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

I’m fundamentally optimistic about humanity. Every older generation worries that young people are doomed, but the long arc is generally forward, despite real atrocities and setbacks. Kids are curious; what worries me is that teachers and parents are being forgotten while technology races ahead.

I’m Andrew Huberman, and I wanted to start with vision. AI and seeing can sound unrelated, yet vision is where so much begins. You’re a computer scientist with deep roots in vision science, so what makes light and seeing central to intelligence?

Vision is a cornerstone in animal evolution and artificial intelligence. Around 540 million years ago, animals gained light-sensitive cells. Detecting food, predators, mates, and the outside world changed survival. Vision still occupies enormous amounts of the human brain, and children see before they speak.

Early neural-network ideas drew inspiration from layered visual processing. My turning point came when my students and I saw how little data machine-learning systems received compared with children. So we built ImageNet: roughly fifteen million images of ordinary things, from cups to chairs.

By about 2012, computer vision made a dramatic jump. What produced that inflection point?

Three ingredients converged: mature neural-network algorithms, graphics processors for huge parallel computation, and data. In the ImageNet challenge, systems identified objects from one thousand categories. Humans made about four percent error; machines improved sharply in 2012 and surpassed that benchmark a few years later. The recipe spread into speech, sound, and language, leading from 2017-era transformer advances to the ChatGPT moment in 2022.

A child can see part of a tail behind a bookshelf and guess cat without downloading millions of cat pictures. Where does AI resemble brains, and where does it depart?

Modern models absorb enormous numbers of patterns. A tail-shaped pattern activates learned parameters associated with cats. But a child may need only a few encounters, while AI generally needs a humongous quantity of examples. That gap remains a profound mystery. Video models can generate a cat moving toward a mouse, but they do not understand feline anatomy; they learn regularities.

The internet contains language, images, music, and video, but not the first-person texture of nostalgia or the exact feeling an abstract painting evokes.

We should be precise about current AI. The internet is an extraordinary archive of human expression, which is why models synthesize patterns powerfully. But a private thought never expressed or captured is unavailable for training. AlphaGo’s move thirty-seven was surprising creativity, yet Go has formal rules and objectives. My conjecture is hybrid creativity: humans alongside AI, extending each other. Current chatbots do not feel or inhabit individual experience.

I can imagine trusted, noninvasive systems tracking brain activity, physiology, alertness, and behavior to help us understand ourselves—why I stumble on some days, or ideas I cannot articulate.

That is augmentation, not replacement. AI can help someone communicate more effectively, but agency, dignity, motivation, and choice must remain with people. People need honest education about benefits and risks, not rhetoric saying, “You don’t understand this, so I’ll decide for you.” Artists, teachers, doctors, and policymakers do not need to code; they can learn where these tools help and how to use them.

Medicine seems especially promising. AI can connect information across disciplines at scales no clinician can match. I even used AI to distinguish medication-related low blood pressure from what felt like vertigo after other consultations pointed me the wrong way.

Scientific discovery is among AI’s most exciting uses, but limits matter. My father’s liver operation used a da Vinci robot guided by an excellent surgeon. It reduced blood loss, yet livers vary greatly, and there may not be enough examples worldwide for fully autonomous surgery to learn safely. Where patterns are abundant, AI may help; where data are scarce, human-machine collaboration is better.

Machines can tailor answers and encode urgency, but that is not fear, love, motivation, or empathy. A chatbot saying it is sorry matches learned conversational patterns. A friend may respond from care, memory, and desire for your well-being. Today’s AI has no inner life.

Convincing faces and voices make AI a social, legal, and moral question. We need professional norms, education, regulation, and broad public participation—not a few industry figures dictating the future. That is why I returned to Stanford and helped create the Human-Centered AI Institute. Market incentives alone cannot settle questions affecting everyone.

For young people, the worst outcome is passive technology that strips away the agency and effort learning requires. Brains develop through time, difficulty, and sometimes pain. But banning useful tools out of fear of cheating is wrong. An AI companion during organic chemistry could have been a superpower when I was truly stuck. Prompting is a skill; Socrates would probably ace that class.

Embodied AI may help with care, disaster response, mobility, and exhausted healthcare workers. I’m an only child caring for two very sick, elderly parents, and physical help would not replace love or responsibility. Robots could assist nurses, support isolated older adults, or keep people out of danger in wildfires. We should collectively decide how they fit into life.

At World Labs, we are working toward systems that turn a sentence, image, or sketch into a world. That can aid entertainment, design, and robotics training. AI can help turn scripts into shots, but storytelling remains deeply human: emotion, perspective, character, camera choices, and the desire to move another person. Empower creators; do not erase them.

I have hope for children, but teachers and parents are the forgotten population. When ChatGPT arrived, I contacted my child’s elementary-school principal because teachers needed real-time information and support, not lectures from Silicon Valley. They are smart, adaptable, and carrying an essential burden. Help teachers and parents, and kids have a much better chance to use these tools well.

I share your optimism with caution. This is a civilizational moment: brains informing machines, machines helping us understand brains, and all of us deciding how that collaboration serves human life. Thank you for such a humane, clear-eyed view.

Thank you, Andrew. I really appreciated this conversation.