back to index

Kai-Fu Lee: AI Superpowers - China and Silicon Valley | Lex Fridman Podcast #27


small model | large model

link |
00:00:00.000
The following is a conversation with Kai Fu Li.
link |
00:00:02.960
He's the chairman and CEO of Sinovation Ventures
link |
00:00:06.520
that manages a $2 billion dual currency investment fund
link |
00:00:10.560
with a focus on developing the next generation
link |
00:00:13.160
of Chinese high tech companies.
link |
00:00:15.440
He's the former president of Google China
link |
00:00:17.840
and the founder of what is now called Microsoft Research
link |
00:00:20.880
Asia, an institute that trained many
link |
00:00:24.160
of the artificial intelligence leaders in China,
link |
00:00:26.520
including CTOs or AI execs at Baidu, Tencent, Alibaba,
link |
00:00:32.080
Lenovo, and Huawei.
link |
00:00:34.840
He was named one of the 100 most influential people
link |
00:00:38.520
in the world by Time Magazine.
link |
00:00:40.680
He's the author of seven bestselling books in Chinese
link |
00:00:43.880
and most recently, the New York Times bestseller called
link |
00:00:47.080
AI Superpowers, China, Silicon Valley,
link |
00:00:50.600
and the New World Order.
link |
00:00:52.760
He has unparalleled experience in working across major tech
link |
00:00:57.200
companies and governments on applications of AI.
link |
00:01:00.120
And so he has a unique perspective
link |
00:01:02.440
on global innovation in the future of AI
link |
00:01:05.080
that I think is important to listen to and think about.
link |
00:01:09.000
This is the Artificial Intelligence Podcast.
link |
00:01:11.960
If you enjoy it, subscribe on YouTube and iTunes,
link |
00:01:15.240
support it on Patreon, or simply connect with me on Twitter
link |
00:01:18.880
at Lex Freedman.
link |
00:01:21.040
And now, here's my conversation with Kaifu Li.
link |
00:01:26.120
I immigrated from Russia to US when I was 13.
link |
00:01:29.440
You immigrated to US at about the same age.
link |
00:01:32.480
The Russian people, the American people, the Chinese people,
link |
00:01:35.920
each have a certain soul, a spirit,
link |
00:01:39.440
that permeates throughout the generations.
link |
00:01:42.080
So maybe it's a little bit of a poetic question,
link |
00:01:45.120
but could you describe your sense of what
link |
00:01:49.240
defines the Chinese soul?
link |
00:01:52.080
I think the Chinese soul of people today, right,
link |
00:01:56.160
we're talking about people who have had centuries of burden
link |
00:02:02.000
because of the poverty that the country has gone through
link |
00:02:05.240
and suddenly shined with hope of prosperity
link |
00:02:10.560
in the past 40 years as China opened up
link |
00:02:13.440
and embraced market economy.
link |
00:02:16.440
And undoubtedly, there are two sets of pressures
link |
00:02:20.200
on the people, that of the tradition,
link |
00:02:24.160
that of facing difficult situations,
link |
00:02:28.040
and that of hope of wanting to be the first
link |
00:02:31.160
to become successful and wealthy,
link |
00:02:33.840
so that it's a very strong hunger and strong desire
link |
00:02:38.360
and strong work ethic that drives China forward.
link |
00:02:41.160
And is there roots to not just this generation,
link |
00:02:43.960
but before, that's deeper than just
link |
00:02:47.880
the new economic developments?
link |
00:02:50.080
Is there something that's unique to China
link |
00:02:52.520
that you could speak to that's in the people?
link |
00:02:54.960
Yeah.
link |
00:02:56.000
Well, the Chinese tradition is about excellence,
link |
00:03:00.280
dedication, and results.
link |
00:03:02.680
And the Chinese exams and study subjects in schools
link |
00:03:07.240
have traditionally started from memorizing 10,000 characters,
link |
00:03:11.080
not an easy task to start with.
link |
00:03:13.600
And further by memorizing historic philosophers,
link |
00:03:17.640
literature, poetry.
link |
00:03:19.000
So it really is probably the strongest road
link |
00:03:22.480
learning mechanism created to make sure people had good memory
link |
00:03:26.920
and remembered things extremely well.
link |
00:03:30.080
That, I think, at the same time suppresses
link |
00:03:33.720
the breakthrough innovation.
link |
00:03:37.360
And also enhances the speed execution get results.
link |
00:03:42.520
And that, I think, characterizes the historic basis of China.
link |
00:03:47.400
That's interesting, because there's echoes of that
link |
00:03:49.160
in Russian education as well as rote memorization.
link |
00:03:52.080
So you memorize a lot of poetry.
link |
00:03:53.800
I mean, there's just an emphasis on perfection in all forms
link |
00:03:59.240
that's not conducive to perhaps what you're speaking to,
link |
00:04:02.240
which is creativity.
link |
00:04:03.640
But you think that kind of education
link |
00:04:05.640
holds back the innovative spirit that you
link |
00:04:09.040
might see in the United States?
link |
00:04:10.960
Well, it holds back the breakthrough innovative spirit
link |
00:04:14.840
that we see in the United States.
link |
00:04:16.480
But it does not hold back the valuable execution oriented,
link |
00:04:21.880
result oriented value creating engines, which we see China
link |
00:04:26.320
being very successful.
link |
00:04:27.960
So is there a difference between a Chinese AI engineer
link |
00:04:32.320
today and an American AI engineer perhaps rooted
link |
00:04:35.600
in the culture that we just talked about or the education
link |
00:04:38.320
or the very soul of the people or no?
link |
00:04:41.160
And what would your advice be to each
link |
00:04:43.720
if there's a difference?
link |
00:04:45.520
Well, there's a lot that's similar,
link |
00:04:47.120
because AI is about mastering sciences,
link |
00:04:51.240
about using known technologies and trying new things.
link |
00:04:54.880
But it's also about picking from many parts of possible networks
link |
00:04:59.760
to use and different types of parameters to tune.
link |
00:05:02.920
And that part is somewhat rote.
link |
00:05:05.280
And it is also, as anyone who's built AI products,
link |
00:05:09.040
can tell you a lot about cleansing the data.
link |
00:05:12.680
Because AI runs better with more data.
link |
00:05:15.200
And data is generally unstructured, errorful,
link |
00:05:20.160
and unclean.
link |
00:05:22.360
And the effort to clean the data is immense.
link |
00:05:26.280
So I think the better part of the American AI engineering
link |
00:05:32.280
process is to try new things, to do things people haven't done
link |
00:05:36.840
before, and to use technology to solve most, if not all,
link |
00:05:41.840
problems.
link |
00:05:43.480
So to make the algorithm work despite not so great data,
link |
00:05:47.160
find error tolerant ways to deal with the data.
link |
00:05:50.680
The Chinese way would be to basically enumerate,
link |
00:05:55.960
to the fullest extent, all the possible ways
link |
00:05:58.560
by a lot of machines, try lots of different ways
link |
00:06:01.000
to get it to work, and spend a lot of resources and money
link |
00:06:05.320
and time cleaning up data.
link |
00:06:07.720
That means the AI engineer may be writing data cleansing
link |
00:06:11.880
algorithms, working with thousands of people
link |
00:06:15.600
who label or correct or do things with the data.
link |
00:06:19.160
That is the incredible hard work that
link |
00:06:21.920
might lead to better results.
link |
00:06:24.040
So the Chinese engineer would rely on and ask for more and more
link |
00:06:28.240
data and find ways to cleanse them and make them work
link |
00:06:31.120
in the system, and probably less time thinking
link |
00:06:34.200
about new algorithms that can overcome data or other issues.
link |
00:06:39.320
So where's your intuition?
link |
00:06:40.560
What do you think the biggest impact the next 10 years
link |
00:06:43.160
lies?
link |
00:06:43.920
Is it in some breakthrough algorithms?
link |
00:06:47.120
Or is it in just this at scale rigor, a rigorous approach
link |
00:06:53.920
to data, cleaning data, organizing data
link |
00:06:57.120
onto the same algorithms?
link |
00:06:58.440
What do you think the big impact in the applied world is?
link |
00:07:02.600
Well, if you're really in the company
link |
00:07:04.560
and you have to deliver results, using known techniques
link |
00:07:08.400
and enhancing data seems like the more expedient approach
link |
00:07:12.240
that's very low risk and likely to generate
link |
00:07:15.640
better and better results.
link |
00:07:17.200
And that's why the Chinese approach has done quite well.
link |
00:07:20.520
Now, there are a lot of more challenging startups
link |
00:07:24.240
and problems, such as autonomous vehicles,
link |
00:07:28.440
medical diagnosis, that existing algorithms probably
link |
00:07:32.560
won't solve.
link |
00:07:34.240
And that would put the Chinese approach more challenged
link |
00:07:38.680
and give them more breakthrough innovation approach, more
link |
00:07:43.720
of an edge on those kinds of problems.
link |
00:07:45.440
So let me talk to that a little more.
link |
00:07:47.040
So my intuition, personally, is that data
link |
00:07:50.960
can take us extremely far.
link |
00:07:53.680
So you brought up autonomous vehicles and medical diagnosis.
link |
00:07:56.480
So your intuition is that huge amounts of data
link |
00:08:00.080
might not be able to completely help us solve that problem.
link |
00:08:04.000
Right.
link |
00:08:04.600
So breaking that down further, autonomous vehicle,
link |
00:08:08.080
I think huge amounts of data probably
link |
00:08:10.080
will solve trucks driving on highways, which
link |
00:08:13.360
will deliver significant value.
link |
00:08:15.640
And China will probably lead in that.
link |
00:08:19.320
And full L5 autonomous is likely to require new technologies
link |
00:08:24.880
we don't yet know.
link |
00:08:26.320
And that might require academia and great industrial research,
link |
00:08:30.320
both innovating and working together.
link |
00:08:32.480
And in that case, US has an advantage.
link |
00:08:35.360
So the interesting question there is,
link |
00:08:37.040
I don't know if you're familiar on the autonomous vehicle
link |
00:08:39.280
space and the developments with Tesla and Elon Musk,
link |
00:08:43.480
where they are, in fact, a full steam ahead
link |
00:08:49.400
into this mysterious, complex world of full autonomy, L5,
link |
00:08:53.480
L4, L5.
link |
00:08:55.080
And they're trying to solve that purely with data.
link |
00:08:58.800
So the same kind of thing that you're saying
link |
00:09:00.800
is just for highway, which is what a lot of people
link |
00:09:03.200
share your intuition, they're trying to solve with data.
link |
00:09:07.200
It's just to linger on that moment further.
link |
00:09:09.320
Do you think possible for them to achieve success
link |
00:09:13.600
with simply just a huge amount of this training
link |
00:09:17.040
on edge cases, on difficult cases in urban environments,
link |
00:09:20.440
not just highway and so on?
link |
00:09:22.840
I think they'll be very hard.
link |
00:09:24.480
One could characterize Tesla's approach as kind
link |
00:09:27.680
of a Chinese strength approach, gather all the data you can,
link |
00:09:31.600
and hope that will overcome the problems.
link |
00:09:34.000
But in autonomous driving, clearly a lot of the decisions
link |
00:09:38.480
aren't merely solved by aggregating data
link |
00:09:41.480
and having feedback loop.
link |
00:09:43.520
There are things that are more akin to human thinking.
link |
00:09:48.040
And how would those be integrated and built?
link |
00:09:51.680
There has not yet been a lot of success
link |
00:09:54.000
integrating human intelligence or, you know,
link |
00:09:57.200
colored expert systems, if you will,
link |
00:09:58.800
even though that's a taboo word with the machine learning.
link |
00:10:02.960
And the integration of the two types of thinking
link |
00:10:05.600
hasn't yet been demonstrated.
link |
00:10:07.840
And the question is, how much can you
link |
00:10:09.600
push a purely machine learning approach?
link |
00:10:12.440
And of course, Tesla also has an additional constraint
link |
00:10:15.480
that they don't have all the sensors.
link |
00:10:18.520
I know that they think it's foolish to use LIDARS,
link |
00:10:21.120
but that's clearly a one less, very valuable and reliable
link |
00:10:25.920
source of input that they're foregoing, which
link |
00:10:29.200
may also have consequences.
link |
00:10:32.440
I think the advantage, of course,
link |
00:10:33.840
is capturing data that no one has ever seen before.
link |
00:10:37.040
And in some cases, such as computer vision and speech
link |
00:10:41.040
recognition, I have seen Chinese companies accumulate data
link |
00:10:44.800
that's not seen anywhere in the Western world,
link |
00:10:47.320
and they have delivered superior results.
link |
00:10:50.200
But then speech recognition and object recognition
link |
00:10:53.720
are relatively suitable problems for deep learning
link |
00:10:57.080
and don't have the potentially need for the human intelligence
link |
00:11:02.440
analytical planning elements.
link |
00:11:04.440
And the same on the speech recognition side,
link |
00:11:06.400
your intuition that speech recognition and the machine
link |
00:11:09.440
learning approaches to speech recognition
link |
00:11:11.440
won't take us to a conversational system that
link |
00:11:14.600
can pass the Turing test, which is maybe akin to what
link |
00:11:19.160
driving is.
link |
00:11:20.040
So it needs to have something more than just simply simple
link |
00:11:25.120
language understanding, simple language generation.
link |
00:11:27.480
Roughly right, I would say that based on purely machine
link |
00:11:32.000
learning approaches, it's hard to imagine.
link |
00:11:35.160
It could lead to a full conversational experience
link |
00:11:40.520
across arbitrary domains, which is akin to L5.
link |
00:11:44.600
I'm a little hesitant to use the word Turing test,
link |
00:11:46.920
because the original definition was probably too easy.
link |
00:11:50.280
We probably do that.
link |
00:11:52.320
The spirit of the Turing test is what I was referring to.
link |
00:11:55.280
Of course.
link |
00:11:56.520
So you've had major leadership research positions
link |
00:11:59.400
at Apple, Microsoft, Google.
link |
00:12:01.640
So continuing on the discussion of America, Russia, Chinese soul
link |
00:12:06.320
and culture and so on, what is the culture of Silicon
link |
00:12:10.520
Valley in contrast to China and maybe US broadly?
link |
00:12:16.400
And what is the unique culture of each of these three
link |
00:12:19.920
major companies, in your view?
link |
00:12:22.040
I think in aggregate, Silicon Valley companies,
link |
00:12:25.120
we could probably include Microsoft in that,
link |
00:12:27.200
even though they're not in the Valley,
link |
00:12:29.120
is really dream big and have visionary goals
link |
00:12:33.960
and believe that technology will conquer all
link |
00:12:37.920
and also the self confidence and the self entitlement
link |
00:12:42.240
that whatever they produce, the whole world should use
link |
00:12:45.440
and must use.
link |
00:12:47.240
And those are historically important, I think.
link |
00:12:54.080
Steve Jobs's famous quote that he doesn't do focus groups.
link |
00:12:59.120
He looks in the mirror and asks the person in the mirror,
link |
00:13:02.360
what do you want?
link |
00:13:03.520
And that really is an inspirational comment
link |
00:13:07.000
that says the great company shouldn't just ask users
link |
00:13:10.480
what they want, but develop something
link |
00:13:13.240
that users will know they want when they see it,
link |
00:13:16.200
but they could never come up with themselves.
link |
00:13:18.960
I think that is probably the most exhilarating description
link |
00:13:23.880
of what the essence of Silicon Valley is,
link |
00:13:26.560
that this brilliant idea could cause you to build something
link |
00:13:31.840
that couldn't come out of the focus groups or A.B. tests.
link |
00:13:35.520
And iPhone would be an example of that.
link |
00:13:38.040
No one in the age of BlackBerry would write down
link |
00:13:40.560
they want an iPhone or multi touch, a browser,
link |
00:13:43.720
might be another example.
link |
00:13:44.800
No one would say they want that in the days of FTP,
link |
00:13:47.520
but once they see it, they want it.
link |
00:13:49.440
So I think that is what Silicon Valley is best at.
link |
00:13:55.680
But it also came with a lot of success.
link |
00:13:58.920
These products became global platforms,
link |
00:14:01.960
and there were basically no competitors anywhere.
link |
00:14:05.080
And that has also led to a belief
link |
00:14:08.400
that these are the only things that one should do,
link |
00:14:13.240
that companies should not tread on other companies territory,
link |
00:14:17.960
so that a Groupon and a Yelp and an OpenTable
link |
00:14:24.040
and the Grubhub would each feel,
link |
00:14:26.240
okay, I'm not going to do the other companies business
link |
00:14:28.520
because that would not be the pride of innovating
link |
00:14:33.280
what each of these four companies have innovated.
link |
00:14:36.920
But I think the Chinese approach is do whatever it takes to win.
link |
00:14:42.720
And it's a winner take all market.
link |
00:14:45.000
And in fact, in the internet space,
link |
00:14:47.200
the market leader will get predominantly all the value
link |
00:14:50.840
extracted out of the system.
link |
00:14:53.320
And the system isn't just defined as one narrow category,
link |
00:14:59.600
but gets broader and broader.
link |
00:15:01.360
So it's amazing ambition for success and domination
link |
00:15:07.960
of increasingly larger product categories
link |
00:15:11.760
leading to clear market winner status
link |
00:15:15.080
and the opportunity to extract tremendous value.
link |
00:15:19.120
And that develops a practical, result oriented,
link |
00:15:25.840
ultra ambitious winner take all gladiatorial mentality.
link |
00:15:31.520
And if what it takes is to build what the competitors built,
link |
00:15:37.400
essentially a copycat, that can be done without infringing laws.
link |
00:15:41.920
If what it takes is to satisfy a foreign country's need
link |
00:15:46.280
by forking the code base and building something
link |
00:15:48.480
that looks really ugly and different, they'll do it.
link |
00:15:51.440
So it's contrasted very sharply with the Silicon Valley approach.
link |
00:15:56.280
And I think the flexibility and the speed and execution
link |
00:16:00.080
has helped the Chinese approach.
link |
00:16:01.960
And I think the Silicon Valley approach
link |
00:16:05.040
is potentially challenged if every Chinese entrepreneur is
link |
00:16:10.280
learning from the whole world, US and China,
link |
00:16:13.200
and the American entrepreneurs only look internally
link |
00:16:16.280
and write off China as a copycat.
link |
00:16:19.600
And the second part of your question about the three
link |
00:16:22.880
companies.
link |
00:16:23.520
The unique elements of the three companies, perhaps.
link |
00:16:26.000
Yeah.
link |
00:16:26.840
I think Apple represents, while the user, please the user,
link |
00:16:33.080
and the essence of design and brand,
link |
00:16:38.520
and it's the one company and perhaps the only tech company
link |
00:16:44.080
that draws people with a strong, serious desire
link |
00:16:49.920
for the product and the willingness to pay a premium
link |
00:16:53.560
because of the halo effect of the brand, which
link |
00:16:57.160
came from the attention to detail and great respect
link |
00:17:00.960
for user needs.
link |
00:17:03.360
Microsoft represents a platform approach
link |
00:17:09.200
that builds giant products that become very strong modes
link |
00:17:14.280
that others can't do because it's
link |
00:17:17.680
well architected at the bottom level
link |
00:17:21.480
and the work is efficiently delegated to individuals
link |
00:17:26.640
and then the whole product is built
link |
00:17:30.360
by adding small parts that sum together.
link |
00:17:33.560
So it's probably the most effective high tech assembly
link |
00:17:37.760
line that builds a very difficult product
link |
00:17:40.480
that the whole process of doing that
link |
00:17:44.800
is kind of a differentiation and something competitors
link |
00:17:50.800
can't easily repeat.
link |
00:17:52.480
Are there elements of the Chinese approach
link |
00:17:54.800
in the way Microsoft went about assembling those little pieces
link |
00:17:59.280
and essentially dominating the market for a long time?
link |
00:18:03.920
Or do you see those as distinct?
link |
00:18:05.640
I think there are elements that are the same.
link |
00:18:08.240
I think the three American companies
link |
00:18:10.440
that had or have Chinese characteristics,
link |
00:18:13.880
and obviously as well as American characteristics,
link |
00:18:16.080
are Microsoft, Facebook, and Amazon.
link |
00:18:20.400
Yes, that's right, Amazon.
link |
00:18:21.720
Because these are companies that will tenaciously
link |
00:18:25.560
go after adjacent markets, build up strong product offering,
link |
00:18:31.320
and find ways to extract greater value from a sphere that's
link |
00:18:38.200
ever increasing.
link |
00:18:39.960
And they understand the value of the platforms.
link |
00:18:43.520
So that's the similarity.
link |
00:18:45.600
And then with Google, I think it's a genuinely value oriented
link |
00:18:53.760
company that does have a heart and soul
link |
00:18:56.960
and that wants to do great things for the world
link |
00:18:59.760
by connecting information and that has also
link |
00:19:06.040
very strong technology genes and wants to use technology
link |
00:19:13.280
and has found out of the box ways to use technology
link |
00:19:19.080
to deliver incredible value to the end user.
link |
00:19:23.680
We can look at Google, for example.
link |
00:19:25.240
You mentioned heart and soul.
link |
00:19:28.040
There seems to be an element where Google
link |
00:19:31.840
is after making the world better.
link |
00:19:34.840
There's a more positive view.
link |
00:19:36.520
I mean, they used to have the slogan, don't be evil.
link |
00:19:38.960
And Facebook a little bit more has a negative tend to it,
link |
00:19:43.120
at least in the perception of privacy and so on.
link |
00:19:46.000
Do you have a sense of how these different companies can
link |
00:19:51.280
achieve, because you've talked about how much
link |
00:19:53.400
we can make the world better in all these kinds of ways
link |
00:19:55.600
with AI, what is it about a company that can make,
link |
00:19:59.360
give it a heart and soul, gain the trust of the public,
link |
00:20:03.200
and just actually just not be evil and do good for the world?
link |
00:20:08.000
It's really hard.
link |
00:20:09.000
And I think Google has struggled with that.
link |
00:20:13.120
First, they don't do evil.
link |
00:20:15.160
Mantra is very dangerous, because every employee's
link |
00:20:18.880
definition of evil is different.
link |
00:20:20.800
And that has led to some difficult employee situations
link |
00:20:23.800
for them.
link |
00:20:25.240
So I don't necessarily think that's a good value statement.
link |
00:20:29.520
But just watching the kinds of things
link |
00:20:31.840
Google or its parent company Alphabet does in new areas
link |
00:20:36.440
like health care, like eradicating mosquitoes,
link |
00:20:40.440
things that are really not in the business
link |
00:20:42.360
of a internet tech company, I think
link |
00:20:45.040
that shows that there is a heart and soul
link |
00:20:47.200
and desire to do good and willingness to put in the resources
link |
00:20:53.920
to do something when they see it's good, they will pursue it.
link |
00:20:58.280
That doesn't necessarily mean it has
link |
00:21:00.640
all the trust of the users.
link |
00:21:02.520
I realize while most people would view Facebook
link |
00:21:06.400
as the primary target of their recent unhappiness
link |
00:21:09.760
about Silicon Valley companies, many would put Google
link |
00:21:12.720
in that category.
link |
00:21:14.080
And some have named Google's business practices
link |
00:21:16.800
as predatory also.
link |
00:21:19.840
So it's kind of difficult to have the two parts of a body.
link |
00:21:24.240
The brain wants to do what it's supposed to do for a shareholder,
link |
00:21:28.080
maximize profit.
link |
00:21:29.280
And then the heart and soul wants
link |
00:21:30.880
to do good things that may run against what the brain wants to do.
link |
00:21:36.120
So in this complex balancing that these companies have to do,
link |
00:21:40.320
you've mentioned that you're concerned about a future where
link |
00:21:44.520
too few companies like Google, Facebook, Amazon
link |
00:21:47.360
are controlling our data or are controlling too much
link |
00:21:51.560
of our digital lives.
link |
00:21:53.360
Can you elaborate on this concern?
link |
00:21:55.400
Perhaps do you have a better way forward?
link |
00:21:58.640
I think I'm hardly the most vocal complainer of this.
link |
00:22:05.000
There are a lot louder complainers out there.
link |
00:22:07.280
I do observe that having a lot of data
link |
00:22:11.840
does perpetuate their strength and limits
link |
00:22:16.120
competition in many spaces.
link |
00:22:19.400
But I also believe AI is much broader than the internet space.
link |
00:22:24.200
So the entrepreneurial opportunities
link |
00:22:26.280
still exists in using AI to empower
link |
00:22:30.480
financial, retail, manufacturing, education,
link |
00:22:34.160
applications.
link |
00:22:35.480
So I don't think it's quite a case of full monopolistic dominance
link |
00:22:39.800
that totally stifles innovation.
link |
00:22:43.960
But I do believe in their areas of strength
link |
00:22:46.400
it's hard to dislodge them.
link |
00:22:49.760
I don't know if I have a good solution.
link |
00:22:53.280
Probably the best solution is let the entrepreneurial VC
link |
00:22:57.160
ecosystem work well and find all the places that
link |
00:23:00.840
can create the next Google, the next Facebook.
link |
00:23:04.200
So there will always be increasing number of challengers.
link |
00:23:08.560
In some sense, that has happened a little bit.
link |
00:23:11.360
You see Uber, Airbnb having emerged despite the strength
link |
00:23:15.760
of the big three.
link |
00:23:19.040
And I think China as an environment
link |
00:23:22.400
may be more interesting for the emergence.
link |
00:23:25.280
Because if you look at companies between, let's say,
link |
00:23:28.920
$50 to $300 billion, China has emerged more of such companies
link |
00:23:36.320
than the US in the last three to four years.
link |
00:23:39.880
Because of the larger marketplace,
link |
00:23:42.120
because of the more fearless nature of the entrepreneurs.
link |
00:23:47.000
And the Chinese giants are just as powerful as American ones.
link |
00:23:50.840
Tencent Alibaba are very strong.
link |
00:23:52.920
But Bytes Dance has emerged worth $75 billion.
link |
00:23:57.040
And financial, while it's Alibaba affiliated,
link |
00:24:00.120
it's nevertheless independent and worth $150 billion.
link |
00:24:03.920
And so I do think if we start to extend
link |
00:24:08.280
to traditional businesses, we will see very valuable companies.
link |
00:24:12.640
So it's probably not the case that in five or 10 years,
link |
00:24:18.120
we'll still see the whole world with these five companies
link |
00:24:20.920
having such dominance.
link |
00:24:22.680
So you've mentioned a couple of times
link |
00:24:26.040
this fascinating world of entrepreneurship
link |
00:24:27.840
in China of the fearless nature of the entrepreneurs.
link |
00:24:31.080
So can you maybe talk a little bit
link |
00:24:32.640
about what it takes to be an entrepreneur in China?
link |
00:24:35.520
What are the strategies that are undertaken?
link |
00:24:38.240
What are the ways that you success?
link |
00:24:41.120
What is the dynamic of VCF funding,
link |
00:24:43.960
of the way the government helps companies, and so on?
link |
00:24:46.480
What are the interesting aspects here that are distinct from,
link |
00:24:49.520
that are different from the Silicon Valley world
link |
00:24:52.880
of entrepreneurship?
link |
00:24:55.240
Well, many of the listeners probably
link |
00:24:58.080
still would brand Chinese entrepreneur as copycats.
link |
00:25:03.000
And no doubt, 10 years ago, that would not
link |
00:25:06.120
be an inaccurate description.
link |
00:25:09.080
Back 10 years ago, an entrepreneur probably
link |
00:25:12.320
could not get funding if he or she could not
link |
00:25:14.840
describe what product he or she is copying from the US.
link |
00:25:20.400
The first question is, who has proven this business model,
link |
00:25:23.520
which is a nice way of asking, who are you copying?
link |
00:25:27.200
And that reason is understandable,
link |
00:25:29.520
because China had a much lower internet penetration
link |
00:25:34.840
and didn't have enough indigenous experience
link |
00:25:40.920
to build innovative products.
link |
00:25:43.200
And secondly, internet was emerging.
link |
00:25:47.600
Link startup was the way to do things,
link |
00:25:49.800
building a first minimally viable product,
link |
00:25:52.920
and then expanding was the right way to go.
link |
00:25:55.320
And the American successes have given a shortcut
link |
00:25:59.480
that if you build your minimally viable product based
link |
00:26:02.840
on an American product, it's guaranteed
link |
00:26:05.040
to be a decent starting point.
link |
00:26:06.720
Then you tweak it afterwards.
link |
00:26:08.400
So as long as there are no IP infringement, which,
link |
00:26:11.720
as far as I know, there hasn't been in the mobile and AI
link |
00:26:15.080
spaces, that's a much better shortcut.
link |
00:26:19.360
And I think Silicon Valley would view that as still not
link |
00:26:23.720
very honorable, because that's not your own idea to start with.
link |
00:26:29.200
But you can't really, at the same time,
link |
00:26:32.600
believe every idea must be your own
link |
00:26:35.160
and believe in the link startup methodology,
link |
00:26:38.120
because link startup is intended to try many, many things
link |
00:26:41.880
and then converge when that works.
link |
00:26:44.240
And it's meant to be iterated and changed.
link |
00:26:46.720
So finding a decent starting point without legal violations,
link |
00:26:51.240
there should be nothing morally dishonorable about that.
link |
00:26:55.520
So just a quick pause on that.
link |
00:26:57.080
It's fascinating that that's why is that not honorable, right?
link |
00:27:01.920
It's exactly as you formulated.
link |
00:27:04.680
It seems like a perfect start for business
link |
00:27:08.040
is to take a look at Amazon and say, OK,
link |
00:27:12.440
we'll do exactly what Amazon is doing.
link |
00:27:14.560
Let's start there in this particular market.
link |
00:27:16.800
And then let's out innovate them from that starting point.
link |
00:27:20.520
Yes. Come up with new ways.
link |
00:27:22.200
I mean, is it wrong to be, except the word copycat just
link |
00:27:26.520
sounds bad, but is it wrong to be a copycat?
link |
00:27:28.800
It just seems like a smart strategy.
link |
00:27:31.640
But yes, doesn't have a heroic nature to it
link |
00:27:35.800
that Steve Jobs, Elon Musk, sort of in something completely
link |
00:27:42.280
coming up with something completely new.
link |
00:27:43.880
Yeah, I like the way you describe it.
link |
00:27:45.480
It's a nonheroic, acceptable way to start the company.
link |
00:27:50.440
And maybe more expedient.
link |
00:27:52.840
So that's, I think, a baggage for Silicon Valley,
link |
00:27:58.920
that if it doesn't let go, then it
link |
00:28:01.320
may limit the ultimate ceiling of the company.
link |
00:28:05.160
Take Snapchat as an example.
link |
00:28:07.200
I think Evan's brilliant.
link |
00:28:09.840
He built a great product.
link |
00:28:11.480
But he's very proud that he wants
link |
00:28:14.160
to build his own features, not copy others.
link |
00:28:16.800
While Facebook was more willing to copy his features,
link |
00:28:21.000
and you see what happens in the competition.
link |
00:28:23.440
So I think putting that handcuff on the company
link |
00:28:27.440
would limit its ability to reach the maximum potential.
link |
00:28:31.560
So back to the Chinese environment,
link |
00:28:33.800
copying was merely a way to learn from the American masters.
link |
00:28:38.400
Just like if we learned to play piano or painting,
link |
00:28:43.480
you start by copying.
link |
00:28:44.560
You don't start by innovating when
link |
00:28:46.160
you don't have the basic skill sets.
link |
00:28:48.200
So very amazingly, the Chinese entrepreneurs
link |
00:28:51.040
about six years ago started to branch off
link |
00:28:56.160
with these lean startups built on American ideas
link |
00:28:59.520
to build better products than American products.
link |
00:29:02.280
But they did start from the American idea.
link |
00:29:04.960
And today, WeChat is better than WhatsApp.
link |
00:29:08.600
Weibo is better than Twitter.
link |
00:29:10.520
Zihu is better than Quora and so on.
link |
00:29:12.920
So that, I think, is Chinese entrepreneurs
link |
00:29:17.000
going to step two.
link |
00:29:18.480
And then step three is once these entrepreneurs have
link |
00:29:21.760
done one or two of these companies,
link |
00:29:23.720
they now look at the Chinese market and the opportunities
link |
00:29:27.400
and come up with ideas that didn't exist elsewhere.
link |
00:29:30.600
So products like and financial under which includes Alipay,
link |
00:29:36.320
which is mobile payments, and also the financial products
link |
00:29:42.080
for loans built on that, and also in education, VIP kid,
link |
00:29:48.560
and in social video, social network, TikTok,
link |
00:29:54.880
and in social eCommerce, Pinduoduo,
link |
00:29:58.640
and then in ride sharing, Mobike.
link |
00:30:01.720
These are all Chinese innovative products
link |
00:30:05.640
that now are being copied elsewhere.
link |
00:30:08.720
So an additional interesting observation
link |
00:30:13.040
is some of these products are built on unique Chinese
link |
00:30:16.000
demographics, which may not work in the US,
link |
00:30:19.360
but may work very well in Southeast Asia, Africa,
link |
00:30:23.160
and other developing worlds that are a few years behind China.
link |
00:30:27.840
And a few of these products maybe are universal
link |
00:30:31.040
and are getting traction even in the United States,
link |
00:30:33.760
such as TikTok.
link |
00:30:35.360
So this whole ecosystem is supported by VCs
link |
00:30:42.080
as a virtuous cycle, because a large market
link |
00:30:44.920
with innovative entrepreneurs will draw a lot of money
link |
00:30:49.400
and then invest in these companies.
link |
00:30:51.560
As the market gets larger and larger,
link |
00:30:54.480
China market is easily three, four times larger than the US.
link |
00:30:58.400
They will create greater value and greater returns
link |
00:31:01.120
for the VCs, thereby raising even more money.
link |
00:31:05.400
So at Sinovation Ventures, our first fund was $15 million.
link |
00:31:10.000
Our last fund was $500 million.
link |
00:31:12.040
So it reflects the valuation of the companies
link |
00:31:16.520
and our us going multi stage and things like that.
link |
00:31:19.840
It also has government support, but not
link |
00:31:23.840
in the way most Americans would think of it.
link |
00:31:26.080
The government actually leaves the entrepreneurial space
link |
00:31:29.520
as a private enterprise, so the self regulating.
link |
00:31:33.200
And the government would build infrastructures
link |
00:31:36.200
that would around it to make it work better.
link |
00:31:39.320
For example, the mass entrepreneur mass innovation
link |
00:31:41.960
plan builds 8,000 incubators.
link |
00:31:44.880
So the pipeline is very strong to the VCs
link |
00:31:48.360
for autonomous vehicles.
link |
00:31:49.680
The Chinese government is building smart highways
link |
00:31:53.280
with sensors, smart cities that separate pedestrians
link |
00:31:56.680
from cars that may allow initially an inferior autonomous
link |
00:32:01.560
vehicle company to launch a car without increasing,
link |
00:32:05.760
with lower casualty, because the roads or the city is smart.
link |
00:32:11.520
And the Chinese government at local levels
link |
00:32:13.800
would have these guiding funds acting as LPs,
link |
00:32:17.360
passive LPs to funds.
link |
00:32:19.400
And when the fund makes money, part of the money made
link |
00:32:23.240
is given back to the GPs and potentially other LPs
link |
00:32:27.280
to increase everybody's return at the expense
link |
00:32:31.960
of the government's return.
link |
00:32:33.680
So that's an interesting incentive
link |
00:32:36.360
that entrusts the task of choosing entrepreneurs to VCs
link |
00:32:41.640
who are better at it than the government
link |
00:32:43.800
by letting some of the profits move that way.
link |
00:32:46.680
So this is really fascinating, right?
link |
00:32:48.720
So I look at the Russian government as a case study
link |
00:32:51.800
where, let me put it this way, there
link |
00:32:54.480
is no such government driven, large scale
link |
00:32:58.520
support of entrepreneurship.
link |
00:33:00.840
And probably the same is true in the United States.
link |
00:33:04.000
But the entrepreneurs themselves kind of find a way.
link |
00:33:07.640
So maybe in a form of advice or explanation,
link |
00:33:11.680
how did the Chinese government arrive to be this way,
link |
00:33:15.560
so supportive on entrepreneurship,
link |
00:33:17.680
to be in this particular way so forward thinking
link |
00:33:21.520
at such a large scale?
link |
00:33:23.120
And also perhaps, how can we copy it in other countries?
link |
00:33:28.280
How can we encourage other governments,
link |
00:33:29.800
like even the United States government,
link |
00:33:31.600
to support infrastructure for autonomous vehicles
link |
00:33:33.760
in that same kind of way, perhaps?
link |
00:33:36.040
Yes.
link |
00:33:36.680
So these techniques are the result of several key things,
link |
00:33:44.440
some of which may be learnable, some of which
link |
00:33:46.480
may be very hard.
link |
00:33:48.440
One is just trial and error and watching
link |
00:33:51.080
what everyone else is doing.
link |
00:33:52.960
I think it's important to be humble and not
link |
00:33:54.960
feel like you know all the answers.
link |
00:33:56.920
The guiding funds idea came from Singapore,
link |
00:33:59.480
which came from Israel.
link |
00:34:01.440
And China made a few tweaks and turned it into a,
link |
00:34:06.080
because the Chinese cities and government officials kind
link |
00:34:09.600
of compete with each other.
link |
00:34:11.320
Because they all want to make their city more successful,
link |
00:34:14.640
so they can get the next level in their political career.
link |
00:34:20.280
And it's somewhat competitive.
link |
00:34:22.320
So the central government made it a bit of a competition.
link |
00:34:25.200
Everybody has a budget.
link |
00:34:26.840
They can put it on AI, or they can put it on bio,
link |
00:34:29.840
or they can put it on energy.
link |
00:34:32.200
And then whoever gets the results, the city shines,
link |
00:34:35.040
the people are better off, the mayor gets a promotion.
link |
00:34:38.000
So the tools is kind of almost like an entrepreneurial
link |
00:34:41.680
environment for local governments
link |
00:34:44.840
to see who can do a better job.
link |
00:34:47.480
And also, many of them tried different experiments.
link |
00:34:52.440
Some have given award to very smart researchers,
link |
00:34:58.440
just give them money and hope they'll start a company.
link |
00:35:00.840
Some have given money to academic research labs,
link |
00:35:05.840
maybe government research labs, to see
link |
00:35:08.440
if they can spin off some companies from the science
link |
00:35:11.920
lab or something like that.
link |
00:35:14.040
Some have tried to recruit overseas Chinese
link |
00:35:17.080
to come back and start companies.
link |
00:35:18.960
And they've had mixed results.
link |
00:35:20.960
The one that worked the best was the guiding funds.
link |
00:35:23.400
So it's almost like a lean startup idea
link |
00:35:25.840
where people try different things in what works, sticks,
link |
00:35:29.160
and everybody copies.
link |
00:35:30.600
So now every city has a guiding fund.
link |
00:35:32.880
So that's how that came about.
link |
00:35:35.680
The autonomous vehicle and the massive spending
link |
00:35:40.400
in highways and smart cities, that's a Chinese way.
link |
00:35:46.080
It's about building infrastructure to facilitate.
link |
00:35:49.480
It's a clear division of the government's responsibility
link |
00:35:52.840
from the market.
link |
00:35:55.400
The market should do everything in a private freeway.
link |
00:36:00.560
But there are things the market can't afford to do,
link |
00:36:02.920
like infrastructure.
link |
00:36:04.520
So the government always appropriates
link |
00:36:08.000
large amounts of money for infrastructure building.
link |
00:36:12.000
This happens with not only autonomous vehicle and AI,
link |
00:36:16.880
but happened with the 3G and 4G.
link |
00:36:20.840
You'll find that the Chinese wireless reception
link |
00:36:25.320
is better than the US, because massive spending that
link |
00:36:28.760
tries to cover the whole country.
link |
00:36:30.720
Whereas in the US, it may be a little spotty.
link |
00:36:34.360
It's a government driven, because I think
link |
00:36:36.160
they view the coverage of cell access and 3G, 4G access
link |
00:36:44.120
to be a governmental infrastructure spending,
link |
00:36:47.080
as opposed to capitalistic.
link |
00:36:49.880
So of course, the state or enterprise
link |
00:36:52.160
is also publicly traded, but they also
link |
00:36:55.000
carry a government responsibility
link |
00:36:57.720
to deliver infrastructure to all.
link |
00:37:00.240
So it's a different way of thinking
link |
00:37:01.880
that may be very hard to inject into Western countries
link |
00:37:05.400
to say starting tomorrow, bandwidth infrastructure
link |
00:37:09.280
and highways are going to be governmental spending
link |
00:37:13.840
with some characteristics.
link |
00:37:16.240
What's your sense, and sorry to interrupt,
link |
00:37:18.240
but because it's such a fascinating point,
link |
00:37:21.680
do you think on the autonomous vehicle space
link |
00:37:25.600
it's possible to solve the problem of full autonomy
link |
00:37:30.120
without significant investment in infrastructure?
link |
00:37:34.040
Well, that's really hard to speculate.
link |
00:37:36.400
I think it's not a yes, no question,
link |
00:37:38.960
but how long does it take question?
link |
00:37:41.920
15 years, 30 years, 45 years.
link |
00:37:45.120
Clearly with infrastructure augmentation,
link |
00:37:48.960
where there's road, the city, or whole city planning,
link |
00:37:52.320
building a new city, I'm sure that will accelerate
link |
00:37:56.440
the day of the L5.
link |
00:37:59.040
I'm not knowledgeable enough, and it's
link |
00:38:01.520
hard to predict even when we're knowledgeable,
link |
00:38:03.920
because a lot of it is speculative.
link |
00:38:07.120
But in the US, I don't think people
link |
00:38:09.800
would consider building a new city the size of Chicago
link |
00:38:13.240
to make it the AI slash autonomous city.
link |
00:38:15.920
There are smaller ones being built, I'm aware of that.
link |
00:38:18.840
But is infrastructure spend really
link |
00:38:21.280
impossible for US or Western countries?
link |
00:38:23.720
I don't think so.
link |
00:38:25.680
The US highway system was built.
link |
00:38:28.920
Was that during President Eisenhower or Kennedy?
link |
00:38:31.960
Eisenhower, yeah.
link |
00:38:33.160
So maybe historians can study how the President Eisenhower
link |
00:38:38.960
get the resources to build this massive infrastructure that
link |
00:38:42.960
surely gave US a tremendous amount of prosperity
link |
00:38:47.560
over the next decade, if not century.
link |
00:38:50.800
If I may comment on that, then, it
link |
00:38:53.240
takes us to artificial intelligence
link |
00:38:54.880
a little bit, because in order to build infrastructure,
link |
00:38:58.080
it creates a lot of jobs.
link |
00:39:00.520
So I'll be actually interested if you
link |
00:39:02.840
would say that you're talking in your book about all kinds
link |
00:39:06.120
of jobs that could and could not be automated.
link |
00:39:08.960
I wonder if building infrastructure
link |
00:39:12.000
is one of the jobs that would not be easily automated,
link |
00:39:15.720
something you can think about, because I think you've mentioned
link |
00:39:18.160
somewhere in a talk, or that there
link |
00:39:21.160
might be, as jobs are being automated,
link |
00:39:24.280
a role for government to create jobs that can't be automated.
link |
00:39:28.160
Yes, I think that's a possibility.
link |
00:39:31.040
Back in the last financial crisis,
link |
00:39:34.280
China put a lot of money to basically give this economy
link |
00:39:40.320
a boost, and a lot of it went into infrastructure building.
link |
00:39:45.520
And I think that's a legitimate way, at the government level,
link |
00:39:49.920
to deal with the employment issues as well as build out
link |
00:39:55.680
the infrastructure, as long as the infrastructures are truly
link |
00:39:58.960
needed, and as long as there is an employment problem, which
link |
00:40:03.160
we don't know.
link |
00:40:04.960
So maybe taking a little step back,
link |
00:40:07.920
if you've been a leader and a researcher in AI
link |
00:40:12.840
for several decades, at least 30 years,
link |
00:40:16.200
so how has AI changed in the West and the East
link |
00:40:21.040
as you've observed, as you've been deep in it
link |
00:40:23.120
over the past 30 years?
link |
00:40:25.120
Well, AI began as the pursuit of understanding
link |
00:40:28.520
human intelligence, and the term itself represents that.
link |
00:40:34.160
But it kind of drifted into the one subarea that
link |
00:40:37.680
worked extremely well, which is machine intelligence.
link |
00:40:40.880
And that's actually more using pattern recognition techniques
link |
00:40:45.080
to basically do incredibly well on a limited domain,
link |
00:40:51.280
large amount of data, but relatively simple kinds
link |
00:40:54.840
of planning, tasks, and not very creative.
link |
00:40:58.720
So we didn't end up building human intelligence.
link |
00:41:02.480
We built a different machine that
link |
00:41:04.760
was a lot better than us, some problems,
link |
00:41:08.040
but nowhere close to us on other problems.
link |
00:41:11.840
So today, I think a lot of people still
link |
00:41:14.200
misunderstand when we say artificial intelligence
link |
00:41:18.080
and what various products can do.
link |
00:41:20.720
People still think it's about replicating human intelligence.
link |
00:41:24.160
But the products out there really
link |
00:41:26.160
are closer to having invented the internet or the spreadsheet
link |
00:41:31.680
or the database and getting broader adoption.
link |
00:41:35.360
And speaking further to the fears, near term fears
link |
00:41:38.400
that people have about AI, so you're commenting
link |
00:41:41.240
on the general intelligence that people
link |
00:41:45.680
in the popular culture from sci fi movies
link |
00:41:48.040
have a sense about AI, but there's practical fears
link |
00:41:50.920
about AI, the kind of narrow AI that you're talking about
link |
00:41:54.800
of automating particular kinds of jobs,
link |
00:41:57.280
and you talk about them in the book.
link |
00:41:59.400
So what are the kinds of jobs in your view
link |
00:42:01.520
that you see in the next five, 10 years beginning
link |
00:42:04.840
to be automated by AI systems algorithms?
link |
00:42:09.240
Yes, this is also maybe a little bit counterintuitive
link |
00:42:13.000
because it's the routine jobs that
link |
00:42:15.440
will be displaced the soonest.
link |
00:42:18.360
And they may not be displaced entirely, maybe 50%, 80%
link |
00:42:23.120
of a job, but when the workload drops by that much,
link |
00:42:26.320
employment will come down.
link |
00:42:28.760
And also another part of misunderstanding
link |
00:42:31.520
is most people think of AI replacing routine jobs,
link |
00:42:35.720
then they think of the assembly line, the workers.
link |
00:42:38.760
Well, that will have some effects,
link |
00:42:40.960
but it's actually the routine white collar workers that's
link |
00:42:44.600
easiest to replace because to replace a white collar worker,
link |
00:42:49.280
you just need software.
link |
00:42:50.720
To replace a blue collar worker,
link |
00:42:53.120
you need robotics, mechanical excellence,
link |
00:42:57.200
and the ability to deal with dexterity,
link |
00:43:01.880
and maybe even unknown environments, very, very difficult.
link |
00:43:05.640
So if we were to categorize the most dangerous white collar
link |
00:43:11.200
jobs, they would be things like back office,
link |
00:43:15.600
people who copy and paste and deal with simple computer
link |
00:43:20.800
programs and data, and maybe paper and OCR,
link |
00:43:25.560
and they don't make strategic decisions,
link |
00:43:29.000
they basically facilitate the process.
link |
00:43:32.040
These software and paper systems don't work,
link |
00:43:34.680
so you have people dealing with new employee orientation,
link |
00:43:40.520
searching for past lawsuits and financial documents,
link |
00:43:45.400
and doing reference check, so basic searching and management
link |
00:43:49.800
of data that's the most in danger of being lost.
link |
00:43:52.800
In addition to the white collar repetitive work,
link |
00:43:56.440
a lot of simple interaction work can also
link |
00:43:59.360
be taken care of, such as tele sales, telemarketing,
link |
00:44:02.840
customer service, as well as many physical jobs
link |
00:44:07.280
that are in the same location and don't
link |
00:44:09.880
require a high degree of dexterity,
link |
00:44:12.240
so fruit picking, dishwashing, assembly line, inspection,
link |
00:44:17.840
our jobs in that category.
link |
00:44:20.360
So altogether, back office is a big part,
link |
00:44:25.440
and the other, the blue collar may be smaller initially,
link |
00:44:29.840
but over time, AI will get better.
link |
00:44:32.560
And when we start to get to over the next 15, 20 years,
link |
00:44:36.880
the ability to actually have the dexterity
link |
00:44:39.120
of doing assembly line, that's a huge chunk of jobs.
link |
00:44:42.600
And when autonomous vehicles start
link |
00:44:44.760
to work initially starting with truck drivers,
link |
00:44:47.400
but eventually to all drivers, that's
link |
00:44:49.640
another huge group of workers.
link |
00:44:52.040
So I see modest numbers in the next five years,
link |
00:44:55.560
but increasing rapidly after that.
link |
00:44:58.080
On the worry of the jobs that are in danger
link |
00:45:01.240
and the gradual loss of jobs, I'm not
link |
00:45:04.320
sure if you're familiar with Andrew Yang.
link |
00:45:06.680
Yes, I am.
link |
00:45:07.800
So there's a candidate for president of the United States
link |
00:45:10.560
whose platform, Andrew Yang, is based around, in part,
link |
00:45:14.960
around job loss due to automation,
link |
00:45:17.680
and also, in addition, the need, perhaps,
link |
00:45:21.120
of universal basic income to support jobs that are folks who
link |
00:45:26.120
lose their job due to automation and so on,
link |
00:45:28.560
and in general, support people under complex,
link |
00:45:31.960
unstable job market.
link |
00:45:34.320
So what are your thoughts about his concerns,
link |
00:45:36.720
him as a candidate, his ideas in general?
link |
00:45:40.000
I think his thinking is generally in the right direction,
link |
00:45:44.600
but his approach as a presidential candidate
link |
00:45:48.440
may be a little bit ahead at the time.
link |
00:45:52.240
I think the displacements will happen,
link |
00:45:56.080
but will they happen soon enough for people
link |
00:45:58.280
to agree to vote for him?
link |
00:46:00.480
The unemployment numbers are not very high yet.
link |
00:46:03.760
And I think he and I have the same challenge.
link |
00:46:07.600
If I want to theoretically convince people this is an issue
link |
00:46:11.520
and he wants to become the president,
link |
00:46:13.880
people have to see how can this be the case when
link |
00:46:17.760
unemployment numbers are low.
link |
00:46:19.680
So that is the challenge.
link |
00:46:21.360
And I think I do agree with him on the displacement issue,
link |
00:46:27.360
on universal basic income, at a very vanilla level.
link |
00:46:32.280
I don't agree with it because I think the main issue
link |
00:46:36.800
is retraining.
link |
00:46:38.320
So people need to be incented not by just giving a monthly
link |
00:46:43.200
$2,000 check or $1,000 check and do whatever they want
link |
00:46:47.160
because they don't have the know how
link |
00:46:50.920
to know what to retrain to go into what type of a job
link |
00:46:56.840
and guidance is needed.
link |
00:46:58.640
And retraining is needed because historically
link |
00:47:01.720
in technology revolutions, when routine jobs were displaced,
link |
00:47:05.080
new routine jobs came up.
link |
00:47:06.920
So there was always room for that.
link |
00:47:09.400
But with AI and automation, the whole point
link |
00:47:12.640
is replacing all routine jobs eventually.
link |
00:47:15.320
So there will be fewer and fewer routine jobs.
link |
00:47:17.840
And AI will create jobs, but it won't create routine jobs
link |
00:47:22.640
because if it creates routine jobs,
link |
00:47:24.840
why wouldn't AI just do it?
link |
00:47:26.880
So therefore, the people who are losing the jobs
link |
00:47:30.360
are losing routine jobs.
link |
00:47:32.280
The jobs that are becoming available are nonroutine jobs.
link |
00:47:35.720
So the social stipend needs to be put in place
link |
00:47:39.320
is for the routine workers who lost their jobs
link |
00:47:42.040
to be retrained maybe in six months, maybe in three years.
link |
00:47:46.120
Takes a while to retrain on the nonroutine job
link |
00:47:48.560
and then take on a job that will last
link |
00:47:51.360
for that person's lifetime.
link |
00:47:53.400
Now, having said that, if you look deeply
link |
00:47:56.160
into Andrew's document, he does cater for that.
link |
00:47:58.240
So I'm not disagreeing with what he's trying to do.
link |
00:48:03.280
But for simplification, sometimes he just says UBI,
link |
00:48:06.360
but simple UBI wouldn't work.
link |
00:48:08.760
And I think you've mentioned elsewhere
link |
00:48:10.600
that the goal isn't necessarily to give people enough money
link |
00:48:15.760
to survive or live or even to prosper.
link |
00:48:19.120
The point is to give them a job that gives them meaning.
link |
00:48:22.800
That meaning is extremely important.
link |
00:48:25.600
That our employment, at least in the United States
link |
00:48:28.600
and perhaps it cares across the world,
link |
00:48:31.200
provides something that's, forgive me for saying,
link |
00:48:34.600
greater than money, it provides meaning.
link |
00:48:38.400
So now what kind of jobs do you think can't be automated?
link |
00:48:44.840
You talk a little bit about creativity
link |
00:48:46.600
and compassion in your book.
link |
00:48:48.200
What aspects do you think it's difficult
link |
00:48:50.720
to automate for an AI system?
link |
00:48:52.320
Because an AI system is currently merely optimizing.
link |
00:48:57.360
It's not able to reason, plan,
link |
00:49:00.120
or think creatively or strategically.
link |
00:49:02.920
It's not able to deal with complex problems.
link |
00:49:05.320
It can't come up with a new problem and solve it.
link |
00:49:09.520
A human needs to find the problem
link |
00:49:12.320
and pose it as an optimization problem,
link |
00:49:15.520
then have the AI work at it.
link |
00:49:17.520
So an AI would have a very hard time
link |
00:49:21.320
discovering a new drug
link |
00:49:23.320
or discovering a new style of painting
link |
00:49:27.320
or dealing with complex tasks
link |
00:49:30.320
such as managing a company
link |
00:49:32.320
that isn't just about optimizing the bottom line,
link |
00:49:35.320
but also about employee satisfaction, corporate brand,
link |
00:49:39.320
and many, many other things.
link |
00:49:40.320
So that is one category of things.
link |
00:49:44.320
And because these things are challenging, creative, complex,
link |
00:49:48.320
doing them creates a higher degree of satisfaction
link |
00:49:52.320
and therefore appealing to our desire for working,
link |
00:49:55.320
which isn't just to make the money,
link |
00:49:57.320
make the ends meet,
link |
00:49:58.320
but also that we've accomplished something
link |
00:50:00.320
that others maybe can't do or can't do as well.
link |
00:50:04.320
Another type of job that is much numerous
link |
00:50:07.320
would be compassionate jobs,
link |
00:50:09.320
jobs that require compassion, empathy, human touch, human trust.
link |
00:50:14.320
AI can't do that because AI is cold, calculating,
link |
00:50:18.320
and even if it can fake that to some extent,
link |
00:50:22.320
it will make errors and that will make it look very silly.
link |
00:50:26.320
And also, I think even if AI did okay,
link |
00:50:29.320
people would want to interact with another person,
link |
00:50:33.320
whether it's for some kind of a service or a teacher or a doctor
link |
00:50:38.320
or a concierge or a masseuse or bartender.
link |
00:50:41.320
There are so many jobs where people just don't want to interact
link |
00:50:46.320
with a cold robot or software.
link |
00:50:50.320
I've had an entrepreneur who built an elderly care robot
link |
00:50:53.320
and they found that the elderly really only use it for customer service.
link |
00:50:58.320
But not to service the product,
link |
00:51:00.320
but they click on customer service and the video of a person comes up
link |
00:51:05.320
and then the person says,
link |
00:51:07.320
how come my daughter didn't call me? Let me show you a picture of her grandkids.
link |
00:51:11.320
So people earn for that, people people interaction.
link |
00:51:15.320
So even if robots improved, people just don't want it.
link |
00:51:19.320
And those jobs are going to be increasing
link |
00:51:21.320
because AI will create a lot of value,
link |
00:51:24.320
$16 trillion to the world in next 11 years according to PWC
link |
00:51:29.320
and that will give people money to enjoy services,
link |
00:51:34.320
whether it's eating a gourmet meal or tourism and traveling
link |
00:51:39.320
or having concierge services.
link |
00:51:41.320
The services revolving around, you know,
link |
00:51:44.320
every dollar of that $16 trillion will be tremendous.
link |
00:51:47.320
It will create more opportunities to service the people who did well
link |
00:51:52.320
through AI with things.
link |
00:51:55.320
But even at the same time, the entire society is very much short
link |
00:52:01.320
in need of many service oriented, compassionate oriented jobs.
link |
00:52:05.320
The best example is probably in healthcare services.
link |
00:52:10.320
There's going to be 2 million new jobs, not counting replacement,
link |
00:52:15.320
just brand new incremental jobs in the next six years in healthcare services.
link |
00:52:20.320
That includes nurses orderly in the hospital,
link |
00:52:24.320
elderly care and also at home care.
link |
00:52:29.320
It's particularly lacking.
link |
00:52:31.320
And those jobs are not likely to be filled.
link |
00:52:34.320
So there's likely to be a shortage.
link |
00:52:36.320
And the reason they're not filled is simply because they don't pay very well
link |
00:52:41.320
and that the social status of these jobs are not very good.
link |
00:52:47.320
So they pay about half as much as a heavy equipment operator,
link |
00:52:52.320
which will be replaced a lot sooner.
link |
00:52:55.320
And they pay probably comparably to someone on the assembly line.
link |
00:52:59.320
And so if we're ignoring all the other issues
link |
00:53:03.320
and just think about satisfaction from one's job,
link |
00:53:07.320
someone repetitively doing the same manual action at an assembly line,
link |
00:53:11.320
that can't create a lot of job satisfaction.
link |
00:53:14.320
But someone taking care of a sick person
link |
00:53:17.320
and getting a hug and thank you from that person and the family,
link |
00:53:21.320
I think is quite satisfying.
link |
00:53:24.320
So if only we could fix the pay for service jobs,
link |
00:53:28.320
there are plenty of jobs that require some training or a lot of training
link |
00:53:33.320
for the people coming off the routine jobs to take.
link |
00:53:36.320
We can easily imagine someone who was maybe a cashier at the grocery store,
link |
00:53:43.320
at stores become automated, learns to become a nurse or at home care.
link |
00:53:49.320
Also, I do want to point out the blue collar jobs are going to stay around a bit longer,
link |
00:53:54.320
some of them quite a bit longer.
link |
00:53:57.320
AI cannot be told, go clean an arbitrary home.
link |
00:54:01.320
That's incredibly hard.
link |
00:54:03.320
Arguably is an L5 level of difficulty.
link |
00:54:07.320
And then AI cannot be a good plumber,
link |
00:54:09.320
because plumber is almost like a mini detective
link |
00:54:12.320
that has to figure out where the leak came from.
link |
00:54:15.320
So yet AI probably can be an assembly line and auto mechanic and so on.
link |
00:54:22.320
So one has to study which blue collar jobs are going away
link |
00:54:26.320
and facilitate retraining for the people to go into the ones that won't go away
link |
00:54:30.320
or maybe even will increase.
link |
00:54:32.320
I mean, it is fascinating that it's easier to build a world champion chess player
link |
00:54:39.320
than it is to build a mediocre plumber.
link |
00:54:41.320
Yes, very true.
link |
00:54:43.320
And to AI, and that goes counterintuitive to a lot of people's understanding
link |
00:54:47.320
of what artificial intelligence is.
link |
00:54:49.320
So it sounds, I mean, you're painting a pretty optimistic picture
link |
00:54:53.320
about retraining, about the number of jobs
link |
00:54:56.320
and actually the meaningful nature of those jobs once we automate repetitive tasks.
link |
00:55:01.320
So overall, are you optimistic about the future
link |
00:55:07.320
where much of the repetitive tasks are automated,
link |
00:55:11.320
that there is a lot of room for humans, for the compassionate,
link |
00:55:15.320
for the creative input that only humans can provide?
link |
00:55:19.320
I am optimistic if we start to take action.
link |
00:55:23.320
If we have no action in the next five years,
link |
00:55:27.320
I think it's going to be hard to deal with the devastating losses that will emerge.
link |
00:55:33.320
So if we start thinking about retraining, maybe with the low hanging fruits,
link |
00:55:39.320
explaining to vocational schools why they should train more plumbers than auto mechanics,
link |
00:55:45.320
maybe starting with some government subsidy for corporations to have more training positions.
link |
00:55:53.320
We start to explain to people why retraining is important.
link |
00:55:57.320
We start to think about what the future of education,
link |
00:56:00.320
how that needs to be tweaked for the era of AI.
link |
00:56:04.320
If we start to make incremental progress,
link |
00:56:06.320
and the greater number of people understand,
link |
00:56:09.320
then there's no reason to think we can't deal with this,
link |
00:56:12.320
because this technological revolution is arguably similar to
link |
00:56:16.320
what electricity, industrial revolutions, and internet brought about.
link |
00:56:20.320
Do you think there's a role for policy, for governments to step in
link |
00:56:24.320
to help with policy to create a better world?
link |
00:56:27.320
Absolutely, and the governments don't have to believe
link |
00:56:32.320
that unemployment will go up, and they don't have to believe automation will be this fast to do something.
link |
00:56:39.320
Revamping vocational school would be one example.
link |
00:56:42.320
Another is if there's a big gap in healthcare service employment,
link |
00:56:47.320
and we know that a country's population is growing older and more longevity living older,
link |
00:56:54.320
because people over 80 require five times as much care as those under 80,
link |
00:56:59.320
then it is a good time to incent training programs for elderly care,
link |
00:57:04.320
to find ways to improve the pay.
link |
00:57:07.320
Maybe one way would be to offer as part of Medicare or the equivalent program
link |
00:57:13.320
for people over 80 to be entitled to a few hours of elderly care at home,
link |
00:57:18.320
and then that might be reimbursable,
link |
00:57:21.320
and that will stimulate the service industry around the policy.
link |
00:57:28.320
Do you have concerns about large entities,
link |
00:57:32.320
whether it's governments or companies, controlling the future of AI development in general?
link |
00:57:38.320
So we talked about companies.
link |
00:57:40.320
Do you have a better sense that governments can better represent the interest of the people
link |
00:57:48.320
than companies, or do you believe companies are better at representing the interest of the people?
link |
00:57:54.320
Or is there no easy answer?
link |
00:57:56.320
I don't think there's an easy answer because it's a double edged sword.
link |
00:57:59.320
The companies and governments can provide better services with more access to data and more access to AI,
link |
00:58:06.320
but that also leads to greater power, which can lead to uncontrollable problems,
link |
00:58:13.320
whether it's monopoly or corruption in the government.
link |
00:58:17.320
So I think one has to be careful to look at how much data that companies and governments have,
link |
00:58:24.320
and some kind of checks and balances would be helpful.
link |
00:58:29.320
So again, I come from Russia.
link |
00:58:33.320
There's something called the Cold War.
link |
00:58:36.320
So let me ask a difficult question here, looking at conflict.
link |
00:58:40.320
Steven Pinker wrote a great book that conflict all over the world is decreasing in general.
link |
00:58:45.320
But do you have a sense that having written the book AI Superpowers,
link |
00:58:51.320
do you see a major international conflict potentially arising between major nations,
link |
00:58:57.320
whatever they are, whether it's Russia, China, European nations, United States,
link |
00:59:02.320
or others in the next 10, 20, 50 years around AI, around the digital space, cyber space?
link |
00:59:09.320
Do you worry about that?
link |
00:59:12.320
Is that something we need to think about and try to alleviate or prevent?
link |
00:59:19.320
I believe in greater engagement.
link |
00:59:22.320
A lot of the worries about more powerful AI are based on an arms race metaphor.
link |
00:59:33.320
And when you extrapolate into military kinds of scenarios,
link |
00:59:41.320
AI can automate autonomous weapons that needs to be controlled somehow.
link |
00:59:48.320
And autonomous decision making can lead to not enough time to fix international crises.
link |
00:59:57.320
So I actually believe a Cold War mentality would be very dangerous
link |
01:00:02.320
because should two countries rely on AI to make certain decisions
link |
01:00:07.320
and they don't even talk to each other, they do their own scenario planning,
link |
01:00:11.320
then something could easily go wrong.
link |
01:00:14.320
I think engagement, interaction, some protocols to avoid inadvertent disasters is actually needed.
link |
01:00:24.320
So it's natural for each country to want to be the best,
link |
01:00:28.320
whether it's in nuclear technologies or AI or bio.
link |
01:00:34.320
But I think it's important to realize if each country has a black box AI
link |
01:00:40.320
and don't talk to each other, that probably presents greater challenges to humanity
link |
01:00:48.320
than if they interacted.
link |
01:00:50.320
I think there can still be competition, but with some degree of protocol for interaction.
link |
01:00:56.320
Just like when there was a nuclear competition,
link |
01:01:01.320
there were some protocol for deterrence among US, Russia, and China.
link |
01:01:07.320
And I think that engagement is needed.
link |
01:01:10.320
So of course, we're still far from AI presenting that kind of danger.
link |
01:01:15.320
But what I worry the most about is the level of engagement seems to be coming down.
link |
01:01:22.320
The level of distrust seems to be going up,
link |
01:01:25.320
especially from the US towards other large countries such as China and Russia.
link |
01:01:32.320
Is there a way to make that better?
link |
01:01:34.320
So that's beautifully put, level of engagement and even just basic trust and communication
link |
01:01:40.320
as opposed to making artificial enemies out of particular countries.
link |
01:01:52.320
Do you have a sense how we can make it better, actionable items that as a society we can take on?
link |
01:02:01.320
I'm not an expert at geopolitics, but I would say that we look pretty foolish as humankind
link |
01:02:10.320
when we are faced with the opportunity to create $16 trillion for humanity.
link |
01:02:19.320
And yet we're not solving fundamental problems with parts of the world still in poverty.
link |
01:02:29.320
And for the first time, we have the resources to overcome poverty and hunger.
link |
01:02:34.320
We're not using it on that, but we're fueling competition among superpowers.
link |
01:02:38.320
And that's a very unfortunate thing.
link |
01:02:41.320
If we become utopian for a moment, imagine a benevolent world government that has this $16 trillion
link |
01:02:54.320
and maybe some AI to figure out how to use it to deal with diseases and problems and hate and things like that.
link |
01:03:02.320
World would be a lot better off.
link |
01:03:04.320
So what is wrong with the current world?
link |
01:03:07.320
I think the people with more skill than I should think about this.
link |
01:03:13.320
And then the geopolitics issue with superpower competition is one side of the issue.
link |
01:03:19.320
There's another side which I worry maybe even more, which is as the $16 trillion all gets made by U.S. and China
link |
01:03:29.320
and a few of the other developed countries, the poorer country will get nothing
link |
01:03:34.320
because they don't have technology and the wealth disparity and inequality will increase.
link |
01:03:42.320
So a poorer country with a large population will not only benefit from the AI boom or other technology booms
link |
01:03:50.320
but they will have their workers who previously had hoped they could do the China model and do outsource manufacturing
link |
01:03:57.320
or the India model so they could do the outsource process or call center
link |
01:04:02.320
while all those jobs are going to be gone in 10 or 15 years.
link |
01:04:05.320
So the individual citizen may be a net liability, I mean financially speaking, to a poorer country
link |
01:04:14.320
and not an asset to claw itself out of poverty.
link |
01:04:19.320
So in that kind of situation, these large countries with not much tech are going to be facing a downward spiral
link |
01:04:29.320
and it's unclear what could be done and then when we look back and say there's $16 trillion being created
link |
01:04:37.320
and it's all being kept by U.S. China and other developed countries, it just doesn't feel right.
link |
01:04:43.320
So I hope people who know about geopolitics can find solutions that's beyond my expertise.
link |
01:04:50.320
So different countries that we've talked about have different value systems.
link |
01:04:54.320
If you look at the United States to an almost extreme degree, there is an absolute desire for freedom of speech.
link |
01:05:02.320
If you look at a country where I was raised, that desire just amongst the people is not as elevated as it is to basically fundamental level
link |
01:05:14.320
to the essence of what it means to be America, right?
link |
01:05:17.320
And the same is true with China, there's different value systems.
link |
01:05:20.320
There is some censorship of internet content that China and Russia and many other countries undertake.
link |
01:05:30.320
Do you see that having effects on innovation, other aspects of some of the tech stuff, AI development we talked about
link |
01:05:40.320
and maybe from another angle, do you see that changing in different ways over the next 10 years, 20 years, 50 years as China continues to grow
link |
01:05:52.320
as it does now in its tech innovation?
link |
01:05:55.320
There's a common belief that full freedom of speech and expression is correlated with creativity, which is correlated with entrepreneurial success.
link |
01:06:08.320
I think empirically we have seen that is not true and China has been successful.
link |
01:06:15.320
That's not to say the fundamental values are not right or not the best, but it's just that perfect correlation isn't there.
link |
01:06:25.320
It's hard to read the tea leaves on opening up or not in any country and I've not been very good at that in my past predictions.
link |
01:06:36.320
But I do believe every country shares some fundamental value, a lot of fundamental values for the long term.
link |
01:06:46.320
So, you know, China is drafting its privacy policy for individual citizens and they don't look that different from the American or European ones.
link |
01:07:02.320
So, people do want to protect their privacy and have the opportunity to express and I think the fundamental values are there.
link |
01:07:13.320
The question is in the execution and timing, how soon or when will that start to open up?
link |
01:07:21.320
So, as long as each government knows, ultimately people want that kind of protection, there should be a plan to move towards that.
link |
01:07:31.320
As to when or how, again, I'm not an expert.
link |
01:07:35.320
On the point of privacy to me, it's really interesting.
link |
01:07:38.320
So, AI needs data to create a personalized awesome experience.
link |
01:07:44.320
I'm just speaking generally in terms of products.
link |
01:07:47.320
And then we have currently, depending on the age and depending on the demographics of who we're talking about,
link |
01:07:53.320
some people are more or less concerned about the amount of data they hand over.
link |
01:07:58.320
So, in your view, how do we get this balance right?
link |
01:08:03.320
That we provide an amazing experience to people that use products.
link |
01:08:09.320
You look at Facebook, you know, the more Facebook knows about you, yes, it's scary to say.
link |
01:08:15.320
The better it can probably, a better experience it can probably create.
link |
01:08:20.320
So, in your view, how do we get that balance right?
link |
01:08:24.320
Yes, I think a lot of people have a misunderstanding that it's okay and possible to just rip all the data out from a provider and give it back to you.
link |
01:08:38.320
So, you can deny them access to further data and still enjoy the services we have.
link |
01:08:43.320
If we take back all the data, all the services will give us nonsense.
link |
01:08:48.320
We'll no longer be able to use products that function well in terms of, you know, right ranking, right products, right user experience.
link |
01:08:57.320
So, yet I do understand we don't want to permit misuse of the data.
link |
01:09:04.320
From legal policy standpoint, I think there can be severe punishment for those who have egregious misuse of the data.
link |
01:09:16.320
That's, I think, a good first step.
link |
01:09:19.320
Actually, China on this aspect has very strong laws about people who sell or give data to other companies.
link |
01:09:27.320
And that over the past few years, since that law came into effect, pretty much eradicated the illegal distribution sharing of data.
link |
01:09:40.320
Additionally, I think giving, I think technology is often a very good way to solve technology misuse.
link |
01:09:52.320
So, can we come up with new technologies that will let us have our cake and eat it too?
link |
01:09:58.320
People are looking into homomorphic encryption, which is letting you keep the data, have it encrypted and train encrypted data.
link |
01:10:07.320
Of course, we haven't solved that one yet, but that kind of direction may be worth pursuing.
link |
01:10:13.320
Also federated learning, which would allow one hospital to train on its hospitals patient data fully because they have a license for that.
link |
01:10:22.320
And then hospitals would then share their models, not data, but models to create a supra AI.
link |
01:10:28.320
And that also maybe has some promise.
link |
01:10:30.320
So I would want to encourage us to be open minded and think of this as not just the policy binary yes no,
link |
01:10:39.320
but letting the technologists try to find solutions to let us have our cake and eat it too, or have most of our cake and eat most of it too.
link |
01:10:48.320
Finally, I think giving each end user a choice is important and having transparency is important.
link |
01:10:55.320
Also, I think that's universal, but the choice you give to the user should not be at a granular level that the user cannot understand.
link |
01:11:04.320
GDPR today causes all these pop ups of yes, no, will you give this site this right to use this part of your data?
link |
01:11:12.320
I don't think any user understands what they're saying yes or no to, and I suspect most are just saying yes because they don't understand it.
link |
01:11:20.320
So while GDPR in its current implementation has lived up to its promise of transparency and user choice,
link |
01:11:30.320
it implemented it in such a way that really didn't deliver the spirit of GDPR.
link |
01:11:39.320
It fit the letter, but not the spirit.
link |
01:11:41.320
So again, I think we need to think about is there a way to fit the spirit of GDPR by using some kind of technology?
link |
01:11:50.320
Can we have a slider?
link |
01:11:52.320
That's an AI trying to figure out how much you want to slide between perfect protection security of your personal data
link |
01:12:01.320
versus high degree of convenience with some risks of not having full privacy.
link |
01:12:07.320
Each user should have some preference and that gives you the user choice,
link |
01:12:11.320
but maybe we should turn the problem on its head and ask can there be an AI algorithm that can customize this
link |
01:12:18.320
because we can understand the slider, but we sure cannot understand every pop up question.
link |
01:12:24.320
And I think getting that right requires getting the balance between what we talked about earlier,
link |
01:12:30.320
which is heart and soul versus profit driven decisions and strategy.
link |
01:12:36.320
I think from my perspective, the best way to make a lot of money in the long term is to keep your heart and soul intact.
link |
01:12:45.320
I think getting that slider right in the short term may feel like you'll be sacrificing profit,
link |
01:12:53.320
but in the long term, you'll be getting user trust and providing a great experience.
link |
01:12:59.320
Do you share that kind of view in general?
link |
01:13:01.320
Yes, absolutely. I sure would hope there is a way we can do long term projects that really do the right thing.
link |
01:13:11.320
I think a lot of people who embrace GDPR, their hearts in the right place.
link |
01:13:16.320
I think they just need to figure out how to build a solution.
link |
01:13:20.320
I've heard utopians talk about solutions that get me excited,
link |
01:13:24.320
but not sure how in the current funding environment they can get started, right?
link |
01:13:29.320
People talk about, imagine this crowdsourced data collection that we all trust,
link |
01:13:37.320
and then we have these agents that we ask them to ask the trusted agent.
link |
01:13:45.320
That agent only, that platform.
link |
01:13:48.320
A trusted joint platform that we all believe is trustworthy that can give us all the close loop personal suggestions
link |
01:14:02.320
by the new social network, new search engine, new ecommerce engine
link |
01:14:07.320
that has access to even more of our data, but not directly but indirectly.
link |
01:14:12.320
I think that general concept of licensing to some trusted engine
link |
01:14:18.320
and finding a way to trust that engine seems like a great idea,
link |
01:14:22.320
but if you think how long it's going to take to implement and tweak and develop it right,
link |
01:14:27.320
as well as to collect all the trust and the data from the people,
link |
01:14:31.320
it's beyond the current cycle of venture capital.
link |
01:14:34.320
How do you do that is a big question.
link |
01:14:37.320
You've recently had a fight with cancer, stage 4 lymphoma,
link |
01:14:44.320
and in a sort of deep personal level, what did it feel like in the darker moments to face your own mortality?
link |
01:14:54.320
Well, I've been the workaholic my whole life,
link |
01:14:57.320
and I've basically worked 9.96, 9am to 9pm, 6 days a week, roughly.
link |
01:15:04.320
And I didn't really pay a lot of attention to my family, friends, and people who loved me,
link |
01:15:10.320
and my life revolved around optimizing for work.
link |
01:15:14.320
While my work was not routine, my optimization really made my life basically a very mechanical process.
link |
01:15:25.320
But I got a lot of highs out of it because of accomplishments that I thought were really important and dear and the highest priority to me.
link |
01:15:36.320
But when I faced mortality and the possible death in matter of months,
link |
01:15:41.320
I suddenly realized that this really meant nothing to me,
link |
01:15:45.320
that I didn't feel like working for another minute,
link |
01:15:48.320
that if I had 6 months left in my life, I would spend it all with my loved ones.
link |
01:15:54.320
And thanking them, giving them love back, and apologizing to them that I lived my life the wrong way.
link |
01:16:02.320
So that moment of reckoning caused me to really rethink that why we exist in this world
link |
01:16:11.320
is something that we might be too much shaped by the society to think that success and accomplishments is why we live.
link |
01:16:22.320
And while that can get you periodic successes and satisfaction,
link |
01:16:29.320
it's really in them facing death, you see what's truly important to you.
link |
01:16:35.320
So as a result of going through the challenges with cancer,
link |
01:16:41.320
I've resolved to live a more balanced lifestyle.
link |
01:16:45.320
I'm now in remission, knock on wood,
link |
01:16:48.320
and I'm spending more time with my family.
link |
01:16:52.320
My wife travels with me.
link |
01:16:54.320
When my kids need me, I spend more time with them.
link |
01:16:57.320
And before, I used to prioritize everything around work.
link |
01:17:02.320
When I had a little bit of time, I would dole it out to my family.
link |
01:17:05.320
Now, when my family needs something, really needs something,
link |
01:17:09.320
I drop everything at work and go to them.
link |
01:17:12.320
And then in the time remaining, I allocate to work.
link |
01:17:15.320
But one's family is very understanding.
link |
01:17:18.320
It's not like they will take 50 hours a week from me.
link |
01:17:22.320
So I'm actually able to still work pretty hard,
link |
01:17:26.320
maybe 10 hours less per week.
link |
01:17:28.320
So I realize the most important thing in my life is really love and the people I love.
link |
01:17:35.320
And I give that the highest priority.
link |
01:17:38.320
It isn't the only thing I do.
link |
01:17:40.320
But when that is needed, I put that at the top priority.
link |
01:17:45.320
And I feel much better and I feel much more balanced.
link |
01:17:49.320
And I think this also gives a hint as to a life of routine work,
link |
01:17:56.320
a life of pursuit of numbers.
link |
01:17:58.320
While my job was not routine, it wasn't pursuit of numbers.
link |
01:18:03.320
Pursuit of, can I make more money?
link |
01:18:05.320
Can I fund more great companies?
link |
01:18:07.320
Can I raise more money?
link |
01:18:09.320
Can I make sure our VC is ranked higher and higher every year?
link |
01:18:13.320
This competitive nature of driving for bigger numbers and better numbers
link |
01:18:20.320
became an endless pursuit of that's mechanical.
link |
01:18:27.320
And bigger numbers really didn't make me happier.
link |
01:18:31.320
And faced with death, I realized bigger numbers really meant nothing.
link |
01:18:36.320
And what was important is that people who have given their heart and their love to me
link |
01:18:42.320
deserve for me to do the same.
link |
01:18:45.320
So there's deep profound truth in that, that everyone should hear and internalize.
link |
01:18:52.320
And that's really powerful for you to say that.
link |
01:18:56.320
I have to ask sort of a difficult question here.
link |
01:19:02.320
So I've competed in sports my whole life, looking historically.
link |
01:19:07.320
I'd like to challenge some aspect of that a little bit on the point of hard work.
link |
01:19:14.320
That it feels that there are certain aspects that is the greatest,
link |
01:19:20.320
the most beautiful aspects of human nature, is the ability to become obsessed,
link |
01:19:26.320
of becoming extremely passionate to the point where, yes, flaws are revealed
link |
01:19:33.320
and just giving yourself fully to a task.
link |
01:19:36.320
That is, in another sense, you mentioned love being important,
link |
01:19:41.320
but in another sense, this kind of obsession, this pure exhibition of passion and hard work
link |
01:19:47.320
is truly what it means to be human.
link |
01:19:50.320
What lessons should we take that's deeper?
link |
01:19:53.320
Because you've accomplished incredible things.
link |
01:19:55.320
Like chasing numbers.
link |
01:19:57.320
But really, there's some incredible work there.
link |
01:20:01.320
So how do you think about that when you look back in your 20s, your 30s?
link |
01:20:07.320
What would you do differently?
link |
01:20:10.320
Would you really take back some of the incredible hard work?
link |
01:20:16.320
I would.
link |
01:20:17.320
But it's in percentages, right?
link |
01:20:20.320
We're both now computer scientists.
link |
01:20:22.320
So I think when one balances one's life, when one is younger,
link |
01:20:27.320
you might give a smaller percentage to family, but you would still give them high priority.
link |
01:20:33.320
And when you get older, you would give a larger percentage to them and still the high priority.
link |
01:20:38.320
And when you're near retirement, you give most of it to them and the highest priority.
link |
01:20:43.320
So I think the key point is not that we would work 20 hours less for the whole life
link |
01:20:50.320
and just spend it aimlessly with the family, but that when the family has a need,
link |
01:20:56.320
when your wife is having a baby, when your daughter has a birthday,
link |
01:21:02.320
or when they're depressed, or when they're celebrating something,
link |
01:21:07.320
or when they have a get together, or when we have family time,
link |
01:21:11.320
that is important for us to put down our phone and PC and be 100% with them.
link |
01:21:18.320
And that priority on the things that really matter isn't going to be so taxing
link |
01:21:26.320
that it would eliminate or even dramatically reduce our accomplishments.
link |
01:21:32.320
It might have some impact, but it might also have other impact
link |
01:21:36.320
because if you have a happier family, maybe you fight less.
link |
01:21:39.320
If you fight less, you don't spend time taking care of all the aftermath of a fight.
link |
01:21:45.320
That's right.
link |
01:21:46.320
And I'm sure that it would take more time.
link |
01:21:48.320
And if it did, I'd be willing to take that reduction.
link |
01:21:53.320
And it's not a dramatic number, but it's a number
link |
01:21:56.320
that I think would give me a greater degree of happiness
link |
01:22:00.320
and knowing that I've done the right thing
link |
01:22:03.320
and still have plenty of hours to get the success that I want to get.
link |
01:22:10.320
So given the many successful companies that you've launched
link |
01:22:14.320
and much success throughout your career,
link |
01:22:17.320
what advice would you give to young people today looking,
link |
01:22:25.320
or it doesn't have to be young, but people today looking to launch
link |
01:22:28.320
and to create the next $1 billion tech startup, or even AI based startup?
link |
01:22:35.320
I would suggest that people understand technology waves move quickly.
link |
01:22:42.320
What worked two years ago may not work today.
link |
01:22:45.320
And that is very much a case in point for AI.
link |
01:22:49.320
I think two years ago, or maybe three years ago,
link |
01:22:53.320
you certainly could say I have a couple of super smart PhDs
link |
01:22:57.320
and we're not sure what we're going to do,
link |
01:22:59.320
but here's how we're going to start and get funding for a very high valuation.
link |
01:23:04.320
Those days are over because AI is going from rocket science towards mainstream.
link |
01:23:11.320
Not yet commodity, but more mainstream.
link |
01:23:14.320
So first, the creation of any company to eventual capitalist
link |
01:23:20.320
has to be creation of business value and monetary value.
link |
01:23:25.320
And when you have a very scarce commodity,
link |
01:23:29.320
VCs may be willing to accept greater uncertainty.
link |
01:23:34.320
But now the number of people who have the equivalent of PhD three years ago
link |
01:23:40.320
because that can be learned more quickly.
link |
01:23:43.320
Platforms are emerging.
link |
01:23:45.320
The cost to become an AI engineer is much lower and there are many more AI engineers.
link |
01:23:51.320
So the market is different.
link |
01:23:53.320
So I would suggest someone who wants to build an AI company
link |
01:23:57.320
be thinking about the normal business questions.
link |
01:24:01.320
What customer cases are you trying to address?
link |
01:24:05.320
What kind of pain are you trying to address?
link |
01:24:08.320
How does that translate to value?
link |
01:24:10.320
How will you extract value and get paid through what channel?
link |
01:24:16.320
And how much business value will get created?
link |
01:24:19.320
That today needs to be thought about much earlier up front than it did three years ago.
link |
01:24:26.320
The scarcity question of AI talent has changed.
link |
01:24:30.320
The number of AI talent has changed.
link |
01:24:32.320
So now you need not just AI but also understanding of business customer and the marketplace.
link |
01:24:41.320
So I also think you should have a more reasonable evaluation expectation
link |
01:24:49.320
and growth expectation.
link |
01:24:51.320
There's going to be more competition.
link |
01:24:53.320
But the good news though is that AI technologies are now more available in open source.
link |
01:25:00.320
TensorFlow, PyTorch and such tools are much easier to use.
link |
01:25:06.320
So you should be able to experiment and get results iteratively faster than before.
link |
01:25:13.320
So take more of a business mindset to this.
link |
01:25:18.320
Think less of this as a laboratory taken into a company because we've gone beyond that stage.
link |
01:25:25.320
The only exception is if you truly have a breakthrough in some technology that really no one has,
link |
01:25:31.320
then the old way still works.
link |
01:25:34.320
But I think that's harder and harder now.
link |
01:25:36.320
So I know you believe as many do that we're far from creating an artificial general intelligence system.
link |
01:25:44.320
But say once we do and you get to ask her one question, what would that question be?
link |
01:25:54.320
What is it that differentiates you and me?
link |
01:26:00.320
Beautifully put, Kaifu, thank you so much for your time today.
link |
01:26:05.320
Thank you.