The world is too loud. Read what matters.

Huberman Lab

Vision Is the Foundation of Intelligence; AI Should Augment Human Agency, Not Replace It

Vision is the foundation of intelligence; AI must augment human agency rather than take it over; and humanity's most precious cognition is precisely what was never uploaded to the internet.

VisionImageNetSpatial intelligenceAI educationEmbodied intelligenceScientific discovery
Fei-Fei Li argues the limits of AI through evolutionary history, ImageNet, and her own experience caring for her parents, without piling on jargon. For anyone who wants a practitioner's view rather than a grand narrative.

The argument · tap a timestamp to hear it

4:37

Intelligence starts with vision, not with language

Li places vision at the core of intelligence, for reasons that come from two levels. In evolutionary history, 540 million years ago marine animals such as trilobites acquired the ability to sense light for the first time, and the Cambrian explosion of life followed. In AI, vision has pushed modern artificial intelligence forward on both the algorithm side and the data side. She also points out that roughly half of the cortical activity in the human brain is tied to visual function. Understanding vision, then, is not a matter of understanding the eye; it is the starting point for understanding intelligence itself.

— Fei-Fei Li
9:24

The bottleneck in AI was never the algorithms, it was data

When Li was an assistant professor at Princeton in 2006, she realized that the real bottleneck for AI algorithms was the absence of data. Drawing inspiration from cognitive neuroscience, she decided to build a large-scale dataset, and the result was ImageNet: 15 million images, with the goal of teaching machines to recognize everyday objects. Before 2012, machines performed far worse than people; the human error rate on the 1,000-category recognition task was about 4%. In 2012 a neural network algorithm cut the error rate sharply, which marked the inflection point, and by around 2016 the algorithms surpassed humans. ImageNet brought algorithms, data and GPUs together, and became the turning point for modern AI.

— Fei-Fei Li
29:54

Sora knows nothing about a cat's muscles, and the cat still runs

Around 2023, several teams began adding video to their training data, and in January 2024 Sora was released: a user types text and gets a few seconds of video, a cat running toward a mouse, for instance. Li stresses that the algorithm does not understand the cat's muscular structure; it has simply learned plausible patterns of cat movement from a large volume of video. This resembles the way humans learn by observation, where biological knowledge does not come first, an enormous amount of data does. The significance of video generation is that it moves AI from static images toward prediction about a dynamic world.

— Fei-Fei Li
37:41

AI's limit is not compute, it is thoughts that were never digitized

Li agrees that AI cannot capture highly individual, unrecorded human cognitive behavior, such as the particular ideas Picasso had while he was creating. The internet is the largest collection of multimodal human behavior there is, but the thinking and feeling that was never uploaded is not in it, and no matter how powerful AI becomes it cannot get access to that. On this basis she holds that humans retain something distinctive, while AI is also able to combine existing information in creative ways. AI's boundary is not compute; it is whether or not human beings have digitized a given cognitive process.

— Fei-Fei Li
49:41

"You don't understand" is the most dangerous sentence in AI

Li stresses that AI should augment human agency rather than strip it away. She criticizes some current leaders in the AI field for talking about AI in condescending language that implies AI will replace human beings, rhetoric she considers neither healthy nor useful. She also argues that public rhetoric of the "you don't understand" variety is dangerous, and makes the case for public education so that people understand the benefits and the harms, with the final choice left in the public's hands. Genetic testing is one example: there was fear early on, but in the end people need to decide for themselves.

— Fei-Fei Li
1:00:45

AI will become research infrastructure, not just a generator of conversation

Li says AI will completely change the way scientific discovery works. Traditional science depends on the intelligence and the speed of the human brain, but AI can store vast amounts of information and synthesize knowledge across disciplines, which creates enormous opportunity in fields such as biomedicine. She uses vision and smell as examples to show that AI can cross the knowledge boundary of any single human expert and connect different fields together. That capability makes AI more than a producer of conversation; it becomes part of the infrastructure of research.

— Fei-Fei Li
1:29:33

The best prompter humanity ever produced was Socrates

Li worries that AI may strip the younger generation of its agency and its motivation to learn, through endless short-video scrolling, for example; but she is against overcorrecting and banning students from using AI altogether. With the right guidance, she believes AI can become a powerful learning tool that makes this generation smarter than we are. She even poses a quiz: who was humanity's best prompter? The answer is Socrates, whose method was itself the pursuit of truth through questions. Prompting is not a technical detail; it is part of public education.

— Fei-Fei Li
1:50:48

Language is not the destination; spatial intelligence is the next frontier

Li founded World Labs in early 2024, focused on spatial intelligence and physical intelligence, which she considers AI's next frontier. The company's foundation models are used to generate 3D and 4D worlds, serving creators, robot training and architectural design, with goals that plainly extend beyond language itself. She also repeatedly stresses that human beings should actively imagine and design AI's future, and cannot leave it to companies or investors to decide on their own. Spatial intelligence is where the judgment that vision is the foundation of intelligence lands next.

— Fei-Fei Li

In their own words · checked verbatim

I see vision as a cornerstone of intelligence.

Fei-Fei Li4:37

we need to think about AI as a tool that helps us in our agency. It it should not take away our agency.

Fei-Fei Li49:41

The absolute bad outcome is that our young generation. There. Agency. And human level motivation of learning and living is taken away by tools.

Fei-Fei Li1:29:33

Who is humanity's best prompter? I'm going to flunk this quiz. Socrates, if he were alive. Because that is the method of prompting. Right, think about it. What is Socrates method? Is prompting and seeking truth by asking questions.

Fei-Fei Li1:35:00

They are the most important people in our society. We should be talking to them. We should be uplifting them. We should be supporting them. We should be providing resources to them.

Fei-Fei Li2:02:02

Figures

When animals first sensed light540 million years ago4:37
Share of human brain cortical activity related to visionabout half6:53
Number of images in ImageNet15 million9:24
Human error rate in the ImageNet challengeabout 4%16:03
Blood loss in her father's surgery versus a typical operation10 times less1:07:02
Protein in a David protein bar20 grams1:19:36
Calories in a David protein bar1501:19:36
When World Labs was foundedearly 20241:50:48
When ChatGPT was releasedNovember 20222:03:03

Glossary

ImageNet
The large-scale visual dataset Li led the construction of: 15 million images across 1,000 object categories.
Spatial intelligence
The ability to understand and generate three-dimensional space, which World Labs treats as AI's next breakthrough.
Embodied intelligence
Intelligence in which AI interacts with the real world through a physical body, for example a robot caring for an elderly person.
Multimodal
Handling text, images, sound and other forms of information at the same time; the internet is regarded as the largest multimodal collection.

How to listen

Who it's for

Educators, parents, AI founders and engineers, and anyone who cares about where technology's limits lie and how that gets debated in public.