back to indexSebastian Thrun: Flying Cars, Autonomous Vehicles, and Education | Lex Fridman Podcast #59
link |
The following is a conversation with Sebastian Thrun.
link |
He's one of the greatest roboticists, computer scientists, and educators of our time.
link |
He led the development of the autonomous vehicles at Stanford
link |
that won the 2005 DARPA Grand Challenge and placed second in the 2007 DARPA Urban Challenge.
link |
He then led the Google self driving car program, which launched the self driving car revolution.
link |
He taught the popular Stanford course on artificial intelligence in 2011,
link |
which was one of the first massive open online courses, or MOOCs as they're commonly called.
link |
That experience led him to co found Udacity, an online education platform.
link |
If you haven't taken courses on it yet, I highly recommend it.
link |
Their self driving car program, for example, is excellent.
link |
He's also the CEO of Kitty Hawk, a company working on building flying cars,
link |
or more technically, EVTOLs, which stands for electric vertical takeoff and landing aircraft.
link |
He has launched several revolutions and inspired millions of people.
link |
But also, as many know, he's just a really nice guy.
link |
It was an honor and a pleasure to talk with him.
link |
This is the Artificial Intelligence Podcast.
link |
If you enjoy it, subscribe on YouTube, give it five stars on Apple Podcast,
link |
follow it on Spotify, support it on Patreon, or simply connect with me on Twitter
link |
at Lex Friedman, spelled F R I D M A N.
link |
If you leave a review on Apple Podcast or YouTube or Twitter,
link |
consider mentioning ideas, people, topics you find interesting.
link |
It helps guide the future of this podcast.
link |
But in general, I just love comments with kindness and thoughtfulness in them.
link |
This podcast is a side project for me, as many people know,
link |
but I still put a lot of effort into it.
link |
So the positive words of support from an amazing community, from you, really help.
link |
I recently started doing ads at the end of the introduction.
link |
I'll do one or two minutes after introducing the episode
link |
and never any ads in the middle that can break the flow of the conversation.
link |
I hope that works for you and doesn't hurt the listening experience.
link |
I provide timestamps for the start of the conversation that you can skip to,
link |
but it helps if you listen to the ad and support this podcast
link |
by trying out the product or service being advertised.
link |
This show is presented by Cash App, the number one finance app in the App Store.
link |
I personally use Cash App to send money to friends,
link |
but you can also use it to buy, sell, and deposit Bitcoin in just seconds.
link |
Cash App also has a new investing feature.
link |
You can buy fractions of a stock, say $1 worth, no matter what the stock price is.
link |
Broker services are provided by Cash App Investing,
link |
a subsidiary of Square, and member SIPC.
link |
I'm excited to be working with Cash App
link |
to support one of my favorite organizations called FIRST,
link |
best known for their FIRST Robotics and LEGO competitions.
link |
They educate and inspire hundreds of thousands of students
link |
in over 110 countries and have a perfect rating on Charity Navigator,
link |
which means the donated money is used to maximum effectiveness.
link |
When you get Cash App from the App Store or Google Play
link |
and use code LEGSPODCAST, you'll get $10,
link |
and Cash App will also donate $10 to FIRST,
link |
which again is an organization that I've personally seen inspire girls and boys
link |
to dream of engineering a better world.
link |
And now, here's my conversation with Sebastian Thrun.
link |
You mentioned that The Matrix may be your favorite movie.
link |
So let's start with a crazy philosophical question.
link |
Do you think we're living in a simulation?
link |
And in general, do you find the thought experiment interesting?
link |
Define simulation, I would say.
link |
Maybe we are, maybe we are not,
link |
but it's completely irrelevant to the way we should act.
link |
Putting aside, for a moment,
link |
the fact that it might not have any impact on how we should act as human beings,
link |
for people studying theoretical physics,
link |
these kinds of questions might be kind of interesting,
link |
looking at the universe as an information processing system.
link |
The universe is an information processing system.
link |
It's a huge physical, biological, chemical computer, there's no question.
link |
But I live here and now.
link |
I care about people, I care about us.
link |
What do you think is trying to compute?
link |
I don't think there's an intention.
link |
I think the world evolves the way it evolves.
link |
And it's beautiful, it's unpredictable.
link |
And I'm really, really grateful to be alive.
link |
Spoken like a true human.
link |
Which last time I checked, I was.
link |
Or that, in fact, this whole conversation is just a touring test
link |
to see if indeed you are.
link |
You've also said that one of the first programs,
link |
or the first few programs you've written was a, wait for it, TI57 calculator.
link |
Maybe that's early 80s.
link |
We don't want to date calculators or anything.
link |
That's early 80s, correct.
link |
So if you were to place yourself back into that time, into the mindset you were in,
link |
could you have predicted the evolution of computing, AI,
link |
the internet technology in the decades that followed?
link |
I was super fascinated by Silicon Valley, which I'd seen on television once
link |
and thought, my god, this is so cool.
link |
They build like DRAMs there and CPUs.
link |
And as a college student a few years later, I decided to really study
link |
intelligence and study human beings.
link |
And found that even back then in the 80s and 90s,
link |
artificial intelligence is what fascinated me the most.
link |
What's missing is that back in the day, the computers are really small.
link |
The brains we could build were not anywhere bigger than a cockroach.
link |
And cockroaches aren't very smart.
link |
So we weren't at the scale yet where we are today.
link |
Did you dream at that time to achieve the kind of scale we have today?
link |
Or did that seem possible?
link |
I always wanted to make robots smart.
link |
And I felt it was super cool to build an artificial human.
link |
And the best way to build an artificial human was to build a robot,
link |
because that's kind of the closest we could do.
link |
Unfortunately, we aren't there yet.
link |
The robots today are still very brittle.
link |
But it's fascinating to study intelligence from a constructive
link |
perspective when you build something.
link |
To understand you build, what do you think it takes to build an intelligent
link |
system, an intelligent robot?
link |
I think the biggest innovation that we've seen is machine learning.
link |
And it's the idea that the computers can basically teach themselves.
link |
Let's give an example.
link |
I'd say everybody pretty much knows how to walk.
link |
And we learn how to walk in the first year or two of our lives.
link |
But no scientist has ever been able to write down the rules of human gait.
link |
We don't understand it.
link |
We have it in our brains somehow.
link |
We can practice it.
link |
But we can't articulate it.
link |
We can't pass it on by language.
link |
And that, to me, is kind of the deficiency of today's computer programming.
link |
When you program a computer, they're so insanely dumb that you have to give them
link |
rules for every contingencies.
link |
Very unlike the way people learn from data and experience,
link |
computers are being instructed.
link |
And because it's so hard to get this instruction set right,
link |
we pay software engineers $200,000 a year.
link |
Now, the most recent innovation, which has been in the make for 30,
link |
40 years, is an idea that computers can find their own rules.
link |
So they can learn from falling down and getting up the same way children can
link |
learn from falling down and getting up.
link |
And that revolution has led to a capability that's completely unmatched.
link |
Today's computers can watch experts do their jobs, whether you're
link |
a doctor or a lawyer, pick up the regularities, learn those rules,
link |
and then become as good as the best experts.
link |
So the dream of in the 80s of expert systems, for example, had at its core
link |
the idea that humans could boil down their expertise on a sheet of paper,
link |
so to sort of reduce, sort of be able to explain to machines
link |
how to do something explicitly.
link |
So do you think, what's the use of human expertise into this whole picture?
link |
Do you think most of the intelligence will come from machines learning
link |
from experience without human expertise input?
link |
So the question for me is much more how do you express expertise?
link |
You can express expertise by writing a book.
link |
You can express expertise by showing someone what you're doing.
link |
You can express expertise by applying it by many different ways.
link |
And I think the expert systems was our best attempt in AI
link |
to capture expertise and rules.
link |
But someone sat down and said, here are the rules of human gait.
link |
Here's when you put your big toe forward and your heel backwards
link |
and you always stop stumbling.
link |
And as we now know, the set of rules, the set of language that we can command
link |
is incredibly limited.
link |
The majority of the human brain doesn't deal with language.
link |
It deals with subconscious, numerical, perceptual things
link |
that we don't even self aware of.
link |
Now, when an AI system watches an expert do their job and practice their job,
link |
it can pick up things that people can't even put into writing,
link |
into books or rules.
link |
And that's where the real power is.
link |
We now have AI systems that, for example, look over the shoulders
link |
of highly paid human doctors like dermatologists or radiologists,
link |
and they can somehow pick up those skills that no one can express in words.
link |
So you were a key person in launching three revolutions,
link |
online education, autonomous vehicles, and flying cars or VTOLs.
link |
So high level, and I apologize for all the philosophical questions.
link |
There's no apology necessary.
link |
How do you choose what problems to try and solve?
link |
What drives you to make those solutions a reality?
link |
I have two desires in life.
link |
I want to literally make the lives of others better.
link |
Or as we often say, maybe jokingly, make the world a better place.
link |
I actually believe in this.
link |
It's as funny as it sounds.
link |
And second, I want to learn.
link |
I want to get new skills.
link |
I don't want to be in a job I'm good at, because if I'm in a job
link |
that I'm good at, the chances for me to learn something interesting
link |
is actually minimized.
link |
So I want to be in a job I'm bad at.
link |
That's really important to me.
link |
So in a bill, for example, what people often
link |
call flying cars, these are electrical, vertical, takeoff,
link |
and landing vehicles.
link |
I'm just no expert in any of this.
link |
And it's so much fun to learn on the job what it actually means
link |
to build something like this.
link |
Now, I'd say the stuff that I've done lately
link |
after I finished my professorship at Stanford,
link |
they really focused on what has the maximum impact on society.
link |
Transportation is something that has transformed the 21st
link |
or 20th century more than any other invention,
link |
in my opinion, even more than communication.
link |
And cities are different.
link |
Workers are different.
link |
Women's rights are different because of transportation.
link |
And yet, we still have a very suboptimal transportation
link |
solution where we kill 1.2 or so million people every year
link |
It's like the leading cause of death for young people
link |
in many countries, where we are extremely inefficient
link |
Just go to your average neighborhood city
link |
and look at the number of parked cars.
link |
That's a travesty, in my opinion.
link |
Or where we spend endless hours in traffic jams.
link |
And very, very simple innovations,
link |
like a self driving car or what people call a flying car,
link |
could completely change this.
link |
I mean, the technology is basically there.
link |
You have to close your eyes not to see it.
link |
So lingering on autonomous vehicles, a fascinating space,
link |
some incredible work you've done throughout your career there.
link |
So let's start with DARPA, I think, the DARPA challenge,
link |
through the desert and then urban to the streets.
link |
I think that inspired an entire generation of roboticists
link |
and obviously sprung this whole excitement
link |
about this particular kind of four wheeled robots
link |
we called autonomous cars, self driving cars.
link |
So you led the development of Stanley, the autonomous car
link |
that won the race to the desert, the DARPA challenge in 2005.
link |
And Junior, the car that finished second
link |
in the DARPA urban challenge, also did incredibly well
link |
What are some painful, inspiring, or enlightening
link |
experiences from that time that stand out to you?
link |
Painful were all these incredibly complicated,
link |
stupid bugs that had to be found.
link |
We had a phase where Stanley, our car that eventually
link |
won the DARPA grand challenge, would every 30 miles
link |
just commit suicide.
link |
And we didn't know why.
link |
And it ended up to be that in the sinking of two computer
link |
clocks, occasionally a clock went backwards
link |
and that negative time elapsed, screwed up
link |
the entire internal logic.
link |
But it took ages to find this.
link |
There were bugs like that.
link |
I'd say enlightening is the Stanford team immediately
link |
focused on machine learning and on software,
link |
whereas everybody else seemed to focus on building better hardware.
link |
Our analysis had been a human being with an existing rental
link |
car can perfectly drive the course
link |
but why do I have to build a better rental car?
link |
I just should replace the human being.
link |
And the human being, to me, was a conjunction of three steps.
link |
We had sensors, eyes and ears, mostly eyes.
link |
We had brains in the middle.
link |
And then we had actuators, our hands and our feet.
link |
Now, the actuators are easy to build.
link |
The sensors are actually also easy to build.
link |
What was missing was the brain.
link |
So we had to build a human brain.
link |
And nothing clearer than to me that the human brain
link |
is a learning machine.
link |
So why not just train our robot?
link |
So we would build massive machine learning
link |
And with that, we were able to not just learn
link |
from human drivers.
link |
We had the entire speed control of the vehicle
link |
was copied from human driving.
link |
But also have the robot learn from experience
link |
where it made a mistake and recover from it
link |
and learn from it.
link |
You mentioned the pain point of software and clocks.
link |
Synchronization seems to be a problem that
link |
continues with robotics.
link |
It's a tricky one with drones and so on.
link |
What does it take to build a thing, a system
link |
with so many constraints?
link |
You have a deadline, no time.
link |
You're unsure about anything really.
link |
It's the first time that people really even exploring.
link |
It's not even sure that anybody can finish
link |
when we're talking about the race to the desert
link |
the year before nobody finish.
link |
What does it take to scramble and finish
link |
a product that actually, a system that actually works?
link |
We were very lucky.
link |
We were a really small team.
link |
The core of the team were four people.
link |
It was four because five couldn't comfortably sit
link |
inside a car, but four could.
link |
And I, as a team leader, my job was
link |
to get pizza for everybody and wash the car and stuff
link |
like this and repair the radiator when it broke
link |
and debug the system.
link |
And we were very open minded.
link |
We had no egos involved.
link |
We just wanted to see how far we can get.
link |
What we did really, really well was time management.
link |
We were done with everything a month before the race.
link |
And we froze the entire software a month before the race.
link |
And it turned out, looking at other teams,
link |
every other team complained if they had just one more week,
link |
they would have won.
link |
And we decided we're not going to fall into that mistake.
link |
We're going to be early.
link |
And we had an entire month to shake the system.
link |
And we actually found two or three minor bugs
link |
in the last month that we had to fix.
link |
And we were completely prepared when the race occurred.
link |
Okay, so first of all, that's such an incredibly rare
link |
achievement in terms of being able to be done on time
link |
What do you, how do you do that in your future work?
link |
What advice do you have in general?
link |
Because it seems to be so rare,
link |
especially in highly innovative projects like this.
link |
People work till the last second.
link |
Well, the nice thing about the DARPA Grand Challenge
link |
is that the problem was incredibly well defined.
link |
We were able for a while to drive
link |
the old DARPA Grand Challenge course,
link |
which had been used the year before.
link |
And then at some reason we were kicked out of the region.
link |
So we had to go to a different desert, the Snorran Desert,
link |
and we were able to drive desert trails
link |
just of the same type.
link |
So there was never any debate about like,
link |
what is actually the problem?
link |
We didn't sit down and say,
link |
hey, should we build a car or a plane?
link |
We had to build a car.
link |
That made it very, very easy.
link |
Then I studied my own life and life of others.
link |
And we realized that the typical mistake that people make
link |
is that there's this kind of crazy bug left
link |
that they haven't found yet.
link |
And it's just, they regret it.
link |
And that bug would have been trivial to fix.
link |
They just haven't fixed it yet.
link |
They didn't want to fall into that trap.
link |
So I built a testing team.
link |
We had a testing team that built a testing booklet
link |
of 160 pages of tests we had to go through
link |
just to make sure we shake out the system appropriately.
link |
And the testing team was with us all the time
link |
and dictated to us today, we do railroad crossings.
link |
Tomorrow we do, we practice the start of the event.
link |
And in all of these, we thought,
link |
oh my God, it's long solved trivial.
link |
And then we tested it out.
link |
Oh my God, it doesn't do a railroad crossing.
link |
Oh my God, it mistakes the rails for metal barriers.
link |
We have to fix this.
link |
So it was really a continuous focus
link |
on improving the weakest part of the system.
link |
And as long as you focus on improving
link |
the weakest part of the system,
link |
you eventually build a really great system.
link |
Let me just pause on that, to me as an engineer,
link |
it's just super exciting that you were thinking like that,
link |
especially at that stage as brilliant,
link |
that testing was such a core part of it.
link |
It may be to linger on the point of leadership.
link |
I think it's one of the first times
link |
you were really a leader
link |
and you've led many very successful teams since then.
link |
What does it take to be a good leader?
link |
I would say most of all, I just take credit.
link |
I put the work of others, right?
link |
That's very convenient turns out
link |
because I can't do all these things myself.
link |
I'm an engineer at heart.
link |
So I care about engineering.
link |
So I don't know what the chicken and the egg is,
link |
but as a kid, I loved computers
link |
because you could tell them to do something
link |
and they actually did it.
link |
And you could like in the middle of the night,
link |
wake up at one in the morning and switch on your computer.
link |
And what he told you to yesterday, it would still do.
link |
That was really cool.
link |
Unfortunately, that didn't quite work with people.
link |
So you go to people and tell them what to do
link |
and they don't do it.
link |
And they hate you for it, or you do it today
link |
and then you go a day later and they stop doing it.
link |
So then the question really became,
link |
how can you put yourself in the brain of people
link |
as opposed to computers?
link |
And in terms of computers, it's super dumb.
link |
If people were as dumb as computers,
link |
I wouldn't want to work with them.
link |
But people are smart and people are emotional
link |
and people have pride and people have aspirations.
link |
So how can I connect to that?
link |
And that's the thing that most of our leadership just fails
link |
because many, many engineers turn manager
link |
believe they can treat their team just the same way
link |
it can treat your computer.
link |
And it just doesn't work this way.
link |
It's just really bad.
link |
So how can I connect to people?
link |
And it turns out as a college professor,
link |
the wonderful thing you do all the time
link |
is to empower other people.
link |
Like your job is to make your students look great.
link |
That's all you do.
link |
You're the best coach.
link |
And it turns out if you do a fantastic job with making
link |
your students look great, they actually love you
link |
and their parents love you.
link |
And they give you all the credit for stuff you don't deserve.
link |
All my students were smarter than me.
link |
All the great stuff invented at Stanford
link |
was their stuff, not my stuff.
link |
And they give me credit and say, oh, Sebastian.
link |
We're just making them feel good about themselves.
link |
So the question really is, can you take a team of people
link |
and what does it take to make them
link |
to connect to what they actually want in life
link |
and turn this into productive action?
link |
It turns out every human being that I know
link |
has incredibly good intentions.
link |
I've really rarely met a person with bad intentions.
link |
I believe every person wants to contribute.
link |
I think every person I've met wants to help others.
link |
It's amazing how much of an urge we have
link |
not to just help ourselves, but to help others.
link |
So how can we empower people and give them
link |
the right framework that they can accomplish this?
link |
In moments when it works, it's magical.
link |
Because you'd see the confluence of people
link |
being able to make the world a better place
link |
and deriving enormous confidence and pride out of this.
link |
And that's when my environment works the best.
link |
These are moments where I can disappear for a month
link |
and come back and things still work.
link |
It's very hard to accomplish.
link |
But when it works, it's amazing.
link |
So I agree with you very much.
link |
It's not often heard that most people in the world
link |
have good intentions.
link |
At the core, their intentions are good
link |
and they're good people.
link |
That's a beautiful message, it's not often heard.
link |
We make this mistake, and this is a friend of mine,
link |
Alex Werder, talking to us, that we judge ourselves
link |
by our intentions and others by their actions.
link |
And I think that the biggest skill,
link |
I mean, here in Silicon Valley, we follow engineers
link |
who have very little empathy and are kind of befuddled
link |
by why it doesn't work for them.
link |
The biggest skill, I think, that people should acquire
link |
is to put themselves into the position of the other
link |
and listen, and listen to what the other has to say.
link |
And they'd be shocked how similar they are to themselves.
link |
And they might even be shocked how their own actions
link |
don't reflect their intentions.
link |
I often have conversations with engineers
link |
where I say, look, hey, I love you, you're doing a great job.
link |
And by the way, what you just did has the following effect.
link |
Are you aware of that?
link |
And then people would say, oh my God, not I wasn't,
link |
because my intention was that.
link |
And I say, yeah, I trust your intention.
link |
You're a good human being.
link |
But just to help you in the future,
link |
if you keep expressing it that way,
link |
then people will just hate you.
link |
And I've had many instances where people say,
link |
oh my God, thank you for telling me this,
link |
because it wasn't my intention to look like an idiot.
link |
It wasn't my intention to help other people.
link |
I just didn't know how to do it.
link |
Very simple, by the way.
link |
There's a book, Dale Carnegie, 1936,
link |
how to make friends and how to influence others.
link |
Has the entire Bible, you just read it and you're done
link |
and you apply it every day.
link |
And I wish I was good enough to apply it every day.
link |
But it's just simple things, right?
link |
Like be positive, remember people's name, smile,
link |
and eventually have empathy.
link |
Really think that the person that you hate
link |
and you think is an idiot,
link |
is actually just like yourself.
link |
It's a person who's struggling, who means well,
link |
and who might need help, and guess what, you need help.
link |
I've recently spoken with Stephen Schwarzman.
link |
I'm not sure if you know who that is, but.
link |
But he said, sort of to expand on what you're saying,
link |
that one of the biggest things you can do
link |
is hear people when they tell you what their problem is
link |
and then help them with that problem.
link |
He says, it's surprising how few people
link |
actually listen to what troubles others.
link |
And because it's right there in front of you
link |
and you can benefit the world the most.
link |
And in fact, yourself and everybody around you
link |
by just hearing the problems and solving them.
link |
I mean, that's my little history of engineering.
link |
That is, while I was engineering with computers,
link |
I didn't care at all what the computer's problems were.
link |
I just told them what to do and to do it.
link |
And it just doesn't work this way with people.
link |
It doesn't work with me.
link |
If you come to me and say, do A, I do the opposite.
link |
But let's return to the comfortable world of engineering.
link |
And can you tell me in broad strokes in how you see it?
link |
Because you're the core of starting it,
link |
the core of driving it,
link |
the technical evolution of autonomous vehicles
link |
from the first DARPA Grand Challenge
link |
to the incredible success we see with the program
link |
you started with Google self driving car
link |
and Waymo and the entire industry that sprung up
link |
of different kinds of approaches, debates and so on.
link |
Well, the idea of self driving car goes back to the 80s.
link |
There was a team in Germany and another team
link |
at Carnegie Mellon that did some very pioneering work.
link |
But back in the day, I'd say the computers were so deficient
link |
that even the best professors and engineers in the world
link |
basically stood no chance.
link |
It then folded into a phase where the US government
link |
spent at least half a billion dollars
link |
that I could count on research projects.
link |
But the way the procurement works,
link |
a successful stack of paper describing lots of stuff
link |
that no one's ever gonna read
link |
was a successful product of a research project.
link |
So we trained our researchers to produce lots of paper.
link |
That all changed with the DARPA Grand Challenge.
link |
And I really gotta credit the ingenious people at DARPA
link |
and the US government and Congress
link |
that took a complete new funding model where they said,
link |
let's not fund effort, let's fund outcomes.
link |
And it sounds very trivial,
link |
but there was no tax code that allowed
link |
the use of congressional tax money for a price.
link |
It was all effort based.
link |
So if you put in a hundred hours in,
link |
you could charge a hundred hours.
link |
If you put in a thousand hours in,
link |
you could build a thousand hours.
link |
By changing the focus instead of making the price,
link |
we don't pay you for development,
link |
we pay for the accomplishment.
link |
They drew in, they automatically drew out
link |
all these contractors who are used to the drug
link |
of getting money per hour.
link |
And they drew in a whole bunch of new people.
link |
And these people are mostly crazy people.
link |
They were people who had a car and a computer
link |
and they wanted to make a million bucks.
link |
The million bucks was their visual price money,
link |
it was then doubled.
link |
And they felt if I put my computer in my car
link |
and program it, I can be rich.
link |
And that was so awesome.
link |
Like half the teams, there was a team that was surfer dudes
link |
and they had like two surfboards on their vehicle
link |
and brought like these fashion girls, super cute girls,
link |
like twin sisters.
link |
And you could tell these guys were not your common
link |
beltway bandit who gets all these big multimillion
link |
and billion dollar countries from the US government.
link |
And there was a great reset.
link |
Universities moved in.
link |
I was very fortunate at Stanford that I just received tenure
link |
so I couldn't get fired no matter what I do,
link |
otherwise I wouldn't have done it.
link |
And I had enough money to finance this thing
link |
and I was able to attract a lot of money from third parties.
link |
And even car companies moved in.
link |
They kind of moved in very quietly
link |
because they were super scared to be embarrassed
link |
that their car would flip over.
link |
But Ford was there and Volkswagen was there
link |
and a few others and GM was there.
link |
So it kind of reset the entire landscape of people.
link |
And if you look at who's a big name
link |
in self driving cars today,
link |
these were mostly people who participated
link |
in those challenges.
link |
Okay, that's incredible.
link |
Can you just comment quickly on your sense of lessons learned
link |
from that kind of funding model
link |
and the research that's going on in academia
link |
in terms of producing papers,
link |
is there something to be learned and scaled up bigger,
link |
having these kinds of grand challenges
link |
that could improve outcomes?
link |
So I'm a big believer in focusing
link |
on kind of an end to end system.
link |
I'm a really big believer in systems building.
link |
I've always built systems in my academic career,
link |
even though I do a lot of math and abstract stuff,
link |
but it's all derived from the idea
link |
of let's solve a real problem.
link |
And it's very hard for me to be an academic
link |
and say, let me solve a component of a problem.
link |
Like with someone there's fields like nonmonetary logic
link |
or AI planning systems where people believe
link |
that a certain style of problem solving
link |
is the ultimate end objective.
link |
And I would always turn it around and say,
link |
hey, what problem would my grandmother care about
link |
that doesn't understand computer technology
link |
and doesn't wanna understand?
link |
And how could I make her love what I do?
link |
Because only then do I have an impact on the world.
link |
I can easily impress my colleagues.
link |
That is much easier,
link |
but impressing my grandmother is very, very hard.
link |
So I would always thought if I can build a self driving car
link |
and my grandmother can use it
link |
even after she loses her driving privileges
link |
or children can use it,
link |
or we save maybe a million lives a year,
link |
that would be very impressive.
link |
And then there's so many problems like these,
link |
like there's a problem with curing cancer,
link |
or whatever it is, live twice as long.
link |
Once a problem is defined,
link |
of course I can't solve it in its entirety.
link |
Like it takes sometimes tens of thousands of people
link |
to find a solution.
link |
There's no way you can fund an army of 10,000 at Stanford.
link |
So you gotta build a prototype.
link |
Let's build a meaningful prototype.
link |
And the DARPA Grand Challenge was beautiful
link |
because it told me what this prototype had to do.
link |
I didn't have to think about what it had to do,
link |
I just had to read the rules.
link |
And that was really beautiful.
link |
And it's most beautiful,
link |
you think what academia could aspire to
link |
is to build a prototype that's the systems level,
link |
that solves or gives you an inkling
link |
that this problem could be solved with this prototype.
link |
First of all, I wanna emphasize what academia really is.
link |
And I think people misunderstand it.
link |
First and foremost, academia is a way
link |
to educate young people.
link |
First and foremost, a professor is an educator.
link |
No matter where you are at,
link |
a small suburban college,
link |
or whether you are a Harvard or Stanford professor,
link |
that's not the way most people think of themselves
link |
in academia because we have this kind of competition
link |
going on for citations and publication.
link |
That's a measurable thing,
link |
but that is secondary to the primary purpose
link |
of educating people to think.
link |
Now, in terms of research,
link |
most of the great science,
link |
the great research comes out of universities.
link |
You can trace almost everything back,
link |
including Google, to universities.
link |
So there's nothing really fundamentally broken here.
link |
It's a good system.
link |
And I think America has the finest university system
link |
We can talk about reach
link |
and how to reach people outside the system.
link |
It's a different topic,
link |
but the system itself is a good system.
link |
If I had one wish, I would say it'd be really great
link |
if there was more debate about
link |
what the great big problems are in society
link |
and focus on those.
link |
And most of them are interdisciplinary.
link |
Unfortunately, it's very easy to fall
link |
into an interdisciplinary viewpoint
link |
where your problem is dictated
link |
by what your closest colleagues believe the problem is.
link |
It's very hard to break out and say,
link |
well, there's an entire new field of problems.
link |
So to give an example,
link |
prior to me working on self driving cars,
link |
I was a roboticist and a machine learning expert.
link |
And I wrote books on robotics,
link |
something called probabilistic robotics.
link |
It's a very methods driven kind of viewpoint of the world.
link |
I built robots that acted in museums as tour guides,
link |
that let children around.
link |
It is something that at the time was moderately challenging.
link |
When I started working on cars,
link |
several colleagues told me,
link |
Sebastian, you're destroying your career
link |
because in our field of robotics,
link |
cars are looked like as a gimmick
link |
and they're not expressive enough.
link |
They can only push the throttle and the brakes.
link |
There's no dexterity.
link |
There's no complexity.
link |
It's just too simple.
link |
And no one came to me and said,
link |
wow, if you solve that problem,
link |
you can save a million lives, right?
link |
Among all robotic problems that I've seen in my life,
link |
I would say the self driving car, transportation,
link |
is the one that has the most hope for society.
link |
So how come the robotics community wasn't all over the place?
link |
And it was because we focused on methods and solutions
link |
and not on problems.
link |
Like if you go around today and ask your grandmother,
link |
What really makes you upset?
link |
I challenge any academic to do this
link |
and then realize how far your research
link |
is probably away from that today.
link |
At the very least, that's a good thing
link |
for academics to deliberate on.
link |
The other thing that's really nice in Silicon Valley is,
link |
Silicon Valley is full of smart people outside academia.
link |
So there's the Larry Pages and Mark Zuckerbergs in the world
link |
who are anywhere smarter, smarter
link |
than the best academics I've met in my life.
link |
And what they do is they are at a different level.
link |
They build the systems,
link |
they build the customer facing systems,
link |
they build things that people can use
link |
without technical education.
link |
And they are inspired by research.
link |
They're inspired by scientists.
link |
They hire the best PhDs from the best universities
link |
So I think this kind of vertical integration
link |
between the real product, the real impact
link |
and the real thought, the real ideas,
link |
that's actually working surprisingly well in Silicon Valley.
link |
It did not work as well in other places in this nation.
link |
So when I worked at Carnegie Mellon,
link |
we had the world's finest computer science university,
link |
but there wasn't those people in Pittsburgh
link |
that would be able to take these
link |
very fine computer science ideas
link |
and turn them into massive, impactful products.
link |
That symbiosis seemed to exist
link |
pretty much only in Silicon Valley
link |
and maybe a bit in Boston and Austin.
link |
Yeah, with Stanford, that's really interesting.
link |
So if we look a little bit further on
link |
from the DARPA Grand Challenge
link |
and the launch of the Google self driving car,
link |
what do you see as the state,
link |
the challenges of autonomous vehicles as they are now
link |
is actually achieving that huge scale
link |
and having a huge impact on society?
link |
I'm extremely proud of what has been accomplished.
link |
And again, I'm taking a lot of credit for the work of others.
link |
And I'm actually very optimistic.
link |
And people have been kind of worrying,
link |
is it too fast? Is it too slow?
link |
Why is it not there yet? And so on.
link |
It is actually quite an interesting, hard problem.
link |
And in that a self driving car,
link |
to build one that manages 90% of the problems
link |
encountered in everyday driving is easy.
link |
We can literally do this over a weekend.
link |
To do 99% might take a month.
link |
Then there's 1% left.
link |
So 1% would mean that you still have a fatal accident
link |
every week, very unacceptable.
link |
So now you work on this 1%
link |
and the 99% of that, the remaining 1%
link |
is actually still relatively easy,
link |
but now you're down to like a hundredth of 1%.
link |
And it's still completely unacceptable in terms of safety.
link |
So the variety of things you encounter are just enormous.
link |
And that gives me enormous respect for human being
link |
that we're able to deal with the couch on the highway,
link |
or the deer in the headlights, or the blown tire
link |
that we've never been trained for.
link |
And all of a sudden have to handle it
link |
in an emergency situation
link |
and often do very, very successfully.
link |
It's amazing from that perspective,
link |
how safe driving actually is given how many millions
link |
of miles we drive every year in this country.
link |
We are now at a point where I believe the technology
link |
is there and I've seen it.
link |
I've seen it in Waymo, I've seen it in Aptiv,
link |
I've seen it in Cruise and in a number of companies
link |
and in Voyage where vehicles now driving around
link |
and basically flawlessly are able to drive people around
link |
in limited scenarios.
link |
In fact, you can go to Vegas today
link |
and order a Summon and Lift.
link |
And if you get the right setting of your app,
link |
you'll be picked up by a driverless car.
link |
Now there's still safety drivers in there,
link |
but that's a fantastic way to kind of learn
link |
what the limits are of technology today.
link |
And there's still some glitches,
link |
but the glitches have become very, very rare.
link |
I think the next step is gonna be to down cost it,
link |
to harden it, the entrapment, the sensors
link |
are not quite an automotive grade standard yet.
link |
And then to really build the business models,
link |
to really kind of go somewhere and make the business case.
link |
And the business case is hard work.
link |
It's not just, oh my God, we have this capability,
link |
people are just gonna buy it.
link |
You have to make it affordable.
link |
You have to find the social acceptance of people.
link |
None of the teams yet has been able to or gutsy enough
link |
to drive around without a person inside the car.
link |
And that's the next magical hurdle.
link |
We'll be able to send these vehicles around
link |
completely empty in traffic.
link |
And I think, I mean, I wait every day,
link |
wait for the news that Waymo has just done this.
link |
So, interesting you mentioned gutsy.
link |
Let me ask some maybe unanswerable question,
link |
maybe edgy questions.
link |
But in terms of how much risk is required,
link |
some guts in terms of leadership style,
link |
it would be good to contrast approaches.
link |
And I don't think anyone knows what's right.
link |
But if we compare Tesla and Waymo, for example,
link |
Elon Musk and the Waymo team,
link |
there's slight differences in approach.
link |
So on the Elon side, there's more,
link |
I don't know what the right word to use,
link |
but aggression in terms of innovation.
link |
And on Waymo side, there's more sort of cautious,
link |
safety focused approach to the problem.
link |
What do you think it takes?
link |
What leadership at which moment is right?
link |
Which approach is right?
link |
Look, I don't sit in either of those teams.
link |
So I'm unable to even verify like somebody says correct.
link |
In the end of the day, every innovator in that space
link |
will face a fundamental dilemma.
link |
And I would say you could put aerospace titans
link |
into the same bucket,
link |
which is you have to balance public safety
link |
with your drive to innovate.
link |
And this country in particular in the States
link |
has a hundred plus year history
link |
of doing this very successfully.
link |
Air travel is what a hundred times a safe per mile
link |
than ground travel, than cars.
link |
And there's a reason for it because people have found ways
link |
to be very methodological about ensuring public safety
link |
while still being able to make progress
link |
on important aspects, for example,
link |
like air and noise and fuel consumption.
link |
So I think that those practices are proven
link |
and they actually work.
link |
We live in a world safer than ever before.
link |
And yes, there will always be the provision
link |
that something goes wrong.
link |
There's always the possibility
link |
that someone makes a mistake
link |
or there's an unexpected failure.
link |
We can never guarantee to a hundred percent
link |
absolute safety other than just not doing it.
link |
But I think I'm very proud of the history of the United States.
link |
I mean, we've dealt with much more dangerous technology
link |
like nuclear energy and kept that safe too.
link |
We have nuclear weapons and we keep those safe.
link |
So we have methods and procedures
link |
that really balance these two things very, very successfully.
link |
You've mentioned a lot of great autonomous vehicle companies
link |
that are taking sort of the level four, level five,
link |
they jump in full autonomy with a safety driver
link |
and take that kind of approach
link |
and also through simulation and so on.
link |
There's also the approach that Tesla Autopilot is doing,
link |
which is kind of incrementally taking a level two vehicle
link |
and using machine learning
link |
and learning from the driving of human beings
link |
and trying to creep up,
link |
trying to incrementally improve the system
link |
until it's able to achieve level four autonomy.
link |
So perfect autonomy in certain kind of geographical regions.
link |
What are your thoughts on these contrasting approaches?
link |
Well, so first of all, I'm a very proud Tesla owner
link |
and I literally use the Autopilot every day
link |
and it literally has kept me safe.
link |
It is a beautiful technology specifically
link |
for highway driving when I'm slightly tired
link |
because then it turns me into a much safer driver.
link |
And I'm 100% confident that's the case.
link |
In terms of the right approach,
link |
I think the biggest change I've seen
link |
since I went to Waymo team is this thing called deep learning.
link |
I think deep learning was not a hot topic
link |
when I started Waymo or Google self driving cars.
link |
It was there, in fact, we started Google Brain
link |
at the same time in Google X.
link |
So I invested in deep learning,
link |
but people didn't talk about it, it wasn't a hot topic.
link |
And now it is, there's a shift of emphasis
link |
from a more geometric perspective
link |
where you use geometric sensors
link |
that give you a full 3D view
link |
when you do a geometric reasoning about,
link |
oh, this box over here might be a car
link |
towards a more human like, oh, let's just learn about it.
link |
This looks like the thing I've seen 10,000 times before.
link |
So maybe it's the same thing, machine learning perspective.
link |
And that has really put, I think,
link |
all these approaches on steroids.
link |
At Udacity, we teach a course in self driving cars.
link |
In fact, I think we've graduated over 20,000 or so people
link |
on self driving car skills.
link |
So every self driving car team in the world
link |
now uses our engineers.
link |
And in this course, the very first homework assignment
link |
is to do lane finding on images.
link |
And lane finding images for layman,
link |
what this means is you put a camera into your car
link |
or you open your eyes and you would know where the lane is.
link |
So you can stay inside the lane with your car.
link |
Humans can do this super easily.
link |
You just look and you know where the lane is,
link |
For machines, for a long time, it was super hard
link |
because people would write these kind of crazy rules.
link |
If there's like wine lane markers
link |
and here's what white really means,
link |
this is not quite white enough.
link |
So let's, oh, it's not white.
link |
Or maybe the sun is shining.
link |
So when the sun shines and this is white
link |
and this is a straight line,
link |
I mean, it's not quite a straight line
link |
because the road is curved.
link |
And do we know that there's really six feet
link |
between lane markings or not or 12 feet, whatever it is.
link |
And now what the students are doing,
link |
they would take machine learning.
link |
So instead of like writing these crazy rules
link |
for the lane marker,
link |
they'll say, hey, let's take an hour of driving
link |
and label it and tell the vehicle,
link |
this is actually the lane by hand.
link |
And then these are examples
link |
and have the machine find its own rules,
link |
what lane markings are.
link |
And within 24 hours, now every student
link |
that's never done any programming before in this space
link |
can write a perfect lane finder
link |
as good as the best commercial lane finders.
link |
And that's completely amazing to me.
link |
We've seen progress using machine learning
link |
that completely dwarfs anything
link |
that I saw 10 years ago.
link |
Yeah, and just as a side note,
link |
the self driving car nanodegree,
link |
the fact that you launched that many years ago now,
link |
maybe four years ago, three years ago is incredible
link |
that that's a great example of system level thinking
link |
sort of just taking an entire course
link |
that teaches you how to solve the entire problem.
link |
I definitely recommend people.
link |
It's become super popular
link |
and it's become actually incredibly high quality
link |
really with Mercedes and various other companies
link |
And we find that engineers from Tesla and Waymo
link |
are taking it today.
link |
The insight was that two things,
link |
one is existing universities will be very slow to move
link |
because they're departmentalized
link |
and there's no department for self driving cars.
link |
So between Mac E and double E and computer science,
link |
getting those folks together
link |
into one room is really, really hard.
link |
And every professor listening here will know,
link |
they'll probably agree to that.
link |
And secondly, even if all the great universities
link |
just did this, which none so far has developed
link |
a curriculum in this field,
link |
it is just a few thousand students that can partake
link |
because all the great universities are super selective.
link |
So how about people in India?
link |
How about people in China or in the Middle East
link |
or Indonesia or Africa?
link |
Why should those be excluded
link |
from the skill of building self driving cars?
link |
Are they any dumber than we are?
link |
Are we any less privileged?
link |
And the answer is we should just give everybody the skill
link |
to build a self driving car.
link |
Because if we do this,
link |
then we have like a thousand self driving car startups.
link |
And if 10% succeed, that's like a hundred,
link |
that means hundred countries now
link |
will have self driving cars and be safer.
link |
It's kind of interesting to imagine impossible to quantify,
link |
but the number, the, you know,
link |
over a period of several decades,
link |
the impact that has like a single course,
link |
like a ripple effect of society.
link |
If you, I just recently talked to Andrew
link |
who was creator of Cosmos show.
link |
It's interesting to think about
link |
how many scientists that show launched.
link |
And so it's really, in terms of impact,
link |
I can't imagine a better course
link |
than the self driving car course.
link |
That's, you know, there's other more specific disciplines
link |
like deep learning and so on that Udacity is also teaching,
link |
but self driving cars,
link |
it's really, really interesting course.
link |
And then it came at the right moment.
link |
It came at a time when there were a bunch of Acqui hires.
link |
Acqui hire is a acquisition of a company,
link |
not for its technology or its products or business,
link |
but for its people.
link |
So Acqui hire means maybe that a company of 70 people,
link |
they have no product yet, but they're super smart people
link |
and they pay a certain amount of money.
link |
So I took Acqui hires like GM Cruise and Uber and others,
link |
and did the math and said,
link |
hey, how many people are there and how much money was paid?
link |
And as a lower bound,
link |
I estimated the value of a self driving car engineer
link |
in these acquisitions to be at least $10 million, right?
link |
So think about this, you get yourself a skill
link |
and you team up and build a company
link |
and your worth now is $10 million.
link |
I mean, that's kind of cool.
link |
I mean, what other thing could you do in life
link |
to be worth $10 million within a year?
link |
But to come back for a moment on to deep learning
link |
and its application in autonomous vehicles,
link |
what are your thoughts on Elon Musk's statement,
link |
provocative statement, perhaps that light air is a crutch.
link |
So this geometric way of thinking about the world
link |
may be holding us back if what we should instead be doing
link |
in this robotic space,
link |
in this particular space of autonomous vehicles
link |
is using camera as a primary sensor
link |
and using computer vision and machine learning
link |
as the primary way to...
link |
Look, I have two comments.
link |
I think first of all, we all know
link |
that people can drive cars without lighters in their heads
link |
because we only have eyes
link |
and we mostly just use eyes for driving.
link |
Maybe we use some other perception about our bodies,
link |
accelerations, occasionally our ears,
link |
certainly not our noses.
link |
So the existence proof is there,
link |
that eyes must be sufficient.
link |
In fact, we could even drive a car
link |
if someone put a camera out
link |
and then gave us the camera image with no latency,
link |
we would be able to drive a car that way the same way.
link |
So a camera is also sufficient.
link |
Secondly, I really love the idea that in the Western world,
link |
we have many, many different people
link |
trying different hypotheses.
link |
It's almost like an anthill,
link |
like if an anthill tries to forge for food,
link |
you can sit there as two ants
link |
and agree what the perfect path is
link |
and then every single ant marches
link |
for the most likely location of food is,
link |
or you can even just spread out.
link |
And I promise you the spread out solution will be better
link |
because if the discussing philosophical,
link |
intellectual ants get it wrong
link |
and they're all moving the wrong direction,
link |
they're going to waste a day
link |
and then they're going to discuss again for another week.
link |
Whereas if all these ants go in a random direction,
link |
someone's going to succeed
link |
and they're going to come back and claim victory
link |
and get the Nobel prize or whatever the ant equivalent is.
link |
And then they all march in the same direction.
link |
And that's great about society.
link |
That's great about the Western society.
link |
We're not plan based, we're not central based.
link |
We don't have a Soviet Union style central government
link |
that tells us where to forge.
link |
We started in C Corp.
link |
We get investor money, go out and try it out.
link |
And who knows who's going to win.
link |
In your, when you look at the longterm vision
link |
of autonomous vehicles,
link |
do you see machine learning
link |
as fundamentally being able to solve most of the problems?
link |
So learning from experience.
link |
I'd say we should be very clear
link |
about what machine learning is and is not.
link |
And I think there's a lot of confusion.
link |
What it is today is a technology
link |
that can go through large databases
link |
of repetitive patterns and find those patterns.
link |
So in example, we did a study at Stanford two years ago
link |
where we applied machine learning
link |
to detecting skin cancer in images.
link |
And we harvested or built a data set
link |
of 129,000 skin photo shots
link |
that were all had been biopsied
link |
for what the actual situation was.
link |
And those included melanomas and carcinomas,
link |
also included rashes and other skin conditions, lesions.
link |
And then we had a network find those patterns.
link |
And it was by and large able to then detect skin cancer
link |
with an iPhone as accurately
link |
as the best board certified Stanford level dermatologist.
link |
Now this thing was great in this one thing
link |
and finding skin cancer, but it couldn't drive a car.
link |
So the difference to human intelligence
link |
is we do all these many, many things
link |
and we can often learn from a very small data set
link |
Whereas machines still need very large data sets
link |
and things that will be very repetitive.
link |
Now that's still super impactful
link |
because almost everything we do is repetitive.
link |
So that's gonna really transform human labor
link |
but it's not this almighty general intelligence.
link |
We're really far away from a system
link |
that will exhibit general intelligence.
link |
To that end, I actually commiserate the naming a little bit
link |
because artificial intelligence, if you believe Hollywood
link |
is immediately mixed into the idea of human suppression
link |
and machine superiority.
link |
I don't think that we're gonna see this in my lifetime.
link |
I don't think human suppression is a good idea.
link |
I don't see it coming.
link |
I don't see the technology being there.
link |
What I see instead is a very pointed focused
link |
pattern recognition technology that's able to
link |
extract patterns from large data sets.
link |
And in doing so, it can be super impactful.
link |
Let's take the impact of artificial intelligence
link |
We all know that it takes something like 10,000 hours
link |
to become an expert.
link |
If you're gonna be a doctor or a lawyer
link |
or even a really good driver,
link |
it takes a certain amount of time to become experts.
link |
Machines now are able and have been shown
link |
to observe people become experts and observe experts
link |
and then extract those rules from experts
link |
in some interesting way.
link |
They could go from law to sales to driving cars
link |
to diagnosing cancer.
link |
And then giving that capability to people who are
link |
completely new in their job.
link |
We now can, and that's been done.
link |
It's been done commercially in many, many instantiations.
link |
So that means we can use machine learning
link |
to make people expert on the very first day of their work.
link |
Like think about the impact.
link |
If your doctor is still in their first 10,000 hours,
link |
you have a doctor who is not quite an expert yet.
link |
Who would not want a doctor who is the world's best expert?
link |
And now we can leverage machines to really eradicate
link |
the error in decision making,
link |
error and lack of expertise for human doctors.
link |
That could save your life.
link |
If we can link on that for a little bit,
link |
in which way do you hope machines in the medical field
link |
could help assist doctors?
link |
You mentioned this sort of accelerating the learning curve
link |
or people, if they start a job or in the first 10,000 hours
link |
can be assisted by machines.
link |
How do you envision that assistance looking?
link |
So we built this app for an iPhone that can detect
link |
and classify and diagnose skin cancer.
link |
And we proved two years ago that it does pretty much
link |
as good or better than the best human doctors.
link |
So let me tell you a story.
link |
So there's a friend of mine, let's call him Ben.
link |
Ben is a very famous venture capitalist.
link |
He goes to his doctor and the doctor looks at a mole
link |
and says, hey, that mole is probably harmless.
link |
And for some very funny reason, he pulls out that phone
link |
He's a collaborator in our study.
link |
And the app says, no, no, no, no, this is a melanoma.
link |
And for background, melanomas are,
link |
and skin cancer is the most common cancer in this country.
link |
Melanomas can go from stage zero to stage four
link |
within less than a year.
link |
Stage zero means you can basically cut it out yourself
link |
with a kitchen knife and be safe.
link |
And stage four means your chances of living
link |
five more years in less than 20%.
link |
So it's a very serious, serious, serious condition.
link |
So this doctor who took out the iPhone,
link |
looked at the iPhone and was a little bit puzzled.
link |
He said, I mean, but just to be safe,
link |
let's cut it out and biopsy it.
link |
That's the technical term for let's get
link |
an in depth diagnostics that is more than just looking at it.
link |
And it came back as cancerous, as a melanoma.
link |
And it was then removed.
link |
And my friend, Ben, I was hiking with him
link |
and we were talking about AI.
link |
And I told him I do this work on skin cancer.
link |
And he said, oh, funny.
link |
My doctor just had an iPhone that found my cancer.
link |
So I was like completely intrigued.
link |
I didn't even know about this.
link |
So here's a person, I mean, this is a real human life, right?
link |
Like who doesn't know somebody
link |
who has been affected by cancer.
link |
Cancer is cause of death number two.
link |
Cancer is this kind of disease that is mean
link |
in the following way.
link |
Most cancers can actually be cured relatively easily
link |
if we catch them early.
link |
And the reason why we don't tend to catch them early
link |
is because they have no symptoms.
link |
Like your very first symptom of a gallbladder cancer
link |
or a pancreas cancer might be a headache.
link |
And when you finally go to your doctor
link |
because of these headaches or your back pain
link |
and you're being imaged, it's usually stage four plus.
link |
And that's the time when the occurring chances
link |
might be dropped to a single digit percentage.
link |
So if we could leverage AI to inspect your body
link |
on a regular basis without even a doctor in the room,
link |
maybe when you take a shower or what have you,
link |
I know this sounds creepy,
link |
but then we might be able to save millions
link |
and millions of lives.
link |
You've mentioned there's a concern that people have
link |
about near term impacts of AI in terms of job loss.
link |
So you've mentioned being able to assist doctors,
link |
being able to assist people in their jobs.
link |
Do you have a worry of people losing their jobs
link |
or the economy being affected by the improvements in AI?
link |
Yeah, anybody concerned about job losses,
link |
please come to Gdacity.com.
link |
We teach contemporary tech skills
link |
and we have a kind of implicit job promise.
link |
We often, when we measure,
link |
we spend way over 50% of our graders in new jobs
link |
and they're very satisfied about it.
link |
And it costs almost nothing,
link |
costs like 1,500 max or something like that.
link |
And so there's a cool new program
link |
that you agree with the U.S. government,
link |
guaranteeing that you will help us give scholarships
link |
that educate people in this kind of situation.
link |
Yeah, we're working with the U.S. government
link |
on the idea of basically rebuilding the American dream.
link |
So Gdacity has just dedicated 100,000 scholarships
link |
for citizens of America for various levels of courses
link |
that eventually will get you a job.
link |
And those courses are all somewhat related
link |
to the tech sector because the tech sector
link |
is kind of the hottest sector right now.
link |
And they range from interlevel digital marketing
link |
to very advanced self diving car engineering.
link |
And we're doing this with the White House
link |
because we think it's bipartisan.
link |
It's an issue that if you wanna really make America great,
link |
being able to be a part of the solution
link |
and live the American dream requires us to be proactive
link |
about our education and our skillset.
link |
It's just the way it is today.
link |
And it's always been this way.
link |
And we always had this American dream
link |
to send our kids to college.
link |
And now the American dream has to be
link |
to send ourselves to college.
link |
We can do this very, very, very efficiently
link |
and very, very, we can squeeze in in the evenings
link |
and things to online.
link |
So our learners go from age 11 to age 80.
link |
I just traveled Germany and the guy in the train compartment
link |
next to me was one of my students.
link |
It's like, wow, that's amazing.
link |
Think about impact.
link |
We've become the educator of choice for now,
link |
I believe officially six countries or five countries.
link |
Most in the Middle East, like Saudi Arabia and in Egypt.
link |
In Egypt, we just had a cohort graduate
link |
where we had 1100 high school students
link |
that went through programming skills,
link |
proficient at the level of a computer science undergrad.
link |
And we had a 95% graduation rate,
link |
even though everything's online, it's kind of tough,
link |
but we kind of trying to figure out
link |
how to make this effective.
link |
The vision is very, very simple.
link |
The vision is education ought to be a basic human right.
link |
It cannot be locked up behind ivory tower walls
link |
only for the rich people, for the parents
link |
who might be bribe themselves into the system.
link |
And only for young people and only for people
link |
from the right demographics and the right geography
link |
and possibly even the right race.
link |
It has to be opened up to everybody.
link |
If we are truthful to the human mission,
link |
if we are truthful to our values,
link |
we're gonna open up education to everybody in the world.
link |
So Udacity's pledge of 100,000 scholarships,
link |
I think is the biggest pledge of scholarships ever
link |
in terms of numbers.
link |
And we're working, as I said, with the White House
link |
and with very accomplished CEOs like Tim Cook
link |
from Apple and others to really bring education
link |
to everywhere in the world.
link |
Not to ask you to pick the favorite of your children,
link |
but at this point.
link |
Oh, that's Jasper.
link |
I only have one that I know of.
link |
In this particular moment, what nano degree,
link |
what set of courses are you most excited about at Udacity
link |
or is that too impossible to pick?
link |
I've been super excited about something
link |
we haven't launched yet in the building,
link |
which is when we talk to our partner companies,
link |
we have now a very strong footing in the enterprise world.
link |
And also to our students,
link |
we've kind of always focused on these hard skills,
link |
like the programming skills or math skills
link |
or building skills or design skills.
link |
And a very common ask is soft skills.
link |
Like how do you behave in your work?
link |
How do you develop empathy?
link |
How do you work on a team?
link |
What are the very basics of management?
link |
How do you do time management?
link |
How do you advance your career
link |
in the context of a broader community?
link |
And that's something that we haven't done very well
link |
at Udacity and I would say most universities
link |
are doing very poorly as well
link |
because we are so obsessed with individual test scores
link |
and pays a little attention to teamwork in education.
link |
So that's something I see us moving into as a company
link |
because I'm excited about this.
link |
And I think, look, we can teach people tech skills
link |
and they're gonna be great.
link |
But if you teach people empathy,
link |
that's gonna have the same impact.
link |
Maybe harder than self driving cars, but.
link |
I think the rules are really simple.
link |
You just have to, you have to want to engage.
link |
It's, we literally went in school and in K through 12,
link |
we teach kids like get the highest math score.
link |
And if you are a rational human being,
link |
you might evolve from this education say,
link |
having the best math score and the best English scores
link |
make me the best leader.
link |
And it turns out not to be that case.
link |
It's actually really wrong because making the,
link |
first of all, in terms of math scores,
link |
I think it's perfectly fine to hire somebody
link |
with great math skills.
link |
You don't have to do it yourself.
link |
You can hire someone with good empathy for you.
link |
That's much harder,
link |
but you can always hire someone with great math skills.
link |
But we live in an affluent world
link |
where we constantly deal with other people.
link |
And that's a beauty.
link |
It's not a nuisance.
link |
So if we somehow develop that muscle
link |
that we can do that well and empower others
link |
in the workplace, I think we're gonna be super successful.
link |
And I know many fellow robot assistant computer scientists
link |
that I will insist to take this course.
link |
Not to be named here.
link |
Many, many years ago, 1903,
link |
the Wright brothers flew in Kitty Hawk for the first time.
link |
And you've launched a company of the same name, Kitty Hawk,
link |
with the dream of building flying cars, eVTOLs.
link |
So at the big picture,
link |
what are the big challenges of making this thing
link |
that actually have inspired generations of people
link |
about what the future looks like?
link |
What does it take?
link |
What are the biggest challenges?
link |
So flying cars has always been a dream.
link |
Every boy, every girl wants to fly.
link |
And let's go back in our history
link |
of your dreaming of flying.
link |
I think honestly, my single most remembered childhood dream
link |
has been a dream where I was sitting on a pillow
link |
I was like five years old.
link |
I remember like maybe three dreams of my childhood,
link |
but that's the one I remember most vividly.
link |
And then Peter Thiel famously said,
link |
they promised us flying cars
link |
and they gave us 140 characters pointing as Twitter
link |
at the time, limited message size to 140 characters.
link |
So if you're coming back now to really go
link |
for these super impactful stuff like flying cars
link |
and to be precise, they're not really cars.
link |
They don't have wheels.
link |
They're actually much closer to a helicopter
link |
than anything else.
link |
They take off vertically and they fly horizontally,
link |
but they have important differences.
link |
One difference is that they are much quieter.
link |
We just released a vehicle called Project Heaviside
link |
that can fly over you as low as a helicopter
link |
and you basically can't hear.
link |
It's like 38 decibels.
link |
It's like, if you were inside the library,
link |
you might be able to hear it,
link |
but anywhere outdoors, your ambient noise is higher.
link |
Secondly, they're much more affordable.
link |
They're much more affordable than helicopters.
link |
And the reason is helicopters are expensive
link |
There's lots of single point of figures in a helicopter.
link |
There's a bolt between the blades
link |
that's caused Jesus bolt.
link |
And the reason why it's called Jesus bolt
link |
is that if this bolt breaks, you will die.
link |
There is no second solution in helicopter flight.
link |
Whereas we have these distributed mechanism.
link |
When you go from gasoline to electric,
link |
you can now have many, many, many small motors
link |
as opposed to one big motor.
link |
And that means if you lose one of those motors,
link |
Heaviside, if it loses a motor, has eight of those.
link |
If it loses one of those eight motors,
link |
so it's seven left, it can take off just like before
link |
and land just like before.
link |
We are now also moving into a technology
link |
that doesn't require a commercial pilot
link |
because in some level,
link |
flight is actually easier than ground transportation
link |
like in self driving cars.
link |
The world is full of like children and bicycles
link |
and other cars and mailboxes and curbs and shrubs
link |
and what have you.
link |
All these things you have to avoid.
link |
When you go above the buildings and tree lines,
link |
there's nothing there.
link |
I mean, you can do the test right now,
link |
look outside and count the number of things you see flying.
link |
I'd be shocked if you could see more than two things.
link |
It's probably just zero.
link |
In the Bay Area, the most I've ever seen was six.
link |
And maybe it's 15 or 20,
link |
So the sky is very ample and very empty and very free.
link |
So the vision is, can we build a socially acceptable
link |
mass transit solution for daily transportation
link |
that is affordable?
link |
And we have an existence proof.
link |
Heaviside can fly 100 miles in range
link |
with still 30% electric reserves.
link |
It can fly up to like 180 miles an hour.
link |
We know that that solution at scale
link |
would make your ground transportation
link |
10 times as fast as a car
link |
based on use census or statistics data,
link |
which means you would take your 300 hours of daily,
link |
of yearly commute down to 30 hours
link |
and give you 270 hours back.
link |
Who wouldn't want, I mean, who doesn't hate traffic?
link |
Like I hate, give me the person that doesn't hate traffic.
link |
Every time I'm in traffic, I hate it.
link |
And if we could free the world from traffic,
link |
we have technology.
link |
We can free the world from traffic.
link |
We have the technology.
link |
We have an existence proof.
link |
It's not a technological problem anymore.
link |
Do you think there is a future where tens of thousands,
link |
maybe hundreds of thousands of both delivery drones
link |
and flying cars of this kind, EV talls fill the sky?
link |
I absolutely believe this.
link |
And there's obviously the societal acceptance
link |
is a major question.
link |
And of course, safety is.
link |
I believe in safety,
link |
we're gonna exceed ground transportation safety
link |
as has happened for aviation already, commercial aviation.
link |
And in terms of acceptance,
link |
I think one of the key things is noise.
link |
That's why we are focusing relentlessly on noise
link |
and we build perhaps the quietest electric vehicle
link |
The nice thing about the sky is it's three dimensional.
link |
So any mathematician will immediately recognize
link |
the difference between 1D of like a regular highway
link |
But to make it clear for the layman,
link |
say you wanna make 100 vertical lanes
link |
of highway 101 in San Francisco,
link |
because you believe building 100 vertical lanes
link |
is the right solution.
link |
Imagine how much it would cost to stack 100 vertical lanes
link |
physically onto 101.
link |
That would be prohibitive.
link |
That would be consuming the world's GDP for an entire year
link |
just for one highway.
link |
It's amazingly expensive.
link |
In the sky, it would just be a recompilation
link |
of a piece of software because all these lanes are virtual.
link |
That means any vehicle that is in conflict
link |
with another vehicle would just go to different altitudes
link |
and then the conflict is gone.
link |
And if you don't believe this,
link |
that's exactly how commercial aviation works.
link |
When you fly from New York to San Francisco,
link |
another plane flies from San Francisco to New York,
link |
they are different altitudes.
link |
So they don't hit each other.
link |
It's a solved problem for the jet space
link |
and it will be a solved problem for the urban space.
link |
There's companies like Google Wing and Amazon
link |
working on very innovative solutions.
link |
How do we have space management?
link |
They use exactly the same principles as we use today
link |
to route today's jets.
link |
There's nothing hard about this.
link |
Do you envision autonomy being a key part of it
link |
so that the flying vehicles are either semi autonomous
link |
semi autonomous or fully autonomous?
link |
You don't want idiots like me flying in the sky,
link |
And if you have 10,000,
link |
watch the movie, The Fifth Element
link |
to get a feel for what will happen if it's not autonomous.
link |
And a centralized, that's a really interesting idea
link |
of a centralized sort of management system
link |
for lanes and so on.
link |
So actually just being able to have
link |
similar as we have in the current commercial aviation,
link |
but scale it up to much, much more vehicles.
link |
That's a really interesting optimization problem.
link |
It is very mathematically, very, very straightforward.
link |
Like the gap we leave between jets is gargantuous.
link |
And part of the reason is there isn't that many jets.
link |
So it just feels like a good solution.
link |
Today, when you get vectored by air traffic control,
link |
someone talks to you, right?
link |
So any ATC controller might have up to maybe 20 planes
link |
on the same frequency.
link |
And then they talk to you, you have to talk back.
link |
And it feels right because there isn't more than 20 planes
link |
around anyhow, so you can talk to everybody.
link |
But if there's 20,000 things around,
link |
you can't talk to everybody anymore.
link |
So we have to do something that's called digital,
link |
like text messaging.
link |
Like we do have solutions.
link |
Like we have what, four or five billion smartphones
link |
in the world now, right?
link |
And they're all connected.
link |
And somehow we solve the scale problem for smartphones.
link |
We know where they all are.
link |
They can talk to somebody and they're very reliable.
link |
They're amazingly reliable.
link |
We could use the same system,
link |
the same scale for air traffic control.
link |
So instead of me as a pilot talking to a human being
link |
and in the middle of the conversation
link |
receiving a new frequency, like how ancient is that?
link |
We could digitize this stuff
link |
and digitally transmit the right flight coordinates.
link |
And that solution will automatically scale
link |
to 10,000 vehicles.
link |
We talked about empathy a little bit.
link |
Do you think we will one day build an AI system
link |
that a human being can love
link |
and that loves that human back, like in the movie, Her?
link |
Look, I'm a pragmatist.
link |
For me, AI is a tool.
link |
It's like a shovel.
link |
And the ethics of using the shovel are always
link |
with us, the people.
link |
And it has to be this way.
link |
In terms of emotions,
link |
I would hate to come into my kitchen
link |
and see that my refrigerator spoiled all my food,
link |
then have it explained to me
link |
that it fell in love with the dishwasher
link |
and it wasn't as nice as the dishwasher.
link |
So as a result, it neglected me.
link |
That would just be a bad experience
link |
and it would be a bad product.
link |
I would probably not recommend this refrigerator
link |
And that's where I draw the line.
link |
I think to me, technology has to be reliable
link |
and has to be predictable.
link |
I want my car to work.
link |
I don't want to fall in love with my car.
link |
I just want it to work.
link |
I want it to compliment me, not to replace me.
link |
I have very unique human properties
link |
and I want the machines to make me,
link |
turn me into a superhuman.
link |
Like I'm already a superhuman today,
link |
thanks to the machines that surround me.
link |
And I give you examples.
link |
I can run across the Atlantic
link |
at near the speed of sound at 36,000 feet today.
link |
That's kind of amazing.
link |
I can, my voice now carries me all the way to Australia
link |
using a smartphone today.
link |
And it's not the speed of sound, which would take hours.
link |
It's the speed of light.
link |
My voice travels at the speed of light.
link |
That makes me superhuman.
link |
I would even argue my flushing toilet makes me superhuman.
link |
Just think of the time before flushing toilets.
link |
And maybe you have a very old person in your family
link |
that you can ask about this
link |
or take a trip to rural India to experience it.
link |
It makes me superhuman.
link |
So to me, what technology does, it compliments me.
link |
It makes me stronger.
link |
Therefore, words like love and compassion
link |
have very little interest in this for machines.
link |
I have interest in people.
link |
You don't think, first of all, beautifully put,
link |
beautifully argued,
link |
but do you think love has use in our tools?
link |
I think love is a beautiful human concept.
link |
And if you think of what love really is,
link |
love is a means to convey safety, to convey trust.
link |
I think trust has a huge need in technology as well,
link |
We want to trust our technology the same way,
link |
in a similar way we trust people.
link |
In human interaction, standards have emerged
link |
and feelings, emotions have emerged,
link |
maybe genetically, maybe biologically,
link |
that are able to convey sense of trust, sense of safety,
link |
sense of passion, of love, of dedication
link |
that makes the human fabric.
link |
And I'm a big slacker for love.
link |
I want to be loved.
link |
I want to be trusted.
link |
I want to be admired.
link |
All these wonderful things.
link |
And because all of us, we have this beautiful system,
link |
I wouldn't just blindly copy this to the machines.
link |
When you look at, say, transportation,
link |
you could have observed that up to the end
link |
of the 19th century, almost all transportation used
link |
any number of legs, from one leg to two legs
link |
to a thousand legs.
link |
And you could have concluded that is the right way
link |
to move about the environment.
link |
We've been made the exception of birds
link |
who use flapping wings.
link |
In fact, there are many people in aviation
link |
that flap wings to their arms and jump from cliffs.
link |
Most of them didn't survive.
link |
Then the interesting thing is that the technology solutions
link |
are very different.
link |
Like in technology, it's really easy to build a wheel.
link |
In biology, it's super hard to build a wheel.
link |
There's very few perpetually rotating things in biology
link |
and they usually run cells and things.
link |
In engineering, we can build wheels.
link |
And those wheels gave rise to cars.
link |
Similar wheels gave rise to aviation.
link |
Like there's no thing that flies
link |
that wouldn't have something that rotates,
link |
like a jet engine or helicopter blades.
link |
So the solutions have used very different physical laws
link |
than nature, and that's great.
link |
So for me to be too much focused on,
link |
oh, this is how nature does it, let's just replicate it.
link |
If you really believed that the solution
link |
to the agricultural evolution was a humanoid robot,
link |
you would still be waiting today.
link |
Again, beautifully put.
link |
You said that you don't take yourself too seriously.
link |
You want me to say that?
link |
You're not taking me seriously.
link |
I'm not, that's right.
link |
Good, you're right, I don't wanna.
link |
I just made that up.
link |
But you have a humor and a lightness about life
link |
that I think is beautiful and inspiring to a lot of people.
link |
Where does that come from?
link |
The smile, the humor, the lightness
link |
amidst all the chaos of the hard work that you're in,
link |
where does that come from?
link |
I just love my life.
link |
I love the people around me.
link |
I'm just so glad to be alive.
link |
Like I'm, what, 52, hard to believe.
link |
People say 52 is a new 51, so now I feel better.
link |
But in looking around the world,
link |
looking around the world, just go back 200, 300 years.
link |
Humanity is, what, 300,000 years old?
link |
But for the first 300,000 years minus the last 100,
link |
our life expectancy would have been
link |
plus or minus 30 years roughly, give or take.
link |
So I would be long dead now.
link |
That makes me just enjoy every single day of my life
link |
because I don't deserve this.
link |
Why am I born today when so many of my ancestors
link |
died of horrible deaths, like famines, massive wars
link |
that ravaged Europe for the last 1,000 years
link |
mystically disappeared after World War II
link |
when the Americans and the Allies
link |
did something amazing to my country
link |
that didn't deserve it, the country of Germany.
link |
This is so amazing.
link |
And then when you're alive and feel this every day,
link |
then it's just so amazing what we can accomplish,
link |
We live in a world that is so incredibly,
link |
vastly changing every day.
link |
Almost everything that we cherish from your smartphone
link |
to your flushing toilet, to all these basic inventions,
link |
your new clothes you're wearing, your watch, your plane,
link |
penicillin, I don't know, anesthesia for surgery,
link |
penicillin have been invented in the last 150 years.
link |
So in the last 150 years, something magical happened.
link |
And I would trace it back to Gutenberg
link |
and the printing press that has been able
link |
to disseminate information more efficiently than before
link |
that all of a sudden we were able to invent agriculture
link |
and nitrogen fertilization that made agriculture
link |
so much more potent that we didn't have to work
link |
in the farms anymore and we could start reading and writing
link |
and we could become all these wonderful things
link |
we are today, from airline pilot to massage therapist
link |
to software engineer.
link |
It's just amazing.
link |
Like living in that time is such a blessing.
link |
We should sometimes really think about this, right?
link |
Steven Pinker, who is a very famous author and philosopher
link |
whom I really adore, wrote a great book called
link |
Enlightenment Now.
link |
And that's maybe the one book I would recommend.
link |
And he asks the question,
link |
if there was only a single article written
link |
in the 20th century, it's only one article, what would it be?
link |
What's the most important innovation,
link |
the most important thing that happened?
link |
And he would say this article would credit
link |
a guy named Karl Bosch.
link |
And I challenge anybody, have you ever heard
link |
of the name Karl Foch?
link |
There's a Bosch Corporation in Germany,
link |
but it's not associated with Karl Bosch.
link |
So I looked it up.
link |
Karl Bosch invented nitrogen fertilization.
link |
And in doing so, together with an older invention
link |
of irrigation, was able to increase the yields
link |
per agricultural land by a factor of 26.
link |
So a 2,500% increase in fertility of land.
link |
And that, so Steve Pinker argues,
link |
saved over 2 billion lives today.
link |
2 billion people who would be dead
link |
if this man hadn't done what he had done, okay?
link |
Think about that impact and what that means to society.
link |
That's the way I look at the world.
link |
I mean, it's so amazing to be alive and to be part of this.
link |
And I'm so glad I lived after Karl Bosch and not before.
link |
I don't think there's a better way to end this, Sebastian.
link |
It's an honor to talk to you,
link |
to have had the chance to learn from you.
link |
Thank you so much for talking to me.
link |
Thanks for coming out.
link |
It's been a real pleasure.
link |
Thank you for listening to this conversation
link |
with Sebastian Thrun.
link |
And thank you to our presenting sponsor, Cash App.
link |
Download it, use code LexPodcast,
link |
you'll get $10 and $10 will go to FIRST,
link |
a STEM education nonprofit that inspires
link |
hundreds of thousands of young minds
link |
to learn and to dream of engineering our future.
link |
If you enjoy this podcast, subscribe on YouTube,
link |
get five stars on Apple Podcast, support it on Patreon,
link |
or connect with me on Twitter.
link |
And now, let me leave you with some words of wisdom
link |
from Sebastian Thrun.
link |
It's important to celebrate your failures
link |
as much as your successes.
link |
If you celebrate your failures really well,
link |
if you say, wow, I failed, I tried, I was wrong,
link |
but I learned something, then you realize you have no fear.
link |
And when your fear goes away, you can move the world.
link |
Thank you for listening and hope to see you next time.