Latest / Radio Davos / What I tell my Stern Business School students on how to thrive in the age of AI

Transcript

What I tell my Stern Business School students on how to thrive in the age of AI · Radio Davos

0:03
Welcome to Radio Davos, the podcast
0:05
from the World Economic Forum that
0:06
looks at the biggest challenges and
0:08
how we might solve them.
0:10
On this episode, this is a joint
0:11
episode with China's
0:13
CGTN, and I'm joined
0:15
by an anchor and
0:17
chief business news editor at CGTN,
0:19
Xin Guan.
0:20
Guan, how are you?
0:21
I'm doing great, Robin, thank you
0:23
for having me, and thank you the
0:24
introduction.
0:25
CGTN is a very important broadcaster
0:27
in China, just tell us what it is.
0:29
Thank you.
0:30
Well, we are an international
0:32
broadcasting organisation and
0:35
our team is the business
0:37
news department.
0:39
So our work is actually
0:41
making sense of what's happening in
0:43
the business and economy
0:46
with the particular focus on
0:48
China.
0:49
So a big part of my role
0:51
is mostly about
0:53
how global audience
0:55
to understand China's economic
0:57
stories and how
0:58
it connects with global economic
1:00
trends.
1:01
Great.
1:02
And you're joining me
1:04
to interrogate
1:06
our guest here, Arun Sundararajan.
1:09
You are the Harold Price Professor
1:12
of Entrepreneurship and Professor
1:14
of Technology Operations and
1:15
Statistics at New York University's
1:18
Stern School of Business.
1:19
Did I get that correct, Arun?
1:20
Yes, we academics,
1:22
after a point, stop having
1:24
new honorifics and so
1:26
we keep extending our titles as
1:28
sort of a feeling of like momentum.
1:31
Okay, we're going to get into this
1:32
discussion about things you know a
1:34
lot about, about employment and the
1:36
change in the
1:38
world of work.
1:40
And I think everyone has lots of
1:42
questions about that.
1:43
You published a book ten years ago,
1:45
we are not going to dwell on that because
1:46
it was ten years ago, but it was
1:48
called The Sharing Economy, The End
1:49
of Employment and the Rise of
1:51
Crowd-Based Capitalism.
1:53
Just remind us what that book was
1:54
about.
1:55
First off, thank you for having me
1:57
on this. I'm really looking forward
1:58
to this conversation.
1:59
The book documented a
2:01
new way of organising economic
2:03
activity that I call crowd-based
2:05
capitalism.
2:06
It described how platforms
2:09
ranging from platforms like
2:11
YouTube at the time to Uber
2:13
and Didi and Airbnb,
2:15
Etsy, were reorganising
2:17
how we provided things like
2:20
retail, like transportation,
2:22
like advertising.
2:23
And instead of having large
2:25
companies that employed people full
2:27
time, paid them salaries and
2:29
produced goods and services,
2:31
a shift towards platform-mediated
2:33
commerce, where you had customers
2:35
on one side, crowds of
2:37
providers on the other and the
2:39
platform sitting in the middle.
2:41
Did that come to pass, that was a
2:42
genuine shift in the world of
2:44
work?
2:44
I believe so.
2:46
I mean, you know, all of the
2:47
statistics and a lot of people, when
2:49
they saw that subtitle,
2:51
The End of Employment, they
2:53
assumed that I was talking about
2:55
humans not having anything to do.
2:57
That wasn't the implica- I mean
2:58
maybe that's what my publisher
2:59
wanted people to think and pick
3:01
up the book with, but it was really
3:03
about a change in the arrangement of
3:05
work. Prediction was that work
3:07
would be arranged less and less
3:09
in the form of jobs or
3:11
employment.
3:12
And increasingly as non-employment
3:15
work arrangements.
3:16
We've seen that come to pass.
3:18
When I wrote the book about 40
3:19
million Americans had
3:21
some form of non- employment work
3:24
arrangement.
3:25
Last year, it was over 70 million.
3:27
I was just told this morning by the
3:29
chief economist of JD.com
3:31
that over 40%
3:33
of China's workers today
3:35
are in non- unemployment work
3:37
arrangements of some kind.
3:38
We call it flexible employment,
3:41
because it's
3:45
actually becoming a
3:47
very important part of China's
3:49
employment landscape due to
3:51
the importance of the sharing
3:52
economy.
3:53
We really need to keep up with
3:55
this new form of employment,
3:58
from social policies to
4:00
training and social
4:02
safety net.
4:03
And I think it will be very
4:05
interesting to watch the development
4:07
in China because China have
4:09
a very strong sharing economy
4:11
development.
4:11
Absolutely. And I think that, you
4:14
know, a lot of those questions
4:15
around how do we protect the
4:17
workforce?
4:18
How do we give people a
4:20
structure for progressing, you
4:22
know, in their quote unquote
4:23
careers, if it isn't provided
4:25
by the organisation?
4:27
All of these questions are actually,
4:29
you know, this is not something that
4:31
I explicitly predicted in my book,
4:32
but they're definitely being
4:33
amplified by the emergence
4:36
of AI and, you know, sort
4:37
of this sustained kind
4:39
of attack of sorts
4:41
on the structures of
4:43
work that we had gotten used
4:45
to over the 20th century.
4:47
Right, I mean, if you were to
4:48
publish a book today with that
4:49
provocative title, The End of
4:51
Employment, everyone would assume
4:53
you're talking about AI because
4:55
there's a lot of fear about the
4:56
displacement of jobs, particularly
4:58
perhaps for entry level jobs.
5:00
I'd really like to hear your take on
5:02
that. Is it the end of employment
5:04
because of AI?
5:05
Well, I certainly think we're going
5:06
to see an acceleration in
5:09
the change of work arrangements.
5:11
I mean, let's sort of break
5:13
this down into two or three
5:14
different threads.
5:15
First is this
5:17
feeling that because this
5:19
new generation of AI has
5:22
capabilities that start
5:24
to mimic what was the
5:26
exclusive provenance of humans.
5:28
To some extent, like the non-routine
5:31
cognitive tasks, which have sort
5:32
of spurred a lot of the income
5:34
growth and benefits to people
5:36
with college degrees.
5:38
Now, AI can produce that kind
5:40
of like non-outine cognitive work,
5:42
but just because AI can
5:44
do what a human does or a machine
5:46
can do or a human doesn't mean
5:48
that it will instantly start to
5:50
do it in the place of the humans.
5:52
That's the fallacy of technological
5:53
determinism, that just because a
5:55
machine can do it, it will
5:56
immediately.
5:57
So I think that we've got some time.
5:59
There are lots of other factors that
6:01
make this transition more viscous.
6:03
In some ways, part of why we're
6:06
seeing some impact on
6:08
entry-level work is because
6:10
there's so much uncertainty that
6:12
organisations are feeling about
6:15
what will the role of
6:17
human beings be in the future,
6:19
that a lot of them are just saying,
6:20
well, let's hold back.
6:22
Let's not hire people
6:24
right now.
6:25
Let's wait for this to shake out
6:27
over the next two, three years and
6:29
see where things land, what kinds
6:31
of capabilities we want in humans.
6:35
And so I don't think
6:36
there's a lot of debate, at
6:38
least in the United States, that
6:40
there has been a slowdown
6:42
in entry-level hiring.
6:44
I think there is a very vibrant
6:46
debate about what's causing it.
6:48
There's a plausible story that
6:50
it's generative AI.
6:51
It's sort of broken the contract
6:53
where a company would hire someone
6:56
who was promising in
6:58
an entry-level coding
7:00
role, banking role, consulting role.
7:02
They do fairly simple things,
7:05
PowerPoint spreadsheets, routine
7:07
coding, in exchange for
7:09
being sort of like these
7:11
investments that would be made into
7:13
them to create the next generation
7:15
of partners of expert
7:17
coders of leaders.
7:18
And now, because the
7:20
generative AI can do...
7:21
A lot of that sort of
7:24
somewhat routine cognitive work,
7:26
that contract is fragmenting.
7:28
It's being compounded by the fact
7:30
that the people who are acquiring
7:32
these capabilities in college
7:34
are not sure what the returns
7:35
on these human capital investments
7:37
are going to be.
7:38
I think that there's a third factor
7:40
which is we
7:42
reorganised and restructured
7:44
a lot of work during
7:46
the COVID shutdowns because
7:48
we had to sort of transition to
7:50
work from home, remote work.
7:53
And so we have to sort of structure
7:55
things a lot.
7:56
And I think that in some ways
7:58
has lowered the return on
8:00
these investments in entry-level
8:02
workers because it's one thing
8:04
if you're in the organisation
8:06
five days a week and sort of
8:07
absorbing from the senior people.
8:10
It's another if you are in sort of
8:12
strategically two or three days a
8:13
weak. So there's a confluence of
8:15
factors.
8:16
I think, you know, anyone who tells
8:18
you they know exactly.
8:20
What kinds of work AI is
8:22
displacing and what's going to
8:23
happen sort of in the long run.
8:25
I would not take too seriously, but
8:27
I think that there's no doubt that
8:29
there is going to be a lot of
8:30
displacement over the next decade.
8:31
I want to put it in the context of
8:33
Asia's development, because we
8:35
are very pragmatic
8:38
about this, because
8:40
Asia has the problem
8:42
of ageing society, so on
8:44
one hand we have this shrinking
8:46
labour force, on the other hand, we
8:48
have AI automation,
8:50
so how will this interaction
8:53
play out in the long term, could it
8:55
be a good thing
8:57
to complement the
8:59
ageing society?
9:00
I think in the context of
9:02
certain Asian economies,
9:04
China, Japan, South Korea,
9:07
where there has been this threat of
9:09
a Japan, where has been
9:12
both a potential and realised
9:14
threat of an economic
9:16
slowdown because of a labour
9:17
shortage, AI and
9:20
in particular the embodied AI,
9:22
the robotic AI, is probably
9:24
going to be a boon
9:26
rather than a threat.
9:28
Because it's going to play a
9:30
big role in not,
9:33
let me put that differently.
9:35
The fact that we are
9:37
seeing the kind of rapid progress
9:39
that we're seeing in AI and
9:41
robotics today is going to mean
9:43
that the GDP slowdowns
9:45
that would have occurred are not
9:47
going to occur in these economies.
9:49
And so it's actually very good
9:51
news for Japan.
9:53
Which has a long history of
9:55
embracing robotics for China,
9:57
which is far
9:59
and away the world's leader
10:01
in adopting industrial
10:03
robotics.
10:04
I read recently that more
10:06
than half of the world industrial
10:08
robots that
10:10
were installed last year were
10:12
installed in China.
10:13
So there's sort of no question that
10:16
there's both the need and the
10:18
capability.
10:19
I think that, you
10:21
know, for the white collar workers,
10:24
the people who do knowledge work,
10:26
what I expect to see is
10:28
a role compression of
10:30
sorts, perhaps in Western economies
10:32
as well, but more so in Asian
10:34
economies, where a lot of
10:35
things are delegated to AI.
10:38
The set of things that the human
10:40
does shifts to
10:43
verification, problem
10:45
formulation, certain kinds
10:47
of judgement, certain kinds
10:49
accountability.
10:50
And it's the people
10:52
who are able to
10:55
demonstrate that
10:57
they're good at this new
10:59
class of non-routine
11:01
skills.
11:02
100 years ago, muscle got
11:04
automated.
11:05
There was a certain kind of
11:06
de-skilling as well, right?
11:08
I mean, you had the skilled
11:09
machinist, you had sort
11:11
of the craft manufacturer,
11:13
and that kind of skill
11:15
was sort of put out of business by
11:17
the machines and work became more
11:19
routine. But there were other
11:21
sort of cognitive capabilities
11:23
that started to command the premium.
11:26
So we're going to see a new cluster
11:28
of capabilities that
11:30
will command the premium.
11:31
We haven't put our finger on exactly
11:33
what that bundle is, but verify,
11:36
you know, being able to know
11:38
when to trust the machine,
11:40
what to delegate to the machine.
11:42
You know, how to exert the judgement
11:44
that allows you to compliment
11:46
rather than be substituted by
11:48
the machine These will be
11:50
the skills that the technological
11:51
change will be biassed in favour of.
11:53
What do you say to your students
11:55
at the University of New York
11:57
who are trying to prepare?
11:59
They'll want, obviously, the degree
12:01
certificate, they'll get that
12:02
presumably, you know, if they do
12:03
their homework and pass the exams.
12:05
But the skills they need and the
12:06
kind of, when they go to those
12:08
employers, if this period of
12:09
uncertainty, as you're suggesting,
12:11
it's not the end of employment
12:13
for young people, but it's
12:16
quite difficult at the moment and
12:18
we don't know.
12:19
What the outcome is, what the
12:20
world's going to look like in five
12:21
years time. So what do you tell them
12:23
in terms of, okay, you've really got
12:24
to nail this, this and this skill?
12:26
I tell them to run for the hills.
12:29
No, I don't.
12:30
I tell them that the way
12:32
that I see artificial intelligence
12:34
is as a technology
12:36
that is going to create
12:38
an incredible amount
12:40
of value.
12:41
And this is simply because
12:43
the barriers to creating
12:45
things of value have never been
12:47
lowered so rapidly by
12:49
any other technology.
12:51
And I know that this sounds like a
12:52
simple idea.
12:54
Lowering the barriers to creating
12:56
things of value, but that's really
12:58
at the core of what's going to
13:00
grow the economy.
13:01
So how do they prepare themselves
13:03
and how do the sort of make sure
13:05
that they're on the sides of the
13:06
winners rather than the losers?
13:08
You know, and I've been telling them
13:09
this for a while.
13:10
One is to think
13:12
like entrepreneurs and be
13:14
entrepreneurs.
13:15
If the barriers to creating thing
13:17
of value have been lowered, go out
13:19
and start to create them while
13:21
you're a student.
13:22
This will do two things.
13:23
One, it creates a portfolio
13:26
of what you're capable of.
13:28
It sets you up to do something
13:30
on your own, if in fact,
13:32
the structure of work is not going
13:34
to be employment.
13:35
But it also sort of builds that
13:37
muscle that allows you to be
13:39
flexible and resilient
13:41
and adaptable.
13:42
And so if you're building that
13:43
between the age of 18 and 22,
13:46
that's going to...
13:47
Like serve you in the long run.
13:48
I also tell them simultaneously to
13:51
position themselves to be AI
13:53
force multipliers.
13:54
So it's not like companies
13:56
will not want any humans.
13:58
In the short run, they may need
14:00
fewer entry-level humans,
14:02
but the ones that they need are
14:04
going to be the ones who
14:06
know how to dramatically
14:08
increase their output using
14:11
AI, and there will
14:12
actually be a premium on
14:14
that kind of human.
14:16
And so make yourself an AI force
14:18
multiplier, be an
14:20
entrepreneur, sort of be adaptable,
14:22
be flexible and build some
14:24
resilience.
14:25
I also tell them that networking
14:27
has never been more important.
14:29
Like, you know, a big part of your
14:31
value in the future is
14:33
going to be your set of
14:35
connections.
14:36
And so invest in that very
14:38
heavily.
14:39
There's never a better time to do
14:40
that when you're a student, because
14:41
you can always go up to someone and
14:43
say, I'm a student.
14:44
I'd like to learn from you.
14:45
Once you're no longer a student that
14:47
doesn't work as well.
14:48
And so these are some of the things
14:50
that I tell my students.
14:51
It's called micro-entrepreneurship,
14:53
right?
14:54
Absolutely, yeah.
14:55
I think that is exactly what's
14:56
happening in China because we have
14:58
this open-sourced AI agent
15:01
called OpenClaw and
15:02
it just went viral in
15:04
China and people
15:06
start their own company just
15:09
of his or her own
15:11
and we call it a one-person company
15:13
and he deployed several
15:15
AI agents working for
15:17
him. And what's interesting is
15:21
in the past, if you are an IT
15:22
student, it's easier to start a
15:24
business because they know how the
15:26
internet works. They know how to
15:27
code. And now if you study
15:29
literature, philosophy,
15:31
but you can just tell AI
15:33
agents what you want to do.
15:35
Tell it about your vision.
15:36
And they're just starting a company.
15:38
I think that is a very
15:40
interesting trend.
15:41
So maybe we can be less
15:43
concerned about the job losses
15:45
and thinking about new
15:47
ways to create value for the
15:49
society.
15:50
Absolutely, and I think that there's
15:52
going to be a lot more people who
15:54
earn their livings as not
15:56
working for someone else,
15:58
but creating things of value,
16:00
working by themselves.
16:01
And the early indications are the
16:03
people who can harness the agents
16:05
certainly have a leg
16:07
up. I do think that it's going be
16:09
a very substantial adjustment
16:11
for countries.
16:12
It's going require a rethinking
16:14
of industrial policy.
16:15
It's going to require a
16:17
rethinking of the
16:19
structures that we put in place
16:21
to allow people to feel
16:23
like they are progressing, the
16:25
structures to provide people with
16:27
benefits and a safety net.
16:29
And, at least over the next decade,
16:31
I think that there may
16:33
be a call to come
16:35
up with new ways to
16:37
redistribute.
16:38
And I'm hoping we're going to get
16:40
beyond the ideas of an AI
16:42
tax or a universal basic income.
16:44
We can get to sort of some other
16:46
ideas where I feel like the
16:48
economists have to do better than,
16:49
you know, tax or
16:51
UBI.
16:52
But I think redistributive
16:54
mechanisms are going to be needed
16:56
in the near term to
16:58
make sure that as we
17:00
get to that eventual destination
17:03
of massive value creation,
17:06
the process is not too
17:08
painful for a significant fraction
17:10
of the workforce.
17:11
It'll change education as well, do
17:12
you think?
17:13
I think it's already started to
17:14
change education, certainly
17:16
college education.
17:18
I mean, every university I speak to
17:20
is in some ways thinking about
17:22
what can be delivered using
17:24
AI.
17:25
And then of the time that's
17:27
freed up, how do we double down
17:30
on, you know, sort of
17:31
smaller group experience,
17:33
more experiential learning,
17:35
more entrepreneurial type things
17:37
that prepare students better for
17:39
the future.
17:40
It's quite an optimistic view,
17:43
I think.
17:44
We won't all be put out of work,
17:46
but things are going to change.
17:48
Yep, there's going to be a
17:50
not-so-nice
17:52
process of adjustment.
17:54
But we saw this in
17:56
many ways at
17:58
the turn of the 20th century.
18:01
There was a whole sort of
18:03
hay farming and oats kind
18:05
of industrial complex that
18:08
there's
18:08
The oil of its day, I assume,
18:10
it was fuelling transport.
18:12
Absolutely.
18:13
And so that infrastructure had
18:15
to be, you know, there were people
18:17
who over a generation, you
18:19
know had to find something else
18:21
to do. The gasoline powered,
18:23
you know, machines that
18:25
put a lot of farmers, that
18:26
mechanised a lot of farming and
18:29
dramatically reduced the number of
18:31
people who could earn a living as an
18:32
independent farmer.
18:34
And so it wasn't a painless process
18:37
and we think of where it took
18:39
us, but we are beginning a
18:41
more accelerated version of that.
18:43
What role should the government play
18:45
in this process to manage
18:47
these challenges in the transition?
18:49
That's a great question.
18:50
I think it depends on what part of
18:52
the world you're talking about and
18:54
what role the government
18:55
historically does play, I think
18:57
because the answer for China is
18:59
probably going to be different than
19:00
the answer from Germany and the
19:02
answer from the United States.
19:03
At a high level I think
19:05
government should be thinking about
19:08
ways in which they can
19:10
facilitate transition.
19:12
You know, educational
19:14
infrastructures have generally
19:16
been to prepare people
19:18
for the beginning of their work
19:19
life, K-12 education,
19:22
college.
19:22
What we need are infrastructures
19:25
that allow people to shift,
19:27
because entry-level work is what's
19:28
being threatened now, but there's
19:30
going to be a lot of mid-career
19:31
transition in the pipe.
19:33
And most of the glory comes from
19:35
creating Tsinghua University,
19:37
right? Not creating a network of
19:38
community colleges that transitions
19:41
people in their 30s and 40s.
19:43
But I think governments need to
19:44
invest very heavily in
19:46
making sure that people can
19:48
make that shift.
19:49
I do think the innovative
19:51
governments will think carefully
19:53
about how do we make sure
19:55
that the value that is going to
19:57
be created by AI is
19:59
equitably distributed,
20:01
the income and wealth value
20:03
that is created by Ai is
20:05
equidably distributed to some
20:07
extent.
20:07
I think it's also useful for
20:09
a government to champion
20:12
the non-income,
20:14
non-wealth.
20:15
Equalising effects of
20:17
a technology like AI in
20:19
education, in health care,
20:22
in access to opportunity, much like
20:24
digital technologies equalised,
20:26
you know, access to entertainment,
20:28
access to communication, AI
20:30
will equalise access to a whole
20:32
bunch of different things.
20:34
And so having the
20:36
population having a mindset that
20:38
this is helping them.
20:40
While also making sure that
20:42
there's some redistributive
20:43
mechanism that is preventing
20:45
too much income and wealth
20:46
inequality from emerging.
20:48
You mentioned Guan, government's
20:50
role, so maybe we can move the
20:52
conversation on now to the
20:54
governance of AI.
20:55
Absolutely. A critical conversation.
20:57
Go ahead. What would you like to
20:58
know from Arun about governance?
21:00
Yes, as Arun
21:02
just mentioned, there are different
21:03
approaches around the
21:05
world and China definitely has
21:07
its own approach.
21:08
I know you are an advisor
21:10
to the Internet Society of China.
21:13
You've been observing the
21:14
development of China's sharing
21:15
economy. Will
21:17
these unique characteristics
21:19
carrying on to the age of AI?
21:21
I think they will.
21:22
Let me put that into context.
21:24
In many countries that are
21:26
not China, over the last 15
21:28
years, we've seen
21:30
a redistribution of
21:32
governance power between
21:34
governments and private entities.
21:37
And so a lot of the roles that used
21:39
to be played by government,
21:41
whether it is on different
21:43
things around what kind of content
21:45
is accessible to people
21:47
who can observe.
21:49
Who provides infrastructure,
21:51
who is responsible for
21:53
the technologies that facilitate
21:56
security.
21:57
A lot of those roles shifted away
21:59
from government and towards
22:01
platforms.
22:03
And this happened sort of de facto
22:05
in many ways.
22:06
And as a consequence,
22:08
if I look at the US today,
22:10
a lot of governance has to be done
22:12
by the companies because we're
22:14
in a world where that is
22:16
the system.
22:18
Now, if you look at China on the
22:19
other hand, there has been
22:22
much more of a top-down
22:24
approach to platform governance.
22:26
There's been more active algorithmic
22:29
governance, where there's
22:31
better visibility into how the
22:32
algorithms are worked and governed.
22:34
I think that there's been much more
22:36
robust industrial policy,
22:39
much stronger sort of
22:40
planned industrial policy that
22:43
is much more broad-based.
22:44
So that it's not just a
22:46
lot of innovation in a
22:48
few areas, but it's across the
22:50
board, robotics, solar,
22:53
electric vehicles, not just
22:55
sort of generative AI.
22:56
And so I expect that
22:58
that blueprint of top-down,
23:02
of sort of active algorithmic
23:04
governance, of
23:06
situating it in a
23:08
broader industrial policy so
23:10
that there's not over-capitalization
23:12
in any particular sort of
23:15
slice of technological
23:17
innovation.
23:18
So no disorderly allocation
23:20
of capital in some sense.
23:23
And then finally, I guess, the
23:24
system where a lot of
23:26
what happens is through
23:28
administrative directives
23:30
rather than through a courtroom
23:31
battle, which is much more
23:33
in other countries.
23:34
I think a lot that is going to
23:36
carry over to AI governance.
23:38
And so you'll probably see
23:41
AI governance structure
23:43
settle in China
23:45
much sooner than in the EU
23:47
and certainly much
23:49
sooner than the United States.
23:53
In the United States we had this Anthropic
23:54
Mythos affair, what's
23:56
the word, where
23:59
it seemed to be that the current
24:01
US administration had a pretty much
24:03
laissez-faire attitude,
24:05
or to put it another way, was
24:06
leaving it up to these platforms to
24:08
work it out, not trying
24:10
not to put regulatory barriers in
24:11
their way. And then you had Mythos
24:13
from Anthropic,
24:15
which the company itself said
24:17
was potentially too dangerous just
24:19
to release widely.
24:20
And then it released it to certain
24:22
people, released it the US
24:23
government. And now the US
24:25
government has imposed certain
24:27
restrictions.
24:28
On it, if you remind us exactly,
24:30
you'll probably follow this closer
24:31
than I have, what's happened there
24:33
and what does this say about US
24:35
AI governance?
24:36
Okay, I think it's a fascinating
24:38
case study and it's one of
24:40
the most important early case
24:42
studies on how AI
24:44
governance will play out in the
24:45
United States.
24:47
So as you point out, the original,
24:49
quote unquote, governance of Mythos
24:51
had to do with its code
24:53
generating capabilities being
24:55
so advanced that
24:57
it posed a threat
24:59
to the computer security.
25:01
Of like not just the country but of
25:03
the globe in terms of this
25:04
capability could be harnessed to
25:06
hack into any system.
25:08
And so over there you saw
25:10
an illustration of this platform
25:12
government talents where
25:14
the platform proactively
25:16
said as a responsible actor
25:18
we are not going to release it.
25:20
Now we're in phase two
25:21
of Mythos where the
25:23
US government has decided
25:25
that it may a threat
25:27
to national security for non-U.S.
25:30
citizens. Not just people in other
25:31
countries, but non-U.S.
25:32
citizens to have access to it.
25:34
And Anthropic has made the choice to
25:36
simply not make it available to
25:38
anyone, in part because
25:39
it's complicated.
25:41
They have a number of non-US
25:42
citizens who work for them.
25:44
So technically these people are not
25:46
supposed to have to access to Mythos
25:48
at this point.
25:48
So I think what this is telling us
25:50
is that we're in some ways
25:52
in uncharted territory
25:55
when it comes to who's
25:57
going to be making the decisions
25:59
and what these decisions are going
26:01
to be. I think the U.S.
26:03
government certainly feels
26:05
like we're at
26:07
a moment where we need
26:09
a different approach to
26:11
controlling what gets
26:14
put out there in terms of
26:15
frontier models, not just from a
26:17
national security point of view, but
26:19
just from, you know, we don't
26:21
know the unintended
26:23
consequences of
26:25
widespread access to technology
26:28
as powerful as this.
26:29
And so I expect that over the next
26:31
year or two we're going to have more
26:33
incidents like this where
26:36
we're going to be making it up as we
26:38
go along in a sense and
26:40
it's going to both the government
26:42
doing that and the platforms
26:44
doing it.
26:45
I think from China's perspective,
26:47
we heard too much national
26:49
security concerns of
26:51
the United States, from
26:53
export controls to
26:55
advanced chips and now advanced
26:57
AI models.
26:58
And at G7, and they were
27:00
talking about trusted allies will
27:02
be granted access to their most
27:04
advanced AI model.
27:06
So that really speaks
27:08
to the necessity of
27:10
AI sovereignty issue.
27:13
This cross-border technology
27:15
cooperation is facing the challenge
27:17
of more fragmentation,
27:19
especially due to geopolitical
27:21
tensions.
27:22
So how do you see the future
27:24
about AI cooperation
27:27
going forward?
27:28
It seems the United States is
27:30
building walls around it.
27:32
China on the other side is
27:34
advocating open-sourced global
27:36
collaboration.
27:38
So your thoughts on that?
27:40
It's helpful to think about AI
27:42
sovereignty in the broader context
27:44
of what's happened with digital
27:46
sovereignty over the last
27:48
10 or 15 years.
27:50
There has been a recognition
27:52
by many countries that
27:53
part of a nation's strategic
27:56
autonomy is going to be shaped
27:58
by having some
28:00
indigenous
28:02
digital capabilities or
28:05
not being too dependent
28:07
on a different nation state
28:09
for critical digital infrastructure.
28:12
And AI happens to be the latest such
28:14
digital infrastructure.
28:15
And I think both the US and
28:17
China and to some extent the EU.
28:20
I mean, the US has,
28:22
you know, without explicitly
28:25
saying we want to be digitally
28:27
sovereign has, gotten
28:29
used to being the leader in
28:32
different technologies and just
28:34
assuming that they will always have
28:35
access to these technologies because
28:37
it's their companies producing them.
28:39
I think China has proactively
28:41
invested and
28:43
implemented industrial
28:45
policy that has
28:47
made it a separate digital
28:49
leader. If you look at the rest of
28:50
the world, it's very unlikely
28:52
that any other
28:54
country is going to have
28:56
anything resembling
28:58
blanket AI sovereignty,
29:01
where everything is produced by
29:03
the country. I personally don't
29:05
think that that's necessary or
29:07
pragmatic.
29:08
I don't any country has
29:11
full stack AI sovereignty
29:13
anyway.
29:15
We have semiconductor dependencies,
29:17
manufacturing dependencies, cloud
29:19
dependencies.
29:20
I think the US and China
29:22
will come up with
29:24
their own ways of
29:27
drawing a circle around
29:29
what is autonomous and
29:32
what preserves their strategic
29:34
autonomy. For most of the rest
29:36
of the world, it's going
29:38
to have to be making
29:40
choices about what
29:42
layer I'm going to
29:44
gain autonomy in, because nobody
29:46
else is going to be able to
29:49
make the semiconductors themselves.
29:51
It may not be in the best interest
29:53
of most countries to say
29:55
we want our frontier models
29:57
to be domestically produced.
29:59
This may not the best thing
30:00
for their citizens, right?
30:02
And so it's really a question
30:04
of what is the layer of
30:06
AI that is most
30:08
important for me to have strategic
30:11
autonomy?
30:12
Should I do it completely by myself
30:14
or should I form an alliance?
30:16
I think for smaller countries, this
30:17
is certainly sort of a
30:19
valid question.
30:21
And an aspect that I've
30:23
seen surprisingly be
30:25
of considerable importance to a lot
30:27
of countries is the issue of
30:29
cultural autonomy, which has less
30:31
to do with
30:33
sort of, I want the technology
30:35
to be domestically produced so
30:37
I can preserve my
30:39
bargaining power, but more about
30:41
I don't want a foreign technology
30:44
teaching my second graders in
30:47
AI-enabled classrooms, because
30:49
I want the culture
30:51
that they learn when they're in
30:53
school to be my country's culture.
30:55
So there are no easy answers
30:57
here. I think that there's an
30:59
argument to be made for openness.
31:01
There's also an argument to be
31:03
against openness from
31:05
a security point of view.
31:07
So it all remains to be seen.
31:08
Just going back to the jobs for a
31:10
moment, there is a concern
31:12
that we might lose
31:13
a generation to AI.
31:15
What do you say about that?
31:16
Well, for
31:18
the generation that is
31:20
in high school, in college,
31:22
just starting their careers, it's a
31:24
difficult time.
31:25
Whether we lose this generation
31:27
to AI is going to depend
31:29
in part on the generation.
31:31
Unfortunately for them compared
31:33
to people who were in their position
31:35
10 years ago, there
31:37
isn't a predictable path
31:39
to starting their career.
31:41
It's something they're going to have
31:42
to make up as they go along.
31:44
So it's going to take a lot more
31:46
resilience and adaptability.
31:48
I do think that during this period
31:50
of adjustment, the uncertainty
31:52
will be overwhelming for a fraction
31:54
of them. But I think for a faction
31:56
of them, they will look back in
31:58
10 years and say,
32:00
this was a period of
32:02
such a great opportunity
32:04
compared to 2016
32:06
where everything was stable and
32:08
predictable.
32:09
Arun Sundararajan, thanks so much
32:10
for joining us on Radio Davos
32:12
and on CGTN.
32:14
Thank you for having me.
32:15
This was such a fabulous
32:16
conversation.
32:19
And thanks to Guan.
32:21
It's been great to collaborate with
32:23
you.
32:23
Yes, it's very interesting
32:25
and I hope I can join you
32:27
again in the future.
32:30
It would be great. You can find Radio
32:31
Davos wherever you get podcasts
32:33
or go to wef.ch/podcasts
32:36
where you'll find all the forums
32:37
podcasts including Meet the Leader
32:39
and Agenda Dialogues.
32:42
Thanks very much to Arun,
32:44
to Guan.
32:44
Thanks to you for listening and
32:47
Radio Davo will be back very soon.
32:49
Goodbye.