# Yann LeCun 氏　ビデオメッセージ

*FRONTia_ch* · https://www.youtube.com/watch?v=FXXds6hf_gA

## Summary

This is a short video message from Yann LeCun recorded for an event launching in Tokyo on July 16th. He introduces his background at NYU, AMI, and Meta FAIR, and argues that AI is at a turning point from disembodied generative models toward physical AI. He explains that world models will let AI systems predict, plan, and act in the real world, and that Japan is uniquely positioned to lead this shift thanks to its manufacturing and robotics expertise.

## Key points

- Yann LeCun is a professor at New York University, chairman and co-founder of AMI (Advanced Machine Intelligence), and a Turing Award recipient.
- He spent 12 years at Meta, where he founded FAIR and served as Chief AI Scientist.
- He was inspired early in his career by the foundational neural network work of Professor Amari and Fukushima from the 1970s and 1980s.
- Generative AI over the last 5 years has been largely disembodied, strong on language, code, and mathematics but unable to apprehend the physical world.
- The next frontier is physical AI: systems that perceive, reason about, and act in any environment, understanding causality and physical reality.
- AMI's mission is building agent AI systems around world models that predict the evolution of the environment and the consequences of their own actions.
- Applications include manufacturing, autonomous robots in complex unstructured environments, infrastructure operation, and real-time grid management for sustainable energy.
- Japan's decades of mastery in precision manufacturing, industrial robotics, and engineering make it uniquely positioned to lead in physical AI.

## Notes

### Speaker Background
- Yann LeCun is a professor at New York University, chairman and co-founder of Advanced Machine Intelligence (AMI), and a Turing Award recipient.
- He spent 12 years at Meta, founded FAIR (Meta's Fundamental AI Research Lab), and served as Chief AI Scientist.
- Early in his career he was inspired by the foundational neural network work of Professor Amari and Fukushima from the 1970s and 1980s.

### The Turning Point in AI
- For the last 5 years, AI through generative models has been largely a disembodied technology.
- It is extraordinarily good at processing human language, computer code, and mathematics, but fundamentally unable to apprehend the physical world.
- The next frontier is physical AI: systems that can perceive, reason about, and act in any environment, understanding causality, physical reality, and real-world constraints.
- LeCun calls this possibly the most important scientific and engineering challenge of our generation.

### World Models and New Architectures
- Getting there requires new architectures that learn from real-world sensor data that is continuous, high-dimensional, and noisy, such as video.
- AI models must build an internal representation of the world and predict the consequences of their actions.
- AMI's mission is building agent AI systems around the concept of world models.
- With world models, systems can predict the evolution of their environment, predict the consequences of their own actions, and plan action sequences to accomplish tasks.

### Applications
- Industry applications such as manufacturing.
- A new generation of autonomous robots operating in complex unstructured environments, going beyond scripted recipes to handle new situations without prior training.
- Operating infrastructure, managing grids in real time, and accelerating the transition to sustainable, resilient energy at scale.

### Japan's Role
- Japan's decades of mastery in precision manufacturing, industrial robotics, and engineering excellence provide exactly the foundation physical AI demands.
- The factory floor and industrial environments are training grounds for the next generation of intelligent systems.
- In physical AI, Japan does not need to catch up; it needs to lead, through genuine international collaboration among researchers, industries, and governments.

## Transcript

[00:03] Konnichiwa. Bonjour. Hello everyone. It is a great honor to address you today, and I'm truly sorry I cannot be there in person. What we're launching in Tokyo on July 16th is exactly the kind of bold, forward-looking initiative that the world needs at this precise moment in the history of AI. My name is Yann LeCun. I am a professor at New York University, uh chairman and co-founder of Advanced Machine Intelligence, which we call AMI, and a recipient of the Turing Award. Prior to AMI, I spent 12 years at Meta, where I founded FAIR, Meta's Fundamental AI Research Lab, and served as Chief AI Scientist. For many years, I've been fortunate to contribute to the development of modern AI and deep learning. Early in my

[00:52] career, I've been inspired by the foundational work of Professor Amari and Fukushima on neural networks from the 1970s and 1980s. Today, my focus is on helping build the next generation of AI systems capable of understanding and interacting with the real world. We are at a turning point. For the last 5 years, artificial intelligence through generative models has been largely a disembodied technology. It's extraordinarily good at processing human language, computer code, and mathematics, but fundamentally unable to apprehend the physical world. The next frontier in AI will change that. It's about building AI systems that can perceive, reason about, and act in any environment. Systems that understand

[01:43] causality, physical reality, and the constraints of the real world. This is physical AI, and it may well be the most important scientific and engineering challenge of our generation. Getting there requires new architectures, systems that learn not just from text and other sequences of symbols, but from real-world sensor data that is continuous, high-dimensional, and noisy. Like video, for example. AI models that build an internal representation of the world and are capable of predicting the consequences of their action will be a centerpiece of that new revolution. At Emmy, this is precisely the mission that we are pursuing, building agent AI systems around the concept of world models. The world model with world model

[02:32] AI systems can predict the evolution of their environment, predict the consequences of their own actions, and plan a sequence of actions to accomplish a particular task. This will enable countless new applications in industry, such as manufacturing. It will also enable a new generation of autonomous robots capable of operating in complex unstructured environment. These robots will go beyond scripted recipes and will be able to apprehend new situations and accomplish new tasks without prior training. More generally, physical AI systems will help operate infrastructure, manage grids in real time, and accelerate the transition to sustainable resilient energy at scale. These are not distant promises.

[03:20] The foundational work is happening now, and Japan is uniquely positioned to be at the center of it. Decades of mastery in precision manufacturing, industrial robotics, and engineering excellence have given Japan exactly the foundation that physical AI demands. The factory floor, the automated plant, the industrial environment, these are training grounds for the next generation of intelligent systems. In the application of physical AI, Japan does not need to catch up. Japan needs to lead, but no country does this alone. The challenges ahead require genuine international collaboration between researchers, between industries, between governments. What you are initiating today is a signal to the world that Japan is ready

[04:09] to be a full partner in shaping this future. I'm deeply excited about what we can build together, and I look forward to working alongside Japan's scientists and engineers and policymakers to create AI systems that don't just understand the world, but can truly act in it, improve it, and help us all thrive within it. Thank you. Merci beaucoup. Arigato gozaimasu.