Yann Lecun discusses the importance of AI and its potential opportunities
Subtitle language:
00:01 → 00:03
Chairman Warner, Vice Chairman Rubio,
00:03 → 00:05
and distinguished members of the committee,
00:05 → 00:09
thank you for the opportunity to appear before you today
00:09 → 00:12
to discuss important issues regarding AI.
00:12 → 00:13
My name is Yann Lecun.
00:13 → 00:15
I'm currently the server professor
00:15 → 00:18
of computer science and data science at New York University.
00:18 → 00:22
I'm also Meta's chief AI scientist
00:22 → 00:27
and co-founder of Meta's Fundamental AI Research Lab.
00:27 → 00:30
At Meta, I focus on AI research, development strategy,
00:30 → 00:32
and scientific leadership.
00:32 → 00:35
AI has progressed leaps and bounds
00:35 → 00:38
since I began my research career in the 1980s.
00:38 → 00:42
Today, we are witnessing the development of generative AI,
00:42 → 00:44
and in particular, large language models.
00:44 → 00:47
These systems are trained through self-supervised
00:47 → 00:48
learning, or more simply, they are
00:48 → 00:51
trained to fill in the blanks.
00:51 → 00:54
In the process of doing so, those AI models
00:54 → 00:57
learn to represent text or images,
00:57 → 00:59
including the meaning, style, and syntax
00:59 → 01:01
in multiple languages.
01:01 → 01:02
The internal representation can then
01:02 → 01:07
be applied to downstream tasks, such as translation, topic
01:07 → 01:10
classification, et cetera.
01:10 → 01:12
It can also be used to predict the next words in a text, which
01:12 → 01:19
allow LLMs to answer questions or write essays and write code
01:19 → 01:20
as well.
01:20 → 01:23
It is important not to undervalue
01:23 → 01:26
the far-reaching potential opportunities they present.
01:26 → 01:29
The development of AI is as foundational
01:29 → 01:32
as the creation of the microprocessor,
01:32 → 01:36
the personal computer, the internet, and the mobile device.
01:36 → 01:38
Like all foundational technologies,
01:38 → 01:41
there will be a multitude of uses of AI.
01:41 → 01:44
And like every technology, AI will be used by people
01:44 → 01:47
for good and bad ends.
01:47 → 01:49
As AI systems continue to develop,
01:49 → 01:52
I'd like to highlight two defining issues.
01:52 → 01:56
The first one is safety, and the second one is access.
01:56 → 01:58
One way to start to address both of these issues
01:58 → 02:02
is through the open sharing of current technology
02:02 → 02:04
and scientific information.
02:04 → 02:08
The free exchange of scientific papers, code, and trained
02:08 → 02:12
models, in the case of AI, has enabled American leadership
02:12 → 02:15
in science and technology.
02:15 → 02:16
This concept is not new.
02:16 → 02:18
It started a long time ago.
02:18 → 02:22
Open sourcing technology has spurred rapid progress
02:22 → 02:25
in systems we now consider basic infrastructure,
02:25 → 02:30
such as the internet and mobile communication networks.
02:30 → 02:34
This doesn't mean that every model can or should be open.
02:34 → 02:37
There is a role for both proprietary and open source AI
02:37 → 02:38
models.
02:38 → 02:41
But an open source basic model should
02:41 → 02:43
be the foundation on which industry
02:43 → 02:45
can build a vibrant ecosystem.
02:45 → 02:48
An open source model creates an industry standard,
02:48 → 02:52
much like the model of the internet in the mid-'90s.
02:52 → 02:54
Through this collaborative effort,
02:54 → 02:56
AI technology will progress faster,
02:56 → 02:59
more reliably, and more securely.
02:59 → 03:02
Open sourcing also gives businesses and researchers
03:02 → 03:05
access to tools that they could not otherwise
03:05 → 03:08
build by themselves, which helps create
03:08 → 03:11
a vast social and economic set of opportunities.
03:11 → 03:16
In other words, open sourcing democratizes access.
03:16 → 03:17
It gives more people and businesses
03:17 → 03:20
the power to build upon state-of-the-art technology
03:20 → 03:24
and to remedy potential weaknesses.
03:24 → 03:28
This also helps promote democratic values
03:28 → 03:31
and institutions, minimize social disparities,
03:31 → 03:34
and improve competition.
03:34 → 03:37
We want to ensure that the United States and American
03:37 → 03:40
companies, together with other democracies,
03:40 → 03:44
lead in AI development ahead of our adversaries
03:44 → 03:47
so that the foundational models are developed here
03:47 → 03:51
and represent and share our values.
03:51 → 03:54
By open sourcing current AI tools,
03:54 → 03:57
we can develop our research and development ecosystem
03:57 → 04:00
faster than our adversary.
04:00 → 04:02
As AI technology progresses, there
04:02 → 04:05
is an urgent need for governments
04:05 → 04:08
to work together, especially democracies,
04:08 → 04:12
to set common AI standards and governance models.
04:12 → 04:14
This is another valuable area where
04:14 → 04:17
we welcome working with regulators
04:17 → 04:20
to set appropriate transparency requirements, red teaming
04:20 → 04:23
standards, and safety mitigations
04:23 → 04:26
to help ensure those codes of practice, standards,
04:26 → 04:30
and guardrails are consistent across the world.
04:30 → 04:33
The White House's voluntary commitment
04:33 → 04:36
are a critical step in ensuring responsible guardrails,
04:36 → 04:39
and they create a model for other governments to follow.
04:39 → 04:43
Continued US leadership by Congress and the White House
04:43 → 04:45
is important in ensuring that society
04:45 → 04:48
can benefit from innovation in AI
04:48 → 04:50
while striking the right balance with protecting
04:50 → 04:54
rights and freedom, preserving national security
04:54 → 04:58
interests, and mitigating risks where those arise.
04:58 → 05:01
I'd like to close by thanking Chairman Warner, Vice Chairman
05:01 → 05:03
Rubio, and the other members of the committee
05:03 → 05:05
for your leadership.
05:05 → 05:06
At the end of the day, our job is
05:06 → 05:09
to work collaboratively with you, with Congress,
05:09 → 05:11
with other nations, and with other companies
05:11 → 05:14
in order to drive innovation and progress in a manner that
05:14 → 05:18
is safe and secure and consistent with our national security
05:18 → 05:19
interests.
05:19 → 05:19
Thank you.
05:19 → 05:22
I look forward to your questions.
Yann Lecun, Meta's chief AI scientist, highlights the progress and potential of AI, emphasizing safety and access as key issues.
This video in English was translated to English on September 20, 2023, using Targum.video AI translation service.
