinci robot surgery and "10x less blood loss." This is a personal family account, not a cited clinical statistic. Laparoscopic and robotic-assisted surgery does generally reduce blood loss compared to open surgery, but "10 times less" is a single-patient claim with no cited study behind it. Treat it as anecdote, not general medical fact.

• Wealthfront ad claim: 4.05% APY for three months. The transcript's own fine print states this boosted rate is variable, subject to change, applies only to new deposits up to $150,000, lasts three months, and requires eligibility. The base rate as of January 30, 2026 was 3.30%. The ad lead figure of 4.05% is accurate under specific conditions; the conditions are buried. Andrew Huberman receives cash compensation from Wealthfront for this testimonial — a disclosed conflict of interest that the episode's own footnote names.

Why this matters for you

• If you or a parent uses AI for a health question, this episode offers a useful reality check: AI performs best when the clinical scenario is common and well-documented in the literature, and worst when data is sparse or the situation is highly individual. Li's framing — "when patterns are not abundant, we have to be careful" — is a practical rule to apply before acting on AI medical advice.

• If you are a caregiver, Li's frank account of managing two elderly, non-English-speaking parents while running a research lab and a startup names a caregiving burden that is real and widespread. Her argument that domestic and elder-care robots could relieve physical labor without replacing emotional connection is worth tracking as this technology matures — though she places meaningful deployment 30 or more years out.

• If you have grandchildren or children in school, Li's specific concern — that teachers are being ignored in the AI conversation while students are already using these tools — suggests a practical step: ask your child's or grandchild's school what, if any, AI literacy curriculum is in place, and whether teachers have received training rather than just warnings.

• Interesting but nothing to act on this week: The broader philosophical discussion of AI consciousness, creativity, and intuition is intellectually engaging but has no near-term practical consequence for most readers. Treat it as context, not a prompt to change behavior.