back to indexDmitri Dolgov: Waymo and the Future of Self-Driving Cars | Lex Fridman Podcast #147
link |
The following is a conversation with Dimitri Dolgov, the CTO of Waymo, which is an autonomous
link |
driving company that started as Google self driving car project in 2009 and became Waymo in 2016.
link |
Dimitri was there all along. Waymo is currently leading in the fully autonomous vehicle space
link |
in that they actually have an at skill deployment of publicly accessible autonomous vehicles
link |
driving passengers around with no safety driver with nobody in the driver seat. This to me is an
link |
incredible accomplishment of engineering on one of the most difficult and exciting artificial
link |
intelligence challenges of the 21st century. Quick mention of a sponsor followed by some
link |
thoughts related to the episode. Thank you to Trial Labs, a company that helps businesses
link |
apply machine learning to solve real world problems. Blinkist, an app I use for reading
link |
through summaries of books, BetterHelp, online therapy with a licensed professional,
link |
and cash app, the app I use to send money to friends. Please check out the sponsors in the
link |
description to get a discount at the support this podcast. As a side note, let me say that
link |
autonomous and semi autonomous driving was the focus of my work at MIT and as a problem space
link |
that I find fascinating and full of open questions from both a robotics and a human psychology
link |
perspective. There's quite a bit that I could say here about my experiences in academia on this
link |
topic that revealed to me, let's say the less admirable sides of human beings. But I choose
link |
to focus on the positive, on solutions. I'm brilliant engineers like Dimitri and the team
link |
at Waymo, who work tirelessly to innovate and to build amazing technology that will define our
link |
future. Because of Dimitri and others like him, I'm excited for this future. And who knows,
link |
perhaps I too will help contribute something of value to it. If you enjoy this thing,
link |
subscribe on YouTube, review it with five stars and up a podcast, follow on Spotify,
link |
support on Patreon, or connect with me on Twitter, Alex Friedman. And now here's my
link |
conversation with Dimitri Dolgov. When did you first fall in love with robotics or even computer
link |
science more in general? Computer science first at a fairly young age. Then robotics happened
link |
much later. I think my first interesting introduction to computers was in the late 80s,
link |
when we got our first computer, I think it was an IBM, I think IBM AT. Remember those things
link |
that had like a turbo button in the front? You press it and make the thing goes faster.
link |
Did they already have floppy disks? Yeah, yeah, yeah. Like the 5.4 inch ones.
link |
I think there was a bigger inch. So when something, then five inches and three inches.
link |
Yeah, I think that was the five. I don't know, maybe that was before that was the giant plates,
link |
then it didn't get that. But it was definitely not the three inch ones.
link |
Anyway, so we got that computer. I spent the first few months just playing video games,
link |
as you would expect. I got bored of that. So I started messing around and trying to figure
link |
out how to make the thing do other stuff. Got into exploring programming. And a couple of years
link |
later, I got to a point where I actually wrote a game, a lot of games, and a game developer,
link |
a Japanese game developer actually offered to buy it for me for a few hundred bucks,
link |
but for a kid in Russia. That's a big deal. That's a big deal. Yeah. I did not take the deal.
link |
Wow, integrity. Yeah, I instead used the bid. Yes, that was not the most acute financial move
link |
that I made in my life. Looking back at it now, I instead put it, well, I had a reason. I put it
link |
online. It was, what did you call it back in the days? It was a freeware thing. It was not open
link |
source, but you could upload the binaries, you would put the game online. And the idea was that
link |
people like it and then they contribute and they send you a little donations. So I did my quick
link |
math of like, of course, thousands and millions of people are going to play my game, send me a
link |
couple of bucks a piece. You should definitely do that. As I said, not the best find. You're
link |
already playing business models at that age. Remember what language it was? What programming
link |
it was? Pascal. Pascal. Pascal. Pascal. And they had a graphical component. So it's not text based.
link |
Yeah, it was like, I think there are 320 by 200, whatever it was, I think kind of the earlier
link |
version. That's the resolution. That's the resolution, right? And I actually think the reason
link |
why this company wanted to buy it is not like the fancy graphics or the implementation. It was
link |
maybe the idea of my actual game. The idea of the game. Well, one of the things, it's so funny,
link |
I used to play this game called Golden Axe. And the simplicity of the graphics and something
link |
about the simplicity of the music, like it still haunts me. I don't know if that's a childhood
link |
thing. I don't know if that's the same thing for Call of Duty these days for young kids. But
link |
I still think that the simple one of the games are simple, that simple purity makes for like
link |
allows your imagination to take over and thereby creating a more magical experience. Like now
link |
with better and better graphics, it feels like your imagination doesn't get to create worlds,
link |
which is kind of interesting. It could be just an old man on a porch waving at kids these days
link |
that have no respect. But I still think that graphics almost get in the way of the experience.
link |
I don't know. Flippy Bird. Yeah. Why don't I don't know if the game is closed.
link |
I don't yet. But that's more about games that like that's more like Tetris World where they
link |
optimally, masterfully like create a fun short term dopamine experience versus I'm more referring to
link |
like role playing games where there's like a story you can live in it for months or years.
link |
Like there's an Elder Scrolls series, which is probably my favorite set of games.
link |
That was a magical experience. And then the graphics are terrible. The characters were
link |
all randomly generated. But they're, I don't know, it pulls you in. There's a story. It's like
link |
an interactive version of an Elder Scrolls Tolkien world and you get to live in it. I don't know. I
link |
miss it. It's one of the things that suck about being an adult is there's no, you have to live
link |
in the real world as opposed to the Elder Scrolls world. You know, whatever brings you joy, right?
link |
Minecraft, right? Minecraft is a great example. You create like it's not the fancy graphics,
link |
but it's the creation of your own worlds. Yeah, that one is crazy. You know, one of the pitches
link |
for being a parent that people tell me is that you can like use the excuse of parenting to
link |
go back into the video game world. And like, like that's like, you know, father, son, father,
link |
father, daughter time, but really you just get to play video games with your kids. So anyway,
link |
at that time, did you have any ridiculous, ambitious dreams of where as a creator you might go
link |
as an engineer? What did you think of yourself as an engineer, as a tinker, or did you want to be
link |
like an astronaut or something like that? You know, I'm tempted to make something up about,
link |
you know, robots, engineering or, you know, mysteries of the universe. But that's not the
link |
actual memory that pops into my mind when you when you asked me about childhood dreams. So I
link |
actually share the real thing. When I was maybe four or five years old, I, as we all do, I thought
link |
about no one I wanted to do when I grow up. And I had this dream of being a traffic control cup.
link |
You know, they don't have those todays, I think, but you know, back in the 80s and you know, in
link |
Russia, you probably are familiar with athletics, they had these, you know, police officers that
link |
would stand in the middle of an intersection all day, and they would have their like striped back
link |
back and wide batons that they would use to, you know, control the flow of traffic. And you know,
link |
for whatever reason, I was strangely infatuated with this whole process. And like that, that was
link |
my dream. That's what I wanted to do when I grew up. And you know, my parents, both physics profs,
link |
by the way, I think we're, you know, a little concerned with that level of ambition coming
link |
from their child at that age. Well, it's an interesting, I don't know if you can relate,
link |
but I very much love that idea. I have an OCD nature that I think lends itself
link |
very close to the engineering mindset, which is you want to kind of optimize,
link |
you know, solve a problem by creating an automated solution, like a set of rules,
link |
that set of rules you can follow, and then thereby make it ultra efficient. I don't know if that's,
link |
it was of that nature. I certainly have that. There's like factor like SimCity and factory
link |
building games, all those kinds of things kind of speak to that engineering mindset. Or did you
link |
just like the uniform? I think it was more of the latter. I think it was the uniform and, you know,
link |
the, the striped baton that made cars go in the right directions. But I guess, you know, it is,
link |
I did end up, I guess, you know, working on the transportation industry one way or another.
link |
No uniform, no, but that's right. Maybe it was my, you know, deep inner infatuation with the,
link |
you know, traffic control batons that led to this career. Okay, when did you,
link |
when was the leap from programming to robotics? That happened later. That was after grad school.
link |
After, and actually the, you know, cell driving cars was, I think, my first real hands on introduction
link |
to robotics. But I never really had that much hands on experience in school and training. I,
link |
you know, worked on applied math and physics. Then in college, I did more abstract computer science.
link |
And it was after grad school that I really got involved in robotics, which was actually cell
link |
driving cars. And, you know, that was a big, big flip. What, what grad school? So I went to
link |
grad school in Michigan, and then I did a postdoc at Stanford, which is that was the postdoc where
link |
I got to play with cell driving cars. Yeah, so we'll return there. Let's go back to, to Moscow.
link |
So I, you know, for episode 100, I talked to my dad. And also I grew up with my dad, I guess.
link |
So I had to put up with them for many years. And he, he went to the Fistyev or MIPT.
link |
It's weird to say in English, because I've heard all this in Russian, Moscow Institute of Physics
link |
and Technology. And to me, that was like, I met some super interesting, as a child, I met some
link |
super interesting characters. It felt to me like the greatest university in the world, the most
link |
elite university in the world. And just the people that I met that came out of there were like,
link |
not only brilliant, but also special humans. It seems like that place really tested the soul,
link |
both like in terms of technically and like spiritually. So that could be just the romanticization
link |
of that place. I'm not sure. But so maybe you can speak to it. But is it correct to say that you
link |
spent some time at Fistyev? Yeah, that's right. Six years. I got my bachelor's and master's and
link |
physics and math there. And it was interesting because my dad, actually both my parents,
link |
went there. And I think all the stories that I heard, like, just like you, Alex, growing up
link |
about the place and, you know, how interesting and special and magical it was, I think that was
link |
a significant, maybe the main reason I wanted to go there for college, enough so that I actually
link |
went back to Russia from the U.S. I graduated high school in the U.S. And you went back there.
link |
I went back there. Yeah. Wow. Exactly the reaction most of my peers in college had,
link |
but, you know, perhaps a little bit stronger that would point me out as this crazy kid.
link |
Were your parents supportive of that? Yeah. Yeah. My games, your previous question, they
link |
supported me in letting me pursue my passions and the things that I was there.
link |
That's a bold move. Wow. What was it like there?
link |
It was interesting. You know, definitely fairly hardcore on the fundamentals of, you know,
link |
math and physics and, you know, lots of good memories from, you know, from those times.
link |
So, okay. So, Stanford, how'd you get into autonomous vehicles?
link |
I had the great fortune and great honor to join Stanford's DARPA urban challenge team
link |
in 2006. This was a third in the sequence of the DARPA challenges. There were two
link |
grand challenges prior to that. And then in 2007, they held the DARPA urban challenge.
link |
So, you know, I was doing my postdoc I had. I joined the team and worked on motion planning for,
link |
you know, that competition. So, okay. So, for people who might not know,
link |
I know from a certain perspective, autonomous vehicles is a funny world.
link |
In a certain circle of people, everybody knows everything. And in a certain circle,
link |
nobody knows anything in terms of general public. So, it's interesting. It's a good question of
link |
what to talk about. But I do think that the urban challenge is worth revisiting. It's a fun little
link |
challenge. One that, first of all, like sparked so many incredible minds to focus on one of the
link |
hardest problems of our time in artificial intelligence. So, that's a success from a
link |
perspective of a single little challenge. But can you talk about, like, what did the challenge
link |
involve? So, were there pedestrians? Were there other cars? What was the goal? Who was on the
link |
team? How long did it take? Any fun sort of specs? Sure, sure, sure. So, the way that the challenge
link |
was constructed in just a little bit of background, as I mentioned, this was the third
link |
competition in that series. The first two were the grand challenge called the grand challenge.
link |
The goal there was to just drive in a completely static environment. You had to drive in a desert.
link |
That was very successful. So, then DARPA followed with what they called the urban challenge, where
link |
the goal was to build vehicles that could operate in more dynamic environments and share them with
link |
other vehicles. There were no pedestrians there. But what DARPA did is they took over an abandoned
link |
air force base. And it was kind of like a little fake city that they built out there. And they had
link |
a bunch of robots, you know, cars that were autonomous in there all at the same time,
link |
mixed in with other vehicles driven by professional drivers. And each car had a mission. And so,
link |
there's a crude map that they received at the beginning. And they had a mission and go, you
link |
know, here and then there and over here. And they kind of all were sharing this environment at the
link |
same time. They had to interact with each other. They had to interact with the human drivers.
link |
So, it's this very first, very rudimentary version of a self driving car that, you know,
link |
could operate in an environment shared with other dynamic actors. That, as you said,
link |
you know, really, in many ways, you know, kickstarted this whole industry.
link |
Okay. So, who was on the team? And how'd you do? I forget.
link |
I came in second. Perhaps that was my contribution to the team. I think the staffer team came in
link |
first in the DARPA challenge. But then I joined the team and, you know, you were the one with the
link |
bug in the code. I mean, do you have sort of memories of some particularly challenging things?
link |
Or, you know, one of the cool things is not, you know, this isn't a product. This isn't the thing
link |
that, you know, there's, you have a little bit more freedom to experiment. So, you can take risks.
link |
And there's, so you can make mistakes. So, is there interesting mistakes? Is there interesting
link |
challenges that stand out to you as something that taught you a good technical lesson or a
link |
good philosophical lesson from that time? Yeah. You know, definitely, definitely a very memorable
link |
time. Not really a challenge, but like one of the most vivid memories that I have from the time. And
link |
I think that was actually one of the days that, you know, really got me hooked on this whole field
link |
was the first time I got to run my software on the car. And I was working on a part of our planning
link |
algorithm that had to navigate in parking lots. So it was, you know, something that, you know,
link |
called free space motion planning. So the very first version of that was, you know, we tried on
link |
the car. It was on Stanford's campus in the middle of the night. And, you know, I had this little,
link |
you know, course constructed with cones in the middle of a parking lot. So we're there like
link |
3am, you know, by the time we got the code to, you know, you know, compile and turn over. And,
link |
you know, it drove. I could actually did something quite reasonable. And, you know, it was, of course,
link |
very buggy at the time and had all kinds of problems. But it was pretty darn magical. I remember
link |
going back and, you know, later at night and trying to fall asleep and just being unable to
link |
fall asleep for the rest of the night. Just my mind was blown. And that's what I've been doing
link |
ever since for more than a decade. In terms of challenges and, you know, interesting memories
link |
like on the day of the competition, it was pretty nerve wrecking. I remember standing there with
link |
Mike Montemarillo, who was the software lead and wrote most of the code. I think I did one little
link |
part of the planner, Mike, you know, incredibly did pretty much the rest of it with a bunch of
link |
other incredible people. But I remember standing on the day of the competition, you know, watching
link |
the car, you know, with Mike and your cars are completely empty, right? They're all there,
link |
lined up in the beginning of the race. And then, you know, DARPA sends them, you know, on their
link |
mission, one by one, something leave. And Mike, you just, they had these sirens. They all had
link |
their different silence, right? Each siren had its own personality, if you will. So, you know,
link |
off they go and you don't see them. You just kind of, and then every once in a while, they, you know,
link |
come a little bit closer to where the audience is. And you can kind of hear, you know, the sound of
link |
your car. And, you know, it seems to be moving along. So that gives you hope. And then, you know,
link |
it goes away and you can't hear it for too long. You start getting anxious, right? So it's a little
link |
bit like, you know, sending your kids to college and like, you know, kind of you invested in them.
link |
You hope you, you build it properly, but like, it's still anxiety inducing. So that was an
link |
incredibly fun few days. In terms of, you know, bugs, as we mentioned, you know, one that was my bug
link |
that caused us the loss of the first place is still a debate that, you know, occasionally have
link |
with people on the CMU team, CMU came first, I should mention that. See, you haven't heard of
link |
them. But yeah, it's something, you know, it's a small school, it's, it's, it's, you know,
link |
it's really a glitch that, you know, they happen to succeed at something robotics related.
link |
Very scenic though. So most people go there for the scenery. Yeah, it's a beautiful campus.
link |
Unlike Stanford. So for people, yeah, that's true. Unlike Stanford. For people who don't
link |
know, CMU is one of the great robotics and sort of artificial intelligence universities in the
link |
world. CMU Carnegie Mellon University. Okay, sorry, go ahead. Good, good PSA. So in the part that
link |
I contributed to, which was navigating parking lots, and the way, you know, that part of the
link |
mission work is you in a parking lot, you would get from DARPA an outline of the map. You basically
link |
get this giant polygon that defined the perimeter of the parking lot. And there would be an entrance
link |
and maybe multiple entrances or exits to it. And then you would get a goal within that open space,
link |
XY heading, where the car had to park. It had no information about the obstacles that the car
link |
might encounter there. So it had to navigate kind of completely free space from the entrance to the
link |
parking lot into that parking space. And then once you're parked there, it had to exit the
link |
parking lot. And you know, while of course, encountering and reasoning about all the obstacles
link |
that it encounters in real time. So our interpretation, or at least my interpretation
link |
of the rules was that you had to reverse out of the parking spot. And that's what our cars did,
link |
even if there's no obstacle in front. That's not what CMU's car did. And it just kind of
link |
drove right through. So there's still a debate. And of course, you know, if you stop and then
link |
reverse out and go out the different way, that costs you some time. And so there's still a debate
link |
whether, you know, it was my poor implementation that cost us extra time, or whether it was,
link |
you know, CMU violating an important rule of the competition. And, you know, I have my own
link |
opinion here. In terms of other bugs, and like, I have to apologize to Mike Montemarillo for
link |
sharing this on air. But it is actually one of the more memorable ones. And it's something that's
link |
kind of become a bit of a metaphor in the industry since then, I think, at least in some
link |
circles, it's called the victory circle or victory lap. And our cars did that. So in one of the
link |
missions in the urban challenge, in one of the courses, there was this big oval right by the
link |
start and finish of the race. So the ARPA head, a lot of the missions would finish in that same
link |
location. And it was pretty cool because you could see the cars come by, you know, kind of finish
link |
that part leg of the trip, that leg of the mission, and then go on and finish the rest of it. And
link |
other vehicles would, you know, come hit their waypoint and exit the oval and off they would go.
link |
Our car in the hand would hit the checkpoint, and then it would do an extra lap around the
link |
oval and only then, you know, leave and go in its merry way. So over the course of the full day,
link |
it accumulated some extra time. And the problem was that we had a bug where it wouldn't, you know,
link |
start reasoning about the next waypoint and plan a route to get to that next point until it
link |
hit a previous one. And in that particular case, by the time you hit that one, it was too late
link |
for us to consider the next one and kind of make a lane change so that every time we would do like
link |
an extra lap. So, you know, that's the Stanford victory lap. The victory lap. Oh, that's, there's,
link |
I feel like there's something philosophically profound in there somehow. But I mean, ultimately,
link |
everybody is a winner in that kind of competition. And it led to sort of famously
link |
to the creation of Google self driving car project and now Waymo. So can we give an overview of
link |
how is Waymo born? How is the Google self driving car project born? What is the mission? What is
link |
the hope? What is it is the engineering kind of set of milestones that it seeks to accomplish?
link |
There's a lot of questions in there. Yeah. I don't know.
link |
But you're right. Kind of the DARPA urban challenge and the previous DARPA grand challenges
link |
kind of led, I think, to a very large, you know, degree to that next step. And then, you know,
link |
Larry and Sergey, Larry Page and Sergey Brin, Google hunter scores, I saw that competition
link |
and believed in the technology that the Google self driving car project was born, you know, at
link |
that time, and we started in 2009, it was a pretty small group of us, about a dozen people
link |
who came together to work on this project at Google. At that time, we saw an incredible early
link |
result in the DARPA urban challenge. I think we're all incredibly excited about
link |
where we got to. And we believed in the future of the technology, but we still had a very
link |
rudimentary understanding of the problem space. So the first goal of this project in 2009 was to
link |
really better understand what we're up against. And, you know, with that goal in mind, when we
link |
started the project, we created a few milestones for ourselves that maximized learning. Well,
link |
the two milestones were, you know, one was to drive 100,000 miles in autonomous mode, which was,
link |
at that time, orders of magnitude that more than anybody has ever done. And the second milestone
link |
was to drive 10 routes. Each one was 100 miles long. They were specifically chosen to be kind of
link |
extra spicy, extra complicated and sampled the full complexity of that domain. And you had to
link |
drive each one from beginning to end with no intervention, no human intervention. So you'd
link |
get to the beginning of the course, you would press the button that would engage in autonomy,
link |
and you had to go for 100 miles beginning to end with no interventions. And it sampled,
link |
again, the full complexity of driving conditions. Some were on freeways. We had one route that
link |
went all through all the freeways and all the bridges in the Bay Area. You know, we had some
link |
that went around Lake Tahoe and kind of mountains roads. We had some that drove through dense urban
link |
environments like downtown Palo Alto and through San Francisco. So it was incredibly
link |
interesting to work on. And it took us just under two years, about a year and a half, a little bit
link |
more to finish both of these milestones. And in that process, it was an incredible amount of fun,
link |
probably the most fun I had in my professional career. And you're just learning so much. You are,
link |
you know, the goal here is to learn a prototype. You're not yet starting to build a production
link |
system, right? So you just, you were, you know, this is when you're kind of, you know, working 24
link |
seven and you're hacking things together. And you also don't know how hard this is. I mean,
link |
it's the point, like, so, I mean, that's an ambitious, if I put myself in that mindset, even
link |
still, that's a really ambitious set of goals, like just those two to picking, picking 10 different,
link |
different, difficult, spicy challenges, and then having zero interventions. So like not saying,
link |
gradually, we're going to like, you know, over a period of 10 years, we're going to have a bunch
link |
of routes and gradually reduce the number of interventions, you know, that literally says,
link |
like, by as soon as possible, we want to have zero and on hard roads. So like to me, if I was
link |
facing that, it's unclear that whether that takes two years or whether that takes 20 years.
link |
I mean, under two, I guess that speaks to a really big difference between doing something
link |
once and having a prototype where you're going after, you know, learning about the problem versus
link |
how you go about engineering a product that where you look at, you know, you properly do
link |
evaluation, you look at metrics, you know, drive dog, and you're confident that you can do that.
link |
And I guess that's why it took a dozen people, you know, 16 months or a little bit more than that,
link |
back in 2009 and 2010, with the technology of, you know, the more than a decade ago,
link |
that amount of time to achieve that milestone of 10 routes, 100 miles each in no interventions.
link |
And, you know, it took us a little bit longer to get to, you know, a full driverless product
link |
that customers use. That's another really important moment. Is there some
link |
memories of technical lessons? Or just one, like, what did you learn about the problem
link |
of driving from that experience? I mean, we can now talk about, like, what you learned from
link |
modern day Waymo. But I feel like you may have learned some profound things in those early days,
link |
even more so, because it feels like what Waymo is now is to trying to, you know, how to do scale,
link |
how to make sure you create a product, how to make sure it's like safe, you know, those things,
link |
which is all fascinating challenges. But, like, you were facing the more fundamental
link |
philosophical problem of driving in those early days, like, what the hell is driving?
link |
As an autonomous, or maybe I'm again romanticizing it, but is there a, is there some
link |
valuable lessons you picked up over there at those two years?
link |
A ton. The most important one is probably that we believe that it's doable. And we've gotten
link |
far enough into the problem that, you know, we had a, I think, only glimpse of the true complexity
link |
of that domain. And it's a little bit like, you know, climbing a mountain where you kind of,
link |
you know, see the next peak and you think that's kind of the summit, but then you get to that and
link |
you kind of see that this is just the start of the journey. But we've tried, we've sampled enough
link |
of the problem space, and we've made enough rapid success, even, you know, with technology
link |
of 2009, 2010, that it gave us confidence to then pursue this as a real product.
link |
So, okay, so the next step, you mentioned the milestones that you had in those two years.
link |
What are the next milestones that then led to the creation of Waymo and beyond?
link |
Yeah, it was a really interesting journey. And, you know, Waymo came a little bit later.
link |
Then, you know, we completed those milestones in 2010. That was the pivot when we decided to
link |
focus on actually building a product using this technology. The initial couple years after that,
link |
we were focused on a freeway, you know, what you would call a driver assist, maybe, you know,
link |
an L3 driver assist program. Then around 2013, we've learned enough about the space and if
link |
thought more deeply about, you know, the product that we wanted to build that we pivoted. We pivoted
link |
towards this vision of, you know, building a driver and deploying it fully driverless vehicles
link |
without a person. And that's the path that we've been on since then. And it was exactly the right
link |
decision for us. So there was a moment where you're also considered like, what is the right
link |
trajectory here? What is the right role of automation in the task of driving? It wasn't
link |
from the early days, obviously, you want to go fully autonomous. From the early days,
link |
it was not. I think it was around 2013, maybe, that we've, that became very clear and we made
link |
that pivot and it also became very clear. And that it's, even the way you go building a driver
link |
assist system is fundamentally different from how you go building a fully driverless vehicle. So,
link |
you know, we've pivoted towards the ladder and that's what we've been working on ever since.
link |
And so that was around 2013. Then there's a sequence of really meaningful for us,
link |
really important defining milestones since then. In 2015, we had our first, actually,
link |
the world's first fully driverless ride on public roads. It was in a custom build vehicle that we
link |
had. We must have seen those. We called them the Firefly, that, you know, funny looking marshmallow
link |
looking thing. And we put a passenger, his name was Steve Mann, a great friend of our project
link |
from the early days. The man happens to be blind. So we put him in that vehicle. The car had no
link |
steering wheel, no pedals. It was an uncontrolled environment. You know, no, you know, lead or
link |
chase cars, no police escorts. And, you know, we did that trip a few times in Austin, Texas.
link |
So that was a really big milestone. But that was in Austin. Yeah. Okay. And, you know, we only,
link |
but at that time, we're only, it took a tremendous amount of engineering. It took a tremendous amount
link |
of validation to get to that point. But, you know, we only did it a few times. We only did that. It
link |
was a fixed route. It was not kind of a controlled environment, but it was a fixed route. And we
link |
only did a few times. Then in 2016, end of 2016, beginning of 2017 is when we founded Waymo,
link |
the company. That's when we kind of, that was the next phase of the project where I wanted,
link |
we believed in kind of the commercial vision of this technology. And it made sense to create an
link |
independent empty, you know, within that alphabet umbrella to pursue this product at scale.
link |
Beyond that, in 2017, later in 2017, was another really huge step for us, really big milestone
link |
where we started, it was October of 2017, where when we started regular driverless operations
link |
on public roads, that first day of operations, we drove in one day, in that first day, 100 miles
link |
in driverless fashion. And then the most important thing about that milestone was not that 100
link |
miles in one day, but that it was the start of kind of regular, ongoing driverless operations.
link |
And when you say driverless, it means no driver.
link |
That's exactly right. So on that first day, we actually hit a mix. And in some,
link |
we didn't want to like, you know, be on YouTube and Twitter that same day. So in many of the rides,
link |
we had somebody in the driver's seat, but they could not disengage, like the car could not
link |
disengage. But actually, on that first day, some of the miles were driven and just completely
link |
empty driver's seat. And this is the key distinction that I think people don't realize,
link |
it's, you know, that oftentimes when you talk about autonomous vehicles, there's often a driver
link |
in the seat that's ready to take over what's called a safety driver. And then Waymo is really one of
link |
the only companies that I'm aware of, or at least as like boldly and carefully and all that is
link |
actually has cases. And now we'll talk about more and more, where there's literally no driver.
link |
So that's another interesting case of where the driver is not supposed to disengage. That's like
link |
a nice middle ground. They're still there, but they're not supposed to disengage. But really,
link |
there's the case when there's no, okay, there's something magical about there being nobody in
link |
the driver's seat. Like, just like to me, you mentioned the first time you wrote some code
link |
for free space navigation of the parking lot, that was like a magical moment. To me, just sort of
link |
as an observer of robots, the first magical moment is seeing an autonomous vehicle turn,
link |
like make a left turn, like, apply sufficient torque to the steering wheel to where like,
link |
there's a lot of rotation. And for some reason, and there's nobody in the driver's seat, for some
link |
reason that, that communicates that here's a being with power that makes a decision. There's
link |
something about like the steering wheel, because we perhaps romanticize the notion of the steering
link |
wheel, it's so essential to our conception, our 20th century conception of a car. And it turning
link |
the steering wheel with nobody in driver's seat, that to me, I think maybe to others, it's really
link |
powerful, like this thing is in control. And then there's this leap of trust that you give,
link |
like, I'm going to put my life in the hands of this thing that's in control. So in that sense,
link |
when there's no driver in the driver's seat, that's a magical moment for robots. So I got
link |
the chance to last year to take a ride in a Williamsville vehicle. And that was the magical
link |
moment. There's like, nobody in the driver's seat. It's like the little details, you would think it
link |
doesn't matter whether there's a driver or not. But like, if there's no driver, and the steering
link |
wheel is turning on its own, I don't know, that's magical. It's absolutely magical. I've taken many
link |
of these rides in a completely empty car. No human in the car pulls up. You call it on your
link |
cell phone, it pulls up, you get in, it takes you on its way. There's nobody in the car but you,
link |
right? That's something called fully driverless, our rider only mode of operation. Yeah, it is
link |
magical. It is transformative. This is what we hear from our riders. It really changes your
link |
experience. And that really is what unlocks the real potential of this technology. But coming
link |
back to our journey, that was 2017 when we started truly driverless operations. Then in 2018, we've
link |
launched our public commercial service that we called Waymo One in Phoenix. In 2019, we started
link |
offering truly driverless rider only rides to our early rider population of users. And then in
link |
2020, it's also been a pretty interesting year, one of the first months. I less about technology,
link |
but more about the maturing and the growth of Waymo as a company. We raised our first
link |
round of external financing this year. We were part of Alphabet. So obviously, we have access to
link |
significant resources. But on the journey of Waymo maturing as a company, it made sense for us to
link |
partially go externally in this round. So we raised about $3.2 billion from that round.
link |
We've also started putting our fifth generation of our driver, our hardware that is on the new
link |
vehicle. But it's also a qualitatively different set of self driving hardware that is now on the
link |
JLR pace. So that was a very important step for us. The hardware specs, fifth generation,
link |
I think it would be fun to maybe, I apologize if I'm interrupting, but maybe talk about maybe the
link |
generations with the focus on what we're talking about on the fifth generation in terms of hardware
link |
specs, like what's on this car. Sure. So we separated on the actual car that we are driving
link |
from the self driving hardware we put on it. Right now, we have, so this is, as I mentioned,
link |
the fifth generation, we've gone through, we started building our own hardware many,
link |
many years ago. And that Firefly vehicle also had the hardware suite that was mostly
link |
designed and engineered and built in house. Lighters are one of the more important components
link |
that we design and build from the ground up. So on the fifth generation,
link |
after drivers of our software and hardware that we're switching to right now, we have,
link |
as with previous generations, in terms of sensing, we have lighters, cameras, and radars. And we
link |
have a pretty beefy computer that processes all that information and makes decisions and real
link |
time on board the car. So in all of the, and it's really a qualitative jump forward in terms of
link |
the capabilities and the various parameters and specs of the hardware compared to what we had
link |
before and compared to what you can get off the shelf in the market today. Meaning from fifth
link |
to fourth or from fifth to first? Definitely from first to fifth, but also from the fourth.
link |
World's dumbest question. Definitely from fourth to fifth, as well as the last step is a big step
link |
forward. So everything's in house. So like lighters built in house and cameras are built in house?
link |
It's different. We work with partners. There's some components that we get from our manufacturing
link |
and supply chain partners. What exactly is in house is a bit different. We do a lot of custom
link |
design on all of our sensing material. There's lighters, radars, cameras. Exactly. There's
link |
lighters are almost exclusively in house. And some of the technologies that we have, some of the
link |
fundamental technologies there are completely unique to Waymo. That is also largely true about
link |
radars and cameras. It's a little bit more of a mix in terms of what we do ourselves versus what
link |
we get from partners. Is there something super sexy about the computer that you can use?
link |
The computer that you can mention? That's not top secret. Like for people who enjoy computers,
link |
I mean, there's a lot of machine learning involved, but there's a lot of just basic
link |
computer. You have to probably do a lot of signal processing on all the different sensors.
link |
You have to integrate everything. It has to be in real time. There's probably some kind of redundancy
link |
type of situation. Is there something interesting you can say about the computer for the people who
link |
love hardware? It does have all of the characteristics, all the properties that you just mentioned.
link |
Redundancy, very beefy compute for general processing as well as inference and ML models.
link |
It is some of the more sensitive stuff that I don't want to get into for IP reasons, but
link |
we've shared a little bit in terms of the specs of the sensors that we have on the car. We've
link |
actually shared some videos of what our lighters see in the world. We have 20 line cameras. We have
link |
five lighters. We have six radars on these vehicles. You can get a feel for the amount of data that
link |
they're producing that all has to be processed in real time to do perception, to do complex
link |
reasoning. It gives you some idea of how beefy those computers are, but I don't want to get
link |
into specifics of exactly how we build them. Okay. Let me try some more questions that you
link |
can get into the specifics of like GPU wise. Is that something you can get into? I know that
link |
Google works with GPUs and so on. I mean, for machine learning folks, it's kind of interesting.
link |
How do I ask it? I've been talking to people in the government about UFOs and they don't
link |
answer any questions. This is how I feel right now asking about GPUs.
link |
But is there something interesting that you could reveal or would leave it up to our imagination
link |
some of the compute? Is there any fun trickery? Like I talked to Chris Latner for a second time
link |
and he was a key person about TPUs and there's a lot of fun stuff going on in Google in terms of
link |
hardware that optimizes for machine learning. Is there something you can reveal in terms of how
link |
much you mentioned customization, how much customization there is for hardware for machine
link |
learning purposes? I'm going to be like that government, you know, you've got a person who
link |
bought UFOs, but I guess I will say that it's really compute. It's really important. We have
link |
very data hungry and compute hungry ML models all over our stack and this is where, you know,
link |
both being part of Alphabet as well as designing our own sensors and the entire hardware suite
link |
together where on one hand you get access to really rich raw sensor data that you can pipe from your
link |
sensors into your compute platform and build the whole pipe from sensor raw sensor data
link |
to the big compute as then have the massive compute to process all that data. This is where
link |
we're finding that having a lot of control of that hardware part of the stack is really
link |
advantageous. One of the fascinating magical places to me, again, might not be able to speak to
link |
the details, but it is the other compute, which is like, we're just talking about a single car,
link |
but the driving experience is a source of a lot of fascinating data and you have a huge
link |
amount of data coming in on the car and the infrastructure of storing some of that data
link |
to then train or to analyze or so on, that's a fascinating piece of it that I understand
link |
a single car. I don't understand how you pull it all together in a nice way. Is that something
link |
that you could speak to in terms of the challenges of seeing the network of cars and then bringing
link |
the data back and analyzing things that went like edge cases of driving, be able to learn on them,
link |
to improve the system, to see where things went wrong, where things went right and analyze all
link |
that kind of stuff? Is there something interesting there from an engineering perspective?
link |
Oh, there's an incredible amount of really interesting work that's happening there,
link |
both in the real time operation of the fleet of cars and the information that they exchange
link |
with each other in real time to make better decisions, as well as the off board component
link |
where you have to deal with massive amounts of data for training your ML models, evaluating
link |
the ML models for simulating the entire system and for evaluating your entire system. And this is
link |
where being part of Alphabet has, once again, been tremendously advantageous. We consume an
link |
incredible amount of compute for ML infrastructure. We build a lot of custom frameworks to get good
link |
on data mining, finding the interesting edge cases for training and for evaluation of the system,
link |
for both training and evaluating some components and your sub parts of the system in various ML
link |
models, as well as the evaluating the entire system and simulation.
link |
Okay, that first piece that you mentioned that cars communicating to each other, essentially,
link |
I mean, through perhaps through a centralized point. But what, that's fascinating too.
link |
How much does that help you? Like, if you imagine, you know, right now the number of
link |
way more vehicles is whatever x, I don't know if you can talk to what that number, but it's not in
link |
the hundreds of millions yet. And imagine if the whole world is way more vehicles,
link |
like that changes potentially the power of connectivity, like the more cars you have,
link |
I guess actually, if you look at Phoenix, because there's enough vehicles, there's enough,
link |
when there's like some level of density, you can start to probably do some really interesting
link |
stuff with the fact that cars can negotiate, can be, can communicate with each other and thereby
link |
make decisions. Is there something interesting there that you can talk to about like, how does
link |
that help with the driving problem from as compared to just a single car solving the driving problem
link |
by itself? Yeah, it's a spectrum. I first to say that, you know, it's it helps. And it helps in
link |
various ways, but it's not required. Right now, the way we build our system, like each cars can
link |
operate independently, they can operate with no connectivity. So I think it is important that,
link |
you know, you have a fully autonomous, you know, fully capable driver that, you know, computerized
link |
driver that each car has. Then, you know, they do share information. And they share information
link |
in real time, and it really helps. So the way we do this today is, you know, whenever one car
link |
encounters something interesting in the world, whether it might be an accident or a new construction
link |
zone, that information immediately gets, you know, uploaded over the air and is propagated to the rest
link |
of the fleet. So and that's kind of how we think about maps as priors in terms of the knowledge
link |
of our drivers, of our fleet of drivers that is distributed across the fleet. And it's updated
link |
in real time. So that's one use case. You know, you can imagine as the, you know, the density
link |
of these vehicles go up that they can exchange more information in terms of what they're planning
link |
to do and start influencing how they interact with each other, as well as, you know, potentially
link |
sharing some observations, right, to help with, you know, if you have enough density of these
link |
vehicles where, you know, one car might be seeing something that another is relevant to another
link |
car that is very dynamic, you know, it's not part of kind of your updating your static prior
link |
of the map of the world, but it's more of a dynamic information that could be relevant to the
link |
decisions that another car is making real time. So you can see them exchanging that information
link |
and you can build on that. But again, I see that as an advantage, but it's, you know, not a requirement.
link |
So what about the human in the loop? So when I got a chance to drive with a
link |
ride in a Waymo, you know, there's customer service.
link |
So like there is somebody that's able to dynamically like tune in and help you out.
link |
What, what role does the human play in that picture? That's a fascinating like, you know,
link |
the idea of teleoperation, be able to remotely control a vehicle. So here what we're talking about
link |
is like, like frictionless, like a human being able to in a, in a frictionless way, sort of
link |
help you out. I don't know if they're able to actually control the vehicle. Is that something
link |
you could talk to? Yes. Okay. To be clear, we don't do teleoperation. I gotta believe in
link |
teleoperation for various reasons. That's not what we have in our cars. We do, as you mentioned,
link |
have, you know, version of, you know, customer support, you know, we call it live health. In
link |
fact, we find it that it's very important for our rider experience, especially if it's your
link |
first trip, you've never been in a fully driverless, right, or only Waymo vehicle, you get in,
link |
there's nobody there. And so you can imagine having all kinds of, you know, questions in your head,
link |
like how this thing works. So we've put a lot of thought into kind of guiding our,
link |
our riders, our customers through that experience, especially for the first time,
link |
they get some information on the phone. If the fully driverless vehicle is used to service their
link |
trip, when you get into the car, we have an in car, you know, screen and audio that kind of guides
link |
them and explains what to expect. They also have a button that they can push that will connect them
link |
to, you know, a real life human being that they can talk to, right, about this whole process.
link |
So that's one aspect of it. There is, you know, I should mention that there is another function
link |
that humans provide to our cars, but it's not teleoperation. You can think of it a little bit
link |
more like, you know, fleet assistance, kind of like, you know, traffic control that you have,
link |
where our cars, again, they're responsible on their own for making all of the decisions,
link |
all of the driving decisions that don't require connectivity. They, you know, anything that is
link |
safety or latency critical is done, you know, purely autonomously by onboard system. But there
link |
are situations where, you know, if connectivity is available, kind of car encounters a particularly
link |
challenging situation, you can imagine like a super hairy scene of an accident, the cars will
link |
do their best, they will recognize that it's an off nominal situation, they will, you know, do their
link |
best to come up, you know, with the right interpretation, the best course of action in
link |
that scenario. But if the connectivity is available, they can ask for confirmation
link |
from, you know, a human assistant to kind of confirm those actions and, you know,
link |
perhaps provide a little bit of kind of contextual information and guidance.
link |
So October 8th was when you're talking about the, was Waymo launched the fully self,
link |
the public version of its fully driverless, that's the right term, I think,
link |
service in Phoenix. Is that October 8th? That's right. It was the introduction of fully driverless
link |
rider only vehicles into our public Waymo One service. Okay, so that's, that's amazing. So
link |
it's like anybody can get into Waymo in Phoenix? That's right. So we previously had early people
link |
in our early rider program taking fully driverless rides in Phoenix. And just this a little while
link |
ago, we opened on October 8th, we opened that mode of operation to the public. So I can,
link |
you know, download the app and, you know, go on a ride. There is a lot more demand right now
link |
for that service. And then we have capacity. So we're kind of managing that. But that's
link |
exactly the way you describe it. Yeah, well, this is interesting. So there's more demand than you can,
link |
you can handle. Like what, what has been the reception so far? Like what, I mean, okay, so,
link |
you know, that's, this is a product, right? That's a whole nother discussion of like how
link |
compelling of a product it is. Great. But it's also like one of the most kind of transformational
link |
technologies of the 21st century. So it's also like a tourist attraction. Like it's fun to,
link |
you know, to be a part of it. So it'd be interesting to see like, what do people say? What do people,
link |
what, what have been the feedback so far? You know, still early days, but so far the feedback has
link |
been incredible, incredibly positive. They, you know, we asked them for feedback during the ride,
link |
we asked them for feedback after the ride as part of their trip. We asked them some questions,
link |
we asked them to rate the performance of our driver. Most by far, you know, most of our
link |
drivers give us five stars in our app, which is absolutely great to see. And, you know, that's,
link |
and we're, they're also giving us feedback on, you know, things we can improve. And, you know,
link |
that's one of the main reasons we're doing this is Phoenix. And, you know, over the last couple
link |
years, and every day today, we are just learning a tremendous amount of new stuff from our users.
link |
There's, there's no substitute for actually doing the real thing, actually having a fully
link |
driverless product out there in the field with, you know, users that are actually, you know,
link |
paying us money to get from point A to point B. So this is a legitimate, like, there's a paid
link |
service. That's right. And the idea is you use the app to go from point A to point B. And then
link |
what, what are the A's? What are the, what's the freedom of the, of the starting and ending places?
link |
It's an area of geography where that service is enabled. It's a, you know, decent size of
link |
geography of territory. It's actually larger than, you know, than size of San Francisco.
link |
And, you know, within that, you have, you know, full freedom of, you know, selecting where you
link |
want to go. You know, of course, there's some and you, on your app, you get a map, you tell the car
link |
where you want to be picked up, you know, and where you want, you know, the car to pull over
link |
and pick you up. And then you tell it where you want to be dropped off. Right. And of course,
link |
there are some exclusions, right? You want to be, you know, you wear it in terms of where the car
link |
is allowed to pull over, right? So, you know, that you can do, but, you know, besides that,
link |
it's amazing. It's not like a fixed, just would be very, I guess, I don't know, maybe that's
link |
what's the question behind your question, but it's not a, you know, preset set of.
link |
Yes, I guess. So within the geographic constraints with that, within that area,
link |
anywhere out, it can be, you can be picked up and dropped off anywhere.
link |
That's right. And, you know, people use them on like all kinds of trips they,
link |
we have, and we have an incredible spectrum of riders. I think the youngest,
link |
actually have car seats in them. And we have, you know, people taking their kids and rides. I
link |
think the youngest riders we had on the cars are, you know, one or two years old, you know,
link |
and the full spectrum of use cases. People, you can take them to, you know, schools,
link |
to, you know, go grocery shopping, to restaurants, to bars, you know, run errands, you know, go
link |
shopping, et cetera, et cetera, you go to your office, right? Like the full spectrum of use cases.
link |
And people, you know, use them in their daily lives to get around. And we see all kinds of,
link |
you know, really interesting use cases and that, that, that's providing us incredibly valuable
link |
experience that we then, you know, used to improve our product.
link |
So as somebody who's been on, done a few long rants with Joe Rogan and others about the toxicity
link |
of the internet and the comments and the negativity in the comments, I'm fascinated by feedback. I,
link |
I believe that most people are good and kind and intelligent and can provide, like,
link |
even in disagreement, really fascinating ideas. So on a product side, it's fascinating to me,
link |
like, how do you get the richest possible user feedback, like, to improve? What's,
link |
what are the channels that you use to measure? Because, like, you're, you're no longer,
link |
you're, it's one of the magical things about autonomous vehicles. It's not,
link |
like, it's frictionless interaction with the human. So, like, you don't get to,
link |
you know, it's just giving a ride. So, like, how do you get feedback from people to, in order to
link |
improve? Yeah, great question. Various mechanisms. So as part of the normal flow, we ask people for
link |
feedback. They, as the car is driving around, you know, we have on the phone and in the car,
link |
and we have a touch screen in the car, you can actually click some buttons and provide
link |
real time feedback on how the car is doing and how the car is handling a particular situation,
link |
you know, both positive and negative. So that's one channel. We have, as we discussed, customer
link |
support or life help, where, you know, if a customer wants to, has a question, or he has
link |
some sort of concern, they can talk to a person in real time. So that, that is another mechanism
link |
that gives us feedback. At the end of a trip, you know, we also ask them how things went. They
link |
give us comments and, you know, a star rating. And, you know, if it's, we also, you know, ask them to
link |
explain what, you know, what went well and, you know, what could be improved. And we, we have
link |
our writers providing, you know, very rich feedback there. A lot, a large fraction is
link |
very passionate and very excited about this technology. So we get really good feedback.
link |
We also run UXR studies, right? You know, specific and that are kind of more, you know,
link |
go more in depth and we will run both kind of lateral and longitudinal studies,
link |
where we have, you know, deeper engagement with our customers. You know, we have our
link |
user experience research team tracking over time. That's things about longitudinal is cool.
link |
That's, that's exactly right. And, you know, that's another really valuable feedback, source of feedback.
link |
And we're just covering a tremendous amount, right? People go grocery shopping and they like
link |
want to load, you know, 20 bags of groceries in our cars. And like that's one workflow that you
link |
maybe don't, you know, think about, you know, getting just right when you're building the
link |
driverless product. I have people like, you know, who bike as part of their trip. So they,
link |
you know, bike somewhere, then they get in our cars, they take a path or their bike,
link |
they load into our vehicle, then they go, and that's, you know, how they, you know,
link |
where we want to pull over and how that, you know, get in and get out process works,
link |
provides very useful feedback in terms of, you know, what makes a good pickup and drop off location?
link |
We get really valuable feedback. And in fact, we had to do some really interesting work with
link |
high definition maps and thinking about walking directions. And if you imagine you're in a store,
link |
in some giant space, and then, you know, you want to be picked up somewhere, like if you just drop
link |
pin at the current location, which is maybe in the middle of a shopping mall, like what's the best
link |
location for the car to come pick you up. And you can, you know, have simple heuristics where
link |
you just kind of take your, you know, you clean in distance and find the nearest spot where the
link |
car can pull over as closest to you. But oftentimes, that's not the most convenient one,
link |
you know, I have many anecdotes where that heuristic breaks in horrible ways. I want example
link |
that, you know, I often mentioned is somebody wanted to be, you know,
link |
dropped off and Phoenix and, you know, weak car picked location that was close and the closest to
link |
there, you know, where the pin was dropped on the map in terms of, you know, latitude and
link |
longitude. But it happened to be on the other side of a parking lot that had this row of cacti
link |
and the poor person had to like walk all around the parking lot to get to where they wanted to be
link |
in 110 degree heat. So that, you know, that was about. So then, you know, we took all take all of
link |
these, all that feedback from our users and incorporated into our system and improve it.
link |
Yeah, I feel like that's like requires AGI to solve the problem of like, when you're, which is a
link |
very common case when you're in a big space of some kind, like apartment building, it doesn't
link |
matter. It's not some large space. And then you call the like the Waymo from there, right?
link |
Like, whatever that doesn't matter, right? Chair vehicle. And like, where is the pin supposed to
link |
drop? I feel like that's, you don't think I think that requires AGI. I'm going to
link |
okay, the alternative, which I think the Google search engine is taught is like,
link |
there's something really valuable about the perhaps slightly dumb answer, but a really powerful one,
link |
which is like, what was done in the past by others? Like, what was the choice made by others?
link |
That seems to be like in terms of Google search, when you have like billions of searches,
link |
that you could, you could see which, like when they recommend what you might possibly mean,
link |
they suggest based on not some machine learning thing, which they also do, but like on what
link |
was successful for others in the past and finding a thing that they were happy with.
link |
Is that integrated at all? Waymo, like what, what pickups work for others?
link |
It is. I think you're exactly right. So there's a real, it's an interesting problem.
link |
Naive solutions have, you know, interesting failure modes. So there's definitely lots of
link |
things that can be done to improve. And both learning from, you know, what works, what doesn't
link |
work in actual hail from, you know, getting richer data and getting more information about the
link |
environment and, you know, richer maps. But you're absolutely right that there's something,
link |
I think there's some properties of solutions that in terms of the effect that they have on
link |
users so much, much, much, much better than others, right? And predictability and
link |
understandability is important. So you can have maybe something that is not quite as optimal,
link |
but is very natural and predictable to the user and kind of works the same way all the time.
link |
And that matters. That matters a lot for the user experience. And, but, you know, to get to the basics,
link |
the pretty fundamental property is that the car actually arrives where you told it, right? Like,
link |
you can always, you know, change it, see it on the map and you can move it around if you don't
link |
like it. And, but like that property that the car actually shows up on pin is critical, which,
link |
you know, where compared to some of the human driven analogs, I think, you know, you can have
link |
more predictability. It's actually the fact, if I did a little bit of a detail here, I think the
link |
fact that it's, you know, your phone and the cars two computers talking to each other can lead to
link |
some really interesting things we can do. And in terms of the user interfaces, you know, both in
link |
terms of function, like the car actually shows up exactly where you told it, you want it to be.
link |
But also some, you know, really interesting things on the user interface, like as the car is driving,
link |
as you, you know, call it, and it's on the way to come pick you up. And of course, you get the
link |
position of the car and the route on the map. But, and they actually follow that route, of course.
link |
But it can also share some really interesting information about what is doing. So, you know,
link |
our cars, as they are coming to pick you up, if it's come, if a car is coming up to a stop sign,
link |
it will actually show you that, like, it's there sitting, because it's at a stop sign,
link |
or a traffic light will show you that it's sitting at a red light. So, you know, they look
link |
little things, right? But I find those little touch touches really interesting, really magical.
link |
And it's just, you know, little things like that that you can do to kind of delight your users.
link |
You know, this makes me think of there's some products that I just love,
link |
like, there's a, there's a company called Rev, rev.com, where I like for this podcast,
link |
for example, I can drag and drop a video. And then they do all the captioning,
link |
it's humans doing the captioning, but they connect you, they automate everything of connecting you
link |
to the humans, and they do the captioning and transcription, it's all effortless. And it like,
link |
I remember when I first started using them, it was like, life's good. Like, because it was so
link |
painful to figure that out earlier. The same thing with something called Isotope RX, this
link |
company I use for cleaning up audio, like the sound cleanup they do, it's like drag and drop,
link |
and it just cleans everything up very nicely. Another experience like that had with Amazon,
link |
one click purchase, first time, I mean, other places do that now, but just the effortlessness
link |
of purchasing, making it frictionless. It kind of communicates to me, like, I'm a fan of design,
link |
I'm a fan of products, that you can just create a really pleasant experience, the simplicity
link |
of it, the elegance just makes you fall in love with it. So I know, do you think about this kind
link |
of stuff? I mean, it's exactly what we've been talking about. It's like, the little details
link |
that somehow make you fall in love with the product, is that we went from like urban challenge days,
link |
where love was not part of the conversation, probably. And to this point where there's human
link |
beings, and you want them to fall in love with the experience, is that something you're trying to
link |
optimize for, trying to think about like how do you create experience that people love? Absolutely.
link |
That's the vision is removing any friction or complexity from getting our users, our writers,
link |
to where they want to go. Making that as simple as possible. And then beyond that,
link |
just transportation, making things and goods get to their destination as seamlessly as
link |
possible. I talked about a drag and drop experience where I kind of express your intent. And then,
link |
it just magically happens. And for our writers, that's what we're trying to get to is you download
link |
an app, and you click and car shows up. It's the same car. It's very predictable. It's a safe and
link |
high quality experience. And then it gets you in a very reliable, very convenient,
link |
uh, frictionless way to where you want to be. And along the journey,
link |
I think we also want to do little things to delight our users.
link |
Like the ride sharing companies, because they don't control the experience, I think,
link |
they can't make people fall in love necessarily with the experience,
link |
or maybe they haven't put in the effort. But I think if I would just speak to the ride
link |
sharing experience that currently have, it's just very, it's just very convenient.
link |
But there's a lot of room for like falling in love with it. Like we can speak to sort of car
link |
companies. Car companies do this well. You can fall in love with a car, right? And be like a loyal
link |
car person, like whatever. Like I like badass hot rods, I guess 69 Corvette. And at this point,
link |
you know, you can't really, cars are so owning a car. It's so 20th century, man. But is there
link |
something about the way more experience where you hope that people will fall in love with it?
link |
Because that, is that part of it? Or is it part of, is it just about making a convenient ride,
link |
not ride sharing? I don't know what the right term is, but just a convenient A to B autonomous
link |
transport. Or like, do you want them to fall in love with Waymo? Maybe elaborate a little bit.
link |
I mean, almost like from a business perspective, I'm curious, like,
link |
how do you want to be in the background invisible? Or do you want to be
link |
like a source of joy that's in the very much in the foreground?
link |
I want to provide the best, most enjoyable transportation solution. And that means building
link |
it, building our product and building our service in a way that people do, kind of use in a very
link |
seamless, frictionless way in their day to day lives. And I think that does mean,
link |
you know, in some way falling in love in that product, right? It just kind of becomes part
link |
of your routine. It comes down, in my mind, to safety, predictability of the experience,
link |
and privacy, I think, aspects of it, right? Our cars, you get the same car, you get very
link |
predictable behavior. And that is important. And if you're going to use it in your daily life.
link |
Privacy. And when you're in a car, you can do other things. You're spending a bunch of just
link |
another space where you're spending a significant part of your life. So not having to share it
link |
with other people who you don't want to share it with, I think is a very nice property. Maybe
link |
you want to take a phone call or do something else in the vehicle. And safety on the quality
link |
of the driving, as well as the physical safety of not having to share that ride is important
link |
to a lot of people. What about the idea that when there's somebody, like a human driving,
link |
and they do a rolling stop on a stop sign, like sometimes you get an Uber or a Lyft or whatever,
link |
like human driver, and they can be a little bit aggressive as drivers.
link |
It feels like there is not all aggression is bad. Now, that may be a wrong, again, 20th century
link |
conception of driving. Maybe it's possible to create a driving experience. Like, if you're in
link |
the back, busy doing something, maybe aggression is not a good thing. It's a very different kind
link |
of experience, perhaps. But it feels like in order to navigate this world, you need to,
link |
how do I phrase this? You need to kind of bend the rules a little bit, or at least I can test
link |
the rules. I don't know what language politicians use to discuss this, but whatever language they
link |
use, you like flirt with the rules. I don't know. But like you sort of have a bit of an aggressive
link |
way of driving that asserts your presence in this world, thereby making other vehicles and people
link |
respect your presence, and thereby allowing you to navigate through intersections in a timely
link |
fashion. I don't know if any of that made sense, but how does that fit into the experience of
link |
driving autonomously? This is your hitting on a very important point of a number of behavioral
link |
components and parameters that make your driving feel assertive and natural, comfortable,
link |
predictable. Our cars will follow rules. They will do the safest thing possible in all situations,
link |
let me be clear on that. But if you think of really, really good drivers, think about professional
link |
limo drivers. They will follow the rules. They're very, very smooth, and yet they're very efficient,
link |
but they're assertive. They're comfortable for the people in the vehicle. They're predictable for the
link |
other people outside the vehicle that they share the environment with, and that's the kind of driver
link |
that we want to build. Maybe there's a sport analogy there. You can do in many sports the
link |
true professionals are very efficient in their movements. They don't do hectic flailing. They're
link |
smooth and precise, and they get the best results. That's the kind of driver that we want to build.
link |
In terms of aggressiveness, you can roll through the stop signs. You can do crazy lane changes.
link |
Typically, it doesn't get you to your destination faster, typically not the safest or most predictable,
link |
the most comfortable thing to do. But there is a way to do both. That's what we're doing,
link |
we're trying to build the driver that is safe, comfortable, smooth, and predictable.
link |
Yeah, that's a really interesting distinction. I think in the early days of autonomous vehicles,
link |
the vehicles felt cautious as opposed to efficient. But when I rode in the Waymo,
link |
it was quite assertive. It moved pretty quickly. One of the surprising feelings was that it
link |
actually went fast and it didn't feel awkwardly cautious than autonomous vehicles. I've also
link |
programmed autonomous vehicles, and everything I've ever built felt awkwardly either overly
link |
aggressive, especially when it was my code, or awkwardly cautious as the way I would put it.
link |
Waymo's vehicle felt assertive, and I think efficient is the right terminology here.
link |
I also like the professional limo driver. We often think an Uber driver or a bus driver or
link |
taxi. People think taxi drivers are professionals. That's like saying I'm a professional walker
link |
just because I've been walking all my life. I think there's an art to it. If you take it
link |
seriously as an art form, then there's a certain way that mastery looks like. It's interesting
link |
to think about what does mastery look like in driving? Perhaps what we associate with aggressiveness
link |
is unnecessary. It's not part of the experience of driving. It's like unnecessary fluff,
link |
that efficiency. You can create a good driving experience within the rules.
link |
You're the first person to tell me this, so it's kind of interesting. I need to think about this,
link |
but that's exactly what it felt like with Waymo. I kind of had this intuition. Maybe it's the
link |
Russian thing. I don't know that you have to break the rules in life to get anywhere,
link |
but maybe it's possible that that's not the case in driving. I have to think about that,
link |
but it certainly felt that way on the streets of Phoenix when I was there in Waymo,
link |
that that was a very pleasant experience. It wasn't frustrating in that come on,
link |
move already kind of feeling. That wasn't there. That's what we're going after.
link |
I don't think you have to pick one. I think truly good driving gives you both efficiency,
link |
assertiveness, but also comfort and predictability and safety. That's what fundamental
link |
improvements in the core capabilities truly unlock. You can kind of think of it as a precision and
link |
recoil tradeoff. You have certain capabilities of your model, and then it's very easy when you
link |
have some curve of precision and recoil. You can move things around. You can choose your operating
link |
point and you're trading off precision versus recoil, false positives versus false negatives.
link |
You can tune things on that curve and be kind of more cautious or more aggressive,
link |
but then aggressive is bad or cautious is bad. True capabilities come from actually
link |
moving the whole curve up, and then you are kind of on a very different plane of those tradeoffs.
link |
That's what we're trying to do here is to move the whole curve up.
link |
Before I forget, let's talk about trucks a little bit. I also got a chance to check out some of
link |
the Waymo trucks. I'm not sure if we want to go too much into that space, but it's a fascinating one,
link |
so maybe you can mention it at least briefly. Waymo is also now doing autonomous trucking,
link |
and how different philosophically and technically is that whole space of problems?
link |
It's one of our two big products and commercial applications of our driver, right hailing and
link |
deliveries. We have Waymo 1 and Waymo Via, moving people and moving goods. Trucking is
link |
an example of moving goods. We've been working on trucking since 2017. It is a very interesting
link |
space. Another question, how different is it? It has this really nice property that the first order
link |
challenges, like the science, the hard engineering, whether it's hardware or onboard software or
link |
offboard software, all of the systems that you build for training your ML models for evaluating
link |
your entire system. Those fundamentals carry over. The true challenges of driving, perception,
link |
semantic understanding, prediction, decision making, planning, evaluation, the simulator,
link |
ML infrastructure, those carry over. The data and the application and the domains might be
link |
different, but the most difficult problems all of that carries over between the domains. That's
link |
very nice. That's how we approach it. We build investing in the core, the technical core,
link |
and then there's specialization of that core technology to different product lines, to different
link |
commercial applications. Just to tease it apart a little bit, on trucks, starting with the hardware.
link |
The configuration of the sensors is different. They're different physically,
link |
geometrically, different vehicles. For example, we have two of our main lasers on the trucks
link |
on both sides so that we don't have the blind spots, whereas on the JLR iPace, we have one of it
link |
sitting at the very top. The actual sensors are almost the same, or largely the same.
link |
So all of the investment that over the years we've put into building our custom lighters,
link |
custom radars, putting the whole system together, that carries over very nicely.
link |
Then on the perception side, the fundamental challenges of seeing, understanding the world,
link |
whether it's object detection, classification, tracking, semantic understanding, all that
link |
carries over. Yes, there's some specialization when you're driving on freeways, range becomes
link |
more important. The domain is a little bit different. But again, the fundamentals carry over
link |
very, very nicely. Same, you get into prediction or decision making. The fundamentals of what it
link |
takes to predict what other people are going to do, to find the long tail, to improve your system
link |
and that long tail of behavior prediction and response, that carries over and so on and so on.
link |
So, I mean, that's pretty exciting. By the way, does Waymovia include using the smaller vehicles
link |
for transportation of goods? That's an interesting distinction. So I would say
link |
there's three interesting modes of operation. So one is moving humans, one is moving goods,
link |
and one is like moving nothing, zero occupancy, meaning like you're going to
link |
the destination, your empty vehicle. I mean, it's... The third is the last way,
link |
if that's the entirety of it, it's the last exciting from the commercial perspective.
link |
Well, I mean, in terms of like, if you think about what's inside a vehicle as it's moving,
link |
because it does some significant fraction of the vehicle's movement has to be empty.
link |
I mean, it's kind of fascinating. Maybe just on that small point, is there different
link |
control and like policies that are applied for a zero occupancy vehicle? So a vehicle with nothing
link |
in it, or is it just move as if there is a person inside? What was with some subtle differences?
link |
As a first order approximation, there are no differences. And if you think about safety and
link |
comfort and quality of driving, only part of it has to do with the people or the goods inside
link |
of the vehicle. But you don't want to be... You want to drive smoothly as we discussed,
link |
not purely for the benefit of whatever you have inside the car. It's also for the benefit of
link |
the people outside kind of feeding, feeding naturally and predictably into that whole
link |
environment. So yes, there's some second order things you can do. You can change
link |
your route and optimize maybe kind of your fleet things at the fleet scale. And you would take
link |
into account whether some of your cars are actually serving a useful trip, whether with people or
link |
with goods, whereas other cars are driving completely empty to that next valuable trip
link |
that they're going to provide. But those are mostly second order effects.
link |
Okay, cool. So Phoenix is an incredible place. And what you've announced in Phoenix
link |
is kind of amazing. But that's just like one city. How do you take over the world?
link |
I mean, I'm asking for a friend.
link |
One step at a time.
link |
Is that the cartoon pinky in the brain? Yeah. Okay.
link |
But gradually is a true answer. So I think the heart of your question is...
link |
Can you ask a better question than I ask?
link |
You're asking a great question. Answer that one.
link |
I'm just going to phrase it in the terms that I want to answer.
link |
That's perfect. That's exactly right. Brilliant. Please.
link |
Please. Where are we today? And what happens next? And what does it take to go beyond Phoenix?
link |
And what does it take to get this technology to more places and more people around the world?
link |
All right. So our next big area of focus is exactly that larger scale commercialization and just scaling up.
link |
If I think about the main... And Phoenix gives us that platform.
link |
It gives us that foundation of upon which we can build. And it's...
link |
There are a few really challenging aspects of this whole problem that you have to pull together
link |
in order to build the technology, in order to deploy it into the field, to go from a driverless car
link |
to a fleet of cars that are providing a service and then all the way to commercialization.
link |
This is what we have in Phoenix. We've taken the technology from a proof point to an actual
link |
deployment and have taken our driver from one car to a fleet that can provide a service.
link |
Beyond that, if I think about what it will take to scale up and deploy in more places with more
link |
customers, I tend to think about three main dimensions, three main axes of scale. One
link |
is the core technology, the hardware and software, core capabilities of our driver.
link |
The second dimension is evaluation and deployment. And the third one is the product, commercial,
link |
and operational excellence. So you can talk a bit about where we are along each one of those
link |
three dimensions about where we are today and what will happen next. On the core technology,
link |
on the hardware and software, together comprised of driver, we obviously have that foundation
link |
that is providing fully driverless trips to our customers as we speak, in fact.
link |
And we've learned a tremendous amount from that. So now what we're doing is we are incorporating all
link |
those lessons into some pretty fundamental improvements in our core technology, both on
link |
the hardware side and on the software side, to build a more general, more robust solution
link |
that then will enable us to massively scale and be young Phoenix. So on the hardware side,
link |
all of those lessons are now incorporated into this fifth generation hardware platform
link |
that is being deployed right now. And that's the platform, the fourth generation, the thing that
link |
we have right now driving in Phoenix, it's good enough to operate fully driverlessly,
link |
night and day, in various speeds and various conditions. But the fifth generation is the
link |
platform upon which we want to go to massive scale. We've really made qualitative improvements in
link |
terms of the capability of the system, the simplicity of the architecture, the reliability
link |
of the redundancy. It is designed to be manufacturable at very large scale and provides the right
link |
unit economics. So that's the next big step for us on the hardware side. That's already there for
link |
scale, the version five. That's right. Is that a coincidence or should we look into a conspiracy
link |
theory that it's the same version as the Pixel phone? Is that what's the hardware thing?
link |
I can neither confirm nor deluxe. All right, cool. So sorry. So that's the, okay, that's
link |
that axis. What else? So similarly, hardware is a very discreet jump. But you know,
link |
similar to how we're making that change from the fourth generation hardware to the fifth,
link |
we're making similar improvements on the software side to make it more robust and
link |
more general and allow us to quickly scale beyond Phoenix. So that's the first dimension
link |
of core technology. The second dimension is evaluation and deployment. How do you
link |
measure your system? How do you evaluate it? How do you build the release and deployment process
link |
where with confidence, you can regularly release new versions of your driver into a fleet?
link |
How do you get good at it so that it is not a huge tax on your researchers and engineers?
link |
So how do you build all these processes, the frameworks, the simulation, the evaluation,
link |
the data science, the validation so that people can focus on improving the system
link |
and kind of the releases just go out the door and get deployed across the fleet.
link |
So we've gotten really good at that in Phoenix. That's been a tremendously difficult problem.
link |
But that's what we have in Phoenix right now that gives us that foundation. And now we're
link |
working on kind of incorporating all the lessons that we've learned to make it more efficient to
link |
go to new places and scale up and just kind of stamp things out. So that's that second dimension
link |
of evaluation and deployment. And the third dimension is product commercial and operational
link |
excellence. And again, Phoenix there is providing an incredibly valuable platform.
link |
That's why we're doing things end to end in Phoenix. We're learning as we discussed a little
link |
earlier today, tremendous amount of really valuable lessons from our users getting really
link |
incredible feedback. And we'll continue to iterate on that and incorporate all those lessons into
link |
making our product even better and more convenient for our users.
link |
So you're converting this whole process of Phoenix in Phoenix into something that could
link |
be copy and pasted elsewhere. So like, perhaps you didn't think of it that way when you were doing
link |
the experimentation in Phoenix. But so how long did you basically, you can correct me, but you've,
link |
I mean, it's still early days, but you've taken a full journey in Phoenix, right?
link |
As you were saying, of like what it takes to basically automate, I mean, it's not the entirety of
link |
Phoenix, right? But I imagine it can encompass the entirety of Phoenix at some, some near term
link |
date, but that's not even perhaps important. Like as long as it's a large enough geographic area.
link |
So what, how copy pasteable is that process currently? And how like, you know, like when
link |
you copy and paste in, in, in Google docs, I think now in, or in Word, you can like apply source
link |
formatting or apply destination formatting. So when you copy and paste the Phoenix into like,
link |
say, Boston, how do you apply the destination formatting? Like how much of the core of the
link |
entire process of bringing an actual public transportation, autonomous transportation
link |
service to a city is there in Phoenix that you understand enough to copy and paste into Boston
link |
or wherever. So we're not quite there yet. We're not at a point where we're kind of massively
link |
copy and pasting all over the place. But Phoenix, what we did in Phoenix, and we very intentionally
link |
have chosen Phoenix as our first full deployment area, you know, exactly for that reason to kind
link |
of tease the problem apart, look at each dimension, you know, focus on the fundamentals of complexity
link |
and de risking those dimensions, and then bringing the entire thing together to get all the way and
link |
force ourselves to learn all those hard lessons on this technology hardware and software on the
link |
evaluation deployment on operating a service operating a business using actually serving
link |
our customers all the way so that we're fully informed about the most difficult, most important
link |
challenges to get us to that next step of massive copy and pasting, as, as you said. And
link |
that's what we're doing right now. We're incorporating all those things that we learned
link |
into that next system that then will allow us to kind of copy and paste all over the place and to
link |
massively scale to, you know, more users and more locations. I mean, you know, just talked a little
link |
bit about, you know, what does that mean along those different dimensions. So on the hardware
link |
side, for example, again, it's that switch from the fourth to the fifth generation. And the fifth
link |
generation is designed to kind of have that property. Can you say what other cities you're
link |
thinking about? Like, I'm thinking about, sorry, when San Francisco now, I thought I want to move
link |
to San Francisco. But I'm thinking about moving to Austin. I don't know why people are not being
link |
very nice about San Francisco currently. For maybe it's a small, it's maybe it's in vogue right now.
link |
But Austin seems I visited there and it was, I was in a Walmart. It's funny, these moments,
link |
like turn your life, there's this very nice woman with kind eyes, just like stopped and said,
link |
you look so handsome in that tie, honey, to me, this is never happening to me in my life, but
link |
just the sweetness of this woman is something I've never experienced, certainly in the streets of
link |
Boston. But even in San Francisco, where people wouldn't, that's just not how they speak or think.
link |
I don't know, there's a warmth to Austin that love. And since Waymo does have a little bit of a
link |
history there, is that a possibility? Is this your version of asking the question of like,
link |
you know, Dimitri, I know you can't share your commercial and deployment roadmap,
link |
but I'm thinking about moving to San Francisco of Austin, like in a blink twice, if you think I
link |
should move to it. Yeah, that's true. That's true. You got me. Well, you know, we've been testing
link |
in all over the place. I think we've been testing in more than 25 cities. We drive in San Francisco,
link |
we drive in, you know, Michigan for snow. We are doing significant amount of testing
link |
in the Bay Area, including San Francisco. Which is not like, because we're talking about the
link |
very different thing, which is like a full on large geographic area, public service.
link |
You can't share. Okay.
link |
What about Moscow? When is that happening? Take on Yandex. I'm not paying attention to those folks.
link |
They're doing, you know, there's a lot of fun. I mean, maybe as a way of a question,
link |
you didn't speak to sort of like policy or like, is there tricky things with government and so on?
link |
Like, is there other friction that you've encountered, except sort of technological friction
link |
of solving this very difficult problem? Is there other stuff that you have to overcome
link |
when deploying a public service in a city? That's interesting.
link |
It's very important. So we put significant effort in creating those partnerships and,
link |
you know, those relationships with governments at all levels, you know, local governments,
link |
municipalities, you know, state level, federal level. We've been engaged in very deep conversations
link |
from the earliest days of our projects. Whenever at all of these levels, you know, whenever we go
link |
to test or, you know, operate in a new area, you know, we always lead with the conversation
link |
with the local officials. But the result of that investment is that, no, it's not challenges we
link |
have to overcome, but it is a very important that we continue to have this conversation.
link |
Oh, yeah. I love politicians too. Okay. So Mr. Elon Musk said that LiDAR is a crutch.
link |
What are your thoughts?
link |
I wouldn't characterize it exactly that way. I know I think LiDAR is very important. It is a key
link |
sensor that, you know, we use just like other modalities, right? As we discussed, our cars use
link |
cameras, LiDARs and radars. They are all very important. They are at kind of the physical level.
link |
They are very different. They have very different, you know, physical characteristics.
link |
Cameras are passive. LiDARs and radars are active. You use different wavelengths.
link |
So that means they complement each other very nicely. And they, you know, together combine.
link |
They can be used to build a much safer and much more capable system. So, you know,
link |
to me, it's more of a question, you know, why the heck would you handicap yourself and not use one
link |
or more of those sensing modalities when they, you know, undoubtedly just make your system
link |
more capable and safer. Now, it, you know, what might make sense for one product or one business
link |
might not make sense for another one. So if you're talking about driver assist technologies,
link |
you make certain design decisions and you make certain tradeoffs. And you make different ones
link |
if you are, you know, building a driver that you deploy in fully driverless vehicles.
link |
And, you know, and LiDAR specifically, when this question comes up, I, you know, typically the
link |
criticisms that I hear are, you know, the counterpoints that cost and aesthetics. And I don't
link |
find either of those honestly very compelling. So on the cost side, there's nothing fundamentally
link |
prohibitive about, you know, the cost of LiDARs. You know, radars used to be very expensive before
link |
people started, you know, before people made certain advances in technology and you started to
link |
manufacture them massive scale and deploy them in vehicles, right? You know, similar with LiDARs.
link |
And this is where the LiDARs that we have on our car, especially the fifth generation,
link |
you know, we've been able to make some pretty qualitative discontinuous jumps in terms of the
link |
fundamental technology that allow us to, you know, manufacture those things at very significant scale
link |
and add a fraction of the cost of your both our previous generation, as well as a fraction of
link |
the cost of, you know, what might be available on the market, you know, off the shelf right now.
link |
And, you know, that improvement will continue. So I think, you know, cost is not a real issue.
link |
Second one is, you know, aesthetics. You know, I don't think that's, you know, a real issue either.
link |
Beauty is an eye of the beholder. You can make LiDAR sexy again.
link |
I think you're exactly right. I think it is sexy. Like, honestly, I think form is a function.
link |
Well, okay. You know, I was actually somebody brought this up to me. I mean, all forms of
link |
LiDAR, even, even like the ones that are like big, you can make look, I mean, you can make look
link |
beautiful. Like there's no sense in which you can't integrate into design. Like there's all
link |
kinds of awesome designs. I don't think small and humble is beautiful. It could be like,
link |
you know, brutalism or like it could be like harsh corners. I mean, like I said, like hot rods,
link |
like I don't like, I don't necessarily like, like, oh man, I'm going to start so much controversy
link |
with this. I don't like Porsches. Okay. The Porsche 911, like everyone says the most beautiful.
link |
No, it, no, it's like, it's like a baby car. It doesn't make any sense. But everyone, it's
link |
beauties in the eye of the beholder. You're already looking at me like, what's this kid talking about?
link |
I'm happy to talk about you're digging your own hole. The form and function and my take on the beauty
link |
of the hardware that we put on our vehicles. You know, I will not comment on the Porsche.
link |
You know, Porsche monologues. Okay. All right. So, but aesthetics, fine. But there's an underlying
link |
like philosophical question behind the kind of LiDAR question is like, how much of the problem
link |
can be solved with computer vision, with machine learning? So I think without sort of disagreements
link |
and so on, it's nice to put it on the spectrum because Waymo is doing a lot of machine learning as
link |
well. It's interesting to think how much of driving if we look at five years, 10 years, 50 years down
link |
the road would can be learned in almost more and more and more end to end way. If you look at what
link |
Tesla is doing with as a machine learning problem, they're doing a multitask learning
link |
thing where it's just they break up driving into a bunch of learning tasks and they have
link |
one single neural network and they're just collecting huge amounts of data that's training that.
link |
I've recently hung out with George Hots. I don't know if you know George.
link |
I love him so much. He's just an entertaining human being. We were off mic talking about
link |
Hunter S. Thompson. He's the Hunter S. Thompson of the time I was driving. Okay. So he I didn't
link |
realize this with comma AI, but they're like really trying to end to end. They're the machine
link |
like looking at the machine learning problem. They're really not doing multitask learning,
link |
but it's it's it's computing the drivable area as a machine learning task and hoping that like
link |
down the line, this level two system as driver assistance will eventually lead to allowing
link |
you to have a fully autonomous vehicle. Okay. There's an underlying deep philosophical question
link |
there, technical question of how much of driving can be learned. So LiDAR is an effective tool today
link |
for actually deploying a successful service in Phoenix, right? That's safe, that's reliable,
link |
etc, etc. But the the question and I'm not saying you can't do machine learning on LiDAR,
link |
but the question is that like how much of driving can be learned eventually? Can we do fully
link |
autonomous that's learned? Yeah. You know, learning is all over the place and play is the
link |
key role in every part of our system. As you said, I would, you know, decouple the sensing
link |
modalities from the, you know, ML and the software parts of it. LiDAR, radar, cameras,
link |
it's all machine learning. All of the object detection classification, of course, I go that
link |
that's what, you know, these modern deep nuts and con nuts are very good at. You feed them raw data,
link |
massive amounts of raw data. And that's actually what our custom build LiDARs and
link |
radars are really good at. And radars, they don't just give you point estimates of, you know,
link |
objects in space, they give you raw, like physical observations. And then you take all
link |
of that raw information, you know, there's colors of the pixels, whether it's, you know,
link |
LiDAR's returns and some auxiliary information, it's not just distance, right? And, you know,
link |
angle and distance is much richer information that you get from those returns, plus really rich
link |
information from the radars, you fuse it all together and you feed it into those massive ML
link |
models that then, you know, lead to the best results in terms of, you know, object detection,
link |
classification, you know, state estimation. So there's a side to interrupt, but there is a fusion.
link |
I mean, that's something that people didn't do for a very long time, which is like at the
link |
sensor fusion level, I guess, like early on fusing the information together, whether
link |
so that the the sensory information that the vehicle receives from the different modalities,
link |
or even from different cameras is combined before it is fed into the machine learning models.
link |
Yeah. So I think this is one of the trends. You're seeing more of that. You mentioned N10.
link |
There's different interpretations of N10. There is kind of the purest interpretation. I'm going
link |
to like have one model that goes from raw sensor data to like, you know, steering torque and,
link |
you know, gas brakes. But, you know, that's too much. I don't think that's the right way to do it.
link |
There's more, you know, smaller versions of N10, where you're kind of doing more end to end learning
link |
or core training or depropagation of kind of signals back and forth across the different stages
link |
of your system. There's, you know, really good ways. It gets into some fairly complex design
link |
choices, where on one hand you want modularity and the composability, the composability of your
link |
system. But on the other hand, you don't want to create interfaces that are too narrow or too
link |
brittle to engineered, where you're giving up on the generality of a solution, or you're unable
link |
to properly propagate signal, you know, reach signal forward and losses and, you know, back
link |
so you can optimize the whole system jointly. So I would decouple, and I guess what you're seeing
link |
in terms of the fusion of the sensing data from different modalities, as well as kind of fusion
link |
at in the temporal level, going more from, you know, frame by frame, where, you know, you would
link |
have one net that would do frame by frame detection and camera and then, you know, something that does
link |
frame by frame and lighter and then radar. And then you fuse it, you know, in a weaker engineered
link |
way later, like the field over the last decade has been evolving in more kind of joint fusion,
link |
more end to end models that are solving some of these tasks, you know, jointly. And there's
link |
tremendous power in that. And, you know, that's the progression that you kind of are, you know,
link |
our stack has been on as well. Now, you know, so I would decouple the kind of sensing and how
link |
that information is used from the role of ML and the entire stack. And, you know, I guess it's,
link |
there's tradeoffs and modularity and how do you inject inductive bias into your system?
link |
All right, this is, there's tremendous power in being able to do that. So, you know, we have,
link |
there's no part of our system that is not heavily, that does not heavily, you know, leverage,
link |
you know, data driven development or, you know, state of the art of ML. But there's mapping,
link |
there's simulator, or there's perception, you know, object level, you know, perception,
link |
whether it's semantic understanding, prediction, decision making, you know, so forth and so on.
link |
It's, and of course, object detection and classification, like finding pedestrians and
link |
cars and cyclists and, you know, cones and signs and vegetation and being very good at
link |
estimating kind of detection classification and state estimation, there's just stable stakes,
link |
like, like that's step zero of this whole stack. You can be incredibly good at that,
link |
whether you use cameras or light as a radar, but that's just, you know, that's stable stakes,
link |
that's just step zero. Beyond that, you get into the really interesting challenges of
link |
semantic understanding of the perception level, you get into scene level reasoning,
link |
you get into very deep problems that have to do with prediction and joint prediction and
link |
interaction, so the interaction between all the actors in the environment, pedestrians, cyclists,
link |
other cars, and you get into decision making, right? So how do you build a lot of systems?
link |
So we leverage ML very heavily in all of these components. I do believe that the best results
link |
you achieve by kind of using a hybrid approach and having different types of ML,
link |
having different models with different degrees of inductive bias that you can have,
link |
and combining kind of model, you know, free approaches with some, you know, model based
link |
approaches and some rule based, physics based systems. So, you know, one example I can give you
link |
is traffic lights. There's a problem of the detection of traffic light state and obviously
link |
that's a great problem for, you know, computer vision, confinates are, you know, that's their
link |
bread and butter, right? That's how you build that. But then the interpretation of, you know,
link |
of a traffic light that you're gonna need to learn that, right? You read, you don't need to
link |
build some, you know, complex ML model that, you know, infers with some, you know, precision
link |
and recall that read means stop. Like, it was a, it's a very clear engineered signal with very
link |
clear semantics, right? So you want to induce that bias, like how you induce that bias and that,
link |
whether, you know, it's a constraint or a cost, you know, function in your stack. But like,
link |
it is important to be able to inject that like clear semantic signal into your stack. And,
link |
you know, that's what we do. And, but then the question of like, and that's when you apply it
link |
to yourself, when you are making decisions, whether you want to stop for a red light,
link |
you know, or not. But if you think about how other people treat traffic lights, we're back to the
link |
ML version of that. Because, you know, they're supposed to stop for a red light, but that doesn't
link |
mean they will. So then you're back in the like very heavy ML domain where you're picking up on
link |
like very subtle keys about, you know, they have to do with the behavior of objects and pedestrians,
link |
cyclists, cars, and the whole thing, you know, the entire configuration of the scene that allow
link |
you to make accurate predictions on whether they will in fact stop or run a red light.
link |
So it sounds like already for Waymo, like machine learning is a huge part of the stack. So it's a
link |
huge part of like, not just, so obviously, the first level zero or whatever you said, which is
link |
like just the object detection of things that, you know, know that machine learning can do,
link |
but also starting to do prediction behavior and so on to model the what other or the other parties
link |
in the scene, entities in the scene are going to do. So machine learning is more and more
link |
playing a role in that as well. Of course. Oh, absolutely. I think we've been going back to
link |
the earliest days, like DARPA, or even the DARPA grand challenge, and team was leveraging, you
link |
know, machine learning was like pre, you know, image nut and I was very different type of ML,
link |
but and I think actually was it was before my time, but the Stanford team on during the grand
link |
challenge had a very interesting machine learned system that would, you know, use lighter and camera
link |
when driving in the desert. And it we had build the model where it would kind of extend the range
link |
of free space reasoning, we get a clear signal from lighter. And then it had a model said,
link |
hey, like this stuff and camera kind of sort of looks like this stuff and lighter. And I know
link |
this stuff and that I've seen in lighter, I'm very confident of this free space. So let me extend
link |
that free space zone into the camera range that would allow the vehicle to drive faster.
link |
And then we've been building on top of that and kind of staying and pushing the state of the art
link |
in ML in all kinds of different ML over the years. And in fact, from the earliest days, I think,
link |
you know, 2010, probably the year where Google, maybe 2011, probably got pretty heavily involved
link |
in machine learning, kind of deep nuts. And at that time, it's probably the only company that
link |
was very heavily investing in kind of state of the art ML and self driving cars, right? And they
link |
go hand in hand. And we've been on that journey ever since we're doing pushing a lot of these
link |
areas in terms of research, you know, at Waymo and we collaborate very heavily with the researchers
link |
in alphabet. And I call kinds of ML, supervised ML, unsupervised ML, you know, publish some
link |
interesting research papers in the space, especially recently, it's just a super super
link |
active learning as well. Yeah, super super active. Of course, there's kind of the more
link |
mature stuff, like, you know, convenants for, you know, object detection. But there's some
link |
really interesting, really active work that's happening in more and bigger models and models
link |
that have more structure to them, you know, not just large bitmaps and reason about temporal
link |
sequences. And some of the interesting breakthroughs that you've, you know, we've seen in language
link |
models, right, you know, transformers, you know, GPT three and friends. There's some really
link |
interesting applications of some of the core breakthroughs to those problems of, you know,
link |
behavior prediction, as well as, you know, decision making and planning, right? You can
link |
think about it kind of the behavior, how, you know, the path of trajectories, the how people drive,
link |
they have kind of a share a lot of the fundamental structure, you know, this problem. There's,
link |
you know, sequential, you know, nature, there's a lot of structure. In this representation,
link |
there is a strong locality, kind of like in sentences, you know, words that follow each other,
link |
they're strongly connected. But there's also kind of larger context that doesn't have that
link |
locality. And you also see that in driving, right? What's happening in the scene as a whole
link |
has very strong implications on, you know, the kind of the next step in that sequence where
link |
whether you're predicting what other people are going to do, whether you're making your own decisions,
link |
or whether in the simulator, you're building generative models of, you know,
link |
humans walking, cyclists riding and other cars driving.
link |
Oh, that's, that's all really fascinating. Like how it's fascinating to think that, uh,
link |
transforming models and all this, all the breakthroughs in language and NLP that might
link |
be applicable to like driving at the higher level at the behavioral level. That's kind of fascinating.
link |
Let me ask about pesky little creatures called pedestrians and cyclists.
link |
They seem so humans are a problem if we can get rid of them, I would.
link |
But unfortunately, they're also a source of joy and love and beauty. So let's keep them around.
link |
They're also our customers. Oh, for your perspective, yes, yes, for sure.
link |
There's also some money. Very good. Um, but I don't even know where I was going. Oh,
link |
yes, pedestrians and cyclists. Uh, I, you know, they're a fascinating injection into the system of
link |
uncertainty of, um, of like a game theoretic dance of what to do. And also,
link |
uh, they have perceptions of their own and they can tweet about your product. So you
link |
don't want to run them over from that perspective. Uh, I mean, I don't know, I'm, I'm joking a lot,
link |
but I think in seriousness, like, you know, pedestrians are a complicated, um, uh, computer
link |
vision problem, a complicated behavioral problem. Is there something interesting you could say about
link |
we, what you've learned from a machine learning perspective, from also an autonomous vehicle
link |
and a product perspective about just interacting with the humans in this world?
link |
Yeah. Just, you know, to state on record, we care deeply about the safety of pedestrians,
link |
you know, even the ones that don't have Twitter accounts. Um, thank you. All right, cool. Not me.
link |
But yes, I, I'm glad, I'm glad somebody does. Okay.
link |
Uh, but you know, in all seriousness, safety of, uh, vulnerable road users, uh, pedestrians
link |
or cyclists is one of our highest priorities. Uh, we do a tremendous amount of testing, uh,
link |
and validation and put a very significant emphasis on, you know, the capabilities of our systems that
link |
have to do with safety around those unprotected vulnerable road users. Um, you know, cars just,
link |
you know, we discussed earlier in Phoenix, we have completely empty cars, completely driverless
link |
cars, you know, driving in this very large area. Uh, and you know, some people use them to, you
link |
know, go to school. So they'll drive through school zones, right? So, uh, kids are kind of the very
link |
special class of those vulnerable user road users, right? You want to be super, super safe, uh, and
link |
super, super cautious around those. So we take it very, very, very seriously. Um, and you know,
link |
what does it take, uh, to, uh, be good at it? Uh, you know, an incredible amount of, uh, performance
link |
across your whole stack, you know, starts with hardware. Uh, and again, you want to use all
link |
something of modalities available to you. Imagine driving on a residential road at night and kind
link |
of making a turn and you don't have, you know, headlights covering some part of the space and
link |
like, you know, a kid might run out and you know, lighters are amazing at that. Uh, they see just
link |
as well in complete darkness as they do during the day, right? So just again, it gives you that
link |
extra, uh, uh, you know, margin in terms of, you know, capability and performance and safety and
link |
quality. Uh, and in fact, we oftentimes, uh, in these kinds of situations, we have our system
link |
detect something in some cases even earlier than our trained operators in the car might do,
link |
especially, you know, in conditions like, you know, very dark nights. Um, so starts with sensing,
link |
then, you know, perception has to be incredibly good. And you have to be very, very good at kind
link |
of detecting, uh, pedestrians, uh, in all kinds of situations and all kinds of environments,
link |
including, you know, people in weird poses, uh, people kind of running, uh, around and, you know,
link |
being partially occluded. Um, uh, so, you know, that, that's stuff number one, right? Then you
link |
have to have in very high accuracy and very low latency in terms of your reactions, uh, to, you
link |
know, what, you know, these, uh, actors might do, right? And we've put a tremendous amount of
link |
engineering and tremendous amount of validation in to make sure our system performs, uh, properly.
link |
And, you know, oftentimes it does require very strong reaction to do the safe thing. And, you
link |
know, we actually see a lot of cases like that. That's the long tail of really rare, you know,
link |
really, uh, you know, crazy events, uh, that, um, contribute to the safety around pedestrians. Like
link |
one, one example that comes to mind that we actually got happened, uh, in Phoenix where we
link |
were, uh, driving, uh, along and I think it was a 45 mile per hour road. So, you know, pretty high
link |
speed traffic and there was a sidewalk next to it and there was a cyclist on the sidewalk. And
link |
as, uh, we were in the right lane and right next to the site, uh, so it was a multi lane road.
link |
Uh, so as we got close, uh, to the cyclist on the sidewalk, uh, it was a woman, you know, she
link |
tripped and fell, just, you know, fell right into the path of our vehicle. Right. Um, and our, you
link |
know, car, uh, uh, you know, this was actually with a test driver, our test drivers, uh, uh, did
link |
exactly the right thing. Uh, they kind of reacted and came to stop and requires both very strong
link |
steering and, uh, you know, strong application of the brake. Uh, and then we simulated what our
link |
system would have done in that situation and it did, you know, exactly the same thing. It, uh,
link |
and that, that speaks to all of those components of really good, uh, state estimation and tracking
link |
and like imagine, you know, a person on a bike and they're falling over and they're doing that
link |
right in front of you. Right. So you have to be really like, things are changing. The appearance
link |
of that whole, uh, thing is changing. Right. And a person goes one way. They're falling on the road.
link |
They're, you know, being flat on the ground in front of you, you know, the bike goes flying the
link |
other direction. Like the two objects that used to be one, they're now, uh, are splitting apart
link |
and the car has to like detect all of that. Uh, like milliseconds matter and it doesn't, you know,
link |
it's not good enough to just break. You have to like steer and break and there's traffic around
link |
you. So like it all has to come together and it was really great, uh, to see in this case in other
link |
cases like that, uh, that we're actually seeing in the wild that our system is, you know, performing
link |
exactly the way, uh, that we would have liked and is able to, you know, avoid, uh, collisions like
link |
this. It's such an exciting space for robotics. Like in the, in that split second to make decisions
link |
of life and death, I don't know if the stakes are high in a sense, but it's also beautiful that,
link |
um, um, for somebody who loves artificial intelligence, the possibility that an AI system
link |
might be able to save a human life. Uh, that's kind of exciting as a, as a problem. Like to wake up,
link |
you get, it's terrifying probably from an, for an engineer to wake up and to think about,
link |
but it's also exciting because it's like, it's, it's in your hands. Let me try to ask a question
link |
that's often brought up about autonomous vehicles and, uh, it might be fun to see if you have
link |
anything, anything interesting to say, which is about the trolley problem. So, uh, the trolley
link |
problem is a interesting philosophical construct of, uh, that highlights and there's many others
link |
like it of the difficult ethical decisions that, uh, we humans have before us in this complicated
link |
world. Uh, so the specifically is the choice between if you were forced to choose, uh, to kill
link |
a group X of people versus a group Y of people, like one person, if you didn't, if you did nothing,
link |
you would kill one person. But if you, you would kill five people. And if you decide to
link |
swerve out of the way, you would only kill one person. Do you do nothing or you choose to do
link |
something and you can construct all kinds of sort of ethical experiments of this kind that,
link |
uh, I think at least on a positive note, inspire you to think about like introspect
link |
what are the, the physics of our morality. And there's usually not good answers there.
link |
I think people love it because it's just an exciting thing to think about. I think people
link |
who build autonomous vehicles usually roll their eyes because, uh, this is not,
link |
this one as constructed, this like literally never comes up in reality. You never have to
link |
choose between killing one or like one of two groups of people. But I wonder if you can speak to,
link |
is there some, something interesting to use an engineer of autonomous vehicles that's within
link |
the trolley problem? Or maybe more generally, are there difficult ethical decisions that you find
link |
that, uh, algorithm must make on the specific version of the trolley problem? Which one would
link |
you do if you're driving? The question itself is a profound question because we humans ourselves
link |
cannot answer it. And that's the very point. Uh, I will kill both. Um, you know, humans,
link |
I think you're exactly right. And that, you know, humans are not particularly good. I think
link |
they kind of phrased as a, like, what would a computer do? But like humans are not very good.
link |
And actually oftentimes think that, you know, freezing and kind of not doing anything because
link |
like you've taken a few extra milliseconds to just process and then you end up like doing the worst
link |
of the possible outcomes, right? So I do think that as you've pointed out, it can be a bit of
link |
a distraction and it can be a bit of a kind of red herring. I think it's an interesting
link |
philosophy, you know, discussion in the realm of philosophy, right? But in terms of what,
link |
you know, how that affects the actual engineering and deployment of self driving vehicles, I,
link |
um, it's not how you go about building a system, right? We have talked about how you engineer a
link |
system, how you go about evaluating the different components and the, you know, the safety of the
link |
entire thing. How do you kind of inject the, you know, various model based safety based
link |
arguments and like, yes, your reason at parts of the system, you know, your reason about the
link |
probability of a collision, the severity of that collision, right? And that is incorporated. And
link |
there's, you know, you have to properly reason about the uncertainty that flows through the system,
link |
right? So, you know, those, you know, factors definitely play a role in how the cars don't
link |
behave, but they tend to be more of like the immersion behavior. And what you see, like,
link |
you're absolutely right, that these, you know, clear theoretical problems that they, you know,
link |
you don't occur to that in the system and really kind of being back to our previous discussion
link |
of like, what, you know, what, you know, which one do you choose? Well, you know, oftentimes,
link |
like, you made a mistake earlier, like, you shouldn't be in that situation in the first place,
link |
right? And in reality, the system comes up. If you build a very good safe and capable driver,
link |
you have enough, you know, clues in the environment that you drive defensively,
link |
so you don't put yourself in that situation, right? And again, you know, it has, you know,
link |
this, if you go back to that analogy of, you know, precision and recoil, like, okay, you can make a,
link |
you know, very hard tradeoff off, but like neither answer is really good. But what instead you focus
link |
on is kind of moving the whole curve up, and then you focus on building the right capability and
link |
the right defensive driving so that, you know, you don't put yourself in a situation like this.
link |
I don't know if you have a good answer for this, but people love it when I ask this question
link |
about books. Are there books in your life that you've enjoyed philosophical fiction,
link |
technical, that had a big impact on you as an engineer or as a human being, you know,
link |
everything from science fiction to a favorite textbook? Is there three books that stand out
link |
that you can think of? Three books. So I would, you know, that impacted me. I would say,
link |
and this one is, you probably know it well, but not generally well known. I think in the U.S.
link |
or kind of internationally, The Master and Margarita. It's one of actually my favorite books. It is,
link |
you know, by Russian, it's a novel by Russian author Mikhail Bulgakov. And it's just, it's a
link |
great book. It's one of those books that you can like reread your entire life. And it's very
link |
accessible. You can read it as a kid. And like, it's, you know, the plot is interesting. It's,
link |
you know, the devil, you know, visiting the Soviet Union. But it, like, you read it, reread it at
link |
different stages of your life. And you enjoy it for different, very different reasons. And you
link |
keep finding like deeper and deeper meaning. And, you know, it kind of affected, you know,
link |
had a, definitely had an imprint on me, you know, mostly from the, probably kind of the cultural,
link |
stylistic aspect, like it makes you one of those books that, you know, is good and makes you
link |
think, but also has like this really, you know, silly, quirky, dark sense of, you know, humor.
link |
Okay. It captures the Russian soul. That's more than many, perhaps many other books. On that,
link |
like, slight note, just out of curiosity, one of the saddest things is I've read that book
link |
in English. Did you, by chance, read it in English or in Russian?
link |
In Russian, only in Russian. And I actually, that is a question I had.
link |
Post myself every once in a while. I wonder how well it translates, if it translates at all.
link |
And there's the language aspect of it. And then there's the cultural aspect. So I,
link |
and actually, I'm not sure if, you know, either of those would work well in English.
link |
Now, I forget their names, but so when the COVID lifts a little bit, I'm traveling to Paris,
link |
for, for several reasons. One is just, I've never been to Paris. I want to go to Paris,
link |
but there's the most famous translators of the Stiyovsky Tolstoy of most of Russian literature
link |
live there. There's a couple, they're famous, a man and a woman. And I'm going to sort of have
link |
a series of conversations with them. And in preparation for that, I'm starting to read
link |
Stiyovsky in Russian. So I'm really embarrassed to say that I've read this, everything I've
link |
read in Russian literature of like serious depth has been in English, even though I can also read,
link |
I mean, obviously in Russian, but for some reason, it seemed in the optimization of life,
link |
it seemed the improper decision to do, to read in Russian, like, you know,
link |
like I don't need to, I need to think in English, not in Russian, but now I'm changing my mind
link |
on that. And so the question of how well it translates is a really fundamental one, like
link |
even with Dostoyevsky. So for what I understand, Dostoyevsky translates easier. Others don't
link |
as much. Obviously the poetry doesn't translate as well. I'm also the music of big fan of Vladimir
link |
Vosotsky. He doesn't obviously translate well. People have tried. But mastermind, I don't know,
link |
I don't know about that one. I just know it in English, you know, as fun as hell in English.
link |
So, so, but it's a curious question. And I want to study it rigorously from both the machine
link |
learning aspect. And also because I want to do a couple of interviews in Russia, that I'm still
link |
unsure of how to properly conduct an interview across a language barrier. It's a fascinating
link |
question that ultimately communicates to an American audience. There's a few Russian people that
link |
I think are truly special human beings. And I feel like I sometimes encounter this with
link |
some incredible scientists and maybe you encounter this as well at some point in your life that
link |
it feels like because of language barrier, their ideas are lost to history. It's a sad thing. I
link |
think about like Chinese scientists or even authors that like, that we don't in English
link |
speaking world don't get to appreciate some like the depth of the culture because it's
link |
lost in translation. And I feel like I would love to show that to the world. Like I'm just
link |
some idiot. But because I have this, like at least some semblance of skill in speaking Russian,
link |
I feel like and I know how to record stuff on a video camera. I feel like I want to catch like
link |
Gregorio Perlman who's a mathematician. I'm not sure if you're familiar with him. I want to talk
link |
to him like he's a fascinating mind and to bring him to a wider audience in English speaking,
link |
it'll be fascinating. But that requires to be rigorous about this question of how well
link |
Bogakov translates. I mean, I know it's a silly concept, but it's a fundamental one
link |
because how do you translate? And that's the thing that Google Translate is also facing
link |
as a more machine learning problem. But I wonder as a more bigger problem for AI,
link |
how do we capture the magic that's there in the language?
link |
I think that's a really interesting, really challenging problem. If you do read it,
link |
master in Russian and be curious to get your opinion. And I think part of it is language,
link |
but part of it is just centuries of culture. The cultures are different, so it's hard to
link |
connect that. Okay, so that was my first one, right? You had two more. The second one I would
link |
probably pick the science fiction by the Strogosky brothers. It's up there with Isaac
link |
Asimov and Ray Bradbury and Company. The Strogosky brothers kind of appealed more to me. I think
link |
more, it made more of an impression on me growing up. I apologize if I'm showing my complete ignorance.
link |
I'm so weak on sci fi. What are they, right? Oh, roadside picnic. Hard to be a god. Beedle
link |
in an anthill. Monday starts on Saturday. It's not just science fiction. It also has very
link |
interesting interpersonal and societal questions. And some of the language is just completely
link |
hilarious. That's right. That's right. Oh, interesting. Monday starts on Saturday.
link |
So I need to read. Oh, boy. You put that in the category of science fiction.
link |
That one is, I mean, this was more of a silly, humorous work. I mean, there is kind of
link |
profound too, right? Science fiction, right, is about this research institute. It has
link |
deep parallels to serious research, but the setting, of course, is that they're working on
link |
magic, right? And that's their style, right? And other books are very different, right? Hard
link |
to be a god, right? It's about this higher society being injected into this primitive world and how
link |
they operate there. Some of the very deep ethical questions there, right? And they've got this full
link |
spectrum. Some is more about a kind of more adventurous style. But I enjoy all of their
link |
books. There's probably a couple. Actually, one, I think that they consider their most important
link |
work. I think it's the snail on a hill. I don't know exactly how it translates. I tried reading
link |
a couple of times. I still don't get it. But everything else I fully enjoyed. And like for
link |
one of my birthdays as a kid, I got their entire collection occupied a giant shelf in my room.
link |
And then over the holidays, my parents couldn't drag me out of the room. And I read the whole
link |
thing cover to cover. And I really enjoyed it. And that's one more. For the third one,
link |
maybe a little bit darker. But it comes to mind is Orwell's 1984.
link |
And I asked what made an impression on me and the books that people should read. That one,
link |
I think, falls in the category of both. Now, definitely, it's one of those books that you
link |
read and you just kind of put it down and you stare in space for a while.
link |
Yeah, that kind of work. I think there's lessons there people should not ignore.
link |
Nowadays, with everything that's happening in the world, I can't help it. But I have my mind jump
link |
to some parallels with what Orwell described. And there's this whole concept of double think
link |
and ignoring logic and holding completely contradictory opinions in your mind and not
link |
have that not bother you and stick into the party line at all costs. There's something there.
link |
If anything, 2020 has taught me. And I'm a huge fan of Animal Farm, which is a kind of friendly
link |
as a friend of 1984 by Orwell. It's kind of another thought experiment of how our society
link |
may go in directions that we wouldn't like it to go. But if anything, that's been kind of
link |
heartbreaking to an optimist about 2020, is that that society is kind of fragile. Like, we have
link |
this. This is a special little experiment we have going on. And not it's not unbreakable.
link |
Like, we should be careful to preserve whatever the special thing we have going on. I mean,
link |
I think 1984 and in these books, Brave New World, they're helpful in thinking stuff can go wrong
link |
in non obvious ways. And it's up to us to preserve it. And it's a responsibility. It's been weighing
link |
heavy on me. Because like, for some reason, like, more than my mom follows me on Twitter. And I feel
link |
like I have I have like now somehow responsibility to do this world. And it dawned on me that like
link |
me and millions of others are like the little ants that maintain this little colony. Right.
link |
So we have a responsibility not to be, I don't know what the right analogy is, but put a flamethrower
link |
to the place. We want to not do that. And there's interesting, complicated ways of doing that as
link |
1984 shows. It could be through bureaucracy, it could be through incompetence, it could be through
link |
misinformation, it could be through division and toxicity. I'm a huge believer in like that love
link |
will be the somehow the solution. So love and robots. Love and robots. Yeah. I think you're
link |
exactly right. Unfortunately, I think it's less of a flamethrower type of an extra. I think it's
link |
more of a in many cases, can be more of a slow boil. And that that's the danger.
link |
Let me ask, it's a fun thing to make a world class roboticist, engineer, and leader uncomfortable
link |
with a ridiculous question about life. What is the meaning of life? It's a main tree from a robotics
link |
and a human perspective. You only have a couple of minutes or one minute to answer. So
link |
I don't know if that makes it more difficult or easier.
link |
You know, they're very tempted to quote one of the stories by Isaac Asimov, actually titled
link |
appropriately titled The Last Question, short story, where the plot is that humans build this
link |
supercomputer, this AI intelligence, and once it gets powerful enough, they pose this question to
link |
it. How can the entropy in the universe be reduced? So the computer replies, hang on,
link |
as of yet, insufficient information to give a meaningful answer. And then thousands of years
link |
go by and they keep posing the same question. The computer gets more and more powerful and
link |
keeps giving the same answer. As of yet, insufficient information to give a meaningful
link |
answer or something along those lines. And then it keeps happening and happening fast forward
link |
like millions of years into the future and billions of years. And at some point, it's just the only
link |
entity in the universe. It's absorbed all humanity and all knowledge in the universe. And it keeps
link |
posing the same question to itself. And finally, it gets to the point where it is able to answer
link |
that question. But of course, at that point, there's the heat death of the universe has occurred,
link |
and that's the only entity. And there's nobody else to provide that answer to. So the only thing
link |
it can do is to answer it by demonstration. So it recreates the Big Bang and resets the clock.
link |
Right? But I can try to give a different version of the answer. Maybe not on the
link |
behalf of all humanity. I think that might be a little presumptuous for me to speak about the
link |
meaning of life on the behalf of all humans. But at least, personally, it changes. I think if
link |
you think about what gives you and your life meaning and purpose and what drives you,
link |
it seems to change over time in the lifespan of your existence. When you just enter this world,
link |
it's all about new experiences. You get new smells, new sounds, new emotions. And that's what's
link |
driving you. You're experiencing new amazing things. And that's magical. That's pretty awesome.
link |
That gives you kind of a meaning. Then you get a little bit older. You start more intentionally
link |
learning about things. I guess actually, before you start intentionally learning,
link |
probably fun. Fun is a thing that gives you kind of meaning and purpose and the thing you optimize
link |
for. And fun is good. Then you start learning. And I guess that this joy of
link |
comprehension and discovery is another thing that gives you meaning and purpose and drives you.
link |
You learn enough stuff and you want to give some of it back. So impact and contributions back to
link |
technology or society, people, local or more globally becomes a new thing that drives a lot
link |
of your behavior and something that gives you purpose and that you derive positive feedback
link |
from. Then you go on and so forth. You go through various stages of life. If you have kids,
link |
that definitely changes your perspective on things. I have three that definitely flips
link |
some bits in your head in terms of what you care about and what you optimize for and what
link |
matters, what doesn't matter. And so on and so forth. It seems to me that it's all of those
link |
things. As you go through life, you want these to be additive. New experiences, fun, learning
link |
impact. You want to be accumulating. I don't want to stop having fun or experiencing new things.
link |
And I think it's important that it just becomes additive as opposed to a replacement or subtraction.
link |
Those fewest problems as far as I got, but ask me in a few years, I might have one or two more
link |
to get to the list. And before you know it, time is up just like it is for this conversation.
link |
But hopefully it was a fun ride. It was a huge honor to meet you. As you know, I've been a
link |
fan of yours and a fan of Google self driving car and Waymo for a long time. I can't wait.
link |
I mean, it's one of the most exciting. If we look back in the 21st century, I truly believe
link |
it'll be one of the most exciting things we descendants of apes have created on this earth.
link |
So I'm a huge fan and I can't wait to see what you do next. Thanks so much for talking to me.
link |
Thanks. Thanks for having me. And it's also a huge fan.
link |
Thanks for listening to this conversation. For reading through summaries of books, better help,
link |
online therapy with a licensed professional and cash app, the app I use to send money to friends.
link |
Please check out these sponsors in the description to get a discount and to support this podcast.
link |
If you enjoy this thing, subscribe on YouTube, review it with five stars and upper podcast,
link |
follow on Spotify, support on Patreon or connect with me on Twitter at Lex Friedman.
link |
And now let me leave you with some words from Isaac Asimov. Science can amuse and
link |
fascinate us all, but it is engineering that changes the world. Thank you for listening
link |
and hope to see you next time.