Name: Joe Rogan Experience #1696 - Lex Fridman
Uploaded: 2021-08-17T00:00:00.000Z
Description: Watch the stand-up comedy video 'Joe Rogan Experience #1696 - Lex Fridman' from PowerfulJRE on Comedy Buff. Published 2021-08-17. Transcript available.

Watch the stand-up comedy video 'Joe Rogan Experience #1696 - Lex Fridman' from PowerfulJRE on Comedy Buff. Published 2021-08-17. Transcript available.

Joe Rogan Experience #1696 - Lex Fridman

Hey, I saw that box that you have in your Instagram. Is that a robot?
Yeah, it's a robot.
That's what it said. Consciousness is not included. I'm like, uh, I see what
you're doing here.
What's in the box?
What's in the box, man? What's in the box?
That's a great movie, by the way.
It's a great movie. It's a dark movie.
Yeah, no, it's a, it's a legged robot and I've been involved with those a lot
recently and I'm going to explore, I was going to bring it here, but I thought
this is the wrong, the other robot I have is the wrong atmosphere.
Is it a Boston Dynamics one?
So I had a lot of, I've been closely working with Boston Dynamics and how do I
put it? I put a lot of my love into what they're doing for a few years. I love
the engineers there. We're close. We like each other.
But, I hear a but comment.
Let me politely say that, um, you know, they're also a company that are trying
to make money. And so there's a marketing team, there's PR, and they're
starting getting in the way of engineers.
And whenever marketing people get in the way of engineering, I'm out. And so
there's a lot of robotics companies. It was kind of heartbreaking for me
because how much I love that company.
In what way did they get in the way?
So, uh, very specifically, I'm interested in the problem of human-robot
interaction where there's this, uh, beautiful dance between a human and a robot.
The same way you have a dog that you love playing with. There's a magic there.
Like, I don't know.
There's an excitement, uh, when Marshall looks at you and looks away and then
looks at you again. And like, just that excitement, I want to understand how we
can engineer that into our AI systems.
So that's called human-robot interaction from a perspective of Boston Dynamics.
They want a machine that doesn't have anything to do with humans. They want a
machine that like, uh, um, patrols a factory looking for anything dangerous or
like, uh, does surveillance on a factory floor or helps in dangerous
environments where humans is too dangerous for humans.
So you want a robot to do the work so that you want always there to be a
distance between a human and a robot. For me, I'm interested in exploring when
human-robot close together.
And I think that's actually really important to understand for safety as well.
So a robot should be able to detect and predict the movement of humans really
well in order to avoid hurting them accidentally.
Like that's a robotics AI problem. You know, it's, uh, predicting the movement
of pedestrians, predicting the movement of humans, whether it's the human body
or the human hand on the factory floor, you have to understand, uh, the mind of
humans.
They don't move like, uh, billiard balls. They have, they move in unpredictable,
complicated ways or rather predict, predictable, but complicated ways.
And that's the problem of human-robot interaction. I think it's beautiful. Not
very many people are really studying it carefully. And I wanted to study it
carefully. Uh, one of the things I did, and you always learn this mistake the
hard way, is I asked for permission on everything.
Um, you know, the right way to do it is just to do it and apologize later.
So permission for like what kind of actions?
Uh, so permission for the kind of things I wanted to do. So I asked for
permission to on video touch spot.
Touch it.
Yeah. Like, like as a human. So, uh, I wanted spot to understand when a human
touches it and not and using only vision, uh, only cameras.
Right. Because there's no sensors that can detect touch, right?
Right. So have a sense where you, I mean, the, uh, where gestures and touch can
be part of the communication between the robot and the human.
And telling them, uh, the Boston Dynamics folks that made them very nervous.
They know I know what I'm doing, but they also know I, I, I'm starting to give
less and less fucks.
And they get nervous about that because, uh, they see the positivity. I, I, I
love people. I love robots. I want to show the positive stuff. I want to
inspire and educate the world.
But they're like, is there some evil thing he's going to do?
What do they think you're going to do?
Well, it's like, they only have stuff to lose is from their perspective. Okay.
From my perspective, I, I'm, I love Boston Dynamics. I want to show off some
awesome stuff.
Right.
Probably one of the greatest engineering feats in the space of robotics ever,
which is, uh, both Atlas, the, the bipedal robot and the spot, the legged
robots.
I, I don't know. From my perspective, it's a win-win from their perspective.
They got nervous.
This is why I'm, I'm, I'm, I don't understand why they would get nervous about
touching the robot. That doesn't make any sense to me because they're a
business.
They're trying to sell a lot of robots to companies that have zero interest in
touching.
I think the future of their robot requires them to understand what it's like
for a, uh, robot to be close to a human being.
So you're looking at it in terms of like a consumer product?
It's like, uh, in the gray area between industry and consumer. So like, you
know, uh, a factory worker is a kind of consumer.
So like more and more, I think Boston Dynamics need to pivot that way in order
to be a successful, profitable business.
They were currently purchased by, I think, Hyundai, like a large company.
So they're, I think, trying to justify their worth, all that kind of stuff.
I'm not sure how businesses work. All I know is when marketing people get in
the way of great
engineering, you run into problems. This is why, sorry to interrupt, why Tesla,
I'm a huge fan of what Elon is doing. He's not letting marketing get in the way
of great engineering.
There's no, there's very little marketing people at Tesla. And most of the ones
that do marketing,
first of all, Elon does great marketing on his Twitter, but also the Tesla
Twitter has run really well
marketing wise. They understand that to highlight, uh, the brilliance of a
particular product,
you have to highlight the engineering. You have to highlight the design. And
that means you don't
want to get in the way of the engineers and the designers.
Well, here's the best example of that. There's zero advertisements for Tesla.
Yeah.
And is the number one selling car in America.
There you go.
Yeah.
It's just brilliantly made. That's all you need to know. I mean, they, everyone
knows,
you drive it, you like, holy shit. You see it, it's wild. You see the giant
screen and
you realize that it can move that fast with no sound. It's, it's pretty
phenomenal stuff.
When you consider the fact that the most, I mean, think about most cars that
are sold in America,
you're inundated with advertisement, you know, whether it's Cadillac or, you
know,
Ford or, you know, whatever there, there's constant ads. They make great cars.
Don't get me wrong.
But Tesla doesn't have any ads and they sell more than anybody, which is pretty
incredible.
Create a product that people love and it's word of mouth from there.
I think it's also connected to Elon too. I don't think it would be as
successful if it wasn't
connected to such a bizarre Tony Stark type character because he's such a
unique cultural,
he's like a, he's an iconic figure. He's not, he's not a normal human. You know,
it's like,
he's our super genius.
Well, it's interesting to analyze them, right? Like, are all the memes about Doge
useful
or are they, are they productive or counterproductive? Like, is the humor
help advertise SpaceX and Tesla or does it destroy them? Is the funding secured
420 help or hurt?
Does him smoking weed and on your show and then saying, oops, uh, does that
help or hurt? I don't
know. It does not hurt. Yeah. It feels like it helps. How could it help? I mean,
how could it hurt
if the company's doing so well? Like you tell me you'd be doing even better?
That sounds crazy.
Well, a lot of most CEOs would say yes. They're wrong. Right. That's what the
sense is they're wrong,
but it's that means you have to rethink the way we run companies. You have to
be more open. You have
to have more fun. You have to be more crazy. But you have to be Elon Musk. So
that's the thing.
You can't just get another person to do it that way. You have to be that guy.
Being crazy is not enough.
Well, this whole thing is his, uh, uh, his love of space balls. Yeah. Right?
Yeah. I mean,
that's genuine. Yeah. Genuine love of space balls. The reason why he created
the flamethrower,
not a flamethrower. The reason why he created, uh, the new, uh, model S and
called it the plaid.
This is all space ball stuff. He even changed the shape of his rocket to look
more like space balls.
I mean, it doesn't even make sense. It's awesome. It's awesome. But again,
you're not going to recreate that artificially. Like you have, it has to
actually genuinely,
naturally, organically be that guy. And that's who he is. But you have to allow
people like that
to rise to the top in all spaces of leadership. Like the, I see in politics,
like, I don't know,
Andrew Yang, somebody like that. Somebody who's doesn't look like the past and
Elon Musk certainly
doesn't look like the past in terms of CEOs. By the way, I know you're not a
huge fan of autopilot
or don't use it very often, but they have, uh, cause I want to squeeze in AI a
little bit.
They have a autonomy day in a few days. Uh, Tesla does. And they're going to,
uh, they've been doing,
I don't know if you've been paying attention, but they've been doing a lot of
interesting stuff on the,
like, uh, semi-autonomous autonomous driving side. So one of the crazy things
they're doing
that I never thought would be possible is to use vision only for their FSD, for
their autopilot,
meaning cameras only talking about breaking with the ways of the past. So in
the past,
you always have radar. If you have more quote unquote, reliable sensors for
emergency situations,
for, um, for like, uh, emergency braking. So here Tesla is using only cameras.
So only vision.
The logic there is our roads are all designed for human eyes. Therefore you
should be able to drive
with only visual information. The problem is with rain, with fog, with all
those conditions, with night,
how are you going to be able to do it? And the fact that they're doing so, uh,
much more successful
than I would expect is, is quite incredible from an AI perspective. And they're
just going full steam
ahead there with, uh, something called Dojo, which is, um, I don't know how, if
you know how this whole
process works, but they, they're basically deploying a version of autopilot
software to a fleet of vehicles.
Those cars are driving sometime by humans, sometime by AI. And then whenever AI
runs into trouble,
that's a little data point that that sent back to the mothership. And then it
retrains the system
and the system gets smarter and smarter and then redeploys a smarter version.
So there's this loop.
They call it the data engine. This loop, it sounds maybe trivial, but it's one
of the first large scale
implementation of an AI system in the world that does this kind of loop. So it's
not oftentimes when
you deploy these kinds of systems, you deploy them and the time between
versions is like years. You have
to basically buy a new smartphone or something like that. Here, the time
between new versions is every week.
I think every Friday they're releasing a new version, something like that,
which is a, uh, um, you know,
it's a revolutionary idea. It sounds ridiculously simple, but it's
revolutionary in that as opposed
to deploying a perfect system, you deploy a system that's not perfect and then
it improves over time
and converges towards something that's safer than human drivers. That's
fascinating. And then the Dojo
computer is they're building their own huge system that's doing the training.
So to train neural networks
or AI systems, you have to have specialized hardware. So Elon, this is another
really interesting lesson
for companies, especially car companies, as opposed to outsourcing the work t