Dr. Fei-Fei Li

Self-driving cars are much simpler robots — just metal boxes running on 2D surfaces, and the goal is NOT to touch anything. Robots are 3D things running in a 3D world, and the goal IS to touch things. The journey will be much longer.

The bitter lesson of AI — simple models with tons of data always win — is harder to apply to robotics. You want actions as output, but your training data lacks actions in 3D worlds. That gap is the core unsolved problem.

ImageNet wasn't a research contribution. It was a hypothesis: if AI is going to learn like humans, it needs data at the scale of human experience. We curated 15 million labeled images and open-sourced it. Two GPUs and a neural network later, the modern AI era began.

The 2012 AI breakthrough came down to three ingredients: big data (ImageNet), neural networks, and GPUs. Those same three ingredients still power today's most advanced models — including ChatGPT.

AI can't count chairs in a video. A toddler can. It can't look at a 2D x-ray image and infer a 3D biological structure. Rosalind Franklin's photo plus Watson & Crick's spatial reasoning gave us DNA. We don't have that yet. World models are the next step.

There's nothing artificial about AI. It's inspired by people. It's created by people. And most importantly, it impacts people. I've said this to Congress, to students, to founders. Every decision in this field is a human decision.

In 2016, major tech companies were avoiding the term "AI" in their marketing — worried it was a dirty word. By 2017, everyone wanted to be an "AI company." By 2025, you couldn't not be one. We went from stigma to mandate in under a decade.

If you're a nurse, AI isn't your enemy — it's your colleague who never sleeps, never makes medication errors out of exhaustion, and can flag deteriorating patients before you can. We don't need less human care. We need humans to stop doing what machines can do better.