WEBVTT

00:00:03.520 --> 00:00:06.717
Konnichiwa. Bonjour. Hello everyone.

00:00:06.800 --> 00:00:10.277
It is a great honor to address you today,
and I'm truly sorry I cannot be

00:00:10.360 --> 00:00:12.517
there in person. What we're launching in

00:00:12.600 --> 00:00:15.597
Tokyo on July 16th is exactly the kind

00:00:15.680 --> 00:00:18.477
of bold, forward-looking initiative that

00:00:18.560 --> 00:00:22.837
the world needs at this precise moment
in the history of AI. My name is Yann

00:00:22.920 --> 00:00:25.357
LeCun. I am a professor at New York

00:00:25.440 --> 00:00:28.637
University, uh chairman and co-founder

00:00:28.720 --> 00:00:31.637
of Advanced Machine Intelligence, which

00:00:31.720 --> 00:00:34.397
we call AMI, and a recipient of the

00:00:34.480 --> 00:00:37.837
Turing Award. Prior to AMI, I spent 12

00:00:37.920 --> 00:00:40.637
years at Meta, where I founded FAIR,

00:00:40.720 --> 00:00:43.357
Meta's Fundamental AI Research Lab, and

00:00:43.440 --> 00:00:46.877
served as Chief AI Scientist.

00:00:46.960 --> 00:00:50.597
For many years, I've been fortunate
to contribute to the development of modern

00:00:50.680 --> 00:00:52.837
AI and deep learning. Early in my

00:00:52.920 --> 00:00:56.477
career, I've been inspired by the

00:00:56.560 --> 00:00:58.917
foundational work of Professor Amari and

00:00:59.000 --> 00:01:01.237
Fukushima on neural networks from the

00:01:01.320 --> 00:01:05.037
1970s and 1980s. Today, my focus is on

00:01:05.120 --> 00:01:06.997
helping build the next generation of AI

00:01:07.080 --> 00:01:09.557
systems capable of understanding and

00:01:09.640 --> 00:01:12.637
interacting with the real world.

00:01:12.720 --> 00:01:16.317
We are at a turning point.
For the last 5 years,

00:01:16.400 --> 00:01:20.117
artificial intelligence through
generative models has been largely a

00:01:20.200 --> 00:01:22.677
disembodied technology.

00:01:22.760 --> 00:01:25.357
It's extraordinarily good at processing

00:01:25.440 --> 00:01:27.437
human language, computer code, and

00:01:27.520 --> 00:01:32.717
mathematics, but fundamentally unable
to apprehend the physical world.

00:01:32.800 --> 00:01:36.677
The next frontier in AI will change that.
It's about building AI systems

00:01:36.760 --> 00:01:40.757
that can perceive,
reason about, and act in any

00:01:40.840 --> 00:01:43.117
environment. Systems that understand

00:01:43.200 --> 00:01:45.917
causality, physical reality, and the

00:01:46.000 --> 00:01:48.237
constraints of the real world.

00:01:48.320 --> 00:01:52.477
This is physical AI, and it may well
be the most important scientific and

00:01:52.560 --> 00:01:54.717
engineering challenge of our generation.

00:01:54.800 --> 00:01:59.837
Getting there requires new
architectures, systems that learn not

00:01:59.920 --> 00:02:01.637
just from text and other sequences of

00:02:01.720 --> 00:02:05.357
symbols, but from real-world sensor data

00:02:05.440 --> 00:02:09.077
that is continuous,
high-dimensional, and noisy.

00:02:09.160 --> 00:02:11.957
Like video, for example. AI models that

00:02:12.040 --> 00:02:14.037
build an internal representation of the

00:02:14.120 --> 00:02:16.517
world and are capable of predicting the

00:02:16.600 --> 00:02:19.197
consequences of their action will be a

00:02:19.280 --> 00:02:22.677
centerpiece of that new revolution.

00:02:22.760 --> 00:02:25.037
At Emmy, this is precisely the mission

00:02:25.120 --> 00:02:27.517
that we are pursuing, building agent AI

00:02:27.600 --> 00:02:30.157
systems around the concept of world

00:02:30.240 --> 00:02:35.197
models. The world model with world model
AI systems can predict the evolution of

00:02:35.280 --> 00:02:39.437
their environment, predict
the consequences of their own actions,

00:02:39.520 --> 00:02:43.317
and plan a sequence of actions
to accomplish a particular task.

00:02:43.400 --> 00:02:47.357
This will enable countless new
applications in industry, such as

00:02:47.440 --> 00:02:51.557
manufacturing. It will also
enable a new generation of

00:02:51.640 --> 00:02:54.317
autonomous robots capable of operating

00:02:54.400 --> 00:02:57.677
in complex unstructured environment.

00:02:57.760 --> 00:02:59.717
These robots will go beyond scripted

00:02:59.800 --> 00:03:02.117
recipes and will be able to apprehend

00:03:02.200 --> 00:03:05.357
new situations and accomplish new tasks

00:03:05.440 --> 00:03:07.797
without prior training. More generally,

00:03:07.880 --> 00:03:10.397
physical AI systems will help operate

00:03:10.480 --> 00:03:12.717
infrastructure, manage grids in real

00:03:12.800 --> 00:03:15.157
time, and accelerate the transition to

00:03:15.240 --> 00:03:18.117
sustainable resilient energy at scale.

00:03:18.200 --> 00:03:20.317
These are not distant promises.

00:03:20.400 --> 00:03:22.877
The foundational work is happening now,

00:03:22.960 --> 00:03:25.717
and Japan is uniquely positioned

00:03:25.800 --> 00:03:30.237
to be at the center of it.
Decades of mastery in precision

00:03:30.320 --> 00:03:32.677
manufacturing, industrial robotics, and

00:03:32.760 --> 00:03:35.197
engineering excellence have given Japan

00:03:35.280 --> 00:03:39.877
exactly the foundation that physical
AI demands. The factory floor, the

00:03:39.960 --> 00:03:43.477
automated plant,
the industrial environment, these are

00:03:43.560 --> 00:03:47.077
training grounds for the next
generation of intelligent systems. In the

00:03:47.160 --> 00:03:49.717
application of physical AI, Japan does

00:03:49.800 --> 00:03:51.757
not need to catch up. Japan needs to

00:03:51.840 --> 00:03:55.397
lead, but no country does this alone.

00:03:55.480 --> 00:03:57.717
The challenges ahead require

00:03:57.800 --> 00:04:02.477
genuine international collaboration
between researchers, between industries,

00:04:02.560 --> 00:04:06.917
between governments.
What you are initiating today is a

00:04:07.000 --> 00:04:11.277
signal to the world that Japan is ready
to be a full partner in shaping this

00:04:11.360 --> 00:04:13.717
future. I'm deeply excited about what we

00:04:13.800 --> 00:04:16.677
can build together, and I look forward

00:04:16.760 --> 00:04:20.117
to working alongside Japan's scientists

00:04:20.200 --> 00:04:24.477
and engineers and policymakers
to create AI systems that don't just

00:04:24.560 --> 00:04:26.957
understand the world, but can truly act

00:04:27.040 --> 00:04:30.237
in it, improve it, and help us all

00:04:30.320 --> 00:04:37.160
thrive within it. Thank you.
Merci beaucoup. Arigato gozaimasu.