back to index

Eric Schmidt: Google | Lex Fridman Podcast #8


small model | large model

link |
00:00:00.000
The following is a conversation with Eric Schmidt.
link |
00:00:03.180
He was the CEO of Google for 10 years
link |
00:00:05.140
and a chairman for six more,
link |
00:00:06.780
guiding the company through an incredible period of growth
link |
00:00:10.100
and a series of world changing innovations.
link |
00:00:12.940
He is one of the most impactful leaders
link |
00:00:15.300
in the era of the internet and the powerful voice
link |
00:00:19.340
for the promise of technology in our society.
link |
00:00:22.300
It was truly an honor to speak with him
link |
00:00:24.780
as part of the MIT course
link |
00:00:26.900
on artificial general intelligence
link |
00:00:28.660
and the artificial intelligence podcast.
link |
00:00:31.900
And now here's my conversation with Eric Schmidt.
link |
00:00:37.020
What was the first moment
link |
00:00:38.020
when you fell in love with technology?
link |
00:00:40.900
I grew up in the 1960s as a boy
link |
00:00:44.380
where every boy wanted to be an astronaut
link |
00:00:46.820
and part of the space program.
link |
00:00:48.900
So like everyone else of my age,
link |
00:00:51.340
we would go out to the cow pasture behind my house,
link |
00:00:54.340
which was literally a cow pasture
link |
00:00:56.260
and we would shoot model rockets off.
link |
00:00:58.540
And that I think is the beginning.
link |
00:01:00.820
And of course, generationally today,
link |
00:01:03.540
it would be video games and all the amazing things
link |
00:01:05.760
that you can do online with computers.
link |
00:01:09.100
There's a transformative, inspiring aspect of science
link |
00:01:12.620
and math that maybe rockets would bring
link |
00:01:15.740
would instill in individuals.
link |
00:01:17.420
You've mentioned yesterday that eighth grade math
link |
00:01:20.140
is where the journey through mathematical universe
link |
00:01:22.180
diverges from many people.
link |
00:01:23.780
It's this fork in the roadway.
link |
00:01:26.900
There's a professor of math at Berkeley, Edward Frankel.
link |
00:01:30.260
He, I'm not sure if you're familiar with him.
link |
00:01:32.460
I am.
link |
00:01:33.300
He has written this amazing book
link |
00:01:35.420
I recommend to everybody called Love and Math.
link |
00:01:37.700
Two of my favorite words.
link |
00:01:41.460
He says that if painting was taught like math,
link |
00:01:46.700
then the students would be asked to paint a fence,
link |
00:01:49.620
which is his analogy of essentially how math is taught.
link |
00:01:52.540
And so you never get a chance to discover the beauty
link |
00:01:55.860
of the art of painting or the beauty of the art of math.
link |
00:01:59.260
So how, when, and where did you discover that beauty?
link |
00:02:05.260
I think what happens with people like myself
link |
00:02:08.040
is that your math enabled pretty early
link |
00:02:11.380
and all of a sudden you discover that you can use that
link |
00:02:14.320
to discover new insights.
link |
00:02:16.560
The great scientists will all tell a story,
link |
00:02:19.100
the men and women who are fantastic today,
link |
00:02:22.020
that somewhere when they were in high school or in college,
link |
00:02:24.600
they discovered that they could discover
link |
00:02:26.060
something themselves.
link |
00:02:27.780
And that sense of building something,
link |
00:02:29.860
of having an impact that you own,
link |
00:02:32.260
drives knowledge acquisition and learning.
link |
00:02:35.460
In my case, it was programming.
link |
00:02:37.020
And the notion that I could build things
link |
00:02:39.820
that had not existed that I had built,
link |
00:02:42.300
that it had my name on it.
link |
00:02:44.400
And this was before open source,
link |
00:02:46.160
but you could think of it as open source contributions.
link |
00:02:49.100
So today, if I were a 16 or 17 year old boy,
link |
00:02:51.780
I'm sure that I would aspire as a computer scientist
link |
00:02:54.660
to make a contribution like the open source heroes
link |
00:02:58.060
of the world today.
link |
00:02:58.920
That would be what would be driving me.
link |
00:03:00.380
And I'd be trying and learning and making mistakes
link |
00:03:03.700
and so forth in the ways that it works.
link |
00:03:06.620
The repository that GitHub represents
link |
00:03:09.940
and that open source libraries represent
link |
00:03:12.200
is an enormous bank of knowledge
link |
00:03:14.900
of all of the people who are doing that.
link |
00:03:17.100
And one of the lessons that I learned at Google
link |
00:03:19.540
was that the world is a very big place
link |
00:03:21.500
and there's an awful lot of smart people.
link |
00:03:23.540
And an awful lot of them are underutilized.
link |
00:03:26.300
So here's an opportunity, for example,
link |
00:03:28.940
building parts of programs, building new ideas
link |
00:03:31.700
to contribute to the greater of society.
link |
00:03:36.540
So in that moment in the 70s,
link |
00:03:38.340
the inspiring moment where there was nothing
link |
00:03:40.660
and then you created something through programming,
link |
00:03:42.820
that magical moment.
link |
00:03:44.720
So in 1975, I think you've created a program called Lex,
link |
00:03:49.180
which I especially like because my name is Lex.
link |
00:03:51.460
So thank you, thank you for creating a brand
link |
00:03:54.620
that established a reputation that's long lasting, reliable
link |
00:03:58.260
and has a big impact on the world and still used today.
link |
00:04:01.180
So thank you for that.
link |
00:04:02.820
But more seriously, in that time, in the 70s,
link |
00:04:08.220
as an engineer, personal computers were being born.
link |
00:04:12.540
Do you think you'd be able to predict the 80s, 90s
link |
00:04:15.580
and the aughts of where computers would go?
link |
00:04:18.900
I'm sure I could not and would not have gotten it right.
link |
00:04:23.180
I was the beneficiary of the great work
link |
00:04:25.420
of many, many people who saw it clearer than I did.
link |
00:04:29.060
With Lex, I worked with a fellow named Michael Lesk,
link |
00:04:32.540
who was my supervisor.
link |
00:04:33.980
And he essentially helped me architect
link |
00:04:36.300
and deliver a system that's still in use today.
link |
00:04:39.180
After that, I worked at Xerox Palo Alto Research Center,
link |
00:04:42.220
where the Alto was invented.
link |
00:04:43.660
And the Alto is the predecessor
link |
00:04:46.060
of the modern personal computer or Macintosh and so forth.
link |
00:04:50.180
And the Altos were very rare.
link |
00:04:52.300
And I had to drive an hour from Berkeley to go use them.
link |
00:04:55.260
But I made a point of skipping classes
link |
00:04:57.380
and doing whatever it took to have access
link |
00:05:00.960
to this extraordinary achievement.
link |
00:05:02.500
I knew that they were consequential.
link |
00:05:04.900
What I did not understand was scaling.
link |
00:05:08.260
I did not understand what would happen
link |
00:05:09.860
when you had 100 million as opposed to 100.
link |
00:05:12.820
And so the, since then,
link |
00:05:14.200
and I have learned the benefit of scale,
link |
00:05:16.260
I always look for things
link |
00:05:17.460
which are going to scale to platforms, right?
link |
00:05:19.660
So mobile phones, Android, all those things.
link |
00:05:23.060
There are, the world is in numerous,
link |
00:05:25.820
there are many, many people in the world,
link |
00:05:27.380
people really have needs.
link |
00:05:28.500
They really will use these platforms
link |
00:05:29.940
and you can build big businesses on top of them.
link |
00:05:32.560
So it's interesting.
link |
00:05:33.400
So when you see a piece of technology,
link |
00:05:34.860
now you think, what will this technology look like
link |
00:05:37.300
when it's in the hands of a billion people?
link |
00:05:39.020
That's right.
link |
00:05:39.900
So an example would be that the market is so competitive now
link |
00:05:44.940
that if you can't figure out a way
link |
00:05:46.940
for something to have a million users or a billion users,
link |
00:05:50.780
it probably is not going to be successful
link |
00:05:53.100
because something else will become the general platform
link |
00:05:56.820
and your idea will become a lost idea
link |
00:06:01.060
or a specialized service with relatively few users.
link |
00:06:04.260
So it's a path to generality.
link |
00:06:05.900
It's a path to general platform use.
link |
00:06:07.660
It's a path to broad applicability.
link |
00:06:10.060
Now there are plenty of good businesses that are tiny.
link |
00:06:12.660
So luxury goods, for example.
link |
00:06:14.900
But if you want to have an impact at scale,
link |
00:06:18.500
you have to look for things which are of common value,
link |
00:06:21.340
common pricing, common distribution
link |
00:06:23.300
and solve common problems.
link |
00:06:24.740
They're problems that everyone has.
link |
00:06:26.140
And by the way, people have lots of problems.
link |
00:06:28.100
Information, medicine, health, education and so forth.
link |
00:06:31.140
Work on those problems.
link |
00:06:32.940
Like you said, you're a big fan of the middle class.
link |
00:06:36.780
Because there's so many of them.
link |
00:06:37.820
There's so many of them.
link |
00:06:38.740
By definition.
link |
00:06:40.140
So any product, any thing that has a huge impact
link |
00:06:44.380
and improves their lives is a great business decision
link |
00:06:47.460
and it's just good for society.
link |
00:06:48.860
And there's nothing wrong with starting off in the high end
link |
00:06:52.340
as long as you have a plan to get to the middle class.
link |
00:06:55.420
There's nothing wrong with starting with a specialized
link |
00:06:57.580
market in order to learn and to build and to fund things.
link |
00:07:01.020
So you start with a luxury market
link |
00:07:02.540
to build a general purpose market.
link |
00:07:04.460
But if you define yourself as only a narrow market,
link |
00:07:07.500
someone else can come along with a general purpose market
link |
00:07:10.940
that can push you to the corner,
link |
00:07:12.340
can restrict the scale of operation,
link |
00:07:14.260
can force you to be a lesser impact than you might be.
link |
00:07:17.820
So it's very important to think in terms of broad businesses
link |
00:07:21.020
and broad impact.
link |
00:07:22.340
Even if you start in a little corner somewhere.
link |
00:07:26.260
So as you look to the 70s but also in the decades to come
link |
00:07:30.980
and you saw computers, did you see them as tools
link |
00:07:34.860
or was there a little element of another entity?
link |
00:07:40.260
I remember a quote saying AI began with our dream
link |
00:07:44.660
to create the gods.
link |
00:07:46.140
Is there a feeling when you wrote that program
link |
00:07:48.620
that you were creating another entity,
link |
00:07:51.300
giving life to something?
link |
00:07:52.820
I wish I could say otherwise,
link |
00:07:54.660
but I simply found the technology platforms so exciting.
link |
00:07:58.740
That's what I was focused on.
link |
00:08:00.460
I think the majority of the people that I've worked with,
link |
00:08:03.380
and there are a few exceptions, Steve Jobs being an example,
link |
00:08:06.700
really saw this as a great technological play.
link |
00:08:09.980
I think relatively few of the technical people understood
link |
00:08:13.700
the scale of its impact.
link |
00:08:15.380
So I used NCP, which is a predecessor to TCPIP.
link |
00:08:19.620
It just made sense to connect things.
link |
00:08:21.180
We didn't think of it in terms of the internet
link |
00:08:23.780
and then companies and then Facebook and then Twitter
link |
00:08:27.020
and then politics and so forth.
link |
00:08:29.180
We never did that build.
link |
00:08:30.740
We didn't have that vision.
link |
00:08:32.860
And I think most people, it's a rare person
link |
00:08:35.300
who can see compounding at scale.
link |
00:08:38.020
Most people can see,
link |
00:08:39.060
if you ask people to predict the future,
link |
00:08:40.580
they'll give you an answer of six to nine months
link |
00:08:43.060
or 12 months,
link |
00:08:44.500
because that's about as far as people can imagine.
link |
00:08:47.500
But there's an old saying,
link |
00:08:48.700
which actually was attributed to a professor at MIT
link |
00:08:50.860
a long time ago,
link |
00:08:52.060
that we overestimate what can be done in one year
link |
00:08:56.380
and we underestimate what can be done in a decade.
link |
00:09:00.100
And there's a great deal of evidence
link |
00:09:02.460
that these core platforms at hardware and software
link |
00:09:05.580
take a decade, right?
link |
00:09:07.740
So think about self driving cars.
link |
00:09:09.420
Self driving cars were thought about in the 90s.
link |
00:09:12.100
There were projects around them.
link |
00:09:13.340
The first DARPA Grand Challenge was roughly 2004.
link |
00:09:17.100
So that's roughly 15 years ago.
link |
00:09:19.700
And today we have self driving cars operating
link |
00:09:22.060
in a city in Arizona, right?
link |
00:09:23.940
It's 15 years and we still have a ways to go
link |
00:09:26.620
before they're more generally available.
link |
00:09:31.620
So you've spoken about the importance,
link |
00:09:33.780
you just talked about predicting into the future.
link |
00:09:37.100
You've spoken about the importance of thinking
link |
00:09:39.940
five years ahead and having a plan for those five years.
link |
00:09:42.860
The way to say it is that almost everybody
link |
00:09:45.980
has a one year plan.
link |
00:09:47.500
Almost no one has a proper five year plan.
link |
00:09:50.940
And the key thing to having a five year plan
link |
00:09:52.780
is to having a model for what's going to happen
link |
00:09:55.260
under the underlying platforms.
link |
00:09:56.900
So here's an example.
link |
00:09:59.980
Moore's Law as we know it,
link |
00:10:01.140
the thing that powered improvements in CPUs
link |
00:10:04.260
has largely halted in its traditional shrinking mechanism
link |
00:10:07.580
because the costs have just gotten so high.
link |
00:10:10.340
It's getting harder and harder.
link |
00:10:12.160
But there's plenty of algorithmic improvements
link |
00:10:14.580
and specialized hardware improvements.
link |
00:10:16.580
So you need to understand the nature of those improvements
link |
00:10:19.660
and where they'll go in order to understand
link |
00:10:21.940
how it will change the platform.
link |
00:10:24.300
In the area of network connectivity,
link |
00:10:26.060
what are the gains that are gonna be possible in wireless?
link |
00:10:29.400
It looks like there's an enormous expansion
link |
00:10:33.380
of wireless connectivity at many different bands.
link |
00:10:36.900
And that we will primarily,
link |
00:10:38.660
historically I've always thought
link |
00:10:39.860
that we were primarily gonna be using fiber,
link |
00:10:42.080
but now it looks like we're gonna be using fiber
link |
00:10:43.940
plus very powerful high bandwidth
link |
00:10:47.380
sort of short distance connectivity
link |
00:10:49.320
to bridge the last mile.
link |
00:10:51.460
That's an amazing achievement.
link |
00:10:53.060
If you know that,
link |
00:10:54.440
then you're gonna build your systems differently.
link |
00:10:56.900
By the way, those networks
link |
00:10:57.780
have different latency properties, right?
link |
00:10:59.640
Because they're more symmetric,
link |
00:11:01.620
the algorithms feel faster for that reason.
link |
00:11:04.980
And so when you think about whether it's a fiber
link |
00:11:07.860
or just technologies in general,
link |
00:11:09.860
so there's this barber wooden poem or quote
link |
00:11:14.180
that I really like.
link |
00:11:15.860
It's from the champions of the impossible
link |
00:11:18.240
rather than the slaves of the possible
link |
00:11:20.340
that evolution draws its creative force.
link |
00:11:23.220
So in predicting the next five years,
link |
00:11:25.980
I'd like to talk about the impossible and the possible.
link |
00:11:29.220
Well, and again, one of the great things about humanity
link |
00:11:32.280
is that we produce dreamers, right?
link |
00:11:34.720
We literally have people who have a vision and a dream.
link |
00:11:37.780
They are, if you will, disagreeable
link |
00:11:40.100
in the sense that they disagree with the,
link |
00:11:42.740
they disagree with what the sort of zeitgeist is.
link |
00:11:45.780
They say there is another way.
link |
00:11:48.020
They have a belief, they have a vision.
link |
00:11:50.280
If you look at science, science is always marked
link |
00:11:54.060
by such people who went against some conventional wisdom,
link |
00:11:58.380
collected the knowledge at the time
link |
00:12:00.220
and assembled it in a way that produced a powerful platform.
link |
00:12:03.660
And you've been amazingly honest about,
link |
00:12:08.300
in an inspiring way, about things you've been wrong
link |
00:12:11.260
about predicting and you've obviously been right
link |
00:12:13.860
about a lot of things, but in this kind of tension,
link |
00:12:18.860
how do you balance, as a company,
link |
00:12:21.260
in predicting the next five years,
link |
00:12:23.580
the impossible, planning for the impossible,
link |
00:12:26.300
so listening to those crazy dreamers, letting them do,
link |
00:12:30.380
letting them run away and make the impossible real,
link |
00:12:34.140
make it happen, and slow, you know,
link |
00:12:36.940
that's how programmers often think,
link |
00:12:38.740
and slowing things down and saying,
link |
00:12:41.560
well, this is the rational, this is the possible,
link |
00:12:44.600
the pragmatic, the dreamer versus the pragmatist,
link |
00:12:48.380
so it's helpful to have a model
link |
00:12:51.380
which encourages a predictable revenue stream
link |
00:12:56.020
as well as the ability to do new things.
link |
00:12:58.660
So in Google's case, we're big enough
link |
00:13:00.540
and well enough managed and so forth
link |
00:13:02.340
that we have a pretty good sense of what our revenue will be
link |
00:13:05.200
for the next year or two, at least for a while.
link |
00:13:07.900
And so we have enough cash generation
link |
00:13:11.540
that we can make bets, and indeed,
link |
00:13:14.700
Google has become alphabet,
link |
00:13:16.780
so the corporation is organized around these bets,
link |
00:13:19.500
and these bets are in areas of fundamental importance
link |
00:13:22.740
to the world, whether it's artificial intelligence,
link |
00:13:26.720
medical technology, self driving cars,
link |
00:13:29.700
connectivity through balloons, on and on and on.
link |
00:13:33.300
And there's more coming and more coming.
link |
00:13:35.980
So one way you could express this
link |
00:13:38.020
is that the current business is successful enough
link |
00:13:41.500
that we have the luxury of making bets.
link |
00:13:44.580
And another one that you could say
link |
00:13:45.940
is that we have the wisdom of being able to see
link |
00:13:49.140
that a corporate structure needs to be created
link |
00:13:51.580
to enhance the likelihood of the success of those bets.
link |
00:13:55.260
So we essentially turned ourselves into a conglomerate
link |
00:13:58.860
of bets and then this underlying corporation, Google,
link |
00:14:02.100
which is itself innovative.
link |
00:14:04.280
So in order to pull this off,
link |
00:14:05.900
you have to have a bunch of belief systems,
link |
00:14:08.060
and one of them is that you have to have
link |
00:14:09.580
bottoms up and tops down.
link |
00:14:11.460
The bottoms up we call 20% time,
link |
00:14:13.580
and the idea is that people can spend 20% of the time
link |
00:14:15.780
whatever they want, and the top down
link |
00:14:17.740
is that our founders in particular
link |
00:14:19.700
have a keen eye on technology
link |
00:14:21.740
and they're reviewing things constantly.
link |
00:14:23.880
So an example would be they'll hear about an idea
link |
00:14:26.540
or I'll hear about something and it sounds interesting,
link |
00:14:28.700
let's go visit them.
link |
00:14:30.380
And then let's begin to assemble the pieces
link |
00:14:33.060
to see if that's possible.
link |
00:14:34.780
And if you do this long enough,
link |
00:14:35.980
you get pretty good at predicting what's likely to work.
link |
00:14:39.780
So that's a beautiful balance that struck.
link |
00:14:42.020
Is this something that applies at all scale?
link |
00:14:44.420
It seems to be that Sergey, again, 15 years ago,
link |
00:14:53.060
came up with a concept called 10% of the budget
link |
00:14:56.840
should be on things that are unrelated.
link |
00:14:58.980
It was called 70, 20, 10.
link |
00:15:00.860
70% of our time on core business,
link |
00:15:03.540
20% on adjacent business, and 10% on other.
link |
00:15:06.780
And he proved mathematically,
link |
00:15:08.700
of course he's a brilliant mathematician,
link |
00:15:10.580
that you needed that 10% to make the sum
link |
00:15:13.860
of the growth work.
link |
00:15:14.700
And it turns out he was right.
link |
00:15:18.620
So getting into the world of artificial intelligence,
link |
00:15:20.940
you've talked quite extensively and effectively
link |
00:15:25.380
to the impact in the near term,
link |
00:15:28.780
the positive impact of artificial intelligence,
link |
00:15:32.020
whether it's especially machine learning
link |
00:15:34.140
in medical applications and education,
link |
00:15:38.580
and just making information more accessible, right?
link |
00:15:41.600
In the AI community, there is a kind of debate.
link |
00:15:45.860
There's this shroud of uncertainty
link |
00:15:47.700
as we face this new world
link |
00:15:49.020
with artificial intelligence in it.
link |
00:15:50.460
And there's some people, like Elon Musk,
link |
00:15:54.260
you've disagreed, at least on the degree of emphasis
link |
00:15:57.660
he places on the existential threat of AI.
link |
00:16:00.700
So I've spoken with Stuart Russell,
link |
00:16:02.540
Max Tegmark, who share Elon Musk's view,
link |
00:16:05.340
and Yoshua Bengio, Steven Pinker, who do not.
link |
00:16:09.180
And so there's a lot of very smart people
link |
00:16:11.860
who are thinking about this stuff, disagreeing,
link |
00:16:14.620
which is really healthy, of course.
link |
00:16:17.180
So what do you think is the healthiest way
link |
00:16:19.100
for the AI community to,
link |
00:16:22.020
and really for the general public,
link |
00:16:23.860
to think about AI and the concern
link |
00:16:27.700
of the technology being mismanaged in some kind of way?
link |
00:16:32.920
So the source of education for the general public
link |
00:16:35.060
has been robot killer movies.
link |
00:16:37.380
Right.
link |
00:16:38.220
And Terminator, et cetera.
link |
00:16:40.860
And the one thing I can assure you we're not building
link |
00:16:44.500
are those kinds of solutions.
link |
00:16:46.620
Furthermore, if they were to show up,
link |
00:16:48.420
someone would notice and unplug them, right?
link |
00:16:51.140
So as exciting as those movies are,
link |
00:16:53.140
and they're great movies,
link |
00:16:54.700
were the killer robots to start,
link |
00:16:57.500
we would find a way to stop them, right?
link |
00:17:00.500
So I'm not concerned about that.
link |
00:17:04.060
And much of this has to do
link |
00:17:05.980
with the timeframe of conversation.
link |
00:17:08.540
So you can imagine a situation 100 years from now
link |
00:17:13.300
when the human brain is fully understood
link |
00:17:15.920
and the next generation and next generation
link |
00:17:18.140
of brilliant MIT scientists have figured all this out,
link |
00:17:20.940
we're gonna have a large number of ethics questions, right?
link |
00:17:25.140
Around science and thinking and robots and computers
link |
00:17:28.060
and so forth and so on.
link |
00:17:29.700
So it depends on the question of the timeframe.
link |
00:17:32.260
In the next five to 10 years,
link |
00:17:34.780
we're not facing those questions.
link |
00:17:37.220
What we're facing in the next five to 10 years
link |
00:17:39.100
is how do we spread this disruptive technology
link |
00:17:42.140
as broadly as possible to gain the maximum benefit of it?
link |
00:17:46.500
The primary benefits should be in healthcare
link |
00:17:48.980
and in education.
link |
00:17:50.860
Healthcare because it's obvious.
link |
00:17:52.320
We're all the same even though we somehow believe we're not.
link |
00:17:55.780
As a medical matter,
link |
00:17:57.340
the fact that we have big data about our health
link |
00:17:59.180
will save lives, allow us to deal with skin cancer
link |
00:18:02.700
and other cancers, ophthalmological problems.
link |
00:18:05.500
There's people working on psychological diseases
link |
00:18:08.420
and so forth using these techniques.
link |
00:18:10.260
I can go on and on.
link |
00:18:11.700
The promise of AI in medicine is extraordinary.
link |
00:18:15.840
There are many, many companies and startups
link |
00:18:17.980
and funds and solutions
link |
00:18:19.480
and we will all live much better for that.
link |
00:18:22.140
The same argument in education.
link |
00:18:25.580
Can you imagine that for each generation of child
link |
00:18:28.540
and even adult, you have a tutor educator that's AI based,
link |
00:18:33.020
that's not a human but is properly trained,
link |
00:18:35.900
that helps you get smarter,
link |
00:18:37.140
helps you address your language difficulties
link |
00:18:39.280
or your math difficulties or what have you.
link |
00:18:41.340
Why don't we focus on those two?
link |
00:18:43.300
The gains societally of making humans smarter and healthier
link |
00:18:47.300
are enormous and those translate for decades and decades
link |
00:18:51.460
and we'll all benefit from them.
link |
00:18:53.900
There are people who are working on AI safety,
link |
00:18:56.300
which is the issue that you're describing
link |
00:18:58.060
and there are conversations in the community
link |
00:19:00.660
that should there be such problems,
link |
00:19:02.500
what should the rules be like?
link |
00:19:04.380
Google, for example, has announced its policies
link |
00:19:07.540
with respect to AI safety, which I certainly support
link |
00:19:10.140
and I think most everybody would support
link |
00:19:12.300
and they make sense, right?
link |
00:19:14.140
So it helps guide the research
link |
00:19:16.300
but the killer robots are not arriving this year
link |
00:19:19.540
and they're not even being built.
link |
00:19:22.540
And on that line of thinking, you said the time scale.
link |
00:19:26.720
In this topic or other topics,
link |
00:19:30.440
have you found it useful on the business side
link |
00:19:34.560
or the intellectual side to think beyond five, 10 years,
link |
00:19:37.480
to think 50 years out?
link |
00:19:39.360
Has it ever been useful or productive?
link |
00:19:41.960
In our industry, there are essentially no examples
link |
00:19:45.160
of 50 year predictions that have been correct.
link |
00:19:48.840
Let's review AI, right?
link |
00:19:50.400
AI, which was largely invented here at MIT
link |
00:19:53.060
and a couple of other universities in the 1956, 1957,
link |
00:19:56.440
1958, the original claims were a decade or two.
link |
00:20:01.320
And when I was a PhD student, I studied AI a bit
link |
00:20:05.180
and it entered during my looking at it,
link |
00:20:07.680
a period which is known as AI winter,
link |
00:20:10.360
which went on for about 30 years,
link |
00:20:12.760
which is a whole generation of science,
link |
00:20:14.720
scientists and a whole group of people
link |
00:20:16.640
who didn't make a lot of progress
link |
00:20:18.400
because the algorithms had not improved
link |
00:20:20.160
and the computers had not approved.
link |
00:20:22.060
It took some brilliant mathematicians
link |
00:20:23.840
starting with a fellow named Jeff Hinton
link |
00:20:25.360
at Toronto and Montreal who basically invented
link |
00:20:29.460
this deep learning model which empowers us today.
link |
00:20:33.020
The seminal work there was 20 years ago
link |
00:20:36.060
and in the last 10 years, it's become popularized.
link |
00:20:39.960
So think about the timeframes for that level of discovery.
link |
00:20:43.840
It's very hard to predict.
link |
00:20:45.880
Many people think that we'll be flying around
link |
00:20:47.700
in the equivalent of flying cars, who knows?
link |
00:20:51.160
My own view, if I wanna go out on a limb,
link |
00:20:54.440
is to say that we know a couple of things
link |
00:20:56.840
about 50 years from now.
link |
00:20:57.960
We know that there'll be more people alive.
link |
00:21:00.440
We know that we'll have to have platforms
link |
00:21:02.160
that are more sustainable because the earth is limited
link |
00:21:05.680
in the ways we all know and that the kind of platforms
link |
00:21:09.160
that are gonna get built will be consistent
link |
00:21:11.360
with the principles that I've described.
link |
00:21:13.000
They will be much more empowering of individuals.
link |
00:21:15.720
They'll be much more sensitive to the ecology
link |
00:21:17.720
because they have to be, they just have to be.
link |
00:21:20.520
I also think that humans are gonna be a great deal smarter
link |
00:21:23.760
and I think they're gonna be a lot smarter
link |
00:21:25.040
because of the tools that I've discussed with you
link |
00:21:27.720
and of course, people will live longer.
link |
00:21:29.160
Life extension is continuing apace.
link |
00:21:32.160
A baby born today has a reasonable chance
link |
00:21:34.600
of living to 100, which is pretty exciting.
link |
00:21:37.080
It's well past the 21st century,
link |
00:21:38.580
so we better take care of them.
link |
00:21:40.600
And you mentioned an interesting statistic
link |
00:21:42.600
on some very large percentage, 60, 70% of people
link |
00:21:46.080
may live in cities.
link |
00:21:48.160
Today, more than half the world lives in cities
link |
00:21:50.460
and one of the great stories of humanity
link |
00:21:53.720
in the last 20 years has been the rural to urban migration.
link |
00:21:57.440
This has occurred in the United States,
link |
00:21:59.200
it's occurred in Europe, it's occurring in Asia
link |
00:22:02.760
and it's occurring in Africa.
link |
00:22:04.660
When people move to cities, the cities get more crowded,
link |
00:22:07.760
but believe it or not, their health gets better,
link |
00:22:10.480
their productivity gets better,
link |
00:22:12.280
their IQ and educational capabilities improve.
link |
00:22:15.440
So it's good news that people are moving to cities,
link |
00:22:18.500
but we have to make them livable and safe.
link |
00:22:20.820
So you, first of all, you are,
link |
00:22:25.860
but you've also worked with some of the greatest leaders
link |
00:22:28.300
in the history of tech.
link |
00:22:29.940
What insights do you draw from the difference
link |
00:22:32.940
in leadership styles of yourself,
link |
00:22:35.660
Steve Jobs, Elon Musk, Larry Page,
link |
00:22:39.140
now the new CEO, Sandra Pichai and others?
link |
00:22:42.740
From the, I would say, calm sages to the mad geniuses.
link |
00:22:47.740
One of the things that I learned as a young executive
link |
00:22:50.660
is that there's no single formula for leadership.
link |
00:22:54.500
They try to teach one, but that's not how it really works.
link |
00:22:58.380
There are people who just understand what they need to do
link |
00:23:01.020
and they need to do it quickly.
link |
00:23:02.540
Those people are often entrepreneurs.
link |
00:23:05.060
They just know and they move fast.
link |
00:23:07.340
There are other people who are systems thinkers
link |
00:23:09.100
and planners, that's more who I am,
link |
00:23:11.420
somewhat more conservative, more thorough in execution,
link |
00:23:15.060
a little bit more risk of risk.
link |
00:23:16.740
A little bit more risk averse.
link |
00:23:18.620
There's also people who are sort of slightly insane,
link |
00:23:22.140
in the sense that they are emphatic and charismatic
link |
00:23:26.060
and they feel it and they drive it and so forth.
link |
00:23:28.900
There's no single formula to success.
link |
00:23:31.340
There is one thing that unifies all of the people
link |
00:23:33.620
that you named, which is very high intelligence.
link |
00:23:36.900
At the end of the day, the thing that characterizes
link |
00:23:40.180
all of them is that they saw the world quicker, faster,
link |
00:23:43.620
they processed information faster.
link |
00:23:45.700
They didn't necessarily make the right decisions
link |
00:23:47.300
all the time, but they were on top of it.
link |
00:23:49.940
And the other thing that's interesting
link |
00:23:51.180
about all those people is they all started young.
link |
00:23:54.140
So think about Steve Jobs starting Apple
link |
00:23:56.940
roughly at 18 or 19.
link |
00:23:58.380
Think about Bill Gates starting at roughly 20, 21.
link |
00:24:01.620
Think about by the time they were 30,
link |
00:24:03.700
Mark Zuckerberg, a good example, at 19, 20.
link |
00:24:06.900
By the time they were 30, they had 10 years.
link |
00:24:10.620
At 30 years old, they had 10 years of experience
link |
00:24:13.700
of dealing with people and products and shipments
link |
00:24:16.940
and the press and business and so forth.
link |
00:24:19.740
It's incredible how much experience they had
link |
00:24:22.740
compared to the rest of us who were busy getting our PhDs.
link |
00:24:25.220
Yes, exactly.
link |
00:24:26.060
So we should celebrate these people
link |
00:24:28.460
because they've just had more life experience, right?
link |
00:24:32.180
And that helps inform the judgment.
link |
00:24:34.340
At the end of the day, when you're at the top
link |
00:24:38.220
of these organizations, all the easy questions
link |
00:24:41.380
have been dealt with, right?
link |
00:24:43.500
How should we design the buildings?
link |
00:24:45.620
Where should we put the colors on our product?
link |
00:24:48.180
What should the box look like, right?
link |
00:24:51.300
The problems, that's why it's so interesting
link |
00:24:53.340
to be in these rooms, the problems that they face, right,
link |
00:24:56.420
in terms of the way they operate,
link |
00:24:58.340
the way they deal with their employees,
link |
00:25:00.060
their customers, their innovation,
link |
00:25:01.860
are profoundly challenging.
link |
00:25:03.900
Each of the companies is demonstrably different culturally.
link |
00:25:09.340
They are not, in fact, cut of the same.
link |
00:25:11.700
They behave differently based on input.
link |
00:25:14.180
Their internal cultures are different.
link |
00:25:15.820
Their compensation schemes are different.
link |
00:25:17.460
Their values are different.
link |
00:25:19.340
So there's proof that diversity works.
link |
00:25:24.700
So, so when faced with a tough decision,
link |
00:25:29.780
in need of advice, it's been said that the best thing
link |
00:25:33.500
one can do is to find the best person in the world
link |
00:25:36.740
who can give that advice and find a way to be
link |
00:25:40.780
in a room with them, one on one and ask.
link |
00:25:44.740
So here we are, and let me ask in a long winded way,
link |
00:25:48.060
I wrote this down.
link |
00:25:50.740
In 1998, there were many good search engines,
link |
00:25:53.420
Lycos, Excite, AltaVista, Infoseek, Ask Jeeves maybe,
link |
00:25:59.260
Yahoo even.
link |
00:26:01.860
So Google stepped in and disrupted everything.
link |
00:26:04.660
They disrupted the nature of search,
link |
00:26:06.580
the nature of our access to information,
link |
00:26:08.860
the way we discover new knowledge.
link |
00:26:11.900
So now it's 2018, actually 20 years later.
link |
00:26:16.020
There are many good personal AI assistants,
link |
00:26:18.740
including, of course, the best from Google.
link |
00:26:22.260
So you've spoken in medical and education,
link |
00:26:25.540
the impact of such an AI assistant could bring.
link |
00:26:28.620
So we arrive at this question.
link |
00:26:30.340
So it's a personal one for me,
link |
00:26:32.180
but I hope my situation represents that of many other,
link |
00:26:36.300
as we said, dreamers and the crazy engineers.
link |
00:26:40.580
So my whole life, I've dreamed of creating
link |
00:26:43.900
such an AI assistant.
link |
00:26:45.860
Every step I've taken has been towards that goal.
link |
00:26:48.420
Now I'm a research scientist in human centered AI
link |
00:26:51.060
here at MIT.
link |
00:26:52.300
So the next step for me as I sit here,
link |
00:26:54.860
so facing my passion is to do what Larry and Sergey did
link |
00:26:59.860
in 98, this simple startup.
link |
00:27:04.180
And so here's my simple question.
link |
00:27:06.820
Given the low odds of success, the timing and luck required,
link |
00:27:10.620
the countless other factors that can't be controlled
link |
00:27:12.700
or predicted, which is all the things
link |
00:27:14.660
that Larry and Sergey faced,
link |
00:27:16.460
is there some calculation, some strategy
link |
00:27:20.140
to follow in this step?
link |
00:27:21.580
Or do you simply follow the passion
link |
00:27:23.700
just because there's no other choice?
link |
00:27:26.580
I think the people who are in universities
link |
00:27:29.660
are always trying to study
link |
00:27:31.860
the extraordinarily chaotic nature of innovation
link |
00:27:35.180
and entrepreneurship.
link |
00:27:37.260
My answer is that they didn't have that conversation.
link |
00:27:41.180
They just did it.
link |
00:27:42.820
They sensed a moment when in the case of Google,
link |
00:27:47.220
there was all of this data that needed to be organized
link |
00:27:49.700
and they had a better algorithm.
link |
00:27:51.300
They had invented a better way.
link |
00:27:53.780
So today with human centered AI,
link |
00:27:56.300
which is your area of research,
link |
00:27:58.060
there must be new approaches.
link |
00:28:00.860
It's such a big field.
link |
00:28:02.460
There must be new approaches,
link |
00:28:04.900
different from what we and others are doing.
link |
00:28:07.220
There must be startups to fund.
link |
00:28:09.540
There must be research projects to try.
link |
00:28:11.940
There must be graduate students to work on new approaches.
link |
00:28:15.020
Here at MIT, there are people who are looking at learning
link |
00:28:18.180
from the standpoint of looking at child learning.
link |
00:28:20.580
How do children learn starting at age one and two?
link |
00:28:23.500
And the work is fantastic.
link |
00:28:25.340
Those approaches are different from the approach
link |
00:28:28.180
that most people are taking.
link |
00:28:29.780
Perhaps that's a bet that you should make
link |
00:28:31.940
or perhaps there's another one.
link |
00:28:33.820
But at the end of the day,
link |
00:28:35.860
the successful entrepreneurs are not as crazy as they sound.
link |
00:28:40.100
They see an opportunity based on what's happened.
link |
00:28:43.100
Let's use Uber as an example.
link |
00:28:45.300
As Travis sells the story,
link |
00:28:46.740
he and his co founder were sitting in Paris
link |
00:28:48.940
and they had this idea because they couldn't get a cab.
link |
00:28:52.060
And they said, we have smartphones and the rest is history.
link |
00:28:56.660
So what's the equivalent of that Travis Eiffel Tower,
link |
00:29:00.980
where is a cab moment that you could,
link |
00:29:03.980
as an entrepreneur, take advantage of?
link |
00:29:05.940
Whether it's in human centered AI or something else.
link |
00:29:08.500
That's the next great startup.
link |
00:29:11.260
And the psychology of that moment.
link |
00:29:13.660
So when Sergey and Larry talk about,
link |
00:29:17.540
and listen to a few interviews, it's very nonchalant.
link |
00:29:20.180
Well, here's the very fascinating web data
link |
00:29:23.780
and here's an algorithm we have for,
link |
00:29:27.700
we just kind of want to play around with that data.
link |
00:29:29.420
And it seems like that's a really nice way
link |
00:29:31.020
to organize this data.
link |
00:29:34.180
I should say what happened to remember
link |
00:29:35.580
is that they were graduate students at Stanford
link |
00:29:38.100
and they thought this was interesting.
link |
00:29:39.300
So they built a search engine
link |
00:29:40.540
and they kept it in their room.
link |
00:29:43.020
And they had to get power from the room next door
link |
00:29:46.300
because they were using too much power in the room.
link |
00:29:48.020
So they ran an extension cord over, right?
link |
00:29:51.460
And then they went and they found a house
link |
00:29:53.500
and they had Google world headquarters of five people,
link |
00:29:56.500
right, to start the company.
link |
00:29:57.540
And they raised $100,000 from Andy Bechtolsheim,
link |
00:30:00.460
who was the Sun founder to do this
link |
00:30:02.220
and Dave Cheriton and a few others.
link |
00:30:04.460
The point is their beginnings were very simple
link |
00:30:08.220
but they were based on a powerful insight.
link |
00:30:11.700
That is a replicable model for any startup.
link |
00:30:14.860
It has to be a powerful insight.
link |
00:30:16.500
The beginnings are simple.
link |
00:30:17.620
And there has to be an innovation.
link |
00:30:19.860
In Larry and Sergey's case, it was PageRank,
link |
00:30:22.820
which was a brilliant idea,
link |
00:30:23.980
one of the most cited papers in the world today.
link |
00:30:26.700
What's the next one?
link |
00:30:29.740
So you're one of, if I may say,
link |
00:30:33.500
richest people in the world.
link |
00:30:36.180
And yet it seems that money is simply a side effect
link |
00:30:38.700
of your passions and not an inherent goal.
link |
00:30:42.980
But you're a fascinating person to ask.
link |
00:30:48.220
So much of our society at the individual level
link |
00:30:51.540
and at the company level and as nations
link |
00:30:55.020
is driven by the desire for wealth.
link |
00:30:58.660
What do you think about this drive?
link |
00:31:01.100
And what have you learned about,
link |
00:31:03.140
if I may romanticize the notion,
link |
00:31:05.020
the meaning of life,
link |
00:31:06.860
having achieved success on so many dimensions?
link |
00:31:10.420
There have been many studies of human happiness
link |
00:31:13.580
and above some threshold,
link |
00:31:16.340
which is typically relatively low for this conversation,
link |
00:31:19.500
there's no difference in happiness about money.
link |
00:31:23.580
The happiness is correlated with meaning and purpose,
link |
00:31:27.060
a sense of family, a sense of impact.
link |
00:31:30.020
So if you organize your life,
link |
00:31:31.900
assuming you have enough to get around
link |
00:31:33.620
and have a nice home and so forth,
link |
00:31:35.860
you'll be far happier if you figure out
link |
00:31:38.300
what you care about and work on that.
link |
00:31:41.660
It's often being in service to others.
link |
00:31:44.580
There's a great deal of evidence that people are happiest
link |
00:31:46.860
when they're serving others and not themselves.
link |
00:31:49.540
This goes directly against the sort of
link |
00:31:52.540
press induced excitement about
link |
00:31:56.100
powerful and wealthy leaders of one kind.
link |
00:31:59.220
And indeed these are consequential people.
link |
00:32:01.700
But if you are in a situation
link |
00:32:03.860
where you've been very fortunate as I have,
link |
00:32:06.100
you also have to take that as a responsibility
link |
00:32:09.020
and you have to basically work both to educate others
link |
00:32:12.180
and give them that opportunity,
link |
00:32:13.580
but also use that wealth to advance human society.
link |
00:32:16.700
In my case, I'm particularly interested in
link |
00:32:18.540
using the tools of artificial intelligence
link |
00:32:20.580
and machine learning to make society better.
link |
00:32:22.860
I've mentioned education, I've mentioned inequality
link |
00:32:26.020
and middle class and things like this,
link |
00:32:28.060
all of which are a passion of mine.
link |
00:32:30.100
It doesn't matter what you do,
link |
00:32:31.860
it matters that you believe in it,
link |
00:32:33.700
that it's important to you,
link |
00:32:35.380
and that your life will be far more satisfying
link |
00:32:38.100
if you spend your life doing that.
link |
00:32:40.540
I think there's no better place to end
link |
00:32:43.460
than a discussion of the meaning of life.
link |
00:32:45.220
Eric, thank you so much.