episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Kyle Vogt: Cruise Automation | Lex Fridman Podcast #14
It'll be one of the most defining, the super exciting that you work on it.
https://karpathy.ai/lexicap/0014-large.html#00:54:50.560
Kyle Vogt: Cruise Automation | Lex Fridman Podcast #14
And, uh, the best of luck in 2018, I'm really excited to see what
https://karpathy.ai/lexicap/0014-large.html#00:54:53.680
Kyle Vogt: Cruise Automation | Lex Fridman Podcast #14
Cruz comes up with.
https://karpathy.ai/lexicap/0014-large.html#00:54:58.640
Kyle Vogt: Cruise Automation | Lex Fridman Podcast #14
Thank you.
https://karpathy.ai/lexicap/0014-large.html#00:54:59.680
Kyle Vogt: Cruise Automation | Lex Fridman Podcast #14
Thanks for having me today.
https://karpathy.ai/lexicap/0014-large.html#00:55:00.160
Kyle Vogt: Cruise Automation | Lex Fridman Podcast #14
Thanks, Carl.
https://karpathy.ai/lexicap/0014-large.html#00:55:01.040
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
The following is a conversation with Leslie Kaelbling. She is a roboticist and professor at
https://karpathy.ai/lexicap/0015-large.html#00:00:00.000
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
MIT. She is recognized for her work in reinforcement learning, planning, robot navigation, and several
https://karpathy.ai/lexicap/0015-large.html#00:00:05.360
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
other topics in AI. She won the IJCAI Computers and Thought Award and was the editor in chief
https://karpathy.ai/lexicap/0015-large.html#00:00:12.080
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
of the prestigious Journal of Machine Learning Research. This conversation is part of the
https://karpathy.ai/lexicap/0015-large.html#00:00:18.560
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
Artificial Intelligence podcast at MIT and beyond. If you enjoy it, subscribe on YouTube,
https://karpathy.ai/lexicap/0015-large.html#00:00:24.320
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
iTunes, or simply connect with me on Twitter at Lex Friedman, spelled F R I D.
https://karpathy.ai/lexicap/0015-large.html#00:00:30.400
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
And now, here's my conversation with Leslie Kaelbling.
https://karpathy.ai/lexicap/0015-large.html#00:00:36.960
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
What made me get excited about AI, I can say that, is I read Gödel Escher Bach when I was
https://karpathy.ai/lexicap/0015-large.html#00:00:42.800
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
in high school. That was pretty formative for me because it exposed the interestingness of
https://karpathy.ai/lexicap/0015-large.html#00:00:47.680
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
primitives and combination and how you can make complex things out of simple parts
https://karpathy.ai/lexicap/0015-large.html#00:00:57.200
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
and ideas of AI and what kinds of programs might generate intelligent behavior. So...
https://karpathy.ai/lexicap/0015-large.html#00:01:02.320
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
So you first fell in love with AI reasoning logic versus robots?
https://karpathy.ai/lexicap/0015-large.html#00:01:07.760
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
Yeah, the robots came because my first job, so I finished an undergraduate degree in philosophy
https://karpathy.ai/lexicap/0015-large.html#00:01:12.720
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
at Stanford and was about to finish a master's in computer science. And I got hired at SRI
https://karpathy.ai/lexicap/0015-large.html#00:01:18.160
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
in their AI lab and they were building a robot. It was a kind of a follow on to shaky,
https://karpathy.ai/lexicap/0015-large.html#00:01:25.360
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
but all the shaky people were not there anymore. And so my job was to try to get this robot to
https://karpathy.ai/lexicap/0015-large.html#00:01:30.960
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
do stuff. And that's really kind of what got me interested in robots.
https://karpathy.ai/lexicap/0015-large.html#00:01:35.840
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
So maybe taking a small step back to your bachelor's in Stanford in philosophy,
https://karpathy.ai/lexicap/0015-large.html#00:01:39.280
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
did master's and PhD in computer science, but the bachelor's in philosophy. So what was that
https://karpathy.ai/lexicap/0015-large.html#00:01:44.400
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
journey like? What elements of philosophy do you think you bring to your work in computer science?
https://karpathy.ai/lexicap/0015-large.html#00:01:49.440
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
So it's surprisingly relevant. So the part of the reason that I didn't do a computer
https://karpathy.ai/lexicap/0015-large.html#00:01:55.200
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
science undergraduate degree was that there wasn't one at Stanford at the time,
https://karpathy.ai/lexicap/0015-large.html#00:01:59.840
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
but that there's a part of philosophy and in fact, Stanford has a special submajor in
https://karpathy.ai/lexicap/0015-large.html#00:02:03.440
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
something called now symbolic systems, which is logic, model theory, formal semantics of
https://karpathy.ai/lexicap/0015-large.html#00:02:07.360
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
natural language. And so that's actually a perfect preparation for work in AI and computer science.
https://karpathy.ai/lexicap/0015-large.html#00:02:13.280
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
That's kind of interesting. So if you were interested in artificial intelligence,
https://karpathy.ai/lexicap/0015-large.html#00:02:20.080
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
what kind of majors were people even thinking about taking? What is it in your science?
https://karpathy.ai/lexicap/0015-large.html#00:02:26.000
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
So besides philosophies, what were you supposed to do if you were fascinated by the idea of creating
https://karpathy.ai/lexicap/0015-large.html#00:02:31.840
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
intelligence? There weren't enough people who did that for that even to be a conversation.
https://karpathy.ai/lexicap/0015-large.html#00:02:37.120
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
I mean, I think probably, probably philosophy. I mean, it's interesting in my class,
https://karpathy.ai/lexicap/0015-large.html#00:02:41.920
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
my graduating class of undergraduate philosophers, probably maybe slightly less than half went on in
https://karpathy.ai/lexicap/0015-large.html#00:02:48.320
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
computer science, slightly less than half went on in law and like one or two went on in philosophy.
https://karpathy.ai/lexicap/0015-large.html#00:02:56.800
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
So it was a common kind of connection. Do you think AI researchers have a role
https://karpathy.ai/lexicap/0015-large.html#00:03:02.320
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
to be part time philosophers or should they stick to the solid science and engineering
https://karpathy.ai/lexicap/0015-large.html#00:03:06.960
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
without sort of taking the philosophizing tangents? I mean, you work with robots,
https://karpathy.ai/lexicap/0015-large.html#00:03:11.440
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
you think about what it takes to create intelligent beings. Aren't you the perfect
https://karpathy.ai/lexicap/0015-large.html#00:03:16.240
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
person to think about the big picture philosophy at all? The parts of philosophy that are closest
https://karpathy.ai/lexicap/0015-large.html#00:03:21.200
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
to AI, I think, or at least the closest to AI that I think about are stuff like
https://karpathy.ai/lexicap/0015-large.html#00:03:25.920
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
belief and knowledge and denotation and that kind of stuff. And that's, you know,
https://karpathy.ai/lexicap/0015-large.html#00:03:29.680
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
it's quite formal. And it's like just one step away from the kinds of computer science work that
https://karpathy.ai/lexicap/0015-large.html#00:03:35.040
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
we do kind of routinely. I think that there are important questions still about what you can do
https://karpathy.ai/lexicap/0015-large.html#00:03:40.800
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
with a machine and what you can't and so on. Although at least my personal view is that I'm
https://karpathy.ai/lexicap/0015-large.html#00:03:50.160
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
completely a materialist. And I don't think that there's any reason why we can't make a robot be
https://karpathy.ai/lexicap/0015-large.html#00:03:54.560
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
behaviorally indistinguishable from a human. And the question of whether it's
https://karpathy.ai/lexicap/0015-large.html#00:04:00.960
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
distinguishable internally, whether it's a zombie or not in philosophy terms, I actually don't,
https://karpathy.ai/lexicap/0015-large.html#00:04:06.560
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
I don't know. And I don't know if I care too much about that.
https://karpathy.ai/lexicap/0015-large.html#00:04:12.720
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
Right. But there is a philosophical notions. They're mathematical and philosophical because
https://karpathy.ai/lexicap/0015-large.html#00:04:15.280
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
we don't know so much of how difficult it is. How difficult is the perception problem?
https://karpathy.ai/lexicap/0015-large.html#00:04:20.400
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
How difficult is the planning problem? How difficult is it to operate in this world successfully?
https://karpathy.ai/lexicap/0015-large.html#00:04:25.600
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
Because our robots are not currently as successful as human beings in many tasks.
https://karpathy.ai/lexicap/0015-large.html#00:04:30.800
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
The question about the gap between current robots and human beings borders a little bit
https://karpathy.ai/lexicap/0015-large.html#00:04:35.920
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
on philosophy. You know, the expanse of knowledge that's required to operate in a human world,
https://karpathy.ai/lexicap/0015-large.html#00:04:42.480
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
required to operate in this world and the ability to form common sense knowledge, the ability to
https://karpathy.ai/lexicap/0015-large.html#00:04:50.160
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
reason about uncertainty. Much of the work you've been doing, there's open questions there that,
https://karpathy.ai/lexicap/0015-large.html#00:04:55.520
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
I don't know, required to activate a certain big picture view.
https://karpathy.ai/lexicap/0015-large.html#00:05:02.880
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
To me, that doesn't seem like a philosophical gap at all. To me, there is a big technical gap.
https://karpathy.ai/lexicap/0015-large.html#00:05:07.840
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
There's a huge technical gap, but I don't see any reason why it's more than a technical gap.
https://karpathy.ai/lexicap/0015-large.html#00:05:12.960
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
Perfect. So, when you mentioned AI, you mentioned SRI, and maybe can you describe to me when you
https://karpathy.ai/lexicap/0015-large.html#00:05:19.360
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
first fell in love with robotics, with robots or inspired, so you mentioned Flaky or Shaky Flaky,
https://karpathy.ai/lexicap/0015-large.html#00:05:28.400
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
and what was the robot that first captured your imagination, what's possible?
https://karpathy.ai/lexicap/0015-large.html#00:05:38.400
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
Right. Well, so the first robot I worked with was Flaky. Shaky was a robot that the SRI
https://karpathy.ai/lexicap/0015-large.html#00:05:42.800
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
people had built, but by the time, I think when I arrived, it was sitting in a corner of somebody's
https://karpathy.ai/lexicap/0015-large.html#00:05:47.120
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
office dripping hydraulic fluid into a pan, but it's iconic and really everybody should read the
https://karpathy.ai/lexicap/0015-large.html#00:05:52.960
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
Shaky Tech Report because it has so many good ideas in it. I mean, they invented ASTAR search
https://karpathy.ai/lexicap/0015-large.html#00:06:00.240
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
and symbolic planning and learning macro operators. They had low level kind of
https://karpathy.ai/lexicap/0015-large.html#00:06:06.560
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
configuration space planning for their robot. They had vision. That's the basic ideas of
https://karpathy.ai/lexicap/0015-large.html#00:06:14.240
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
a ton of things. Can you take a step back? Shaky have arms. What was the job? Shaky was a mobile
https://karpathy.ai/lexicap/0015-large.html#00:06:19.360
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
robot, but it could push objects, and so it would move things around. With which actuator? With
https://karpathy.ai/lexicap/0015-large.html#00:06:26.240
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
itself, with its base. Okay, great. And they had painted the baseboards black, so it used vision
https://karpathy.ai/lexicap/0015-large.html#00:06:32.080
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
to localize itself in a map. It detected objects. It could detect objects that were surprising to
https://karpathy.ai/lexicap/0015-large.html#00:06:43.600
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
it. It would plan and replan based on what it saw. It reasoned about whether to look and take
https://karpathy.ai/lexicap/0015-large.html#00:06:49.360
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
pictures. I mean, it really had the basics of so many of the things that we think about now.
https://karpathy.ai/lexicap/0015-large.html#00:06:55.520
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
How did it represent the space around it? So it had representations at a bunch of different levels
https://karpathy.ai/lexicap/0015-large.html#00:07:03.280
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
of abstraction. So it had, I think, a kind of an occupancy grid of some sort at the lowest level.
https://karpathy.ai/lexicap/0015-large.html#00:07:08.960
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
At the high level, it was abstract symbolic kind of rooms and connectivity. So where does flaky
https://karpathy.ai/lexicap/0015-large.html#00:07:14.880
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
come in? Yeah, okay. So I showed up at SRI and we were building a brand new robot. As I said,
https://karpathy.ai/lexicap/0015-large.html#00:07:21.440
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
none of the people from the previous project were kind of there or involved anymore. So we were kind
https://karpathy.ai/lexicap/0015-large.html#00:07:28.240
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
of starting from scratch and my advisor was Stan Rosenstein. He ended up being my thesis advisor
https://karpathy.ai/lexicap/0015-large.html#00:07:33.200
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
and he was motivated by this idea of situated computation or situated automata. And the idea was
https://karpathy.ai/lexicap/0015-large.html#00:07:40.880
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
that the tools of logical reasoning were important, but possibly only for the engineers
https://karpathy.ai/lexicap/0015-large.html#00:07:49.600
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
or designers to use in the analysis of a system, but not necessarily to be manipulated in the head
https://karpathy.ai/lexicap/0015-large.html#00:07:58.480
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
of the system itself. So I might use logic to prove a theorem about the behavior of my robot,
https://karpathy.ai/lexicap/0015-large.html#00:08:04.560
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
even if the robot's not using logic in its head to prove theorems. So that was kind of the
https://karpathy.ai/lexicap/0015-large.html#00:08:10.480
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
distinction. And so the idea was to kind of use those principles to make a robot do stuff. But
https://karpathy.ai/lexicap/0015-large.html#00:08:14.400
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
a lot of the basic things we had to kind of learn for ourselves because I had zero background in
https://karpathy.ai/lexicap/0015-large.html#00:08:23.440
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
robotics. I didn't know anything about control. I didn't know anything about sensors. So we
https://karpathy.ai/lexicap/0015-large.html#00:08:29.600
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
reinvented a lot of wheels on the way to getting that robot to do stuff. Do you think that was
https://karpathy.ai/lexicap/0015-large.html#00:08:33.600
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
an advantage or a hindrance? Oh no, I mean, I'm big in favor of wheel reinvention actually. I mean,
https://karpathy.ai/lexicap/0015-large.html#00:08:37.280
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
I think you learn a lot by doing it. It's important though to eventually have the pointers
https://karpathy.ai/lexicap/0015-large.html#00:08:44.800
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
to, so that you can see what's really going on. But I think you can appreciate much better the
https://karpathy.ai/lexicap/0015-large.html#00:08:49.680
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
good solutions once you've messed around a little bit on your own and found a bad one.
https://karpathy.ai/lexicap/0015-large.html#00:08:56.640
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
Yeah. I think you mentioned reinventing reinforcement learning and referring to
https://karpathy.ai/lexicap/0015-large.html#00:09:00.400
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
rewards as pleasures, pleasure. Yeah. Or I think, which I think is a nice name for it.
https://karpathy.ai/lexicap/0015-large.html#00:09:04.880
Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15
Yeah. It's more fun almost. Do you think you could tell the history of AI machine learning
https://karpathy.ai/lexicap/0015-large.html#00:09:11.440