back to indexKai-Fu Lee: AI Superpowers - China and Silicon Valley | Lex Fridman Podcast #27
link |
The following is a conversation with Kai Fu Lee.
link |
He's the chairman and CEO of Cinovation Ventures
link |
that manages a $2 billion dual currency investment fund
link |
with a focus on developing the next generation
link |
of Chinese high tech companies.
link |
He's the former president of Google China
link |
and the founder of what is now called
link |
Microsoft Research Asia,
link |
an institute that trained many of the artificial
link |
intelligence leaders in China,
link |
including CTOs or AI execs at Baidu,
link |
Tencent, Alibaba, Lenovo, and Huawei.
link |
He was named one of the 100 most influential people
link |
in the world by Time Magazine.
link |
He's the author of seven bestselling books in Chinese
link |
and most recently, the New York Times bestseller
link |
called AI Superpowers, China, Silicon Valley,
link |
and the New World Order.
link |
He has unparalleled experience
link |
in working across major tech companies
link |
and governments and applications of AI,
link |
and so he has a unique perspective on global innovation
link |
and the future of AI that I think is important
link |
to listen to and think about.
link |
This is the Artificial Intelligence Podcast.
link |
If you enjoy it, subscribe on YouTube and iTunes,
link |
support it on Patreon, or simply connect with me
link |
on Twitter at Lex Friedman.
link |
And now, here's my conversation with Kaifu Li.
link |
I immigrated from Russia to US when I was 13.
link |
You immigrated to US at about the same age.
link |
The Russian people, the American people,
link |
the Chinese people each have a certain soul,
link |
a spirit that permeates throughout the generations.
link |
So maybe it's a little bit of a poetic question,
link |
but could you describe your sense
link |
of what defines the Chinese soul?
link |
I think the Chinese soul of people today, right,
link |
we're talking about people who have had centuries of burden
link |
because of the poverty that the country has gone through,
link |
and suddenly shined with hope of prosperity
link |
in the past 40 years as China opened up
link |
and embraced market economy.
link |
And undoubtedly, there are two sets of pressures
link |
on the people, that of the tradition,
link |
that of facing difficult situations,
link |
and that of hope of wanting to be the first
link |
to become successful and wealthy.
link |
So that's a very strong hunger and a strong desire
link |
and strong work ethic that drives China forward.
link |
And is there roots to not just this generation,
link |
but before that's deeper
link |
than just the new economic developments?
link |
Is there something that's unique to China
link |
that you could speak to that's in the people?
link |
Yeah, well, the Chinese tradition
link |
is about excellence, dedication, and results.
link |
And the Chinese exams and study subjects in schools
link |
have traditionally started
link |
from memorizing 10,000 characters,
link |
not an easy task to start with.
link |
And further by memorizing
link |
his historic philosopher's literature poetry.
link |
So it really is probably
link |
the strongest rote learning mechanism created
link |
to make sure people had good memory
link |
and remember things extremely well.
link |
That's, I think at the same time,
link |
suppresses the breakthrough innovation
link |
and also enhances the speed execution get results.
link |
And that I think characterizes
link |
the historic basis of China.
link |
That's interesting,
link |
because there's echoes of that in Russian education
link |
as well as rote memorization.
link |
So you have to memorize a lot of poetry.
link |
I mean, there's just an emphasis on perfection in all forms
link |
that's not conducive to perhaps what you're speaking to,
link |
which is creativity.
link |
But you think that kind of education holds back
link |
the innovative spirit that you might see
link |
in the United States?
link |
Well, it holds back the breakthrough innovative spirits
link |
that we see in the United States,
link |
but it does not hold back the valuable execution oriented,
link |
result oriented value creating engines,
link |
which we see China being very successful.
link |
So is there a difference between a Chinese AI engineer today
link |
and an American AI engineer,
link |
perhaps rooted in the culture that we just talked about
link |
or the education or the very soul of the people or no?
link |
And what would your advice be to each
link |
if there's a difference?
link |
Well, there's a lot that's similar
link |
because AI is about mastering sciences,
link |
about using known technologies and trying new things,
link |
but it's also about picking from many parts
link |
of possible networks to use
link |
and different types of parameters to tune.
link |
And that part is somewhat rote.
link |
And it is also, as anyone who's built AI products
link |
can tell you a lot about cleansing the data
link |
because AI runs better with more data
link |
and data is generally unstructured,
link |
errorful and unclean.
link |
And the effort to clean the data is immense.
link |
So I think the better part of American engineering,
link |
AI engineering process is to try new things,
link |
to do things people haven't done before
link |
and to use technology to solve most if not all problems.
link |
So to make the algorithm work despite not so great data,
link |
find error tolerant ways to deal with the data.
link |
The Chinese way would be to basically enumerate
link |
to the fullest extent all the possible ways
link |
by a lot of machines, try lots of different ways
link |
to get it to work and spend a lot of resources
link |
and money and time cleaning up data.
link |
That means the AI engineer may be writing
link |
data cleansing algorithms, working with thousands of people
link |
who label or correct or do things with the data.
link |
That is the incredible hard work
link |
that might lead to better results.
link |
So the Chinese engineer would rely on
link |
and ask for more and more and more data
link |
and find ways to cleanse them and make them work
link |
in the system and probably less time thinking
link |
about new algorithms that can overcome data
link |
So where's your intuition?
link |
Where do you think the biggest impact
link |
in the next 10 years lies?
link |
Is it in some breakthrough algorithms
link |
or is it in just this at scale rigor,
link |
a rigorous approach to data, cleaning data,
link |
organizing data onto the same algorithms?
link |
What do you think the big impact in the applied world is?
link |
Well, if you're really in the company
link |
and you have to deliver results,
link |
using known techniques and enhancing data
link |
seems like the more expedient approach
link |
that's very low risk and likely to generate
link |
better and better results.
link |
And that's why the Chinese approach has done quite well.
link |
Now, there are a lot of more challenging startups
link |
and problems such as autonomous vehicles,
link |
medical diagnosis that existing algorithms
link |
probably won't solve.
link |
And that would put the Chinese approach more challenged
link |
and give them more breakthrough innovation approach,
link |
more of an edge on those kinds of problems.
link |
So let me talk to that a little more.
link |
So my intuition personally is that data
link |
can take us extremely far.
link |
So you brought up autonomous vehicles and medical diagnosis.
link |
So your intuition is that huge amounts of data
link |
might not be able to completely help us solve that problem.
link |
Right, so breaking that down further in autonomous vehicle,
link |
I think huge amounts of data probably will solve
link |
trucks driving on highways,
link |
which will deliver a significant value
link |
and China will probably lead in that.
link |
And full L5 autonomous is likely to require
link |
new technologies we don't yet know.
link |
And that might require academia
link |
and great industrial research,
link |
both innovating and working together.
link |
And in that case, US has an advantage.
link |
So the interesting question there is,
link |
I don't know if you're familiar
link |
on the autonomous vehicle space
link |
and the developments with Tesla and Elon Musk.
link |
Where they are in fact full steam ahead
link |
into this mysterious complex world of full autonomy, L5,
link |
L4, L5, and they're trying to solve that purely with data.
link |
So the same kind of thing that you're saying
link |
is just for highway,
link |
which is what a lot of people share your intuition.
link |
They're trying to solve with data.
link |
So just to linger on that moment further,
link |
do you think possible for them to achieve success
link |
with simply just a huge amount of this training
link |
on edge cases and difficult cases in urban environments,
link |
not just highway and so on?
link |
I think it would be very hard.
link |
One could characterize Tesla's approach
link |
as kind of a Chinese strength approach, right?
link |
Gather all the data you can
link |
and hope that will overcome the problems.
link |
But in autonomous driving,
link |
clearly a lot of the decisions aren't merely solved
link |
by aggregating data and having feedback loop.
link |
There are things that are more akin to human thinking.
link |
And how would those be integrated and built?
link |
There has not yet been a lot of success
link |
integrating human intelligence
link |
or call it expert systems if you will,
link |
even though that's a taboo word with the machine learning.
link |
And the integration of the two types of thinking
link |
hasn't yet been demonstrated.
link |
And the question is how much can you push
link |
a purely machine learning approach?
link |
And of course, Tesla also has an additional constraint
link |
that they don't have all the sensors.
link |
I know that they think it's foolish to use LIDARs,
link |
but that's clearly a one less very valuable
link |
and reliable source of input that they're foregoing,
link |
which may also have consequences.
link |
I think the advantage of course is capturing data
link |
that no one has ever seen before.
link |
And in some cases such as computer vision
link |
and speech recognition,
link |
I have seen Chinese companies accumulate data
link |
that's not seen anywhere in the Western world
link |
and they have delivered superior results.
link |
But then speech recognition and object recognition
link |
are relatively suitable problems for deep learning
link |
and don't have the potentially need
link |
for the human intelligence analytical planning elements.
link |
And the same on the speech recognition side,
link |
your intuition that speech recognition
link |
and the machine learning approaches to speech recognition
link |
won't take us to a conversational system
link |
that can pass the Turing test,
link |
which is sort of maybe akin to what driving is.
link |
So it needs to have something more than just simply
link |
simple language understanding, simple language generation.
link |
I would say that based on purely machine learning approaches,
link |
it's hard to imagine it could lead
link |
to a full conversational experience
link |
across arbitrary domains, which is akin to L5.
link |
I'm a little hesitant to use the word Turing test
link |
because the original definition was probably too easy.
link |
We probably do that, yeah.
link |
The spirit of the Turing test
link |
is what I was referring to. Of course.
link |
So you've had major leadership research positions
link |
at Apple, Microsoft, Google.
link |
So continuing on the discussion of America, Russia,
link |
Chinese, Seoul and culture and so on.
link |
What is the culture of Silicon Valley
link |
in contrast to China and maybe US broadly?
link |
And what is the unique culture
link |
of each of these three major companies in your view?
link |
I think in aggregate, Silicon Valley companies,
link |
and we could probably include Microsoft in that,
link |
even though they're not in the Valley,
link |
is really dream big and have visionary goals
link |
and believe that technology will conquer all.
link |
And also the self confidence and the self entitlement
link |
that whatever they produce,
link |
the whole world should use and must use.
link |
And those are historically important, I think.
link |
Steve Jobs famous quote that he doesn't do focus groups,
link |
he looks in the mirror and asks the person in the mirror,
link |
And that really is an inspirational comment that says,
link |
the great company shouldn't just ask users what they want,
link |
but develop something that users will know they want
link |
but they could never come up with themselves.
link |
I think that is probably the most exhilarating description
link |
of what the essence of Silicon Valley is,
link |
that this brilliant idea could cause you to build something
link |
that couldn't come out of the focus groups or AB tests.
link |
And iPhone would be an example of that.
link |
No one in the age of Blackberry would write down
link |
they want an iPhone or multi touch.
link |
A browser might be another example.
link |
No one would say they want that in the days of FTP,
link |
but once they see it, they want it.
link |
So I think that is what Silicon Valley is best at.
link |
But it also comes with, it came with a lot of success.
link |
These products became global platforms
link |
and there were basically no competitors anywhere.
link |
And that has also led to a belief
link |
that these are the only things that one should do,
link |
that companies should not tread on other companies territory
link |
so that a Groupon and a Yelp and then OpenTable
link |
and the Grubhub would each feel,
link |
okay, I'm not gonna do the other company's business
link |
because that would not be the pride of innovating
link |
what each of these four companies have innovated.
link |
But I think the Chinese approach
link |
is do whatever it takes to win.
link |
And it's a winner take all market.
link |
And in fact, in the internet space,
link |
the market leader will get predominantly all the value
link |
extracted out of the system.
link |
So, and the system isn't just defined
link |
as one narrow category, but gets broader and broader.
link |
So it's amazing ambition for success and domination
link |
of increasingly larger product categories
link |
leading to clear market winner status
link |
and the opportunity to extract tremendous value.
link |
And that develops a practical, result oriented,
link |
ultra ambitious winner take all gladiatorial mentality.
link |
And if what it takes is to build what the competitors built,
link |
essentially a copycat that can be done
link |
without infringing laws.
link |
If what it takes is to satisfy a foreign country's need
link |
by forking the code base and building something
link |
that looks really ugly and different, they'll do it.
link |
So it's contrasted very sharply
link |
with the Silicon Valley approach.
link |
And I think the flexibility and the speed and execution
link |
has helped the Chinese approach.
link |
And I think the Silicon Valley approach
link |
is potentially challenged if every Chinese entrepreneur
link |
is learning from the whole world, US and China,
link |
and the American entrepreneurs only look internally
link |
and write off China as a copycat.
link |
And the second part of your question
link |
about the three companies.
link |
The unique elements of the three companies perhaps.
link |
Yeah, I think Apple represents
link |
while the user please the user
link |
and the essence of design and brand
link |
and it's the one company and perhaps the only tech company
link |
that draws people with a strong, serious desire
link |
for the product and the willingness to pay a premium
link |
because of the halo effect of the brand
link |
which came from the attention to detail
link |
and great respect for user needs.
link |
Microsoft represents a platform approach
link |
that builds giant products that become very strong modes
link |
that others can't do because it's well architected
link |
at the bottom level and the work is efficiently delegated
link |
to individuals and then the whole product is built
link |
by adding small parts that sum together.
link |
So it's probably the most effective high tech assembly line
link |
that builds a very difficult product
link |
that and the whole process of doing that
link |
is kind of a differentiation and something competitors
link |
can't easily repeat.
link |
Are there elements of the Chinese approach
link |
in the way Microsoft went about assembling
link |
those little pieces and dominating,
link |
essentially dominating the market for a long time
link |
or do you see those as distinct?
link |
I think there are elements that are the same.
link |
I think the three American companies that had
link |
or have Chinese characteristics and obviously
link |
as well as American characteristics are Microsoft,
link |
Facebook and Amazon.
link |
Yes, that's right, Amazon.
link |
Because these are companies that will tenaciously
link |
go after adjacent markets,
link |
build up strong product offering and find ways
link |
to extract greater value from a sphere
link |
that's ever increasing and they understand
link |
the value of the platforms.
link |
So that's the similarity and then with Google,
link |
I think it's a genuinely value oriented company
link |
that does have a heart and soul
link |
and that wants to do great things for the world
link |
by connecting information
link |
and that has also very strong technology genes
link |
and wants to use technology
link |
and has found out of the box ways to use technology
link |
to deliver incredible value to the end user.
link |
If you can look at Google, for example,
link |
you mentioned heart and soul.
link |
There seems to be an element where Google
link |
is after making the world better.
link |
There's a more positive view.
link |
They used to have the slogan, don't be evil.
link |
And Facebook a little bit more has a negative tend to it.
link |
At least in the perception of privacy and so on.
link |
Do you have a sense of how these different companies
link |
can achieve, because you've talked about
link |
how much we can make the world better
link |
in all these kinds of ways with AI.
link |
What is it about a company that can make,
link |
give it a heart and soul, gain the trust of the public
link |
and just actually just not be evil
link |
and do good for the world?
link |
It's really hard and I think Google
link |
has struggled with that.
link |
First, the don't do evil mantra is very dangerous
link |
because every employee's definition of evil is different.
link |
And that has led to some difficult
link |
employee situations for them.
link |
So I don't necessarily think that's a good value statement,
link |
but just watching the kinds of things Google
link |
or its parent company Alphabet does
link |
in new areas like healthcare, like eradicating mosquitoes,
link |
things that are really not in the business
link |
of a internet tech company.
link |
I think that shows that there's a heart and soul
link |
and desire to do good and willingness
link |
to put in the resources to do something
link |
when they see it's good, they will pursue it.
link |
That doesn't necessarily mean
link |
it has all the trust of the users.
link |
I realize while most people would view Facebook
link |
as the primary target of their recent unhappiness
link |
about Silicon Valley companies,
link |
many would put Google in that category.
link |
And some have named Google's business practices
link |
as predatory also.
link |
So it's kind of difficult to have the two parts of a body.
link |
The brain wants to do what it's supposed to do
link |
for a shareholder, maximize profit.
link |
And then the heart and soul wants to do good things
link |
that may run against what the brain wants to do.
link |
So in this complex balancing
link |
that these companies have to do,
link |
you've mentioned that you're concerned
link |
about a future where too few companies
link |
like Google, Facebook, Amazon are controlling our data
link |
or controlling too much of our digital lives.
link |
Can you elaborate on this concern
link |
and perhaps do you have a better way forward?
link |
I think I'm hardly the most vocal complainer of this.
link |
There are a lot louder complainers out there.
link |
I do observe that having a lot of data
link |
does perpetuate their strength
link |
and limits competition in many spaces.
link |
But I also believe AI is much broader
link |
than the internet space.
link |
So the entrepreneurial opportunities still exists
link |
in using AI to empower financial,
link |
retail, manufacturing, education applications.
link |
So I don't think it's quite a case
link |
of full monopolistic dominance
link |
that totally stifles innovation.
link |
But I do believe in their areas of strength
link |
it's hard to dislodge them.
link |
I don't know if I have a good solution.
link |
Probably the best solution is let
link |
the entrepreneurial VC ecosystem work well
link |
and find all the places that can create the next Google,
link |
the next Facebook.
link |
So there will always be increasing number of challengers.
link |
In some sense that has happened a little bit.
link |
You see Uber, Airbnb having emerged
link |
despite the strength of the big three.
link |
And I think China as an environment
link |
may be more interesting for the emergence
link |
because if you look at companies
link |
between let's say 50 to $300 billion,
link |
China has emerged more of such companies
link |
than the US in the last three to four years
link |
because of the larger marketplace,
link |
because of the more fearless nature of the entrepreneurs.
link |
And the Chinese giants are just as powerful
link |
Tencent, Alibaba are very strong,
link |
but ByteDance has emerged worth 75 billion
link |
and financial while it's Alibaba affiliated,
link |
it's nevertheless independent and worth 150 billion.
link |
And so I do think if we start to extend
link |
to traditional businesses,
link |
we will see very valuable companies.
link |
So it's probably not the case that in five or 10 years
link |
we'll still see the whole world
link |
with these five companies having such dominance.
link |
So you've mentioned a couple of times
link |
this fascinating world of entrepreneurship in China
link |
of the fearless nature of the entrepreneur.
link |
So can you maybe talk a little bit about
link |
what it takes to be an entrepreneur in China?
link |
What are the strategies that are undertaken?
link |
What are the ways to achieve success?
link |
What is the dynamic of VCF funding
link |
of the way the government helps companies and so on?
link |
What are the interesting aspects here
link |
that are distinct from, that are different
link |
from the Silicon Valley world of entrepreneurship?
link |
Well, many of the listeners probably still
link |
would brand Chinese entrepreneur as copycats.
link |
And no doubt 10 years ago,
link |
that would not be an inaccurate description.
link |
Back 10 years ago,
link |
an entrepreneur probably could not get funding
link |
if he or she could not describe
link |
what product he or she is copying from the US.
link |
The first question is who has proven this business model
link |
which is a nice way of asking who are you copying?
link |
And that reason is understandable
link |
because China had a much lower internet penetration
link |
and didn't have enough indigenous experience
link |
to build innovative products.
link |
And secondly, internet was emerging.
link |
Link startup was the way to do things,
link |
building a first minimally viable product
link |
and then expanding was the right way to go.
link |
And the American successes have given a shortcut
link |
that if you built your minimally viable product
link |
based on an American product,
link |
it's guaranteed to be a decent starting point.
link |
Then you tweak it afterwards.
link |
So as long as there are no IP infringement,
link |
which as far as I know there hasn't been in the mobile
link |
and AI spaces, that's a much better shortcut.
link |
And I think Silicon Valley would view that
link |
as still not very honorable
link |
because that's not your own idea to start with,
link |
but you can't really at the same time
link |
believe every idea must be your own
link |
and believe in the link startup methodology
link |
because link startup is intended to try many, many things
link |
and then converge when that works.
link |
And it's meant to be iterated and changed.
link |
So finding a decent starting point
link |
without legal violations,
link |
there should be nothing morally dishonorable about that.
link |
Yeah, so just a quick pause on that.
link |
It's fascinating that that's,
link |
why is that not honorable, right?
link |
It's exactly as you formulated.
link |
It seems like a perfect start for business.
link |
Is to take, look at Amazon and say,
link |
okay, we'll do exactly what Amazon is doing.
link |
Let's start there in this particular market
link |
and then let's out innovate them from that starting point.
link |
Come up with new ways.
link |
I mean, is it wrong to be,
link |
except the word copycat just sounds bad,
link |
but is it wrong to be a copycat?
link |
It just seems like a smart strategy,
link |
but yes, it doesn't have a heroic nature to it
link |
that like Steve Jobs, Elon Musk,
link |
sort of in something completely,
link |
coming up with something completely new.
link |
Yeah, I like the way you describe it.
link |
It's a nonheroic, acceptable way to start the company
link |
and maybe more expedient.
link |
So that's, I think, a baggage for Silicon Valley
link |
that if it doesn't let go,
link |
then it may limit the ultimate ceiling of the company.
link |
Take Snapchat as an example.
link |
I think, you know, Evan's brilliant.
link |
He built a great product,
link |
but he's very proud that he wants to build his own features,
link |
While Facebook was more willing to copy his features
link |
and you see what happens in the competition.
link |
So I think putting that handcuff on the company
link |
would limit its ability to reach the maximum potential.
link |
So back to the Chinese environment,
link |
copying was merely a way to learn from the American masters.
link |
Just like we, if we learned to play piano or painting,
link |
you start by copying.
link |
You don't start by innovating
link |
when you don't have the basic skill sets.
link |
So very amazingly, the Chinese entrepreneurs
link |
about six years ago started to branch off
link |
with these lean startups built on American ideas
link |
to build better products than American products.
link |
But they did start from the American idea.
link |
And today WeChat is better than WhatsApp,
link |
Weibo is better than Twitter,
link |
Zhihu is better than Quora and so on.
link |
So that I think is Chinese entrepreneurs going to step two.
link |
And then step three is once these entrepreneurs
link |
have done one or two of these companies,
link |
they now look at the Chinese market and the opportunities
link |
and come up with ideas that didn't exist elsewhere.
link |
So products like Ant Financial,
link |
under which includes Alipay, which is mobile payments,
link |
and also the financial products for loans built on that.
link |
And also in education, VIPKID,
link |
and in social video, social network, TikTok,
link |
and in social eCommerce, Pinduoduo,
link |
and then in ride sharing, Mobike,
link |
these are all Chinese innovated products
link |
that now are being copied elsewhere.
link |
So an additional interesting observation
link |
is some of these products
link |
are built on unique Chinese demographics,
link |
which may not work in the US,
link |
but may work very well in Southeast Asia, Africa,
link |
and other developing worlds
link |
that are a few years behind China.
link |
And a few of these products maybe are universal
link |
and are getting traction even in the United States,
link |
So this whole ecosystem is supported by VCs
link |
as a virtuous cycle,
link |
because a large market with innovative entrepreneurs
link |
will draw a lot of money
link |
and then invest in these companies.
link |
As the market gets larger and larger,
link |
the China market is easily three, four times larger than the US,
link |
they will create greater value and greater returns
link |
for the VCs, thereby raising even more money.
link |
So at Sinovation Ventures, our first fund was 15 million,
link |
our last fund was 500 million.
link |
So it reflects the valuation of the companies
link |
and our us going multi stage and things like that.
link |
It also has government support,
link |
but not in the way most Americans would think of it.
link |
The government actually leaves the entrepreneurial space
link |
as a private enterprise, sort of self regulating,
link |
and the government would build infrastructures
link |
that would around it to make it work better.
link |
For example, the Mass Entrepreneur Mass Innovation Plan
link |
builds 8,000 incubators,
link |
so the pipeline is very strong to the VCs.
link |
For autonomous vehicles,
link |
the Chinese government is building smart highways
link |
with sensors, smart cities
link |
that separate pedestrians from cars
link |
that may allow initially an inferior
link |
autonomous vehicle company to launch a car
link |
without increasing with lower casualty
link |
because the roads or the city is smart.
link |
And the Chinese government at local levels
link |
would have these guiding funds acting as LPs,
link |
passive LPs to funds.
link |
And when the fund makes money,
link |
part of the money made is given back
link |
to the GPs and potentially other LPs
link |
to increase everybody's return
link |
at the expense of the government's return.
link |
So that's an interesting incentive
link |
that entrusts the task of choosing entrepreneurs to VCs
link |
who are better at it than the government
link |
by letting some of the profits move that way.
link |
So this is really fascinating, right?
link |
So I look at the Russian government as a case study
link |
where, let me put it this way,
link |
there's no such government driven
link |
large scale support of entrepreneurship.
link |
And probably the same is true in the United States,
link |
but the entrepreneurs themselves kind of find a way.
link |
So maybe in a form of advice or explanation,
link |
how did the Chinese government arrive to be this way
link |
so supportive on entrepreneurship
link |
to be in this particular way so forward thinking
link |
at such a large scale?
link |
And also perhaps, how can we copy it in other countries?
link |
How can we encourage other governments,
link |
like even the United States government,
link |
to support infrastructure for autonomous vehicles
link |
in that same kind of way, perhaps?
link |
Yes, so these techniques are
link |
the result of several key things,
link |
some of which may be learnable,
link |
some of which may be very hard.
link |
One is just trial and error
link |
and watching what everyone else is doing.
link |
I think it's important to be humble
link |
and not feel like you know all the answers.
link |
The guiding funds idea came from Singapore,
link |
which came from Israel.
link |
And China made a few tweaks and turned it into a,
link |
because the Chinese cities and government officials
link |
kind of compete with each other
link |
because they all want to make their city more successful
link |
so they can get the next level in their political career.
link |
And it's somewhat competitive.
link |
So the central government made it a bit of a competition.
link |
Everybody has a budget.
link |
They can put it on AI or they can put it on bio
link |
or they can put it on energy.
link |
And then whoever gets the results,
link |
the city shines, the people are better off,
link |
the mayor gets a promotion.
link |
So the tools is kind of almost like an entrepreneurial
link |
environment for local governments
link |
to see who can do a better job.
link |
And also many of them try different experiments.
link |
Some have given award to very smart researchers.
link |
Just give them money and hope they'll start a company.
link |
Some have given money to academic research labs,
link |
maybe government research labs
link |
to see if they can spin off some companies
link |
from the science lab or something like that.
link |
Some have tried to recruit overseas Chinese
link |
to come back and start companies.
link |
And they've had mixed results.
link |
The one that worked the best was the guiding funds.
link |
So it's almost like a lean startup idea
link |
where people try different things
link |
and what works sticks and everybody copies.
link |
So now every city has a guiding fund.
link |
So that's how that came about.
link |
The autonomous vehicle and the massive spending
link |
in highways and smart cities, that's a Chinese way.
link |
It's about building infrastructure to facilitate.
link |
It's a clear division of the government's responsibility
link |
The market should do everything in a private freeway,
link |
but there are things the market can't afford to do
link |
like infrastructure.
link |
So the government always appropriates large amounts
link |
of money for infrastructure building.
link |
This happens with not only autonomous vehicle and AI,
link |
but happened with the 3G and 4G.
link |
You'll find that the Chinese wireless reception
link |
is better than the US because massive spending
link |
that tries to cover the whole country,
link |
whereas in the US it may be a little spotty.
link |
It's a government driven because I think they view
link |
the coverage of cell access and 3G, 4G access
link |
to be a governmental infrastructure spending
link |
as opposed to capitalistic.
link |
So that's, of course, the state owned enterprises
link |
are also publicly traded,
link |
but they also carry a government responsibility
link |
to deliver infrastructure to all.
link |
So it's a different way of thinking
link |
that may be very hard to inject into Western countries
link |
to say starting tomorrow, bandwidth infrastructure
link |
and highways are gonna be governmental spending
link |
with some characteristics.
link |
What's your sense, and sorry to interrupt,
link |
but because it's such a fascinating point,
link |
do you think on the autonomous vehicle space
link |
it's possible to solve the problem of full autonomy
link |
without significant investment in infrastructure?
link |
Well, that's really hard to speculate.
link |
I think it's not a yes, no question,
link |
but how long does it take question?
link |
15 years, 30 years, 45 years.
link |
Clearly with infrastructure augmentation,
link |
whether it's road, the city or whole city planning,
link |
building a new city, I'm sure that will accelerate
link |
the day of the L5.
link |
I'm not knowledgeable enough,
link |
and it's hard to predict even when we're knowledgeable
link |
because a lot of it is speculative.
link |
But in the US, I don't think people would consider
link |
building a new city the size of Chicago
link |
to make it the AI slash autonomous city.
link |
There are smaller ones being built, I'm aware of that.
link |
But is infrastructure spend really impossible
link |
for US or Western countries?
link |
The US highway system was built,
link |
was that during President Eisenhower or Kennedy?
link |
So maybe historians can study
link |
how the President Eisenhower get the resources
link |
to build this massive infrastructure
link |
that surely gave US a tremendous amount of prosperity
link |
over the next decade, if not century.
link |
If I may comment on that then,
link |
it takes us to artificial intelligence a little bit
link |
because in order to build infrastructure,
link |
it creates a lot of jobs.
link |
So I'll be actually interested if you would say
link |
that you talk in your book about all kinds of jobs
link |
that could and could not be automated.
link |
I wonder if building infrastructure is one of the jobs
link |
that would not be easily automated.
link |
Something you could think about
link |
because I think you've mentioned somewhere in the talk
link |
or that there might be, as jobs are being automated,
link |
a role for government to create jobs
link |
that can't be automated.
link |
Yes, I think that's a possibility.
link |
Back in the last financial crisis,
link |
China put a lot of money
link |
to basically give this economy a boost
link |
and a lot of it went into infrastructure building.
link |
And I think that's a legitimate way at the government level
link |
to deal with the employment issues
link |
as well as build out the infrastructure
link |
as long as the infrastructures are truly needed
link |
and as long as there is an employment problem,
link |
which no, we don't know.
link |
So maybe taking a little step back,
link |
if you've been a leader and a researcher in AI
link |
for several decades, at least 30 years,
link |
so how has AI changed in the West and the East
link |
as you've observed, as you've been deep in it
link |
over the past 30 years?
link |
Well, AI began as the pursuit
link |
of understanding human intelligence
link |
and the term itself represents that,
link |
but it kind of drifted into the one sub area
link |
that worked extremely well, which is machine intelligence.
link |
And that's actually more using pattern recognition techniques
link |
to basically do incredibly well on a limited domain,
link |
large amount of data,
link |
but relatively simple kinds of planning tasks
link |
and not very creative.
link |
So we didn't end up building human intelligence.
link |
We built a different machine
link |
that was a lot better than us, some problems,
link |
but nowhere close to us on other problems.
link |
So today, I think a lot of people still misunderstand
link |
when we say artificial intelligence
link |
and what various products can do,
link |
people still think it's about replicating
link |
human intelligence,
link |
but the products out there really are closer
link |
to having invented the internet or the spreadsheet
link |
or the database and getting broader adoption.
link |
And speaking further to the fears,
link |
near term fears that people have about AI,
link |
so you're commenting on the sort of general intelligence
link |
that people in the popular culture from sci fi movies
link |
have a sense about AI,
link |
but there's practical fears about AI,
link |
the narrow AI that you're talking about
link |
of automating particular kinds of jobs
link |
and you talk about them in the book.
link |
So what are the kinds of jobs in your view
link |
that you see in the next five, 10 years
link |
beginning to be automated by AI systems algorithms?
link |
Yes, this is also maybe a little bit counterintuitive
link |
because it's the routine jobs
link |
that will be displaced the soonest
link |
and they may not be displaced entirely,
link |
maybe 50%, 80% of a job,
link |
but when the workload drops by that much,
link |
employment will come down.
link |
And also another part of misunderstanding
link |
is most people think of AI replacing routine jobs
link |
than they think of the assembly line, the workers.
link |
Well, that will have some effect,
link |
but it's actually the routine white collar workers
link |
that's easiest to replace
link |
because to replace a white collar worker,
link |
you just need software.
link |
To replace a blue collar worker,
link |
you need robotics, mechanical excellence,
link |
and the ability to deal with dexterity
link |
and maybe even unknown environments, very, very difficult.
link |
So if we were to categorize the most dangerous
link |
white collar jobs,
link |
they would be things like back office,
link |
people who copy and paste
link |
and deal with simple computer programs and data
link |
and maybe paper and OCR,
link |
and they don't make strategic decisions.
link |
They basically facilitate the process.
link |
These softwares and paper systems don't work.
link |
So you have people dealing with new employee orientation,
link |
searching for past lawsuits and financial documents,
link |
and doing reference check.
link |
So basic searching and management of data.
link |
That's the most endangered being lost.
link |
In addition to the white collar repetitive work,
link |
a lot of simple interaction work can also be taken care of
link |
such as telesales, telemarketing, customer service,
link |
as well as many physical jobs
link |
that are in the same location
link |
and don't require a high degree of dexterity.
link |
So fruit picking, dishwashing, assembly line inspection
link |
are jobs in that category.
link |
So altogether, back office is a big part.
link |
And the blue collar may be smaller initially,
link |
but over time, AI will get better.
link |
And when we start to get to over the next 15, 20 years,
link |
the ability to actually have the dexterity
link |
of doing assembly line, that's a huge chunk of jobs.
link |
And when autonomous vehicles start to work,
link |
initially starting with truck drivers,
link |
but eventually to all drivers,
link |
that's another huge group of workers.
link |
So I see modest numbers in the next five years,
link |
but increasing rapidly after that.
link |
On the worry of the jobs that are in danger
link |
and the gradual loss of jobs,
link |
I'm not sure if you're familiar with Andrew Yang.
link |
So there's a candidate for president of the United States
link |
whose platform Andrew Yang is based around,
link |
in part around job loss due to automation.
link |
And also in addition,
link |
the need perhaps of universal basic income
link |
to support jobs that are,
link |
folks who lose their job due to automation and so on.
link |
And in general, support people
link |
under complex, unstable job market.
link |
So what are your thoughts about his concerns,
link |
him as a candidate, his ideas in general?
link |
I think his thinking is generally in the right direction,
link |
but his approach as a presidential candidate
link |
may be a little bit ahead of the time.
link |
And I think the displacements will happen,
link |
but will they happen soon enough
link |
for people to agree to vote for him?
link |
The unemployment numbers are not very high yet.
link |
And I think he and I have the same challenge.
link |
If I want to theoretically convince people this is an issue
link |
and he wants to become the president,
link |
people have to see how can this be the case
link |
when unemployment numbers are low.
link |
So that is the challenge.
link |
And I think I do agree with him on the displacement issue,
link |
on universal basic income.
link |
At a very vanilla level, I don't agree with it
link |
because I think the main issue is retraining.
link |
So people need to be incented
link |
not by just giving a monthly $2,000 check or $1,000 check
link |
and do whatever they want
link |
because they don't have the know how
link |
to know what to retrain to go into what type of a job.
link |
And guidance is needed.
link |
And retraining is needed
link |
because historically when technology revolutions,
link |
when routine jobs were displaced, new routine jobs came up.
link |
So there was always room for that.
link |
But with AI and automation,
link |
the whole point is replacing all routine jobs eventually.
link |
So there will be fewer and fewer routine jobs.
link |
And AI will create jobs, but it won't create routine jobs
link |
because if it creates routine jobs,
link |
why wouldn't AI just do it?
link |
So therefore the people who are losing the jobs
link |
are losing routine jobs.
link |
The jobs that are becoming available are non routine jobs.
link |
So the social stipend needs to be put in place
link |
is for the routine workers who lost their jobs
link |
to be retrained maybe in six months, maybe in three years,
link |
takes a while to retrain on the non routine job
link |
and then take on a job that will last
link |
for that person's lifetime.
link |
Now, having said that,
link |
if you look deeply into Andrew's document,
link |
he does cater for that.
link |
So I'm not disagreeing with what he's trying to do.
link |
But for simplification, sometimes he just says UBI,
link |
but simple UBI wouldn't work.
link |
And I think you've mentioned elsewhere
link |
that the goal isn't necessarily to give people enough money
link |
to survive or live, or even to prosper.
link |
The point is to give them a job that gives them meaning.
link |
That meaning is extremely important.
link |
That our employment, at least in the United States
link |
and perhaps it carries across the world,
link |
provides something that's, forgive me for saying,
link |
greater than money.
link |
It provides meaning.
link |
So now, what kind of jobs do you think can't be automated?
link |
Can you talk a little bit about creativity
link |
and compassion in your book?
link |
What aspects do you think it's difficult to automate
link |
Because an AI system is currently merely optimizing.
link |
It's not able to reason, plan,
link |
or think creatively or strategically.
link |
It's not able to deal with complex problems.
link |
It can't come up with a new problem and solve it.
link |
A human needs to find the problem
link |
and pose it as an optimization problem,
link |
then have the AI work at it.
link |
So an AI would have a very hard time discovering a new drug
link |
or discovering a new style of painting
link |
or dealing with complex tasks such as managing a company
link |
that isn't just about optimizing the bottom line,
link |
but also about employee satisfaction, corporate brand,
link |
and many, many other things.
link |
So that is one category of things.
link |
And because these things are challenging, creative, complex,
link |
doing them creates a high degree of satisfaction
link |
and therefore appealing to our desire for working,
link |
which isn't just to make the money, make the ends meet,
link |
but also that we've accomplished something
link |
that others maybe can't do or can't do as well.
link |
Another type of job that is much numerous
link |
would be compassionate jobs, jobs that require compassion,
link |
empathy, human touch, human trust.
link |
AI can't do that because AI is cold, calculating,
link |
and even if it can fake that to some extent,
link |
it will make errors and that will make it look very silly.
link |
And also, I think even if AI did okay,
link |
people would want to interact with another person,
link |
whether it's for some kind of a service or a teacher
link |
or a doctor or a concierge or a masseuse or a bartender.
link |
There are so many jobs where people just don't want
link |
to interact with a cold robot or software.
link |
I've had an entrepreneur who built an elderly care robot
link |
and they found that the elderly really only use it
link |
for customer service.
link |
And not, but not to service the product,
link |
but they click on customer service
link |
and the video of a person comes up
link |
and then the person says,
link |
how come my daughter didn't call me?
link |
Let me show you a picture of her grandkids.
link |
So people yearn for that people, people interaction.
link |
So even if robots improved, people just don't want it.
link |
And those jobs are going to be increasing
link |
because AI will create a lot of value,
link |
$16 trillion to the world in the next 10 years.
link |
Next 11 years, according to PWC.
link |
And that will give people money to enjoy services,
link |
whether it's eating a gourmet meal or tourism and traveling
link |
or having concierge services,
link |
the services revolving around every dollar
link |
of that $16 trillion will be tremendous.
link |
It will create more opportunities
link |
that are to service the people who did well
link |
through AI with things.
link |
But even at the same time,
link |
the entire society is very much short
link |
in need of many service oriented,
link |
compassionate oriented jobs.
link |
The best example is probably in healthcare services.
link |
There's going to be 2 million new jobs,
link |
not counting replacement,
link |
just brand new incremental jobs
link |
in the next six years in healthcare services.
link |
That includes nurses, orderly in the hospital,
link |
elderly care and also at home care is particularly lacking.
link |
And those jobs are not likely to be filled.
link |
So there's likely to be a shortage.
link |
And the reason they're not filled
link |
is simply because they don't pay very well
link |
and that the social status of these jobs are not very good.
link |
So they pay about half as much
link |
as a heavy equipment operator,
link |
which will be replaced a lot sooner.
link |
And they pay probably comparably
link |
to someone on the assembly line.
link |
And so if we ignoring all the other issues
link |
and just think about satisfaction from one's job,
link |
someone repetitively doing the same manual action
link |
at an assembly line,
link |
that can't create a lot of job satisfaction,
link |
but someone taking care of a sick person
link |
and getting a hug and thank you
link |
from that person and the family,
link |
I think is quite satisfying.
link |
So if only we could fix the pay for service jobs,
link |
there are plenty of jobs that require some training
link |
or a lot of training
link |
for the people coming off the routine jobs to take.
link |
We can easily imagine someone
link |
who was maybe a cashier at the grocery store
link |
as stores become automated,
link |
learns to become a nurse or an at home care.
link |
I also do want to point out the blue collar jobs
link |
are going to stay around a bit longer.
link |
Some of them quite a bit longer.
link |
AI cannot be told go clean an arbitrary home.
link |
That's incredibly hard.
link |
Arguably it's an L5 level of difficulty, right?
link |
And then AI cannot be a good plumber
link |
because plumber is almost like a mini detective
link |
that has to figure out where the leak came from.
link |
So yet AI probably can be an assembly line
link |
and auto mechanic and so on.
link |
So one has to study which blue collar jobs are going away
link |
and facilitate retraining for the people
link |
to go into the ones that won't go away
link |
or maybe even will increase.
link |
I mean, it is fascinating that it's easier
link |
to build a world champion chess player
link |
than it is to build a mediocre plumber.
link |
And to AI and that goes counterintuitive
link |
to a lot of people's understanding
link |
of what artificial intelligence is.
link |
So it sounds, I mean, you're painting
link |
a pretty optimistic picture about retraining
link |
about the number of jobs
link |
and actually the meaningful nature of those jobs
link |
once we automate the repetitive tasks.
link |
So overall, are you optimistic about the future
link |
where much of the repetitive tasks are automated?
link |
That there is a lot of room for humans
link |
for the compassionate, for the creative input
link |
that only humans can provide?
link |
I am optimistic if we start to take action.
link |
If we have no action in the next five years,
link |
I think it's going to be hard to deal
link |
with the devastating losses that will emerge.
link |
So if we start thinking about retraining,
link |
maybe with the low hanging fruits,
link |
explaining to vocational schools
link |
why they should train more plumbers than auto mechanics,
link |
maybe starting with some government subsidy
link |
for corporations to have more training positions.
link |
We start to explain to people why retraining is important.
link |
We start to think about what the future of education,
link |
how that needs to be tweaked for the era of AI.
link |
If we start to make incremental progress
link |
and the greater number of people understand,
link |
then there's no reason to think we can't deal with this
link |
because this technological revolution
link |
is arguably similar to what electricity,
link |
industrial revolutions, and internet brought about.
link |
Do you think there's a role for policy,
link |
for governments to step in,
link |
to help with policy to create a better world?
link |
Absolutely, and the governments don't have to believe
link |
an employment will go up,
link |
and they don't have to believe automation will be this fast
link |
Revamping vocational school would be one example.
link |
Another is if there's a big gap
link |
in healthcare service employment,
link |
and we know that a country's population is growing older,
link |
more longevity, living older,
link |
because people over 80 require five times as much care
link |
as those under 80,
link |
then it is a good time to incent training programs
link |
for elderly care to find ways to improve the pay.
link |
Maybe one way would be to offer as part of Medicare
link |
or the equivalent program for people over 80
link |
to be entitled to a few hours of elderly care at home,
link |
and then that might be reimbursable,
link |
and that will stimulate the service industry
link |
around the policy.
link |
Do you have concerns about large entities,
link |
whether it's governments or companies,
link |
controlling the future of AI development in general?
link |
So we talked about companies.
link |
Do you have a better sense that governments
link |
can better represent the interests of the people
link |
than companies, or do you believe companies
link |
are better at representing the interests of the people?
link |
Or is there no easy answer?
link |
I don't think there's an easy answer
link |
because it's a double edged sword.
link |
The companies and governments can provide better services
link |
with more access to data and more access to AI,
link |
but that also leads to greater power,
link |
which can lead to uncontrollable problems,
link |
whether it's monopoly or corruption in the government.
link |
So I think one has to be careful
link |
to look at how much data that companies and governments have
link |
and some kind of checks and balances would be helpful.
link |
So again, I come from Russia.
link |
There's something called the Cold War.
link |
So let me ask a difficult question here
link |
looking at conflict.
link |
Steven Pinker written a great book
link |
that conflict all over the world is decreasing in general.
link |
But do you have a sense that having written
link |
the book AI Superpowers,
link |
do you see a major international conflict
link |
potentially arising between major nations,
link |
whatever they are, whether it's Russia, China,
link |
European nations, United States or others
link |
in the next 10, 20, 50 years around AI,
link |
around the digital space, cyberspace?
link |
Do you worry about that?
link |
Is that something we need to think about
link |
and try to alleviate or prevent?
link |
I believe in greater engagement.
link |
A lot of the worries about more powerful AI
link |
are based on a arms race metaphor.
link |
And when you extrapolate into military kinds of scenarios,
link |
AI can automate and autonomous weapons
link |
that needs to be controlled somehow
link |
and autonomous decision making
link |
can lead to not enough time to fix international crises.
link |
So I actually believe a Cold War mentality
link |
would be very dangerous
link |
because should two countries rely on AI
link |
to make certain decisions
link |
and they don't even talk to each other,
link |
they do their own scenario planning,
link |
then something could easily go wrong.
link |
I think engagement, interaction, some protocols
link |
to avoid inadvertent disasters is actually needed.
link |
So it's natural for each country to want to be the best,
link |
whether it's in nuclear technologies or AI or bio.
link |
But I think it's important to realize
link |
if each country has a black box AI
link |
and don't talk to each other,
link |
that probably presents greater challenges to humanity
link |
than if they interacted.
link |
I think there can still be competition,
link |
but with some degree of protocol for interaction,
link |
just like when there was a nuclear competition,
link |
there were some protocol for deterrence
link |
among US, Russia, and China.
link |
And I think that engagement is needed.
link |
So of course, we're still far from AI
link |
presenting that kind of danger.
link |
But what I worry the most about
link |
is the level of engagement seems to be coming down.
link |
The level of distrust seems to be going up,
link |
especially from the US towards other large countries
link |
such as China and of course, and Russia, yes.
link |
Is there a way to make that better?
link |
So let's beautifully put level of engagement
link |
and even just basic trust and communication
link |
as opposed to sort of making artificial enemies
link |
out of particular countries.
link |
Do you have a sense how we can make it better?
link |
Actionable items that as a society we can take on?
link |
I'm not an expert at geopolitics,
link |
but I would say that we look pretty foolish as humankind
link |
when we are faced with the opportunity
link |
to create $16 trillion for humanity,
link |
and yet we're not solving fundamental problems
link |
with parts of the world still in poverty.
link |
And for the first time,
link |
we have the resources to overcome poverty and hunger.
link |
We're not using it on that,
link |
but we're fueling competition among superpowers.
link |
And that's a very unfortunate thing.
link |
If we become utopian for a moment,
link |
imagine a benevolent world government
link |
that has this $16 trillion and maybe some AI
link |
to figure out how to use it to deal with diseases
link |
and problems and hate and things like that.
link |
World would be a lot better off.
link |
So what is wrong with the current world?
link |
I think the people with more skill than I
link |
should think about this.
link |
And then the geopolitics issue with superpower competition
link |
is one side of the issue.
link |
There's another side which I worry maybe even more,
link |
which is as the $16 trillion all gets made by US and China
link |
and a few of the other developed countries,
link |
the poorer country will get nothing
link |
because they don't have technology
link |
and the wealth disparity and inequality will increase.
link |
So a poorer country with a large population
link |
will not only benefit from the AI boom
link |
or other technology booms,
link |
but they will have their workers
link |
who previously had hoped they could do the China model
link |
and do outsource manufacturing or the India model
link |
so they could do the outsource process or call center.
link |
Well, all those jobs are gonna be gone in 10 or 15 years.
link |
So the individual citizen may be a net liability,
link |
I mean, financially speaking to a poorer country
link |
and not an asset to claw itself out of poverty.
link |
So in that kind of situation,
link |
these large countries with not much tech
link |
are going to be facing a downward spiral
link |
and it's unclear what could be done.
link |
And then when we look back
link |
and say there's $16 trillion being created
link |
and it's all being kept by US, China
link |
and other developed countries, it just doesn't feel right.
link |
So I hope people who know about geopolitics
link |
can find solutions that's beyond my expertise.
link |
So different countries that we've talked about
link |
have different value systems.
link |
If you look at the United States,
link |
to an almost extreme degree,
link |
there is an absolute desire for freedom of speech.
link |
If you look at a country where I was raised,
link |
that desire just amongst the people
link |
is not as elevated as it is to basically fundamental level
link |
to the essence of what it means to be America, right?
link |
And the same is true with China,
link |
there's different value systems.
link |
There's some censorship of internet content
link |
that China and Russia and many other countries undertake.
link |
Do you see that having effects on innovation,
link |
other aspects of some of the tech stuff,
link |
AI development we talked about,
link |
and maybe from another angle,
link |
do you see that changing in different ways
link |
over the next 10 years, 20 years, 50 years
link |
as China continues to grow as it does now
link |
in its tech innovation?
link |
There's a common belief
link |
that full freedom of speech and expression
link |
is correlated with creativity,
link |
which is correlated with entrepreneurial success.
link |
I think empirically we have seen that is not true
link |
and China has been successful.
link |
That's not to say the fundamental values are not right
link |
but it's just that perfect correlation isn't there.
link |
It's hard to read the tea leaves on opening up or not
link |
and I've not been very good at that in my past predictions,
link |
but I do believe every country
link |
shares a lot of fundamental values for the longterm.
link |
So China is drafting its privacy policy
link |
for individual citizens,
link |
and they don't look that different
link |
from the American or European ones.
link |
So people do want to protect their privacy
link |
and have the opportunity to express
link |
and I think the fundamental values are there.
link |
The question is in the execution and timing,
link |
how soon or when will that start to open up?
link |
So as long as each government knows
link |
ultimately people want that kind of protection,
link |
there should be a plan to move towards that
link |
as to when or how and I'm not an expert.
link |
On the point of privacy to me, it's really interesting.
link |
So AI needs data to create
link |
a personalized awesome experience, right?
link |
I'm just speaking generally in terms of products.
link |
And then we have currently, depending on the age
link |
and depending on the demographics of who we're talking about,
link |
some people are more or less concerned
link |
about the amount of data they hand over.
link |
So in your view, how do we get this balance right
link |
that we provide an amazing experience
link |
to people that use products?
link |
You look at Facebook, the more Facebook knows about you,
link |
yes, it's scary to say, the better it can probably,
link |
better experience it can probably create.
link |
So in your view, how do we get that balance right?
link |
Yes, I think a lot of people have a misunderstanding
link |
that it's okay and possible to just rip all the data out
link |
from a provider and give it back to you.
link |
So you can deny them access to further data
link |
and still enjoy the services we have.
link |
If we take back all the data,
link |
all the services will give us nonsense.
link |
We'll no longer be able to use products that function well
link |
in terms of right ranking, right products,
link |
right user experience.
link |
So yet I do understand we don't want to permit misuse
link |
of the data from legal policy standpoint.
link |
I think there can be severe punishment
link |
for those who have egregious misuse of the data.
link |
That's I think a good first step.
link |
Actually China in this side on this aspect
link |
has very strong laws about people who sell
link |
or give data to other companies.
link |
And that over the past few years,
link |
since that law came into effect,
link |
pretty much eradicated the illegal distribution,
link |
Additionally, I think giving,
link |
I think technology is often a very good way
link |
to solve technology misuse.
link |
So can we come up with new technologies
link |
that will let us have our cake and eat it too?
link |
People are looking into homomorphic encryption,
link |
which is letting you keep the data,
link |
have it encrypted and train on encrypted data.
link |
Of course, we haven't solved that one yet,
link |
but that kind of direction may be worth pursuing.
link |
Also federated learning,
link |
which would allow one hospital
link |
to train on its hospital's patient data fully
link |
because they have a license for that.
link |
And then hospitals would then share their models,
link |
not data, but models to create a super AI.
link |
And that also maybe has some promise.
link |
So I would want to encourage us to be open minded
link |
and think of this as not just the policy binary, yes, no,
link |
but letting the technologists try to find solutions
link |
to let us have our cake and eat it too,
link |
or have most of our cake and eat most of it too.
link |
Finally, I think giving each end user a choice is important
link |
and having transparency is important.
link |
Also, I think that's universal,
link |
but the choice you give to the user
link |
should not be at a granular level
link |
that the user cannot understand.
link |
GDPR today causes all these popups of yes, no,
link |
will you give this site this right
link |
to use this part of your data?
link |
I don't think any user understands
link |
what they're saying yes or no to.
link |
And I suspect most are just saying yes
link |
because they don't understand it.
link |
So while GDPR in its current implementation
link |
has lived up to its promise of transparency and user choice,
link |
it implemented it in such a way
link |
that really didn't deliver the spirit of GDPR.
link |
It fit the letter, but not the spirit.
link |
So again, I think we need to think about
link |
is there a way to fit the spirit of GDPR
link |
by using some kind of technology?
link |
Can we have a slider that's an AI trying to figure out
link |
how much you want to slide between
link |
perfect protection security of your personal data
link |
versus a high degree of convenience
link |
with some risks of not having full privacy?
link |
Each user should have some preference
link |
and that gives you the user choice.
link |
But maybe we should turn the problem on its head
link |
and ask can there be an AI algorithm that can customize this?
link |
Because we can understand the slider,
link |
but we sure cannot understand every popup question.
link |
And I think getting that right
link |
requires getting the balance between
link |
what we talked about earlier,
link |
which is heart and soul
link |
versus profit driven decisions and strategy.
link |
I think from my perspective,
link |
the best way to make a lot of money in the long term
link |
is to keep your heart and soul intact.
link |
I think getting that slider right in the short term
link |
may feel like you'll be sacrificing profit,
link |
but in the long term,
link |
you'll be gaining user trust
link |
and providing a great experience.
link |
Do you share that kind of view in general?
link |
I sure would hope there is a way
link |
we can do long term projects
link |
that really do the right thing.
link |
I think a lot of people who embrace GDPR,
link |
their heart's in the right place.
link |
I think they just need to figure out how to build a solution.
link |
I've heard utopians talk about solutions
link |
that get me excited,
link |
but I'm not sure how in the current funding environment
link |
they can get started.
link |
People talk about,
link |
imagine this crowdsourced data collection
link |
that we all trust.
link |
And then we have these agents
link |
that we ask the trusted agent to...
link |
That agent only, that platform,
link |
so a trusted joint platform
link |
that we all believe is trustworthy,
link |
that can give us all the closed loop personal suggestions
link |
by the new social network, new search engine,
link |
new eCommerce engine that has access
link |
to even more of our data,
link |
but not directly, but indirectly.
link |
So I think that general concept
link |
of licensing to some trusted engine
link |
and finding a way to trust that engine
link |
seems like a great idea.
link |
But if you think how long it's gonna take
link |
to implement and tweak and develop it right,
link |
as well as to collect all the trusts
link |
and the data from the people,
link |
it's beyond the current cycle of venture capital.
link |
So how do you do that is a big question.
link |
You've recently had a fight with cancer,
link |
stage four lymphoma and in a sort of deep personal level,
link |
what did it feel like in the darker moments
link |
to face your own mortality?
link |
Well, I've been the workaholic my whole life
link |
and I've basically worked nine, nine, six,
link |
nine a.m. to nine p.m. six days a week, roughly.
link |
And I didn't really pay a lot of attention
link |
to my family, friends, and people who loved me.
link |
And my life revolved around optimizing for work.
link |
While my work was not routine,
link |
my optimization really what made my life
link |
basically very mechanical process.
link |
But I got a lot of highs out of it
link |
because of accomplishments
link |
that I thought were really important and dear
link |
and the highest priority to me.
link |
But when I faced mortality
link |
and the possible death in matter of months,
link |
I suddenly realized that this really meant nothing to me,
link |
that I didn't feel like working for another minute,
link |
that if I had six months left in my life,
link |
I would spend it all with my loved ones
link |
and thanking them, giving them love back
link |
and apologizing to them that I lived my life the wrong way.
link |
So that moment of reckoning
link |
caused me to really rethink that why we exist in this world
link |
is something that we might be too much shaped by the society
link |
to think that success and accomplishments is why we live.
link |
But while that can get you
link |
periodic successes and satisfaction,
link |
it's really in the facing death
link |
you see what's truly important to you.
link |
So as a result of going through the challenges with cancer,
link |
I've resolved to live a more balanced lifestyle.
link |
I'm now in remission, knock on wood,
link |
and I'm spending more time with my family.
link |
My wife travels with me.
link |
When my kids need me, I spend more time with them.
link |
And before I used to prioritize everything around work.
link |
When I had a little bit of time,
link |
I would dole it out to my family.
link |
Now, when my family needs something, really needs something,
link |
I drop everything at work and go to them.
link |
And then in the time remaining, I allocate to work.
link |
But one's family is very understanding.
link |
It's not like they will take 50 hours a week from me.
link |
So I'm actually able to still work pretty hard,
link |
maybe 10 hours less per week.
link |
So I realized the most important thing in my life
link |
is really love and the people I love.
link |
And I give that the highest priority.
link |
It isn't the only thing I do,
link |
but when that is needed, I put that at the top priority
link |
and I feel much better and I feel much more balanced.
link |
And I think this also gives a hint
link |
as to a life of routine work, a life of pursuit of numbers.
link |
While my job was not routine, it was in pursuit of numbers,
link |
pursuit of can I make more money?
link |
Can I fund more great companies?
link |
Can I raise more money?
link |
Can I make sure our VC is ranked higher and higher
link |
This competitive nature of driving for bigger numbers
link |
and better numbers became a endless pursuit
link |
that's mechanical.
link |
And bigger numbers really didn't make me happier.
link |
And faced with death, I realized bigger numbers
link |
really meant nothing.
link |
And what was important is that people who have given
link |
their heart and their love to me
link |
deserve for me to do the same.
link |
So there's deep, profound truth in that,
link |
that everyone should hear and internalize.
link |
I mean, that's really powerful for you to say that.
link |
I have to ask sort of a difficult question here.
link |
So I've competed in sports my whole life,
link |
looking historically, I'd like to challenge some aspect
link |
of that a little bit on the point of hard work.
link |
That it feels that there are certain aspects
link |
that is the greatest, the most beautiful aspects
link |
of human nature is the ability to become obsessed,
link |
of becoming extremely passionate to the point where yes,
link |
flaws are revealed and just giving yourself fully to a task.
link |
That is, in another sense, you mentioned love
link |
being important, but in another sense,
link |
this kind of obsession, this pure exhibition of passion
link |
and hard work is truly what it means to be human.
link |
What lessons should we take that's deeper?
link |
Because you've accomplished incredible things.
link |
You say it chasing numbers,
link |
but really there's some incredible work there.
link |
So how do you think about that when you look back
link |
in your 20s, your 30s, what would you do differently?
link |
Would you really take back some of the incredible hard work?
link |
I would, but it's in percentages, right?
link |
We're both computer scientists.
link |
So I think when one balances one's life,
link |
when one is younger, you might give a smaller percentage
link |
to family, but you would still give them high priority.
link |
And when you get older, you would give a larger percentage
link |
to them and still the high priority.
link |
And when you're near retirement, you give most of it to them
link |
and the highest priority.
link |
So I think the key point is not that we would work 20 hours
link |
less for the whole life and just spend it aimlessly
link |
with the family, but that's when the family has a need,
link |
when your wife is having a baby,
link |
when your daughter has a birthday or when they're depressed
link |
or when they're celebrating something
link |
or when they have a get together or when we have family time
link |
that it's important for us to put down our phone and PC
link |
and be a hundred percent with them.
link |
And that priority on the things that really matter
link |
isn't going to be so taxing that it would eliminate
link |
or even dramatically reduce our accomplishments.
link |
It might have some impact, but it might also have
link |
other impact because if you have a happier family,
link |
maybe you fight less.
link |
If you fight less, you don't spend time taking care
link |
of all the aftermath of a fight.
link |
So it's unclear that it would take more time.
link |
And if it did, I'd be willing to take that reduction.
link |
And it's not a dramatic number, but it's a number
link |
that I think would give me a greater degree of happiness
link |
and knowing that I've done the right thing
link |
and still have plenty of hours to get the success
link |
that I want to get.
link |
So given the many successful companies that you've launched
link |
and much success throughout your career,
link |
what advice would you give to young people today looking,
link |
or it doesn't have to be young,
link |
but people today looking to launch
link |
and to create the next $1 billion tech startup
link |
or even AI based startup?
link |
I would suggest that people understand
link |
technology waves move quickly.
link |
What worked two years ago may not work today.
link |
And that is very much case in point for AI.
link |
I think two years ago, or maybe three years ago,
link |
you certainly could say I have a couple
link |
of super smart PhDs and we're not sure
link |
what we're gonna do, but here's how we're gonna start
link |
and get funding for a very high valuation.
link |
Those days are over because AI is going
link |
from rocket science towards mainstream,
link |
not yet commodity, but more mainstream.
link |
So first the creation of any company
link |
to a venture capitalists has to be creation
link |
of business value and monetary value.
link |
And when you have a very scarce commodity,
link |
VCs may be willing to accept greater uncertainty.
link |
But now the number of people who have the equivalent
link |
of PhD three years ago, because that can be learned
link |
more quickly, platforms are emerging,
link |
the cost to become a AI engineer is much lower
link |
and there are many more AI engineers.
link |
So the market is different.
link |
So I would suggest someone who wants to build an AI company
link |
be thinking about the normal business questions.
link |
What customer cases are you trying to address?
link |
What kind of pain are you trying to address?
link |
How does that translate to value?
link |
How will you extract value and get paid
link |
through what channel and how much business value
link |
That today needs to be thought about much earlier upfront
link |
than it did three years ago.
link |
The scarcity question of AI talent has changed.
link |
The number of AI talent has changed.
link |
So now you need not just AI, but also understanding
link |
of business customer and the marketplace.
link |
So I also think you should have a more reasonable
link |
valuation expectation and growth expectation.
link |
There's gonna be more competition.
link |
But the good news though, is that AI technologies
link |
are now more available in open source.
link |
TensorFlow, PyTorch and such tools are much easier to use.
link |
So you should be able to experiment and get results
link |
iteratively faster than before.
link |
So take more of a business mindset to this,
link |
think less of this as a laboratory taken into a company,
link |
because we've gone beyond that stage.
link |
The only exception is if you truly have a breakthrough
link |
in some technology that really no one has,
link |
then the old way still works.
link |
But I think that's harder and harder now.
link |
So I know you believe as many do that we're far
link |
from creating an artificial general intelligence system.
link |
But say once we do, and you get to ask her one question,
link |
what would that question be?
link |
What is it that differentiates you and me?
link |
Beautifully put, Kaifu, thank you so much
link |
for your time today.