Lex Fridman Podcast

Sergey Levine: Robotics and Machine Learning | Lex Fridman Podcast #108

Sergey Levine is a professor at Berkeley and a world-class researcher in deep learning, reinforcement learning, robotics, and computer vision, including the development of algorithms for end-to-end training of neural network policies that combine perception and control, scalable algorithms for inverse reinforcement learning, and deep RL algorithms. Support this podcast by signing up with these sponsors: - ExpressVPN at https://www.expressvpn.com/lexpod - Cash App - use code "LexPodcast" and download: - Cash App (App Store): https://apple.co/2sPrUHe - Cash App (Google Play): https://bit.ly/2MlvP5w EPISODE LINKS: Sergey's Twitter: https://twitter.com/svlevine Sergey's Website: http://rail.eecs.berkeley.edu/ Sergey's Papers: https://scholar.google.com/citations?user=8R35rCwAAAAJ PODCAST INFO: Podcast website: https://lexfridman.com/podcast Apple Podcasts: https://apple.co/2lwqZIr Spotify: https://spoti.fi/2nEwCF8 RSS: https://lexfridman.com/feed/podcast/ Full episodes playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOdP_8GztsuKi9nrraNbKKp4 Clips playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOeciFP3CBCIEElOJeitOr41 OUTLINE: 0:00 - Introduction 3:05 - State-of-the-art robots vs humans 16:13 - Robotics may help us understand intelligence 22:49 - End-to-end learning in robotics 27:01 - Canonical problem in robotics 31:44 - Commonsense reasoning in robotics 34:41 - Can we solve robotics through learning? 44:55 - What is reinforcement learning? 1:06:36 - Tesla Autopilot 1:08:15 - Simulation in reinforcement learning 1:13:46 - Can we learn gravity from data? 1:16:03 - Self-play 1:17:39 - Reward functions 1:27:01 - Bitter lesson by Rich Sutton 1:32:13 - Advice for students interesting in AI 1:33:55 - Meaning of life CONNECT: - Subscribe to this YouTube channel - Twitter: https://twitter.com/lexfridman - LinkedIn: https://www.linkedin.com/in/lexfridman - Facebook: https://www.facebook.com/LexFridmanPage - Instagram: https://www.instagram.com/lexfridman - Medium: https://medium.com/@lexfridman - Support on Patreon: https://www.patreon.com/lexfridman

Lex FridmanhostSergey Levineguest

Jul 13, 20201h 37mWatch on YouTube ↗

WHAT IT’S REALLY ABOUT

Sergey Levine on Robots, Learning, and the Path to Real Intelligence

Lex Fridman and Sergey Levine discuss the gap between human and robot intelligence, emphasizing that hardware is nearly there but autonomy and adaptability are far behind.
Levine argues that common sense and flexible behavior likely emerge from lifelong learning and interaction with the real world, not from hand-coded knowledge or pure internet-scale data.
They explore deep reinforcement learning, off-policy and offline RL, reward design, and why robots are a powerful testbed for understanding general intelligence rather than just an application area.
The conversation also touches on issues like explainability, safety, simulation limits, self-play, intrinsic motivation, and the broader societal and philosophical implications of increasingly capable AI systems.

IDEAS WORTH REMEMBERING

5 ideas

The real gap between humans and robots is intelligence, not hardware.

Levine argues we can largely engineer robot bodies comparable to humans, but robots lack the flexible, adaptive autonomy that lets humans quickly handle novel tasks like using a new joystick under pressure.

Common sense likely emerges from massive, structured real-world experience.

He suggests the 'iceberg' of human knowledge is built over a lifetime of interacting with the world and choosing what to try, something current ML—especially trained on IID internet data—still struggles to replicate.

Treating perception and control jointly can outperform modular designs.

End-to-end RL policies that map pixels directly to torques can exploit task structure, reducing the required precision of each subcomponent and outperforming pipelines that separately solve vision and control.

Off-policy and offline RL are key to making RL practical in the real world.

In safety-critical or expensive domains, you can’t freely explore; progress hinges on algorithms that learn effectively from large logs of prior behavior while knowing when their predictions are unreliable.

Simulation is powerful but cannot substitute real-world learning indefinitely.

Any human-designed bottleneck—like a simulator—will eventually limit performance; truly continually improving systems must learn directly from real-world data, despite messiness like broken dishes and unknown rewards.

WORDS WORTH SAVING

5 quotes

The hardware gap we can almost close; the intelligence gap is very wide.

— Sergey Levine

Perhaps the reason our current systems lack common sense is that they simply inhabit a different universe.

— Sergey Levine

If your machine has any bottleneck that is built by humans and doesn’t improve from data, it will eventually be the thing that holds it back.

— Sergey Levine

I’d like to build a machine that runs up against the ceiling of the complexity of the universe.

— Sergey Levine

Reinforcement learning gives us a principled way to optimize behavior even when we don’t know the equations that govern the system.

— Sergey Levine

Human vs. robot intelligence: hardware vs. autonomy and the "intelligence gap"Lifelong learning, common sense, and the role of real-world interactionReinforcement learning fundamentals: policies, value functions, on/off-policy, deep RLOff-policy and offline RL: leveraging large logged datasets safely and effectivelyRobotics as a vehicle to study and advance artificial intelligenceReward design, intrinsic motivation, and unsupervised/curiosity-driven RLSafety, simulation limits, and long-term risks and uses of AI and RL

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.