Dr. Lex Fridman: Machines, Creativity & Love

00:00:00.000 | - Welcome to the Huberman Lab Podcast,

00:00:02.280 | where we discuss science and science-based tools

00:00:04.900 | for everyday life.

00:00:05.900 | I'm Andrew Huberman,

00:00:10.420 | and I'm a professor of neurobiology and ophthalmology

00:00:13.220 | at Stanford School of Medicine.

00:00:15.020 | Today, I have the pleasure of introducing Dr. Lex Friedman

00:00:17.900 | as our guest on the Huberman Lab Podcast.

00:00:21.100 | Dr. Friedman is a researcher at MIT

00:00:23.180 | specializing in machine learning, artificial intelligence,

00:00:26.540 | and human-robot interactions.

00:00:29.340 | I must say that the conversation with Lex

00:00:31.820 | was without question,

00:00:33.800 | one of the most fascinating conversations

00:00:35.660 | that I've ever had, not just in my career,

00:00:37.700 | but in my lifetime.

00:00:39.440 | I knew that Lex worked on these topics,

00:00:41.500 | and I think many of you are probably familiar with Lex

00:00:43.940 | and his interest in these topics

00:00:45.180 | from his incredible podcast, the Lex Friedman Podcast.

00:00:48.260 | If you're not already watching that podcast,

00:00:50.240 | please subscribe to it, it is absolutely fantastic.

00:00:53.740 | But in holding this conversation with Lex,

00:00:56.040 | I realized something far more important.

00:00:58.620 | He revealed to us a bit of his dream,

00:01:00.900 | his dream about humans and robots,

00:01:03.360 | about humans and machines,

00:01:04.980 | and about how those interactions can change the way

00:01:07.520 | that we perceive ourselves

00:01:08.940 | and that we interact with the world.

00:01:10.740 | We discuss relationships of all kinds,

00:01:13.060 | relationships with animals, relationships with friends,

00:01:16.420 | relationships with family, and romantic relationships.

00:01:20.420 | And we discuss relationships with machines,

00:01:23.200 | machines that move and machines that don't move,

00:01:26.500 | and machines that come to understand us in ways

00:01:28.980 | that we could never understand for ourselves,

00:01:31.700 | and how those machines can educate us about ourselves.

00:01:35.720 | Before this conversation,

00:01:37.420 | I had no concept of the ways in which machines

00:01:40.060 | could inform me or anyone about themselves.

00:01:43.700 | By the end, I was absolutely taken with the idea,

00:01:46.780 | and I'm still taken with the idea

00:01:48.540 | that interactions with machines of a very particular kind,

00:01:51.920 | a kind that Lex understands and wants to bring to the world,

00:01:55.180 | can not only transform the self,

00:01:57.360 | but may very well transform humanity.

00:02:00.100 | So whether or not you're familiar

00:02:01.340 | with Dr. Lex Friedman or not,

00:02:03.320 | I'm certain you're going to learn a tremendous amount

00:02:05.340 | from him during the course of our discussion,

00:02:07.500 | and that it will transform the way

00:02:08.940 | that you think about yourself and about the world.

00:02:12.060 | Before we begin, I want to mention

00:02:13.860 | that this podcast is separate

00:02:15.100 | from my teaching and research roles at Stanford.

00:02:17.560 | It is, however, part of my desire and effort

00:02:19.940 | to bring zero cost to consumer information about science

00:02:22.740 | and science-related tools to the general public.

00:02:25.560 | In keeping with that theme,

00:02:26.720 | I'd like to thank the sponsors of today's podcast.

00:02:29.580 | Our first sponsor is Roca.

00:02:31.420 | Roca makes sunglasses and eyeglasses

00:02:33.340 | that are of absolutely phenomenal quality.

00:02:35.620 | The company was founded

00:02:36.500 | by two all-American swimmers from Stanford,

00:02:38.500 | and everything about the sunglasses and eyeglasses

00:02:40.760 | they've designed had performance in mind.

00:02:43.900 | Now, I've spent a career working on the visual system,

00:02:46.180 | and one of the fundamental issues

00:02:47.700 | that your visual system has to deal with

00:02:49.700 | is how to adjust what you see when it gets darker

00:02:52.780 | or brighter in your environment.

00:02:54.600 | With Roca sunglasses and eyeglasses,

00:02:56.820 | whether or not it's dim in the room or outside,

00:02:58.920 | whether or not there's cloud cover,

00:03:00.100 | or whether or not you walk into a shadow,

00:03:01.540 | you can always see the world with absolute clarity.

00:03:04.300 | And that just tells me that they really understand

00:03:06.220 | the way that the visual system works,

00:03:07.620 | processes like habituation and attenuation.

00:03:10.140 | All these things that work at a real mechanistic level

00:03:12.520 | have been built into these glasses.

00:03:14.540 | In addition, the glasses are very lightweight.

00:03:16.740 | You don't even notice really that they're on your face.

00:03:19.060 | And the quality of the lenses is terrific.

00:03:21.780 | Now, the glasses were also designed

00:03:23.500 | so that you could use them not just while working

00:03:25.460 | or at dinner, et cetera, but while exercising.

00:03:28.580 | They don't fall off your face or slip off your face

00:03:30.620 | if you're sweating.

00:03:31.800 | And as I mentioned, they're extremely lightweight,

00:03:33.460 | so you can use them while running,

00:03:34.860 | you can use them while cycling, and so forth.

00:03:37.060 | Also, the aesthetic of Roca glasses is terrific.

00:03:39.680 | Unlike a lot of performance glasses out there,

00:03:41.680 | which frankly make people look like cyborgs,

00:03:44.580 | these glasses look great.

00:03:46.020 | You can wear them out to dinner.

00:03:47.200 | You can wear them for essentially any occasion.

00:03:50.460 | If you'd like to try Roca glasses,

00:03:51.960 | you can go to roca.com, that's R-O-K-A.com,

00:03:55.440 | and enter the code Huberman to save 20% off

00:03:58.080 | your first order.

00:03:59.140 | That's Roca, R-O-K-A.com,

00:04:01.080 | and enter the code Huberman at checkout.

00:04:03.500 | Today's episode is also brought to us by InsideTracker.

00:04:06.880 | InsideTracker is a personalized nutrition platform

00:04:09.400 | that analyzes data from your blood and DNA

00:04:11.980 | to help you better understand your body

00:04:13.560 | and help you reach your health goals.

00:04:15.700 | I am a big believer in getting regular blood work done

00:04:18.220 | for the simple reason that many of the factors

00:04:20.700 | that impact our immediate and long-term health

00:04:23.180 | can only be assessed from a quality blood test.

00:04:25.820 | And now with the advent of quality DNA tests,

00:04:28.460 | we can also get insight into some of our genetic

00:04:30.900 | underpinnings of our current and long-term health.

00:04:34.300 | The problem with a lot of blood and DNA tests out there,

00:04:36.700 | however, is you get the data back

00:04:38.500 | and you don't know what to do with those data.

00:04:40.480 | You see that certain things are high

00:04:41.760 | or certain things are low,

00:04:43.000 | but you really don't know what the actionable items are,

00:04:45.200 | what to do with all that information.

00:04:47.300 | With InsideTracker, they make it very easy

00:04:49.740 | to act in the appropriate ways on the information

00:04:52.580 | that you get back from those blood and DNA tests.

00:04:55.060 | And that's through the use of their online platform.

00:04:57.580 | They have a really easy to use dashboard

00:04:59.860 | that tells you what sorts of things can bring the numbers

00:05:02.880 | for your metabolic factors, endocrine factors, et cetera,

00:05:05.740 | into the ranges that you want and need

00:05:08.040 | for immediate and long-term health.

00:05:10.140 | In fact, I know one individual just by way of example

00:05:13.100 | that was feeling good but decided to go

00:05:15.280 | with an InsideTracker test and discovered

00:05:16.960 | that they had high levels

00:05:17.960 | of what's called C-reactive protein.

00:05:19.800 | They would have never detected that otherwise.

00:05:21.820 | C-reactive protein is associated

00:05:23.360 | with a number of deleterious health conditions,

00:05:25.920 | some heart issues, eye issues, et cetera.

00:05:28.200 | And so they were able to take immediate action

00:05:30.160 | to try and resolve those CRP levels.

00:05:33.400 | And so with InsideTracker, you get that sort of insight.

00:05:35.980 | And as I mentioned before, without a blood or DNA test,

00:05:38.440 | there's no way you're going to get that sort of insight

00:05:40.360 | until symptoms start to show up.

00:05:42.840 | If you'd like to try InsideTracker,

00:05:44.300 | you can go to insidetracker.com/huberman

00:05:47.360 | to get 25% off any of InsideTracker's plans.

00:05:50.260 | You just use the code Huberman at checkout.

00:05:52.680 | That's insidetracker.com/huberman

00:05:55.420 | to get 25% off any of InsideTracker's plans.

00:05:58.700 | Today's podcast is brought to us by Athletic Greens.

00:06:01.560 | Athletic Greens is an all-in-one

00:06:03.200 | vitamin mineral probiotic drink.

00:06:05.640 | I started taking Athletic Greens way back in 2012.

00:06:09.020 | And so I'm delighted that they're sponsoring the podcast.

00:06:11.860 | The reason I started taking Athletic Greens

00:06:13.720 | and the reason I still take Athletic Greens

00:06:15.800 | is that it covers all of my vitamin mineral probiotic bases.

00:06:19.440 | In fact, when people ask me, what should I take?

00:06:22.000 | I always suggest that the first supplement people take

00:06:24.440 | is Athletic Greens for the simple reason

00:06:26.920 | is that the things it contains covers your bases

00:06:29.480 | for metabolic health, endocrine health,

00:06:31.880 | and all sorts of other systems in the body.

00:06:33.960 | And the inclusion of probiotics are essential

00:06:36.520 | for a healthy gut microbiome.

00:06:38.900 | There are now tons of data showing that

00:06:40.840 | we have neurons in our gut

00:06:42.820 | and keeping those neurons healthy

00:06:44.240 | requires that they are exposed

00:06:45.620 | to what are called the correct microbiota,

00:06:47.940 | little microorganisms that live in our gut

00:06:49.880 | and keep us healthy.

00:06:51.100 | And those neurons in turn help keep our brain healthy.

00:06:53.860 | They influence things like mood, our ability to focus,

00:06:56.500 | and many, many other factors related to health.

00:06:59.880 | With Athletic Greens, it's terrific

00:07:01.400 | because it also tastes really good.

00:07:03.160 | I drink it once or twice a day.

00:07:04.700 | I mix mine with water and I add a little lemon juice

00:07:07.040 | or sometimes a little bit of lime juice.

00:07:09.380 | If you want to try Athletic Greens,

00:07:11.080 | you can go to athleticgreens.com/huberman.

00:07:14.080 | And if you do that, you can claim their special offer.

00:07:16.680 | They're giving away five free travel packs,

00:07:18.520 | the little packs that make it easy to mix up

00:07:20.420 | Athletic Greens while you're on the road.

00:07:22.660 | And they'll give you a year supply of vitamin D3 and K2.

00:07:26.260 | Again, go to athleticgreens.com/huberman

00:07:28.940 | to claim that special offer.

00:07:30.880 | And now my conversation with Dr. Lex Friedman.

00:07:34.600 | - We meet again.

00:07:35.540 | - We meet again.

00:07:37.420 | Thanks so much for sitting down with me.

00:07:39.600 | I have a question that I think is on a lot

00:07:42.160 | of people's minds or ought to be on a lot of people's minds

00:07:46.400 | because we hear these terms a lot these days,

00:07:50.040 | but I think most people, including most scientists

00:07:53.100 | and including me, don't know really

00:07:56.900 | what is artificial intelligence and how is it different

00:08:01.900 | from things like machine learning and robotics?

00:08:05.240 | So if you would be so kind as to explain to us

00:08:08.900 | what is artificial intelligence and what is machine learning?

00:08:13.900 | - Well, I think that question is as complicated

00:08:17.100 | and as fascinating as the question of what is intelligence.

00:08:21.780 | So I think of artificial intelligence first

00:08:26.480 | as a big philosophical thing.

00:08:28.740 | Pamela McCormick said AI was,

00:08:32.460 | AI was the ancient wish to forge the gods

00:08:37.460 | or was born as an ancient wish to forge the gods.

00:08:41.740 | So I think at the big philosophical level,

00:08:44.260 | it's our longing to create other intelligence systems,

00:08:48.300 | perhaps systems more powerful than us.

00:08:50.400 | At the more narrow level, I think it's also a set of tools

00:08:56.740 | that are computational mathematical tools

00:08:59.260 | to automate different tasks.

00:09:01.100 | And then also it's our attempt to understand our own mind.

00:09:05.620 | So build systems that exhibit some intelligent behavior

00:09:09.940 | in order to understand what is intelligence

00:09:12.900 | in our own selves.

00:09:14.580 | So all those things are true.

00:09:16.160 | Of course, what AI really means as a community,

00:09:19.340 | as a set of researchers and engineers, it's a set of tools,

00:09:22.640 | a set of computational techniques

00:09:25.280 | that allow you to solve various problems.

00:09:29.560 | There's a long history that approaches the problem

00:09:33.000 | from different perspectives.

00:09:34.220 | What's always been throughout one of the threads,

00:09:37.640 | one of the communities goes under the flag

00:09:40.280 | of machine learning, which is emphasizing in the AI space,

00:09:45.280 | the task of learning.

00:09:48.120 | How do you make a machine that knows very little

00:09:50.520 | in the beginning, follow some kind of process

00:09:53.800 | and learns to become better and better in a particular task.

00:09:58.260 | What's been most very effective in the recent about 15 years

00:10:03.260 | is a set of techniques that fall under the flag

00:10:05.680 | of deep learning that utilize neural networks.

00:10:08.680 | What neural networks are, are these fascinating things

00:10:12.120 | inspired by the structure of the human brain very loosely,

00:10:17.120 | but they have, it's a network of these little basic

00:10:20.540 | computational units called neurons, artificial neurons.

00:10:24.400 | And they have, these architectures have an input and output.

00:10:28.720 | They know nothing in the beginning

00:10:30.120 | and they're tasked with learning something interesting.

00:10:33.500 | What that something interesting is

00:10:35.240 | usually involves a particular task.

00:10:37.580 | There's a lot of ways to talk about this and break this down.

00:10:41.960 | Like one of them is how much human supervision

00:10:45.880 | is required to teach this thing.

00:10:48.340 | So supervised learning is broad category

00:10:51.860 | is the neural network knows nothing in the beginning.

00:10:56.160 | And then it's given a bunch of examples of,

00:10:59.500 | in computer vision, that will be examples of cats,

00:11:02.020 | dogs, cars, traffic signs.

00:11:04.440 | And then you're given the image

00:11:06.080 | and you're given the ground truth of what's in that image.

00:11:09.560 | And when you get a large database of such image examples

00:11:13.040 | where you know the truth,

00:11:15.040 | the neural network is able to learn by example,

00:11:18.640 | that's called supervised learning.

00:11:20.480 | The question, there's a lot of fascinating questions

00:11:22.760 | within that, which is how do you provide the truth?

00:11:26.040 | When you've given an image of a cat,

00:11:28.780 | how do you provide to the computer

00:11:32.920 | that this image contains a cat?

00:11:34.920 | Do you just say the entire image is a picture of a cat?

00:11:37.980 | Do you do what's very commonly been done,

00:11:40.320 | which is a bounding box.

00:11:41.540 | You have a very crude box around the cat's face

00:11:45.040 | saying this is a cat.

00:11:46.520 | Do you do semantic segmentation?

00:11:48.720 | Mind you, this is a 2D image of a cat.

00:11:51.000 | So it's not a, the computer knows nothing

00:11:53.940 | about our three-dimensional world.

00:11:55.680 | It's just looking at a set of pixels.

00:11:57.340 | So semantic segmentation is drawing a nice,

00:12:01.120 | very crisp outline around the cat and saying that's a cat.

00:12:04.920 | That's really difficult to provide that truth.

00:12:07.080 | And one of the fundamental open questions

00:12:09.480 | in computer vision is,

00:12:10.800 | is that even a good representation of the truth?

00:12:13.800 | Now there's another contrasting set of ideas

00:12:18.560 | that are attention, that are overlapping

00:12:21.120 | is what's used to be called unsupervised learning,

00:12:24.340 | what's commonly now called self-supervised learning,

00:12:27.140 | which is trying to get less and less and less

00:12:29.840 | human supervision into the task.

00:12:33.960 | So self-supervised learning is more,

00:12:38.740 | it's been very successful in the domain of a language model,

00:12:42.000 | natural language processing,

00:12:43.180 | and now more and more is being successful

00:12:44.960 | in computer vision tasks.

00:12:46.480 | And what's the idea there is let the machine

00:12:49.640 | without any ground truth annotation,

00:12:53.160 | just look at pictures on the internet

00:12:55.740 | or look at texts on the internet

00:12:57.600 | and try to learn something generalizable about the ideas

00:13:02.600 | that are at the core of language or at the core of vision.

00:13:07.200 | And based on that, we humans at its best

00:13:11.620 | like to call that common sense.

00:13:12.900 | So with this, we have this giant base of knowledge

00:13:16.000 | on top of which we build more sophisticated knowledge,

00:13:18.720 | but we have this kind of common sense knowledge.

00:13:21.400 | And so the idea with self-supervised learning

00:13:23.380 | is to build this common sense knowledge

00:13:26.000 | about what are the fundamental visual ideas

00:13:30.440 | that make up a cat and a dog and all those kinds of things

00:13:33.280 | without ever having human supervision.

00:13:35.780 | The dream there is you just let an AI system

00:13:40.780 | that's self-supervised run around the internet for a while,

00:13:44.840 | watch YouTube videos for millions and millions of hours

00:13:47.520 | and without any supervision be primed and ready

00:13:52.080 | to actually learn with very few examples

00:13:54.640 | once the human is able to show up.

00:13:56.680 | We think of children in this way, human children,

00:14:00.160 | is your parents only give one or two examples

00:14:03.040 | to teach a concept.

00:14:04.600 | The dream of self-supervised learning

00:14:07.040 | is that would be the same with machines,

00:14:10.080 | that they would watch millions of hours of YouTube videos

00:14:13.960 | and then come to a human and be able to understand

00:14:16.760 | when the human shows them, this is a cat.

00:14:19.640 | Remember this is a cat.

00:14:20.800 | They will understand that a cat is not just a thing

00:14:23.540 | with pointy ears or a cat is a thing that's orange

00:14:27.520 | or is furry, they'll see something more fundamental

00:14:30.800 | that we humans might not actually be able to introspect

00:14:33.480 | and understand.

00:14:34.400 | Like if I asked you what makes a cat versus a dog,

00:14:36.800 | you wouldn't probably not be able to answer that.

00:14:39.400 | But if I showed you, brought to you a cat and a dog,

00:14:42.760 | you'll be able to tell the difference.

00:14:44.400 | What are the ideas that your brain uses

00:14:47.060 | to make that difference?

00:14:48.800 | That's the whole dream of self-supervised learning

00:14:51.280 | is it would be able to learn that on its own,

00:14:53.880 | that set of common sense knowledge

00:14:56.080 | that's able to tell the difference.

00:14:57.880 | And then there's like a lot of incredible uses

00:15:01.600 | of self-supervised learning very weirdly

00:15:05.040 | called self-play mechanism.

00:15:07.280 | That's the mechanism behind the reinforcement learning

00:15:11.740 | successes of the systems that want to go at alpha zero,

00:15:16.740 | that want to chess.

00:15:18.840 | - Oh, I see, that play games.

00:15:20.840 | - That play games. - Got it.

00:15:21.880 | - So the idea of self-play, this probably applies

00:15:26.000 | to other domains than just games,

00:15:27.880 | is a system that just plays against itself.

00:15:30.860 | And this is fascinating in all kinds of domains,

00:15:33.580 | but it knows nothing in the beginning.

00:15:36.600 | And the whole idea is it creates a bunch of mutations

00:15:39.540 | of itself and plays against those versions of itself.

00:15:44.540 | And the fascinating thing is when you play against systems

00:15:50.220 | that are a little bit better than you,

00:15:51.820 | you start to get better yourself.

00:15:53.720 | Like learning, that's how learning happens.

00:15:56.100 | That's true for martial arts, that's true in a lot of cases

00:15:59.240 | where you want to be interacting with systems

00:16:02.040 | that are just a little better than you.

00:16:03.880 | And then through this process of interacting with systems

00:16:06.800 | just a little better than you,

00:16:08.120 | you start following this process where everybody

00:16:10.480 | starts getting better and better and better and better

00:16:12.620 | until you are several orders of magnitude better

00:16:15.480 | than the world champion in chess, for example.

00:16:18.080 | And it's fascinating 'cause it's like a runaway system.

00:16:21.040 | One of the most terrifying and exciting things

00:16:23.560 | that David Silver, the creator of AlphaGo and AlphaZero,

00:16:27.200 | one of the leaders of the team said to me

00:16:31.440 | is they haven't found the ceiling for AlphaZero,

00:16:36.640 | meaning it could just arbitrarily keep improving.

00:16:39.360 | Now in the realm of chess, that doesn't matter to us

00:16:41.840 | that it's like it just ran away with the game of chess.

00:16:44.980 | Like it's like just so much better than humans.

00:16:48.360 | But the question is if you can create that in the realm

00:16:52.120 | that does have a bigger, deeper effect on human beings,

00:16:56.060 | on societies, that can be a terrifying process.

00:16:59.940 | To me, it's an exciting process

00:17:01.760 | if you supervise it correctly.

00:17:03.720 | If you inject what's called value alignment,

00:17:08.720 | you make sure that the goals that the AI is optimizing

00:17:13.500 | is aligned with human beings and human societies.

00:17:17.000 | There's a lot of fascinating things to talk about

00:17:19.200 | within the specifics of neural networks

00:17:23.200 | and all the problems that people are working on.

00:17:25.600 | But I would say the really big exciting one

00:17:28.300 | is self-supervised learning,

00:17:29.840 | where trying to get less and less human supervision

00:17:33.840 | of neural networks.

00:17:38.760 | And also just to comment and I'll shut up.

00:17:41.980 | - No, please keep going.

00:17:43.100 | I'm learning.

00:17:44.120 | I have questions, but I'm learning, so please keep going.

00:17:46.480 | - So that to me, what's exciting is not the theory,

00:17:49.420 | it's always the application.

00:17:51.240 | One of the most exciting applications

00:17:52.920 | of artificial intelligence,

00:17:55.120 | specifically neural networks and machine learning

00:17:57.480 | is Tesla Autopilot.

00:17:59.140 | So these are systems that are working in the real world.

00:18:01.600 | This isn't an academic exercise.

00:18:03.600 | This is human lives at stake.

00:18:05.200 | This is safety critical.

00:18:07.200 | - These are automated vehicles, autonomous vehicles.

00:18:09.480 | - Semi-autonomous, we've gone through wars on these topics.

00:18:14.480 | - Semi-autonomous.

00:18:16.200 | - Semi-autonomous.

00:18:17.040 | So even though it's called FSD, full self-driving,

00:18:22.040 | it is currently not fully autonomous,

00:18:24.960 | meaning human supervision is required.

00:18:27.740 | So human is tasked with overseeing the systems.

00:18:30.920 | In fact, liability wise, the human is always responsible.

00:18:35.240 | This is a human factor psychology question,

00:18:37.960 | which is fascinating.

00:18:39.360 | I'm fascinated by the whole space,

00:18:43.000 | which is a whole nother space of human robot interaction.

00:18:46.160 | When AI systems and humans work together

00:18:48.760 | to accomplish tasks.

00:18:49.960 | That dance to me is one of the smaller communities,

00:18:54.960 | but I think it will be one of the most important

00:18:58.280 | open problems once they're solved,

00:19:00.340 | is how do humans and robots dance together.

00:19:03.160 | To me, semi-autonomous driving is one of those spaces.

00:19:07.860 | So for Elon, for example, he doesn't see it that way.

00:19:11.680 | He sees semi-autonomous driving as a stepping stone

00:19:16.520 | towards fully autonomous driving.

00:19:18.320 | Humans and robots can't dance well together.

00:19:22.720 | Like humans and humans dance and robots and robots dance.

00:19:25.680 | We need to, this is an engineering problem,

00:19:28.060 | we need to design a perfect robot that solves this problem.

00:19:31.680 | To me forever, maybe this is not the case with driving,

00:19:34.120 | but the world is going to be full of problems

00:19:37.140 | where it's always humans and robots have to interact.

00:19:40.380 | Because I think robots will always be flawed,

00:19:43.520 | just like humans are going to be flawed, are flawed.

00:19:47.460 | And that's what makes life beautiful, that they're flawed.

00:19:51.000 | That's where learning happens

00:19:52.360 | at the edge of your capabilities.

00:19:55.860 | So you always have to figure out how can flawed robots

00:20:00.320 | and flawed humans interact together

00:20:03.800 | such that the sum is bigger than the whole,

00:20:08.360 | as opposed to focusing on just building the perfect robot.

00:20:11.280 | So that's one of the most exciting applications,

00:20:15.360 | I would say, of artificial intelligence to me,

00:20:17.800 | is autonomous driving and semi-autonomous driving.

00:20:20.640 | And that's a really good example of machine learning,

00:20:23.240 | because those systems are constantly learning.

00:20:26.240 | And there's a process there that maybe I can comment on.

00:20:30.600 | Andrej Karpathy, who's the head of Autopilot,

00:20:34.360 | calls it the data engine.

00:20:36.360 | And this process applies for a lot of machine learning,

00:20:38.900 | which is you build a system

00:20:40.340 | that's pretty good at doing stuff,

00:20:42.200 | you send it out into the real world,

00:20:45.340 | it starts doing the stuff,

00:20:46.820 | and then it runs into what are called edge cases,

00:20:49.180 | like failure cases, where it screws up.

00:20:52.660 | You know, we do this as kids, that's, you know, you have-

00:20:55.520 | - We do this as adults.

00:20:56.360 | - We do this as adults, exactly.

00:20:58.580 | But we learn really quickly.

00:21:00.140 | But the whole point,

00:21:01.340 | and this is the fascinating thing about driving,

00:21:03.660 | is you realize there's millions of edge cases.

00:21:06.120 | There's just like weird situations that you did not expect.

00:21:10.780 | And so the data engine process is

00:21:13.480 | you collect those edge cases,

00:21:15.100 | and then you go back to the drawing board

00:21:17.080 | and learn from them.

00:21:18.440 | And so you have to create this data pipeline

00:21:20.980 | where all these cars, hundreds of thousands of cars,

00:21:23.940 | as you're driving around, and something weird happens.

00:21:27.020 | And so whenever this weird detector fires,

00:21:30.900 | it's another important concept,

00:21:33.380 | that piece of data goes back to the mothership

00:21:37.420 | for the training, for the retraining of the system.

00:21:40.860 | And through this data engine process,

00:21:42.620 | it keeps improving and getting better and better

00:21:44.940 | and better and better.

00:21:45.860 | So basically, you send out a pretty clever AI systems

00:21:49.340 | out into the world,

00:21:50.900 | and let it find the edge cases.

00:21:54.420 | Let it screw up just enough

00:21:56.580 | to figure out where the edge cases are,

00:21:58.560 | and then go back and learn from them,

00:22:00.800 | and then send out that new version

00:22:02.620 | and keep updating that version.

00:22:04.220 | - Is the updating done by humans?

00:22:06.320 | - The annotation is done by humans.

00:22:08.520 | So you have to, the weird examples come back,

00:22:13.420 | the edge cases,

00:22:14.800 | and you have to label what actually happened in there.

00:22:17.700 | There's also some mechanisms for automatically labeling,

00:22:22.700 | but mostly I think you always have to rely on humans

00:22:25.860 | to improve, to understand what's happening

00:22:28.100 | in the weird cases.

00:22:29.980 | And then there's a lot of debate.

00:22:31.780 | And that's the other thing,

00:22:32.660 | what is artificial intelligence?

00:22:34.460 | Which is a bunch of smart people

00:22:36.780 | having very different opinions about what is intelligence.

00:22:39.740 | So AI is basically a community of people

00:22:41.900 | who don't agree on anything.

00:22:43.980 | - It seems to be the case.

00:22:45.820 | And first of all, this is a beautiful description of terms

00:22:48.660 | that I've heard many times among my colleagues at Stanford,

00:22:51.900 | at meetings, in the outside world.

00:22:53.760 | And there's so many fascinating things.

00:22:55.940 | I have so many questions,

00:22:56.940 | but I do want to ask one question about the culture of AI,

00:23:00.220 | because it does seem to be a community where,

00:23:02.620 | at least as an outsider,

00:23:03.980 | where it seems like there's very little consensus

00:23:06.140 | about what the terms

00:23:07.300 | and the operational definitions even mean.

00:23:09.620 | And there seems to be a lot of splitting happening now

00:23:12.340 | of not just supervised and unsupervised learning,

00:23:14.900 | but these sort of intermediate conditions

00:23:18.060 | where machines are autonomous,

00:23:20.780 | but then go back for more instruction.

00:23:22.100 | Like kids go home from college during the summer

00:23:24.020 | and get a little, you know, mom still feeds them.

00:23:26.220 | Then eventually they leave the nest kind of thing.

00:23:28.680 | Is there something in particular about engineers

00:23:32.660 | or about people in this realm of engineering

00:23:35.820 | that you think lends itself to disagreement?

00:23:39.060 | - Yeah, I think so first of all,

00:23:42.180 | the more specific you get, the less disagreement there is.

00:23:44.640 | So there's a lot of disagreement

00:23:45.900 | about what is artificial intelligence,

00:23:47.860 | but there's less disagreement about what is machine learning

00:23:50.620 | and even less when you talk about active learning

00:23:52.680 | or machine teaching or self-supervised learning.

00:23:56.580 | And then when you get into like NLP language models

00:23:59.780 | or transformers,

00:24:00.620 | when you get into specific neural network architectures,

00:24:03.780 | there's less and less and less disagreement

00:24:05.620 | about those terms.

00:24:06.660 | So you might be hearing the disagreement

00:24:08.060 | from the high level terms.

00:24:09.340 | And that has to do with the fact that engineering,

00:24:12.060 | especially when you're talking about intelligent systems

00:24:15.580 | is a little bit of an art and a science.

00:24:20.580 | So the art part is the thing that creates disagreements

00:24:25.320 | because then you start having disagreements

00:24:28.620 | about how easy or difficult the particular problem is.

00:24:33.620 | For example, a lot of people disagree with Elon,

00:24:37.580 | how difficult the problem of autonomous driving is.

00:24:41.120 | And so, but nobody knows.

00:24:43.120 | So there's a lot of disagreement

00:24:44.340 | about what are the limits of these techniques.

00:24:47.160 | And through that, the terminology also contains within it,

00:24:50.560 | the disagreements.

00:24:53.940 | But overall, I think it's also a young science

00:24:56.760 | that also has to do with that.

00:24:58.820 | So like, it's not just engineering,

00:25:01.180 | it's that artificial intelligence truly

00:25:03.960 | as a large scale discipline where it's thousands,

00:25:06.800 | tens of thousands, hundreds of thousands of people

00:25:09.420 | working on it, huge amounts of money being made.

00:25:11.660 | That's a very recent thing.

00:25:13.820 | So we're trying to figure out those terms.

00:25:16.460 | And of course there's egos and personalities

00:25:18.860 | and a lot of fame to be made,

00:25:20.940 | like the term deep learning, for example.

00:25:25.740 | Neural networks have been around for many, many decades,

00:25:28.040 | since the '60s, you can argue since the '40s.

00:25:30.860 | So there was a rebranding of neural networks

00:25:33.440 | into the word deep learning, term deep learning,

00:25:36.880 | that was part of the reinvigoration of the field,

00:25:40.920 | but it's really the same exact thing.

00:25:42.680 | - I didn't know that.

00:25:43.640 | I mean, I grew up in the age of neuroscience

00:25:46.000 | when neural networks were discussed,

00:25:49.060 | computational neuroscience and theoretical neuroscience,

00:25:51.360 | they had their own journals.

00:25:53.060 | It wasn't actually taken terribly seriously

00:25:55.020 | by experimentalists until a few years ago,

00:25:57.220 | I would say about five to seven years ago,

00:26:00.680 | excellent theoretical neuroscientists like Larry Abbott

00:26:03.520 | and other colleagues, certainly at Stanford as well,

00:26:07.460 | that people started paying attention

00:26:08.780 | to computational methods.

00:26:10.340 | But these terms, neural networks, computational methods,

00:26:13.200 | I actually didn't know that neural network works

00:26:15.140 | in deep learning, where those have now become

00:26:18.540 | kind of synonymous.

00:26:19.380 | - No, they were always, no, they're always the same thing.

00:26:22.700 | - Interesting.

00:26:24.160 | - I'm a neuroscientist and I didn't know that.

00:26:25.740 | - So, well, because neural networks probably

00:26:28.060 | mean something else in neuroscience, not something else,

00:26:30.180 | but a little different flavor depending on the field.

00:26:32.620 | And that's fascinating too, because neuroscience and AI,

00:26:36.700 | people have started working together and dancing a lot more

00:26:40.680 | in the recent, I would say probably decade.

00:26:43.020 | - Oh, machines are going into the brain.

00:26:45.020 | I have a couple of questions,

00:26:47.600 | but one thing that I'm sort of fixated on

00:26:49.800 | that I find incredibly interesting is this example you gave

00:26:54.400 | of playing a game with a mutated version of yourself

00:26:58.200 | as a competitor.

00:26:59.420 | - Yeah.

00:27:00.260 | - I find that incredibly interesting as a kind of a parallel

00:27:04.260 | or a mirror for what happens when we try and learn

00:27:06.700 | as humans, which is we generate repetitions

00:27:09.980 | of whatever it is we're trying to learn

00:27:11.920 | and we make errors, occasionally we succeed.

00:27:15.020 | In a simple example, for instance,

00:27:16.580 | of trying to throw bull's eyes on a dartboard.

00:27:18.980 | I'm going to have errors, errors, errors.

00:27:20.460 | I'll probably miss the dartboard

00:27:21.680 | and maybe occasionally hit a bull's eye.

00:27:23.380 | And I don't know exactly what I just did, right?

00:27:26.220 | But then let's say I was playing darts

00:27:28.720 | against a version of myself

00:27:30.340 | where I was wearing a visual prism,

00:27:32.460 | like my visual, I had a visual defect.

00:27:34.840 | You learn certain things in that mode as well.

00:27:38.940 | You're saying that a machine can sort of mutate itself.

00:27:42.880 | Does the mutation always cause a deficiency

00:27:45.100 | that it needs to overcome?

00:27:46.540 | Because mutations in biology

00:27:47.980 | sometimes give us superpowers, right?

00:27:49.580 | Occasionally you'll get somebody

00:27:51.060 | who has better than 20/20 vision

00:27:52.700 | and they can see better than 99.9% of people out there.

00:27:56.420 | So when you talk about a machine playing a game

00:27:59.120 | against a mutated version of itself,

00:28:01.320 | is the mutation always what we call a negative mutation

00:28:04.720 | or an adaptive or a maladaptive mutation?

00:28:07.600 | - No, you don't know until you mutate first

00:28:11.860 | and then figure out and they compete against each other.

00:28:14.460 | - So you're evolving,

00:28:15.780 | the machine gets to evolve itself in real time.

00:28:18.360 | - Yeah, and I think of it, which would be exciting,

00:28:21.800 | if you could actually do with humans.

00:28:23.600 | It's not just, so usually you freeze a version of the system.

00:28:28.600 | So really you take an Andrew of yesterday

00:28:33.920 | and you make 10 clones of them

00:28:35.640 | and then maybe you mutate, maybe not.

00:28:39.000 | And then you do a bunch of competitions

00:28:41.080 | of the Andrew of today.

00:28:42.480 | Like you fight to the death and who wins last.

00:28:45.640 | So I love that idea of creating a bunch of clones of myself

00:28:49.000 | from each of the day for the past year

00:28:52.360 | and just seeing who's going to be better

00:28:54.560 | at like podcasting or science or picking up chicks at a bar

00:28:58.860 | or I don't know, or competing in jujitsu.

00:29:02.040 | That's one way to do it.

00:29:03.120 | I mean, a lot of Lexes would have to die for that process,

00:29:06.360 | but that's essentially what happens

00:29:07.900 | is in reinforcement learning

00:29:09.440 | through the self-playing mechanisms,

00:29:11.480 | it's a graveyard of systems that didn't do that well.

00:29:14.760 | And the surviving, the good ones survive.

00:29:19.760 | - Do you think that, I mean, Darwin's theory of evolution

00:29:22.800 | might have worked in some sense in this way,

00:29:26.300 | but at the population level?

00:29:27.720 | I mean, you get a bunch of birds with different shaped beaks

00:29:29.560 | and some birds have the shape beak

00:29:30.960 | that allows them to get the seeds.

00:29:32.280 | I mean, it's a trivially simple example

00:29:34.880 | of Darwinian evolution, but I think it's correct

00:29:39.280 | even though it's not exhaustive.

00:29:40.800 | Is that what you're referring to?

00:29:42.140 | You essentially that normally this is done

00:29:44.080 | between members of a different species.

00:29:45.560 | Lots of different members of species have different traits

00:29:47.880 | and some get selected for,

00:29:49.440 | but you could actually create multiple versions of yourself

00:29:52.580 | with different traits.

00:29:53.880 | - So with, I should probably have said this,

00:29:56.440 | but perhaps it's implied,

00:29:59.200 | but with machine learning or with reinforcement learning

00:30:01.200 | through these processes,

00:30:02.480 | one of the big requirements is to have an objective function,

00:30:05.660 | a loss function, a utility function.

00:30:07.800 | Those are all different terms for the same thing.

00:30:10.200 | Is there's like an equation that says what's good.

00:30:15.080 | And then you're trying to optimize that equation.

00:30:17.500 | So there's a clear goal for these systems.

00:30:20.980 | - Because it's a game, like with chess, there's a goal.

00:30:23.920 | - But for anything,

00:30:25.280 | anything you want machine learning to solve,

00:30:27.660 | there needs to be an objective function.

00:30:29.940 | In machine learning, it's usually called loss function

00:30:32.880 | that you're optimizing.

00:30:34.340 | The interesting thing about evolution complicated of course,

00:30:38.700 | but the goal also seems to be evolving.

00:30:41.700 | Like it's a, I guess adaptation to the environment

00:30:44.020 | is the goal, but it's unclear.

00:30:46.460 | You can convert that always.

00:30:48.600 | It's like survival of the fittest.

00:30:52.120 | It's unclear what the fittest is.

00:30:53.840 | In machine learning, the starting point,

00:30:56.800 | and this is like what human ingenuity provides,

00:31:00.460 | is that fitness function of what's good and what's bad,

00:31:04.420 | which it lets you know which of the systems is going to win.

00:31:08.340 | So you need to have a equation like that.

00:31:10.920 | One of the fascinating things about humans

00:31:12.840 | is we figure out objective functions for ourselves.

00:31:17.160 | Like we're, it's the meaning of life.

00:31:20.500 | Like why the hell are we here?

00:31:22.960 | And a machine currently has to have

00:31:26.620 | a hard-coded statement about why.

00:31:29.240 | - It has to have a meaning

00:31:30.240 | of artificial intelligence-based life.

00:31:33.240 | - Right, it can't.

00:31:34.520 | So like there's a lot of interesting explorations

00:31:37.580 | about that function being more about curiosity,

00:31:42.400 | about learning new things and all that kind of stuff,

00:31:45.240 | but it's still hard-coded.

00:31:46.680 | If you want a machine to be able to be good at stuff,

00:31:49.600 | it has to be given very clear statements

00:31:53.440 | of what good at stuff means.

00:31:56.080 | That's one of the challenges of artificial intelligence

00:31:58.360 | is you have to formalize the,

00:32:01.600 | in order to solve a problem, you have to formalize it

00:32:04.200 | and you have to provide

00:32:06.080 | both like the full sensory information.

00:32:08.240 | You have to be very clear about

00:32:10.040 | what is the data that's being collected.

00:32:12.600 | And you have to also be clear about the objective function.

00:32:15.860 | What is the goal that you're trying to reach?

00:32:18.840 | And that's a very difficult thing

00:32:20.720 | for artificial intelligence.

00:32:22.100 | - I love that you mentioned curiosity.

00:32:23.940 | I'm sure this definition falls short in many ways,

00:32:26.920 | but I define curiosity as a strong interest

00:32:31.200 | in knowing something,

00:32:33.160 | but without an attachment to the outcome.

00:32:35.680 | You know, it's sort of a, it's not,

00:32:38.040 | it could be a random search,

00:32:39.320 | but there's not really an emotional attachment.

00:32:42.040 | It's really just a desire to discover

00:32:44.040 | and unveil what's there

00:32:45.520 | without hoping it's a gold coin under a rock.

00:32:48.880 | You're just looking under rocks.

00:32:50.640 | Is that more or less how the machine,

00:32:52.800 | you know, within machine learning,

00:32:54.000 | it sounds like there are elements of reward prediction

00:32:57.280 | and, you know, rewards.

00:32:58.560 | The machine has to know when it's done the right thing.

00:33:01.440 | So can you make machines that are curious

00:33:05.280 | or are the sorts of machines that you are describing

00:33:08.000 | curious by design?

00:33:09.360 | - Yeah, curiosity is a kind of a symptom, not the goal.

00:33:15.200 | So what happens is one of the big trade-offs

00:33:21.200 | in reinforcement learning

00:33:22.280 | is this exploration versus exploitation.

00:33:25.540 | So when you know very little,

00:33:27.400 | it pays off to explore a lot,

00:33:29.600 | even suboptimal, like even trajectories

00:33:32.440 | that seem like they're not going to lead anywhere.

00:33:34.620 | That's called exploration.

00:33:36.180 | The smarter and smarter and smarter you get,

00:33:38.600 | the more emphasis you put on exploitation,

00:33:41.820 | meaning you take the best solution.

00:33:44.180 | You take the best path.

00:33:45.780 | Now through that process,

00:33:47.220 | the exploration can look like curiosity by us humans,

00:33:52.220 | but it's really just trying to get out of the local optimal,

00:33:55.600 | the thing that's already discovered.

00:33:57.300 | It's from an AI perspective,

00:34:00.180 | it's always looking to optimize the objective function.

00:34:04.380 | It derives, and we could talk about this a lot more,

00:34:08.160 | but in terms of the tools of machine learning today,

00:34:11.520 | it derives no pleasure from just the curiosity of like,

00:34:16.520 | I don't know, discovery.

00:34:19.320 | - So there's no dopamine for a machine.

00:34:20.840 | - There's no dopamine.

00:34:21.820 | - There's no reward system, chemical,

00:34:24.000 | or I guess electronic reward system.

00:34:26.880 | - That said, if you look at machine learning literature

00:34:30.360 | and reinforcement learning literature,

00:34:32.080 | they will use like deep mind,

00:34:34.020 | we use terms like dopamine.

00:34:35.740 | We're constantly trying to use the human brain

00:34:38.820 | to inspire totally new solutions to these problems.

00:34:41.820 | So they'll think like,

00:34:42.720 | how does dopamine function in the human brain?

00:34:44.940 | And how can that lead to more interesting ways

00:34:49.120 | to discover optimal solutions?

00:34:51.460 | But ultimately, currently,

00:34:53.200 | there has to be a formal objective function.

00:34:57.460 | Now you could argue that humans

00:34:58.660 | also has a set of objective functions

00:35:00.460 | we're trying to optimize.

00:35:01.860 | We're just not able to introspect them.

00:35:04.500 | - Yeah, we don't actually know what we're looking for

00:35:07.800 | and seeking and doing.

00:35:09.320 | - Well, like Lisa Feldman Barrett,

00:35:10.680 | who you've spoken with, at least on Instagram,

00:35:13.420 | I hope you--

00:35:14.260 | - I met her through you, yeah.

00:35:15.080 | - Yeah, I hope you actually have her on this podcast.

00:35:17.580 | - She's terrific.

00:35:18.800 | - So she has a very,

00:35:20.660 | it has to do with homeostasis, like that,

00:35:25.920 | basically there's a very dumb objective function

00:35:28.980 | that the brain is trying to optimize,

00:35:30.860 | like to keep like body temperature the same.

00:35:32.940 | Like there's a very dumb

00:35:34.300 | kind of optimization function happening.

00:35:36.460 | And then what we humans do with our fancy consciousness

00:35:39.540 | and cognitive abilities is we tell stories to ourselves

00:35:42.320 | so we can have nice podcasts,

00:35:44.060 | but really it's the brain trying to maintain

00:35:46.740 | just like healthy state, I guess.

00:35:50.720 | That's fascinating.

00:35:51.860 | I also see the human brain

00:35:55.520 | and I hope artificial intelligence systems

00:35:58.940 | as not just systems that solve problems or optimize a goal,

00:36:03.940 | but are also storytellers.

00:36:06.260 | I think there's a power to telling stories.

00:36:08.820 | We tell stories to each other.

00:36:10.060 | That's what communication is.

00:36:11.660 | Like when you're alone, that's when you solve problems.

00:36:16.660 | That's when it makes sense to talk about solving problems.

00:36:19.040 | But when you're a community,

00:36:20.820 | the capability to communicate, tell stories,

00:36:24.000 | share ideas in such a way that those ideas

00:36:27.580 | are stable over a long period of time,

00:36:29.980 | that's like, that's being a charismatic storyteller.

00:36:33.280 | And I think both humans are very good at this.

00:36:35.860 | Arguably, I would argue that's why we are who we are

00:36:40.220 | is we're great storytellers.

00:36:42.260 | And then AI, I hope will also become that.

00:36:44.780 | So it's not just about being able to solve problems

00:36:47.460 | with a clear objective function.

00:36:49.020 | It's afterwards be able to tell like a way better,

00:36:51.900 | like make up a way better story

00:36:53.360 | about why you did something or why you failed.

00:36:55.760 | - So you think that robots and/or machines of some sort

00:36:59.840 | are going to start telling humans stories?

00:37:02.340 | - Well, definitely.

00:37:03.440 | So the technical field for that is called explainable AI,

00:37:07.340 | explainable artificial intelligence,

00:37:09.320 | is trying to figure out how you get the AI system

00:37:14.160 | to explain to us humans why the hell it failed

00:37:17.520 | or why it succeeded,

00:37:18.860 | or there's a lot of different sort of versions of this,

00:37:22.320 | or to visualize how it understands the world.

00:37:26.300 | That's a really difficult problem,

00:37:28.100 | especially with neural networks that are famously opaque,

00:37:33.100 | that they, we don't understand in many cases

00:37:36.320 | why a particular neural network does what it does so well.

00:37:40.480 | And to try to figure out where it's going to fail,

00:37:43.700 | that requires the AI to explain itself.

00:37:46.240 | There's a huge amount of money,

00:37:49.540 | like there's a huge amount of money in this,

00:37:52.360 | especially from government funding and so on,

00:37:54.560 | because if you want to deploy AI systems in the real world,

00:37:59.460 | we humans at least, want to ask it a question,

00:38:02.640 | like why the hell did you do that?

00:38:04.260 | Like in a dark way, why did you just kill that person, right?

00:38:09.120 | Like if a car ran over a person,

00:38:10.620 | we wouldn't understand why that happened.

00:38:13.000 | And now again, we're sometimes very unfair to AI systems

00:38:17.880 | because we humans can often not explain why very well,

00:38:21.880 | but that's the field of explainable AI.

00:38:25.560 | That's very, people are very interested in,

00:38:28.160 | because the more and more we rely on AI systems,

00:38:31.500 | like the Twitter recommender system, that AI algorithm,

00:38:35.660 | that's, I would say impacting elections,

00:38:39.140 | perhaps starting wars or at least military conflict.

00:38:41.980 | That's that algorithm.

00:38:43.660 | We want to ask that algorithm, first of all,

00:38:46.580 | do you know what the hell you're doing?

00:38:48.500 | Do you know, do you understand

00:38:50.300 | the society level effects you're having?

00:38:52.900 | And can you explain the possible other trajectories?

00:38:55.840 | Like we would have that kind of conversation with a human.

00:38:58.300 | We want to be able to do that with an AI.

00:39:00.020 | And on my own personal level,

00:39:02.020 | I think it would be nice to talk to AI systems

00:39:05.420 | for stupid stuff, like robots, when they fail to-

00:39:10.020 | - Why'd you fall down the stairs?

00:39:12.900 | - Yeah, but not an engineering question,

00:39:15.860 | but almost like a endearing question.

00:39:18.740 | Like, I'm looking for, if I fell

00:39:22.580 | and you and I were hanging out,

00:39:25.020 | I don't think you need an explanation

00:39:28.020 | exactly what were the dynamic,

00:39:29.860 | like what was the under actuated system problem here?

00:39:32.740 | Like what was the texture of the floor or so on?

00:39:36.180 | Or like what was the-

00:39:37.580 | - No, I want to know what you're thinking.

00:39:39.020 | - That, or you might joke about like,

00:39:41.200 | you're drunk again, go home or something.

00:39:43.100 | Like there could be humor in it.

00:39:44.860 | That's an opportunity.

00:39:46.940 | Like storytelling isn't just explanation of what happened.

00:39:51.040 | It's something that makes people laugh,

00:39:54.380 | makes people fall in love,

00:39:56.100 | makes people dream and understand things

00:39:58.980 | in a way that poetry makes people understand things

00:40:01.980 | as opposed to a rigorous log of where every sensor was,

00:40:06.980 | where every actuator was.

00:40:09.540 | - I mean, I find this incredible

00:40:11.460 | because one of the hallmarks of severe autism spectrum

00:40:15.660 | disorders is a report of experience from the autistic person

00:40:20.660 | that is very much a catalog of action steps.

00:40:25.320 | It's like, how do you feel today?

00:40:26.380 | And they'll say, well, I got up and I did this

00:40:27.960 | and then I did this and I did this.

00:40:29.080 | And it's not at all the way that a person with,

00:40:32.180 | who doesn't have autism spectrum disorder would respond.

00:40:35.700 | And the way you describe these machines has so much human,

00:40:41.100 | has so much humanism or so much of a human

00:40:44.020 | and biological element.

00:40:45.420 | But I realized that we were talking about machines.

00:40:48.020 | I want to make sure that I understand

00:40:51.660 | if there's a distinction between a machine that learns,

00:40:56.660 | a machine with artificial intelligence and a robot.

00:41:01.180 | At what point does a machine become a robot?

00:41:03.840 | So if I have a ballpoint pen,

00:41:06.460 | I'm assuming I wouldn't call that a robot.

00:41:08.660 | But if my ballpoint pen can come to me when it's on,

00:41:13.180 | when I moved to the opposite side of the table,

00:41:15.340 | if it moves by whatever mechanism,

00:41:17.940 | at that point, does it become a robot?

00:41:20.660 | - Okay, there's a million ways to explore this question.

00:41:23.420 | It's a fascinating one.

00:41:25.060 | So first of all, there's a question of what is life?

00:41:28.120 | Like how do you know something is a living form and not?

00:41:32.540 | And it's similar to the question of when does sort of a,

00:41:35.460 | maybe a cold computational system becomes a,

00:41:40.140 | well, we're already loading these words

00:41:41.860 | with a lot of meaning, robot and machine.

00:41:44.140 | But so one, I think movement is important,

00:41:49.140 | but that's kind of a boring idea

00:41:52.340 | that a robot is just a machine

00:41:54.580 | that's able to act in the world.

00:41:56.600 | So one, artificial intelligence could be

00:42:00.040 | both just the thinking thing,

00:42:01.800 | which I think is what machine learning is,

00:42:04.040 | and also the acting thing,

00:42:05.720 | which is what we usually think about robots.

00:42:07.880 | So robots are the things that have a perception system

00:42:10.160 | that's able to take in the world,

00:42:11.580 | however you define the world,

00:42:13.160 | is able to think and learn

00:42:14.600 | and do whatever the hell it does inside

00:42:16.700 | and then act on the world.

00:42:18.700 | So that's the difference between maybe an AI system

00:42:21.600 | or a machine and a robot.

00:42:23.220 | It's something that's able,

00:42:24.280 | a robot is something that's able to perceive the world

00:42:27.220 | and act in the world.

00:42:28.160 | - So it could be through language or sound,

00:42:31.080 | or it could be through movement or both.

00:42:32.660 | - Yeah, and I think it could also be in the digital space,

00:42:36.080 | as long as there's a aspect of entity

00:42:39.020 | that's inside the machine

00:42:41.240 | and a world that's outside the machine,

00:42:44.200 | and there's a sense in which the machine

00:42:46.440 | is sensing that world and acting in it.

00:42:49.340 | - So we could, for instance,

00:42:50.920 | there could be a version of a robot,

00:42:52.600 | according to the definition that I think you're providing,

00:42:55.400 | where the robot, where I go to sleep at night

00:42:58.240 | and this robot goes and forages for information

00:43:01.440 | that it thinks I want to see

00:43:03.680 | loaded onto my desktop in the morning.

00:43:05.280 | There was no movement of that machine.

00:43:07.000 | There was no language,

00:43:07.840 | but it essentially has movement in cyberspace.

00:43:11.140 | - Yeah, there's a distinction that I think is important

00:43:16.140 | in that there's an element of it being an entity,

00:43:23.400 | whether it's in the digital or the physical space.

00:43:26.620 | So when you have something like Alexa in your home,

00:43:30.560 | most of the speech recognition,

00:43:35.160 | most of what Alexa is doing

00:43:36.720 | is constantly being sent back to the mothership.

00:43:39.160 | When Alexa is there on its own,

00:43:44.940 | that's to me a robot.

00:43:46.940 | When it's there interacting with the world,

00:43:49.460 | when it's simply a finger of the main mothership,

00:43:54.460 | then Alexa is not a robot.

00:43:56.660 | Then it's just an interaction device.

00:43:58.660 | That then maybe the main Amazon Alexa, AI,

00:44:02.380 | big, big system is the robot.

00:44:04.800 | So that's important because there's some element

00:44:08.840 | to us humans, I think,

00:44:10.600 | where we want there to be an entity,

00:44:12.560 | whether in the digital or the physical space.

00:44:14.740 | That's where ideas of consciousness come in

00:44:16.740 | and all those kinds of things

00:44:18.560 | that we project our understanding

00:44:21.420 | what it means to be a being.

00:44:23.180 | And so to take that further,

00:44:27.020 | when does a machine become a robot?

00:44:28.980 | I think there's a special moment.

00:44:35.220 | There's a special moment in a person's life,

00:44:37.820 | in a robot's life where it surprises you.

00:44:40.760 | I think surprise is a really powerful thing

00:44:44.260 | where you know how the thing works

00:44:46.820 | and yet it surprises you.

00:44:48.420 | That's a magical moment for us humans.

00:44:51.940 | So whether it's a chess playing program

00:44:54.620 | that does something that you haven't seen before

00:44:57.580 | that makes people smile,

00:44:58.980 | like, huh, those moments happen with alpha zero

00:45:03.060 | for the first time in chess playing

00:45:05.540 | or grandmasters were really surprised by a move.

00:45:08.740 | They didn't understand the move

00:45:10.220 | and then they studied and study

00:45:11.580 | and then they understood it.

00:45:13.340 | But that moment of surprise,

00:45:15.260 | that's for grandmasters in chess.

00:45:17.300 | I find that moment of surprise really powerful,

00:45:20.380 | really magical in just everyday life.

00:45:23.300 | - Because it supersedes the human brain in that moment?

00:45:26.460 | - Not supersedes like outperforms,

00:45:30.980 | but surprises you in a positive sense.

00:45:35.380 | Like I didn't think you could do that.

00:45:37.980 | I didn't think that you had that in you.

00:45:40.740 | And I think that moment is a big transition

00:45:44.420 | for a robot from a moment of being a servant

00:45:48.140 | that accomplishes a particular task

00:45:51.140 | with some level of accuracy,

00:45:52.820 | with some rate of failure to an entity,

00:45:57.820 | a being that's struggling just like you are in this world.

00:46:01.900 | And that's a really important moment

00:46:04.420 | that I think you're not gonna find many people

00:46:07.540 | in the AI community that talk like I just did.

00:46:09.900 | I'm not speaking like some philosopher or some hippie.

00:46:14.340 | I'm speaking from purely engineering perspective.

00:46:16.780 | I think it's really important for robots to become entities

00:46:20.500 | and explore that as a real engineering problem,

00:46:23.380 | as opposed to everybody treats robots

00:46:25.860 | in the robotics community.

00:46:27.660 | They don't even call them a he or she.

00:46:29.540 | They don't give them, try to avoid giving them names.

00:46:32.220 | They really want to see it like a system, like a servant.

00:46:36.700 | They see it as a servant that's trying to accomplish a task.

00:46:40.260 | To me, I don't think I'm just romanticizing the notion.

00:46:44.660 | I think it's a being.

00:46:46.180 | It's currently perhaps a dumb being,

00:46:48.880 | but in the long arc of history,

00:46:53.060 | humans are pretty dumb beings too.

00:46:55.180 | - I would agree with that statement.

00:46:57.100 | - So I tend to really want to explore this,

00:47:00.500 | treating robots really as entities.

00:47:04.460 | So like anthropomorphization,

00:47:08.400 | which is the act of looking at an inanimate object

00:47:12.500 | and projecting onto it lifelike features,

00:47:15.460 | I think robotics generally sees that as a negative.

00:47:20.460 | I see it as a superpower.

00:47:23.860 | Like that, we need to use that.

00:47:26.420 | - Well, I'm struck by how that really grabs on

00:47:29.560 | to the relationship between human and machine

00:47:32.800 | or human and robot.

00:47:34.060 | So it's the simple question is,

00:47:37.400 | and I think you've already told us the answer,

00:47:38.880 | but does interacting with a robot change you?

00:47:43.040 | Does it, in other words, do we develop relationships

00:47:46.580 | to robots?

00:47:48.020 | - Yeah, I definitely think so.

00:47:50.240 | I think the moment you see a robot or AI systems

00:47:55.240 | as more than just servants,

00:47:57.680 | but entities, they begin to change.

00:48:01.240 | Just like good friends do, just like relationships,

00:48:04.880 | just like other humans.

00:48:06.760 | I think for that, you have to have certain aspects

00:48:09.900 | of that interaction.

00:48:11.440 | Like the robot's ability to say no,

00:48:15.820 | to have its own sense of identity,

00:48:19.080 | to have its own set of goals

00:48:21.720 | that's not constantly serving you,

00:48:23.120 | but instead trying to understand the world

00:48:24.920 | and do that dance of understanding

00:48:27.200 | through communication with you.

00:48:28.920 | So I definitely think there's a,

00:48:30.720 | I mean, I have a lot of thoughts about this, as you may know,

00:48:35.200 | and that's at the core of my lifelong dream, actually,

00:48:38.720 | of what I want to do,

00:48:39.840 | which is I believe that most people have

00:48:44.840 | a notion of loneliness in them that we haven't discovered,

00:48:50.280 | that we haven't explored, I should say.

00:48:53.120 | And I see AI systems as helping us explore that

00:48:57.860 | so that we can become better humans,

00:49:00.240 | better people towards each other.

00:49:02.280 | So I think that connection between human and AI,

00:49:06.800 | human and robot is not only possible,

00:49:11.280 | but will help us understand ourselves

00:49:14.520 | in ways that are like several orders of magnitude,

00:49:17.520 | deeper than we ever could have imagined.

00:49:21.320 | I tend to believe that,

00:49:22.600 | well, I have very wild levels of belief

00:49:29.360 | in terms of how impactful that would be.

00:49:35.600 | - So when I think about human relationships,

00:49:37.740 | I don't always break them down into variables,

00:49:41.140 | but we could explore a few of those variables

00:49:43.400 | and see how they map to human-robot relationships.

00:49:47.340 | One is just time, right?

00:49:49.160 | If you spend zero time with another person at all

00:49:52.720 | in cyberspace or on the phone or in person,

00:49:55.380 | you essentially have no relationship to them.

00:49:58.120 | If you spend a lot of time, you have a relationship.

00:49:59.600 | This is obvious, but I guess one variable would be time,

00:50:01.800 | how much time you spend with the other entity,

00:50:05.180 | robot or human.

00:50:06.520 | The other would be wins and successes.

00:50:10.000 | You know, you enjoy successes together.

00:50:12.020 | I'll give a absolutely trivial example of this in a moment,

00:50:16.680 | but the other would be failures.

00:50:19.180 | When you struggle with somebody,

00:50:21.520 | whether or not you struggle between one another,

00:50:23.560 | you disagree, like I was really struck by the fact

00:50:25.660 | that you said that robot's saying no.

00:50:27.040 | I've never thought about a robot saying no to me,

00:50:30.080 | but there it is.

00:50:31.540 | - I look forward to you being one of the first people

00:50:34.040 | to send this robot.

00:50:35.180 | - So do I.

00:50:36.320 | So there's struggle.

00:50:37.640 | You know, when you struggle with somebody, you grow closer.

00:50:41.100 | Sometimes the struggles are imposed

00:50:43.600 | between those two people, so-called trauma bonding.

00:50:45.900 | They call it in the whole psychology literature

00:50:48.300 | and pop psychology literature.

00:50:50.120 | But in any case, I could imagine,

00:50:52.200 | so time, successes together, struggle together,

00:50:57.200 | and then just peaceful time, hanging out at home,

00:51:00.960 | watching movies, waking up near one another.

00:51:04.740 | Here, we're breaking down the kind of elements

00:51:08.160 | of relationships of any kind.

00:51:09.840 | So do you think that these elements apply

00:51:13.880 | to robot-human relationships?

00:51:16.400 | And if so, then I could see how if the robot

00:51:21.400 | is its own entity and has some autonomy

00:51:25.880 | in terms of how it reacts to you,

00:51:27.160 | it's not just there just to serve you.

00:51:28.960 | It's not just a servant.

00:51:30.200 | It actually has opinions and can tell you

00:51:32.800 | when maybe your thinking is flawed

00:51:34.440 | or your actions are flawed.

00:51:35.760 | - It can also leave.

00:51:37.160 | - It could also leave.

00:51:39.360 | So I've never conceptualized

00:51:40.920 | robot-human interactions this way.

00:51:42.660 | So tell me more about how this might look.

00:51:46.500 | Are we thinking about a human-appearing robot?

00:51:50.220 | I know you and I have both had intense relationships

00:51:53.600 | to our, we have separate dogs, obviously, but to animals.

00:51:57.240 | This sounds a lot like human-animal interaction.

00:51:59.120 | So what is the ideal human-robot relationship?

00:52:04.120 | - So there's a lot to be said here,

00:52:06.360 | but you actually pinpointed one of the big,

00:52:09.520 | big first steps, which is this idea of time.

00:52:13.760 | And it's a huge limitation

00:52:15.500 | in machine learning community currently.

00:52:17.840 | Now we're back to like the actual details.

00:52:21.460 | Lifelong learning is a problem space

00:52:26.340 | that focuses on how AI systems can learn

00:52:29.580 | over a long period of time.

00:52:30.980 | What's currently most machine learning systems

00:52:35.420 | are not able to do is to,

00:52:37.860 | all of the things you've listed under time,

00:52:39.780 | the successes, the failures,

00:52:41.820 | or just chilling together watching movies,

00:52:44.600 | AI systems are not able to do that,

00:52:47.960 | which is all the beautiful, magical moments

00:52:51.040 | that I believe are the days filled with.

00:52:53.880 | They're not able to keep track of those together with you.

00:52:57.380 | - 'Cause they can't move with you and be with you.

00:52:59.200 | - No, no, like literally we don't have the techniques

00:53:02.060 | to do the learning,

00:53:03.440 | the actual learning of containing those moments.

00:53:07.180 | Current machine learning systems are really focused

00:53:09.960 | on understanding the world in the following way.

00:53:12.480 | It's more like the perception system,

00:53:14.340 | like looking around, understand like what's in the scene,

00:53:19.340 | that there's a bunch of people sitting down,

00:53:22.220 | that there is cameras and microphones,

00:53:24.860 | that there's a table, understand that.

00:53:27.500 | But the fact that we shared this moment of talking today

00:53:30.940 | and still remember that for next time you're,

00:53:34.160 | for like next time you're doing something,

00:53:36.660 | remember that this moment happened.

00:53:38.360 | We don't know how to do that technique wise.

00:53:40.420 | This is what I'm hoping to innovate on

00:53:44.140 | as I think it's a very, very important component

00:53:47.080 | of what it means to create a deep relationship,

00:53:49.940 | that sharing of moments together.

00:53:52.060 | - Could you post a photo of you and the robot,

00:53:54.060 | like selfie with robot and then the robot sees that image

00:53:58.340 | and recognizes that was time spent,

00:54:00.940 | there were smiles or there were tears

00:54:03.620 | and create some sort of metric of emotional depth

00:54:08.620 | in the relationship and update its behavior.

00:54:11.660 | So could it text you in the middle of the night and say,

00:54:15.160 | "Why haven't you texted me back?"

00:54:16.820 | - Well, yes, all of those things,

00:54:18.620 | but we can dig into that.

00:54:21.420 | But I think that time element, forget everything else,

00:54:24.820 | just sharing moments together, that changes everything.

00:54:29.200 | I believe that changes everything.

00:54:30.940 | There's specific things that are more in terms of systems

00:54:33.560 | that I can explain you.

00:54:35.360 | It's more technical and probably a little bit offline

00:54:39.380 | 'cause I have kind of wild ideas

00:54:41.300 | how that can revolutionize social networks

00:54:44.760 | and operating systems.

00:54:47.660 | But the point is that element alone,

00:54:50.620 | forget all the other things we're talking about,

00:54:53.180 | like emotions, saying no, all that,

00:54:56.060 | just remember sharing moments together

00:54:58.600 | would change everything.

00:55:00.260 | We don't currently have systems that share moments together.

00:55:05.260 | Like even just you and your fridge,

00:55:08.100 | just all those times you went late at night

00:55:11.060 | and ate the thing you shouldn't have eaten,

00:55:13.300 | that was a secret moment you had with your refrigerator.

00:55:16.680 | You shared that moment, that darkness

00:55:18.860 | or that beautiful moment where you just,

00:55:21.220 | like heartbroken for some reason,

00:55:24.060 | you're eating that ice cream or whatever,

00:55:26.300 | that's a special moment.

00:55:27.620 | And that refrigerator was there for you.

00:55:29.620 | And the fact that it missed the opportunity

00:55:31.820 | to remember that is tragic.

00:55:35.980 | And once it does remember that,

00:55:37.820 | I think you're gonna be very attached to the refrigerator.

00:55:42.780 | You're gonna go through some hell with that refrigerator.

00:55:45.700 | Most of us have like in a developed world

00:55:49.640 | have weird relationships with food, right?

00:55:51.520 | So you can go through some deep moments

00:55:54.880 | of trauma and triumph with food.

00:55:57.260 | And at the core of that is the refrigerator.

00:55:59.200 | So a smart refrigerator, I believe would change society,

00:56:04.200 | not just the refrigerator,

00:56:06.180 | but these ideas in the systems all around us.

00:56:10.220 | So that, I just wanna comment

00:56:11.900 | on how powerful that idea of time is.

00:56:14.220 | And then there's a bunch of elements of actual interaction

00:56:17.860 | of allowing you as a human to feel like you're being heard,

00:56:22.860 | truly heard, truly understood that we human,

00:56:30.060 | like deep friendship is like that, I think,

00:56:33.020 | but we're still, there's still an element of selfishness.

00:56:36.940 | There's still an element of not really being able

00:56:39.540 | to understand another human.

00:56:40.740 | And a lot of the times when you're going through trauma

00:56:44.440 | together through difficult times and through successes,

00:56:47.580 | you're actually starting to get that inkling

00:56:49.540 | of understanding of each other.

00:56:51.220 | But I think that can be done more aggressively,

00:56:54.040 | more efficiently.

00:56:57.420 | Like if you think of a great therapist,

00:56:59.280 | I think I've never actually been to a therapist,

00:57:01.820 | but I'm a believer.

00:57:03.200 | I used to want to be a psychiatrist.

00:57:04.960 | - Do Russians go to therapists?

00:57:06.340 | - No, they don't, they don't.

00:57:07.980 | And if they do, the therapist don't live to tell the story.

00:57:12.100 | I do believe in talk there, which friendship is to me,

00:57:17.700 | is talk therapy or like, you don't necessarily need to talk.

00:57:22.700 | It's like just connecting in the space of ideas

00:57:27.140 | and the space of experiences.

00:57:28.900 | And I think there's a lot of ideas

00:57:30.820 | of how to make AI systems to be able to ask

00:57:33.580 | the right questions and truly hear another human.

00:57:37.320 | This is what we try to do with podcasting, right?

00:57:40.320 | I think there's ways to do that with AI.

00:57:42.620 | But above all else, just remembering

00:57:45.980 | the collection of moments that make up the day,

00:57:49.820 | the week, the months.

00:57:51.980 | I think you maybe have some of this as well.

00:57:55.680 | Some of my closest friends still

00:57:57.180 | are the friends from high school.

00:57:58.780 | That's time, we've been through a bunch of shit together.

00:58:02.960 | And that like we're very different people,

00:58:06.220 | but just the fact that we've been through that

00:58:07.940 | and we remember those moments.

00:58:09.320 | And those moments somehow create a depth of connection

00:58:12.820 | like nothing else, like you and your refrigerator.

00:58:15.840 | - I love that because my graduate advisor,

00:58:20.640 | unfortunately she passed away,

00:58:21.720 | but when she passed away, somebody said at her memorial,

00:58:25.780 | all these amazing things she had done, et cetera.

00:58:29.400 | And then her kids got up there and she had young children

00:58:32.880 | that I knew when she was pregnant with them.

00:58:35.480 | And so it was really, even now,

00:58:37.900 | I can feel like your heart gets heavy thinking about this.

00:58:39.880 | They're going to grow up without their mother.

00:58:41.820 | And it was really amazing.

00:58:42.660 | Very, very strong young girls and now young women.

00:58:47.220 | And what they said was incredible.

00:58:49.300 | They said what they really appreciated most

00:58:51.720 | about their mother, who was an amazing person,

00:58:54.960 | is all the unstructured time they spent together.

00:58:59.480 | So it wasn't the trips to the zoo.

00:59:00.960 | It wasn't, oh, she woke up at five in the morning

00:59:03.600 | and drove us to school.

00:59:04.440 | She did all those things too.

00:59:05.500 | She had two hour commute in each direction.

00:59:07.360 | It was incredible, ran a lab, et cetera.

00:59:09.360 | But it was the unstructured time.

00:59:11.440 | So on the passing of their mother,

00:59:13.280 | that's what they remembered was the biggest give

00:59:16.520 | and what bonded them to her was all the time

00:59:18.380 | where they just kind of hung out.

00:59:20.400 | And the way you described the relationship

00:59:22.000 | to a refrigerator is so, I want to say human-like,

00:59:27.000 | but I'm almost reluctant to say that

00:59:28.740 | because what I'm realizing as we're talking

00:59:31.420 | is that what we think of as human-like

00:59:34.400 | might actually be a lower form of relationship.

00:59:39.000 | There may be relationships that are far better

00:59:42.360 | than the sorts of relationships that we can conceive

00:59:45.420 | in our minds right now,

00:59:46.820 | based on what these machine relationship interactions

00:59:50.060 | could teach us.

00:59:51.100 | Do I have that right?

00:59:52.820 | - Yeah, I think so.

00:59:54.080 | I think there's no reason to see machines

00:59:55.720 | as somehow incapable of teaching us something

00:59:59.900 | that's deeply human.

01:00:01.580 | I don't think humans have a monopoly on that.

01:00:04.980 | I think we understand ourselves very poorly

01:00:06.920 | and we need to have the kind of prompting from a machine.

01:00:11.920 | And definitely part of that is just remembering the moments.

01:00:16.800 | Remembering the moments.

01:00:17.920 | I think the unstructured time together,

01:00:22.180 | I wonder if it's quite so unstructured.

01:00:27.520 | That's like calling this podcast unstructured time.

01:00:30.520 | - Maybe what they meant was it wasn't a big outing.

01:00:33.860 | It wasn't just, there was no specific goal,

01:00:36.160 | but a goal was created through the lack of a goal.

01:00:39.640 | Like where you just hang out

01:00:40.560 | and then you start playing thumb war

01:00:42.520 | and you end up playing thumb war for an hour there.

01:00:45.080 | So the structure emerges from lack of structure.

01:00:48.900 | - No, but the thing is the moments,

01:00:51.440 | there's something about those times

01:00:54.320 | that create special moments.

01:00:56.440 | And I think that those could be optimized for.

01:01:00.280 | I think we think of like a big outing as, I don't know,

01:01:02.480 | going to Six Flags or something,

01:01:03.800 | or some big, the Grand Canyon, or go into some,

01:01:08.160 | I don't know, I think we would need to,

01:01:11.640 | we don't quite yet understand as humans

01:01:13.800 | what creates magical moments.

01:01:15.680 | I think there's possible to optimize a lot of those things.

01:01:18.320 | And perhaps like podcasting is helping people discover

01:01:21.080 | that like maybe the thing we want to optimize for

01:01:24.200 | isn't necessarily like some sexy, like quick clips.

01:01:29.200 | Maybe what we want is long form authenticity.

01:01:34.520 | - Depth. - Depth.

01:01:36.260 | So we were trying to figure that out.

01:01:38.760 | Certainly from a deep connection

01:01:40.200 | between humans and humans and AI systems,

01:01:44.160 | I think long conversations

01:01:45.920 | or long periods of communication over a series of moments,

01:01:52.280 | like my new, perhaps seemingly insignificant

01:01:56.160 | to the big ones, the big successes, the big failures,

01:02:00.760 | those are all just stitching those together

01:02:03.020 | and talking throughout.

01:02:04.420 | I think that's the formula

01:02:06.600 | for a really, really deep connection

01:02:08.240 | that from like a very specific engineering perspective

01:02:11.940 | is I think a fascinating open problem

01:02:15.560 | that hasn't been really worked on very much.

01:02:18.400 | And for me, if I have the guts

01:02:21.840 | and I mean, there's a lot of things to say,

01:02:24.800 | but one of it is guts is I'll build a startup around it.

01:02:29.340 | - Yeah, so let's talk about this startup

01:02:32.360 | and let's talk about the dream.

01:02:34.800 | You've mentioned this dream before

01:02:36.040 | in our previous conversations,

01:02:37.280 | always as little hints dropped here and there.

01:02:40.120 | Just for anyone listening,

01:02:41.360 | there's never been an offline conversation about this dream.

01:02:43.720 | I'm not privy to anything except what Lex says now.

01:02:48.160 | And I realized that there's no way

01:02:49.760 | to capture the full essence of a dream

01:02:52.840 | in any kind of verbal statement

01:02:54.520 | in a way that captures all of it.

01:02:58.200 | But what is this dream

01:03:00.960 | that you've referred to now several times

01:03:03.840 | when we've sat down together and talked on the phone?

01:03:06.920 | Maybe it's this company, maybe it's something distinct.

01:03:09.100 | If you feel comfortable,

01:03:10.920 | it'd be great if you could share

01:03:12.120 | a little bit about what that is.

01:03:13.320 | - Sure, so the way people express long-term vision,

01:03:19.060 | I've noticed is quite different.

01:03:20.880 | Like Elon is an example of somebody

01:03:22.660 | who can very crisply say exactly what the goal is.

01:03:27.540 | Also has to do with the fact the problems he's solving

01:03:29.820 | have nothing to do with humans.

01:03:31.320 | So my long-term vision is a little bit more difficult

01:03:36.400 | to express in words, I've noticed, as I've tried.

01:03:40.840 | It could be my brain's failure.

01:03:42.380 | But there's a ways to sneak up to it.

01:03:45.400 | So let me just say a few things.

01:03:48.880 | Early on in life and also in the recent years,

01:03:53.000 | I've interacted with a few robots

01:03:55.160 | where I understood there's magic there.

01:03:57.220 | And that magic could be shared by millions

01:04:01.260 | if it's brought to light.

01:04:04.640 | When I first met Spot from Boston Dynamics,

01:04:07.620 | I realized there's magic there that nobody else is seeing.

01:04:10.400 | - Is the dog.

01:04:11.440 | - The dog, sorry.

01:04:12.280 | The Spot is the four-legged robot from Boston Dynamics.

01:04:17.280 | Some people might've seen it, it's this yellow dog.

01:04:20.360 | And sometimes in life,

01:04:25.320 | you just notice something that just grabs you.

01:04:28.440 | And I believe that this is something that,

01:04:31.400 | this magic is something that could be

01:04:34.800 | every single device in the world.

01:04:37.040 | The way that I think maybe Steve Jobs

01:04:41.360 | thought about the personal computer.

01:04:43.160 | Woz didn't think about it, the personal computer this way,

01:04:46.720 | but Steve did, which is like he thought

01:04:49.400 | that the personal computer should be as thin

01:04:50.960 | as a sheet of paper and everybody should have one.

01:04:53.640 | I mean, this idea, I think it is heartbreaking

01:04:58.640 | that the world is being filled up with machines

01:05:03.840 | that are soulless.

01:05:05.000 | And I think every one of them can have that same magic.

01:05:10.280 | One of the things that also inspired me

01:05:14.800 | in terms of a startup is that magic can be engineered

01:05:17.640 | much easier than I thought.

01:05:19.560 | That's my intuition with everything

01:05:21.320 | I've ever built and worked on.

01:05:23.960 | So the dream is to add a bit of that magic

01:05:28.840 | in every single computing system in the world.

01:05:32.360 | So the way that Windows operating system for a long time

01:05:36.480 | was the primary operating system everybody interacted with.

01:05:39.440 | They built apps on top of it.

01:05:41.320 | I think this is something that should be as a layer,

01:05:45.800 | it's almost as an operating system in every device

01:05:49.040 | that humans interact with in the world.

01:05:51.000 | Now what that actually looks like the actual dream

01:05:54.720 | when I was officially a kid,

01:05:56.520 | it didn't have this concrete form of a business.

01:06:01.120 | It had more of a dream of exploring your own loneliness

01:06:08.240 | by interacting with machines, robots.

01:06:11.840 | This deep connection between humans and robots

01:06:15.720 | was always a dream.

01:06:17.160 | And so for me, I'd love to see a world

01:06:20.160 | where there's every home has a robot

01:06:22.440 | and not a robot that washes the dishes or a sex robot,

01:06:27.440 | or I don't know, I think of any kind of activity

01:06:31.440 | the robot can do, but more like a companion.

01:06:33.840 | - A family member.

01:06:34.800 | - A family member, the way a dog is.

01:06:37.680 | But a dog that's able to speak your language too.

01:06:41.960 | So not just connect the way a dog does

01:06:45.160 | by looking at you and looking away

01:06:46.800 | and almost like smiling with its soul in that kind of way,

01:06:51.120 | but also to actually understand what the hell,

01:06:54.560 | like why are you so excited about the successes?

01:06:56.720 | Like understand the details, understand the traumas.

01:06:59.880 | And I just think that has always filled me with excitement

01:07:05.560 | that I could, with artificial intelligence,

01:07:09.280 | bring joy to a lot of people.

01:07:12.320 | More recently, I've been more and more hardbroken

01:07:16.320 | to see the kind of division, derision,

01:07:21.700 | even hate that's boiling up on the internet

01:07:27.080 | through social networks.

01:07:28.600 | And I thought this kind of mechanism is exactly applicable

01:07:32.760 | in the context of social networks as well.

01:07:34.800 | So it's an operating system that serves

01:07:38.440 | as your guide on the internet.

01:07:43.440 | One of the biggest problems with YouTube

01:07:45.760 | and social networks currently

01:07:47.960 | is they're optimizing for engagement.

01:07:49.880 | I think if you create AI systems

01:07:52.560 | that know each individual person,

01:07:54.920 | you're able to optimize for long-term growth,

01:07:59.080 | for a long-term happiness.

01:08:00.800 | - Of the individual?

01:08:01.640 | - Of the individual, of the individual.

01:08:03.720 | And there's a lot of other things to say, which is the,

01:08:08.460 | in order for AI systems to learn everything about you,

01:08:14.460 | they need to collect, they need to,

01:08:17.880 | just like you and I, when we talk offline,

01:08:20.000 | we're collecting data about each other,

01:08:21.520 | secrets about each other.

01:08:23.520 | The same way AI has to do that.

01:08:26.600 | And that allows you to,

01:08:28.720 | and that requires you to rethink ideas

01:08:32.720 | of ownership of data.

01:08:35.480 | I think each individual should own all of their data

01:08:39.920 | and very easily be able to leave.

01:08:41.920 | Just like AI systems can leave,

01:08:43.600 | humans can disappear and delete all of their data

01:08:48.200 | in a moment's notice,

01:08:49.380 | which is actually better than we humans can do.

01:08:54.220 | Once we load the data into each other, it's there.

01:08:56.880 | I think it's very important to be both,

01:09:00.440 | give people complete control over their data

01:09:02.880 | in order to establish trust that they can trust you.

01:09:06.200 | And the second part of trust is transparency.

01:09:08.480 | Whenever the data is used to make it very clear

01:09:11.880 | what it's being used for.

01:09:13.040 | And not clear in a lawyerly legal sense,

01:09:16.120 | but clear in a way that people really understand

01:09:18.800 | what it's used for.

01:09:19.640 | I believe when people have the ability

01:09:21.520 | to delete all their data and walk away

01:09:24.160 | and know how the data is being used, I think they'll stay.

01:09:29.840 | - The possibility of a clean breakup

01:09:31.880 | is actually what will keep people together.

01:09:33.360 | - Yeah, I think so.

01:09:34.440 | I think, exactly.

01:09:36.400 | I think a happy marriage requires the ability

01:09:39.680 | to divorce easily without the divorce industrial complex

01:09:44.680 | or whatever is currently going on.

01:09:48.040 | There's so much money to be made from lawyers and divorce.

01:09:50.720 | But yeah, the ability to leave is what enables love, I think.

01:09:55.220 | - It's interesting, I've heard the phrase

01:09:57.320 | from a semi-cynical friend that marriage

01:09:59.960 | is the leading cause of divorce.

01:10:01.580 | But now we've heard that divorce

01:10:03.560 | or the possibility of divorce

01:10:04.960 | could be the leading cause of marriage.

01:10:06.620 | - Of a happy marriage.

01:10:08.100 | - Good point.

01:10:08.940 | - Of a happy marriage.

01:10:09.920 | So yeah, but there's a lot of details there.

01:10:12.740 | But the big dream is that connection

01:10:14.680 | between AI system and a human.

01:10:17.480 | And I haven't, you know, there's so much fear

01:10:21.040 | about artificial intelligence systems and about robots

01:10:23.880 | that I haven't quite found the right words

01:10:26.400 | to express that vision because the vision I have is one,

01:10:31.320 | it's not like some naive delusional vision

01:10:33.480 | of like technology is going to save everybody.

01:10:36.440 | It's, I really do just have a positive view

01:10:40.320 | of ways AI systems can help humans explore themselves.

01:10:44.720 | - I love that positivity.

01:10:46.040 | And I agree that the stance,

01:10:49.120 | everything is doomed is equally bad

01:10:54.400 | to say that everything's going to turn out all right.

01:10:56.560 | There has to be a dedicated effort.

01:10:58.280 | And clearly you're thinking about

01:11:00.620 | what that dedicated effort would look like.

01:11:02.520 | You mentioned two aspects to this dream.

01:11:06.600 | And I want to make sure that I understand

01:11:07.800 | where they connect if they do

01:11:10.040 | or if these are independent streams.

01:11:12.400 | One was this hypothetical robot family member

01:11:17.240 | or some other form of robot that would allow people

01:11:20.240 | to experience the kind of delight

01:11:23.400 | that you experienced many times

01:11:27.040 | and that you would like the world to be able to have.

01:11:30.280 | And it's such a beautiful idea of this give.

01:11:33.480 | And the other is social media or social network platforms

01:11:38.160 | that really serve individuals and their best selves

01:11:42.340 | and their happiness and their growth.

01:11:44.160 | Is there crossover between those

01:11:45.600 | or are these two parallel dreams?

01:11:47.040 | - It's 100% the same thing.

01:11:48.480 | It's difficult to kind of explain without going through

01:11:51.560 | details, but maybe one easy way to explain

01:11:54.620 | the way I think about social networks

01:11:56.700 | is to create an AI system that's yours.

01:11:59.660 | That's yours.

01:12:00.680 | It's not like Amazon Alexa that's centralized.

01:12:03.240 | You own the data.

01:12:04.560 | It's like your little friend

01:12:07.440 | that becomes your representative on Twitter

01:12:11.320 | that helps you find things that will make you feel good,

01:12:16.200 | that will also challenge your thinking to make you grow,

01:12:19.880 | but not let you get lost in the negative spiral

01:12:24.880 | of dopamine that gets you to be angry

01:12:29.440 | or most just get you to be not open to learning.

01:12:34.240 | And so that little representative

01:12:36.240 | is optimizing your long-term health.

01:12:38.780 | And I believe that that is not only good for human beings,

01:12:45.360 | it's also good for business.

01:12:47.360 | I think long-term you can make a lot of money

01:12:50.480 | by challenging this idea that the only way to make money

01:12:54.440 | is maximizing engagement.

01:12:57.360 | And one of the things that people disagree with me on

01:12:59.680 | is they think Twitter's always going to win.

01:13:02.400 | Like maximizing engagement is always going to win.

01:13:04.880 | I don't think so.

01:13:06.280 | I think people have woken up now to understanding

01:13:09.500 | that like they don't always feel good.

01:13:12.760 | The ones who are on Twitter a lot,

01:13:16.220 | that they don't always feel good at the end of the week.

01:13:19.300 | - I would love feedback from whatever this creature,

01:13:24.000 | whatever, I can't, I don't know what to call it,

01:13:26.720 | as to maybe at the end of the week,

01:13:28.720 | it would automatically unfollow

01:13:30.520 | some of the people that I follow

01:13:31.940 | because it realized through some really smart data

01:13:35.600 | about how I was feeling inside

01:13:37.120 | or how I was sleeping or something

01:13:38.560 | that that just wasn't good for me,

01:13:40.600 | but it might also put things and people in front of me

01:13:43.600 | that I ought to see.

01:13:45.500 | Is that the kind of sliver of what this looks like?

01:13:48.640 | - The whole point, because of the interaction,

01:13:50.540 | because of sharing the moments

01:13:53.580 | and learning a lot about you,

01:13:55.100 | you're now able to understand

01:13:59.560 | what interactions led you to become

01:14:01.960 | a better version of yourself.

01:14:04.280 | Like the person you yourself are happy with.

01:14:07.280 | This isn't, if you're into a flat earth

01:14:10.080 | and you feel very good about it,

01:14:12.800 | that you believe the earth is flat,

01:14:15.440 | like the idea that you should censor, that is ridiculous.

01:14:19.600 | If it makes you feel good

01:14:21.020 | and you're becoming the best version of yourself,

01:14:23.440 | I think you should be getting

01:14:24.600 | as much flat earth as possible.

01:14:26.420 | Now it's also good to challenge your ideas,

01:14:29.320 | but not because the centralized committee decided,

01:14:34.320 | but because you tell to the system

01:14:37.280 | that you like challenging your ideas.

01:14:39.280 | I think all of us do.

01:14:40.800 | And then, which actually YouTube doesn't do that well,

01:14:44.040 | once you go down the flat earth rabbit hole,

01:14:45.900 | that's all you're gonna see.

01:14:47.160 | It's nice to get some really powerful communicators

01:14:51.640 | to argue against flat earth.

01:14:53.600 | And it's nice to see that for you

01:14:57.000 | and potentially at least long-term to expand your horizons.

01:15:01.100 | Maybe the earth is not flat,

01:15:03.240 | but if you continue to live your whole life

01:15:05.160 | thinking the earth is flat, I think,

01:15:08.100 | and you're being a good father or son or daughter

01:15:11.680 | and you're being the best version of yourself

01:15:14.360 | and you're happy with yourself, I think the earth is flat.

01:15:18.940 | So I think this kind of idea,

01:15:21.360 | and I'm just using that kind of silly, ridiculous example

01:15:24.220 | because I don't like the idea of centralized forces

01:15:29.220 | controlling what you can and can't see.

01:15:33.240 | But I also don't like this idea of not censoring anything

01:15:39.660 | because that's always the biggest problem with that

01:15:42.540 | is there's a central decider.

01:15:45.820 | I think you yourself can decide

01:15:47.600 | what you want to see and not.

01:15:49.480 | And it's good to have a companion that reminds you

01:15:54.020 | that you felt shitty last time you did this,

01:15:56.500 | or you felt good last time you did this.

01:15:58.580 | - Man, I feel like in every good story,

01:16:00.120 | there's a guide or a companion that flies out

01:16:03.540 | or forages a little bit further or a little bit differently

01:16:06.440 | and brings back information that helps us

01:16:08.420 | or at least tries to steer us in the right direction.

01:16:11.360 | - So that's exactly what I'm thinking

01:16:16.360 | and what I've been working on.

01:16:17.800 | I should mention there's a bunch of difficulties here.

01:16:20.820 | You see me up and down a little bit recently.

01:16:23.760 | So there's technically a lot of challenges here.

01:16:27.600 | Like with a lot of technologies

01:16:30.480 | and the reason I'm talking about it on a podcast comfortably

01:16:34.360 | as opposed to working in secret is it's really hard

01:16:38.500 | and maybe it's time has not come.

01:16:41.000 | And that's something you have to constantly struggle with

01:16:44.140 | in terms of like entrepreneurially as a startup.

01:16:46.980 | Like I've also mentioned to you maybe offline,

01:16:50.660 | I really don't care about money.

01:16:52.540 | I don't care about business success,

01:16:55.300 | all those kinds of things.

01:16:56.600 | So it's a difficult decision to make how much of your time

01:17:03.400 | do you want to go all in here and give everything to this?

01:17:06.780 | It's a big roll of the dice

01:17:09.780 | because I've also realized that working

01:17:12.540 | on some of these problems, both with the robotics

01:17:16.660 | and the technical side in terms of the machine learning

01:17:20.660 | system that I'm describing, it's lonely, it's really lonely

01:17:25.660 | because both on a personal level and a technical level.

01:17:31.420 | So on a technical level, I'm surrounded by people

01:17:33.580 | that kind of doubt me,

01:17:37.900 | which I think all entrepreneurs go through.

01:17:40.540 | And they doubt you in the following sense.

01:17:42.960 | They know how difficult it is.

01:17:46.800 | Like the people that, the colleagues of mine,

01:17:49.540 | they know how difficult lifelong learning is.

01:17:52.300 | They also know how difficult it is to build a system

01:17:54.740 | like this, to build a competitive social network.

01:17:59.380 | And in general, there's a kind of loneliness

01:18:04.060 | to just working on something on your own

01:18:08.800 | for long periods of time.

01:18:10.260 | And you start to doubt whether given that you don't have

01:18:14.280 | a track record of success, like that's a big one.

01:18:17.980 | When you look in the mirror, especially when you're young,

01:18:20.500 | but I still have that on most things,

01:18:22.660 | you look in the mirror and you have these big dreams.

01:18:26.500 | How do you know you're actually as smart

01:18:30.780 | as you think you are?

01:18:32.480 | Like, how do you know you're going to be able

01:18:34.220 | to accomplish this dream?

01:18:35.420 | You have this ambition.

01:18:36.540 | - You sort of don't, but you're kind of pulling on a string

01:18:40.920 | hoping that there's a bigger ball of yarn.

01:18:42.960 | - Yeah, but you have this kind of intuition.

01:18:45.160 | I think I pride myself in knowing what I'm good at

01:18:51.360 | because the reason I have that intuition

01:18:54.460 | is 'cause I think I'm very good at knowing

01:18:57.540 | all the things I suck at, which is basically everything.

01:19:01.280 | So like whenever I notice like, wait a minute,

01:19:04.520 | I'm kind of good at this, which is very rare for me.

01:19:08.180 | I think like that might be a ball of yarn worth pulling at.

01:19:11.500 | And the thing with the, in terms of engineering systems

01:19:14.760 | that are able to interact with humans,

01:19:16.540 | I think I'm very good at that.

01:19:18.520 | And it's 'cause we're talking about podcasting and so on.

01:19:22.040 | I don't know if I'm very good at podcasts.

01:19:23.620 | - You're very good at podcasting.

01:19:25.200 | - But I certainly don't.

01:19:27.180 | I think maybe it is compelling for people to watch

01:19:31.460 | a kind-hearted idiot struggle with this form.

01:19:35.360 | Maybe that's what's compelling.

01:19:37.180 | But in terms of like actual being a good engineer

01:19:40.580 | of human robot interaction systems, I think I'm good.

01:19:45.540 | But it's hard to know until you do it

01:19:47.980 | and then the world keeps telling you you're not.

01:19:50.380 | And it's just full of doubt.

01:19:52.900 | And it's really hard.

01:19:53.820 | And I've been struggling with that recently.

01:19:55.540 | It's kind of a fascinating struggle,

01:19:57.180 | but then that's where the Goggins thing comes in

01:19:59.780 | is like, aside from the stay hard motherfucker,

01:20:03.900 | is the like, whenever you're struggling,

01:20:07.860 | that's a good sign that if you keep going,

01:20:11.180 | that you're going to be alone in the success, right?

01:20:14.580 | - Well, in your case, however, I agree.

01:20:18.660 | And actually David had a post recently

01:20:20.340 | that I thought was among his many brilliant posts

01:20:23.220 | was one of the more brilliant about how,

01:20:26.120 | you know, he talked about this myth of the light

01:20:27.860 | at the end of the tunnel.

01:20:29.520 | And instead what he replaced that myth with

01:20:33.460 | was a concept that eventually your eyes adapt to the dark.

01:20:38.460 | That the tunnel, it's not about a light at the end,

01:20:40.320 | that it's really about adapting to the dark of the tunnel.

01:20:42.780 | It was very Goggins- - I love him so much.

01:20:44.820 | - Yeah, you guys share a lot in common,

01:20:48.880 | knowing you both a bit, you know, share a lot in common.

01:20:52.500 | But in this loneliness and the pursuit of this dream,

01:20:57.500 | it seems to me it has a certain component to it

01:21:01.320 | that is extremely valuable,

01:21:04.060 | which is that the loneliness itself

01:21:06.700 | could serve as a driver to build the companion

01:21:09.720 | for the journey.

01:21:10.560 | - Well, I'm very deeply aware of that.

01:21:12.800 | So like some people can make,

01:21:17.280 | 'cause I talk about love a lot,

01:21:18.620 | I really love everything in this world,

01:21:21.900 | but I also love humans, friendship and romantic, you know,

01:21:26.620 | like even the cheesy stuff, just-

01:21:31.020 | - You like romantic movies.

01:21:32.460 | - Yeah, not those, not necessarily.

01:21:35.140 | - Well, I got so much shit from Rogan about like,

01:21:38.340 | what is it, the tango scene from "Scent of a Woman."

01:21:41.100 | But yes, I find like there's nothing better

01:21:43.980 | than a woman in a red dress, like, you know,

01:21:47.500 | it's just like classy.

01:21:49.420 | - You should move to Argentina, my friend.

01:21:51.020 | You know, my father is Argentine.

01:21:52.240 | And you know what he said when I went on your podcast

01:21:54.780 | for the first time, he said, "He dresses well."

01:21:58.340 | Because in Argentina, the men go to a wedding

01:22:00.240 | or a party or something.

01:22:01.460 | You know, in the US, by halfway through the night,

01:22:03.660 | 10 minutes in the night, all the jackets are off.

01:22:05.780 | It looks like everyone's undressing for the party

01:22:07.600 | they just got dressed up for.

01:22:08.940 | And he said, "You know, I like the way he dresses."

01:22:11.620 | And then when I started, he was talking about you.

01:22:13.180 | And then when I started my podcast, he said,

01:22:15.740 | "Why don't you wear a real suit like your friend Lex?"

01:22:19.400 | [laughing]

01:22:20.460 | - I remember that.

01:22:21.300 | - Any case.

01:22:22.140 | But let's talk about this pursuit just a bit more,

01:22:27.820 | because I think what you're talking about is building a,

01:22:31.640 | not just a solution for loneliness,

01:22:33.900 | but you've alluded to the loneliness

01:22:36.060 | as itself an important thing.

01:22:37.800 | And I think you're right.

01:22:38.640 | I think within people, there is like caverns of thoughts

01:22:44.100 | and shame, but also just the desire to be,

01:22:47.660 | to have resonance, to be seen and heard.

01:22:51.220 | And I don't even know that it's seen

01:22:52.420 | and heard through language.

01:22:53.780 | But these reservoirs of loneliness, I think,

01:22:58.200 | well, they're interesting.

01:23:01.340 | Maybe you could comment a little bit about it,

01:23:02.620 | because just as often as you talk about love,

01:23:05.160 | I haven't quantified it,

01:23:06.100 | but it seems that you talk about this loneliness.

01:23:08.440 | And maybe you just would, if you're willing,

01:23:10.180 | you could share a little bit more about that

01:23:12.220 | and what that feels like now in the pursuit

01:23:15.300 | of building this robot-human relationship.

01:23:18.360 | And you've been, let me be direct.

01:23:20.460 | You've been spending a lot of time

01:23:21.940 | on building a robot-human relationship.

01:23:25.420 | Where's that at?

01:23:27.220 | - Oh, in terms of business and in terms of systems?

01:23:33.380 | - No, I'm talking about a specific robot.

01:23:35.540 | - Oh, robot.

01:23:37.100 | So, okay, I should mention a few things.

01:23:39.860 | So, one is there's a startup where there's an idea

01:23:42.660 | where I hope millions of people can use.

01:23:46.000 | And then there's my own personal,

01:23:49.260 | almost like Frankenstein explorations

01:23:52.060 | with particular robots.

01:23:55.060 | So, I'm very fascinated with the legged robots

01:23:59.220 | in my own private, sounds like dark,

01:24:03.140 | but n-of-one experiments to see if I can recreate the magic.

01:24:09.620 | And that's been, I have a lot of really good already,

01:24:14.160 | perception systems and control systems

01:24:17.100 | that are able to communicate affection

01:24:19.240 | in a dog-like fashion.

01:24:20.780 | So, I'm in a really good place there.

01:24:22.520 | The stumbling blocks,

01:24:23.560 | which also have been part of my sadness recently,

01:24:27.080 | is that I also have to work with robotics companies.

01:24:30.740 | That I gave so much of my heart, soul,

01:24:34.620 | and love and appreciation towards Boston Dynamics,

01:24:37.940 | but Boston Dynamics is also,

01:24:41.940 | is a company that has to make a lot of money

01:24:43.860 | and they have marketing teams.

01:24:45.740 | And they're like looking at this silly Russian kid

01:24:47.980 | in a suit and tie.

01:24:49.340 | It's like, what's he trying to do with all this love

01:24:51.420 | and robot interaction and dancing and so on?

01:24:53.480 | So, there was a, I think, let's say for now,

01:24:58.480 | it's like when you break up with a girlfriend or something.

01:25:01.180 | Right now, we decided to part ways on this particular thing.

01:25:04.060 | They're huge supporters of mine, they're huge fans,

01:25:05.900 | but on this particular thing,

01:25:08.660 | Boston Dynamics is not focusing on

01:25:12.640 | or interested in human-robot interaction.

01:25:15.060 | In fact, their whole business currently

01:25:17.180 | is keep the robot as far away from humans as possible.

01:25:20.000 | Because it's in the industrial setting

01:25:23.800 | where it's doing monitoring in dangerous environments.

01:25:27.300 | It's almost like a remote security camera,

01:25:29.460 | essentially is its application.

01:25:31.000 | To me, I thought it's still,

01:25:34.560 | even in those applications, exceptionally useful

01:25:37.120 | for the robot to be able to perceive humans, like see humans

01:25:41.440 | and to be able to, in a big map,

01:25:45.520 | localize where those humans are and have human intention.

01:25:48.400 | For example, like this,

01:25:49.820 | I did this a lot of work with pedestrians

01:25:52.080 | for a robot to be able to anticipate

01:25:53.860 | what the hell the human is doing, like where it's walking.

01:25:57.540 | Humans are not ballistics object.

01:25:59.200 | They're not, just because you're walking this way one moment

01:26:01.760 | doesn't mean you'll keep walking that direction.

01:26:03.820 | You have to infer a lot of signals,

01:26:05.400 | especially with the head movement and the eye movement.

01:26:07.720 | And so I thought that's super interesting to explore,

01:26:09.760 | but they didn't feel that.

01:26:11.680 | So I'll be working with a few other robotics companies

01:26:14.500 | that are much more open to that kind of stuff.

01:26:18.360 | And they're super excited and fans of mine.

01:26:20.680 | And hopefully Boston Dynamics, my first love,

01:26:23.440 | that getting back with an ex-girlfriend will come around.

01:26:26.000 | But so the algorithmically, I'm basically done there.

01:26:33.360 | The rest is actually getting some of these companies

01:26:36.700 | to work with.

01:26:37.540 | And then there's, for people who'd work with robots

01:26:41.300 | know that one thing is to write software that works

01:26:44.900 | and the other is to have a real machine that actually works.

01:26:48.440 | And it breaks down in all kinds of different ways

01:26:50.860 | that are fascinating.

01:26:51.720 | And so there's a big challenge there.

01:26:53.940 | But that's almost, it may sound a little bit confusing

01:26:58.760 | in the context of our previous discussion

01:27:01.300 | because the previous discussion was more about the big dream,

01:27:04.580 | how I hoped to have millions of people

01:27:07.700 | enjoy this moment of magic.

01:27:09.420 | The current discussion about a robot

01:27:12.160 | is something I personally really enjoy,

01:27:15.100 | just brings me happiness.

01:27:16.620 | I really try to do now everything that just brings me joy.

01:27:20.960 | I maximize that 'cause robots are awesome.

01:27:24.180 | But two, given my little bit growing platform,

01:27:28.440 | I wanna use the opportunity to educate people.

01:27:31.820 | It's just like robots are cool.

01:27:34.100 | And if I think they're cool, I'll be able to,

01:27:36.620 | I hope be able to communicate why they're cool to others.

01:27:39.620 | So this little robot experiment

01:27:42.140 | is a little bit of a research project too.

01:27:43.980 | There's a couple of publications with MIT folks around that.

01:27:47.020 | But the other is just to make some cool videos

01:27:49.780 | and explain to people how they actually work.

01:27:52.200 | And as opposed to people being scared of robots,

01:27:56.640 | they can still be scared, but also excited.

01:27:59.860 | Like see the dark side, the beautiful side,

01:28:03.120 | the magic of what it means to bring,

01:28:05.540 | for a machine to become a robot.

01:28:09.780 | I want to inspire people with that, but that's less,

01:28:13.860 | it's interesting because I think the big impact

01:28:18.140 | in terms of the dream does not have to do with embodied AI.

01:28:22.340 | So it does not need to have a body.

01:28:24.020 | I think the refrigerator is enough that for an AI system,

01:28:28.680 | just to have a voice and to hear you,

01:28:31.320 | that's enough for loneliness.

01:28:33.400 | The embodiment is just-

01:28:36.520 | - By embodiment you mean the physical structure.

01:28:38.480 | - Physical instantiation of intelligence.

01:28:41.280 | So it's a legged robot or even just a thing.

01:28:44.780 | I have a few other humanoid robot,

01:28:48.520 | a little humanoid robot, maybe I'll keep them on the table.

01:28:51.320 | It's like walks around or even just like a mobile platform

01:28:54.720 | that can just like turn around and look at you.

01:28:57.540 | It's like we mentioned with the pen,

01:28:59.200 | something that moves and can look at you.

01:29:02.060 | It's like that butter robot that asks, what is my purpose?

01:29:07.060 | That is really, it's almost like art.

01:29:15.640 | There's something about a physical entity that moves around,

01:29:19.660 | that's able to look at you and interact with you

01:29:22.620 | that makes you wonder what it means to be human.

01:29:25.920 | It like challenges you to think,

01:29:27.500 | if that thing looks like it has consciousness,

01:29:31.200 | what the hell am I?

01:29:33.940 | And I like that feeling.

01:29:35.100 | I think that's really useful for us.

01:29:36.500 | It's humbling for us humans, but that's less about research.

01:29:40.440 | It's certainly less about business

01:29:42.660 | and more about exploring our own selves

01:29:45.960 | and challenging others to think about what makes them human.

01:29:50.960 | - I love this desire to share the delight

01:29:56.380 | of an interaction with a robot.

01:29:58.040 | And as you describe it,

01:29:58.880 | I actually find myself starting to crave that

01:30:01.340 | because we all have those elements from childhood

01:30:04.920 | or from adulthood where we experience something,

01:30:06.760 | we want other people to feel that.

01:30:09.800 | And I think that you're right.

01:30:10.920 | I think a lot of people are scared of AI.

01:30:12.560 | I think a lot of people are scared of robots.

01:30:14.440 | My only experience of a robotic like thing

01:30:18.840 | is my Roomba vacuum where it goes about,

01:30:21.840 | actually was pretty good at picking up Costello's hair

01:30:24.280 | when he was shed and I was grateful for it.

01:30:28.260 | But then when I was on a call or something

01:30:30.440 | and it would get caught on a wire or something,

01:30:33.200 | I would find myself getting upset with the Roomba.

01:30:35.280 | In that moment, I'm like, what are you doing?

01:30:37.520 | And obviously it's just doing what it does.

01:30:40.320 | But that's a kind of mostly positive

01:30:42.480 | but slightly negative interaction.

01:30:44.760 | But what you're describing has so much more richness

01:30:48.600 | and layers of detail that I can only imagine

01:30:51.640 | what those relationships are like.

01:30:53.660 | - Well, there's a few, just a quick comment.

01:30:55.380 | So I've had, they're currently in Boston,

01:30:57.680 | I have a bunch of Roombas for my robot.

01:31:00.540 | And I did this experiment.

01:31:01.920 | - Wait, how many Roombas?

01:31:03.120 | Sounds like a fleet of Roombas.

01:31:05.880 | - Yeah, so probably seven or eight.

01:31:08.640 | - Wow, that's a lot of Roombas.

01:31:09.960 | So this place is very clean.

01:31:11.780 | (both laughing)

01:31:12.800 | Well, so this, I'm kind of waiting.

01:31:14.740 | This is the place we're currently in in Austin

01:31:18.440 | is way larger than I need.

01:31:20.280 | But I basically got it to make sure I have room for robots.

01:31:26.420 | - So you have these seven or so Roombas,

01:31:30.400 | you deploy all seven at once?

01:31:32.100 | - Oh no, I do different experiments with them.

01:31:36.040 | So one of the things I wanna mention is this is,

01:31:38.860 | I think there was a YouTube video

01:31:40.140 | that inspired me to try this,

01:31:41.780 | is I got them to scream in pain and moan in pain

01:31:47.800 | whenever they were kicked or contacted.

01:31:52.440 | And I did that experiment to see how I would feel.

01:31:56.680 | I meant to do like a YouTube video on it,

01:31:58.800 | but then it just seemed very cruel.

01:32:00.360 | - Did any Roomba rights activists come after you?

01:32:03.420 | - Like, I think if I release that video,

01:32:07.600 | I think it's gonna make me look insane,

01:32:09.680 | which I know people know I'm already insane.

01:32:12.760 | - Now you have to release the video.

01:32:14.360 | (both laughing)

01:32:15.880 | - I think maybe if I contextualize it

01:32:18.300 | by showing other robots,

01:32:20.320 | like to show why this is fascinating,

01:32:23.200 | because ultimately I felt like they were human

01:32:26.020 | almost immediately.

01:32:27.360 | And that display of pain was what did that.

01:32:30.360 | - Giving them a voice.

01:32:31.480 | - Giving them a voice,

01:32:32.360 | especially a voice of dislike of pain.

01:32:37.180 | - I have to connect you to my friend, Eddie Chang.

01:32:39.200 | He studies speech and language.

01:32:40.480 | He's a neurosurgeon and we're lifelong friends.

01:32:42.960 | He studies speech and language,

01:32:46.360 | but he describes some of these more primitive,

01:32:50.160 | visceral vocalizations, cries, groans,

01:32:54.020 | moans of delight, other sounds as well,

01:32:57.480 | use your imagination,

01:32:58.940 | as such powerful rudders for the emotions of other people.

01:33:03.940 | And so I find it fascinating.

01:33:06.080 | I can't wait to see this video.

01:33:07.560 | So is the video available online?

01:33:09.480 | - No, I haven't recorded it.

01:33:12.040 | I just hit a bunch of Roombas

01:33:13.680 | that are able to scream in pain in my Boston place.

01:33:18.240 | - People are ready as well.

01:33:22.320 | - Next podcast episode with Lex,

01:33:24.840 | maybe we'll have that one.

01:33:26.180 | Who knows?

01:33:27.020 | - So the thing is like people I've noticed

01:33:29.060 | because I talk so much about love

01:33:30.960 | and it's really who I am.

01:33:32.360 | I think they want to,

01:33:33.520 | to a lot of people,

01:33:35.800 | it seems like there's got to be a dark person

01:33:38.520 | in there somewhere.

01:33:39.440 | And I thought if I release videos

01:33:40.960 | and Roomba's screaming and they're like, yep, yep,

01:33:43.040 | that guy's definitely insane.

01:33:44.560 | - What about like shouts of glee and delight?

01:33:47.760 | You could do that too, right?

01:33:49.180 | - Well, I don't know how to,

01:33:50.360 | I don't, how to, to me delight is quiet, right?

01:33:54.420 | Like a-

01:33:55.260 | - You're Russian.

01:33:56.080 | Americans are much louder than Russians.

01:33:59.840 | - Yeah, yeah.

01:34:01.000 | But I don't, I mean, unless you're talking about like,

01:34:04.760 | I don't know how you would have sexual relations

01:34:06.480 | with a Roomba.

01:34:07.320 | - Well, I wasn't necessarily saying sexual delight, but-

01:34:10.260 | - Trust me, I tried, I'm just kidding.

01:34:12.680 | That's a joke, internet.

01:34:14.080 | Okay, but I was fascinated in the psychology

01:34:16.760 | of how little it took.

01:34:17.720 | 'Cause you mentioned you had a negative relationship

01:34:20.560 | with a Roomba a little bit.

01:34:21.920 | - Well, I'd find that mostly I took it for granted.

01:34:25.160 | It just served me.

01:34:26.040 | It collected Costello's hair.

01:34:27.560 | And then when it would do something I didn't like,

01:34:29.200 | I would get upset with it.

01:34:30.240 | So that's not a good relationship.

01:34:31.760 | It was taken for granted and I would get upset

01:34:35.000 | and then I'd park it again.

01:34:36.060 | And I just like, you're, you're in the, in the corner.

01:34:39.040 | - Yeah, but there's a way to frame it's,

01:34:43.400 | it being quite dumb as a almost cute, you know,

01:34:48.400 | almost connecting with it for its dumbness.

01:34:51.120 | And I think that's a artificial intelligence problem.

01:34:54.000 | I think flaws are, should be a feature, not a bug.

01:34:59.280 | So along the lines of this, the different sorts

01:35:01.860 | of relationships that one could have with robots

01:35:03.800 | and the fear, but also some of the positive relationships

01:35:06.900 | that one could have, there's so much dimensionality,

01:35:10.320 | there's so much to explore.

01:35:12.400 | But power dynamics in relationships are very interesting

01:35:16.300 | because the obvious ones that the unsophisticated view

01:35:20.640 | of this is, you know, one, there's a master and a servant,

01:35:24.360 | right, but there's also manipulation.

01:35:27.800 | There's benevolent manipulation.

01:35:30.180 | You know, children do this with parents, puppies do this.

01:35:33.120 | Puppies turn their head and look cute

01:35:34.920 | and maybe give out a little noise.

01:35:37.560 | Kids, coo, and parents always think that they're, you know,

01:35:41.000 | they're doing this because, you know,

01:35:43.020 | they love the parent, but in many ways,

01:35:46.520 | studies show that those coos are ways to extract

01:35:49.320 | the sorts of behaviors and expressions

01:35:50.820 | from the parent that they want.

01:35:51.800 | The child doesn't know it's doing this,

01:35:53.160 | it's completely subconscious,

01:35:54.260 | but it's benevolent manipulation.

01:35:56.460 | So there's one version of fear of robots

01:35:59.720 | that I hear a lot about that I think most people

01:36:02.160 | can relate to where the robots take over

01:36:04.520 | and they become the masters and we become the servants.

01:36:08.140 | But there could be another version that, you know,

01:36:12.600 | in certain communities that I'm certainly not a part of,

01:36:15.140 | but they call topping from the bottom,

01:36:17.200 | where the robot is actually manipulating you

01:36:20.780 | into doing things, but you are under the belief

01:36:24.180 | that you are in charge, but actually they're in charge.

01:36:29.120 | And so I think that's one that if we could explore that

01:36:33.920 | for a second, you could imagine it wouldn't necessarily

01:36:36.480 | be bad, although it could lead to bad things.

01:36:39.100 | The reason I want to explore this is I think people

01:36:41.800 | always default to the extreme, like the robots take over

01:36:45.560 | and we're in little jail cells and they're out having fun

01:36:48.100 | and ruling the universe.

01:36:50.600 | What sorts of manipulation can a robot potentially carry out,

01:36:55.260 | good or bad?

01:36:56.500 | - Yeah, so there's a lot of good and bad manipulation

01:36:59.740 | between humans, right?

01:37:00.940 | Just like you said, to me, especially like you said,

01:37:05.940 | topping from the bottom, is that the term?

01:37:12.340 | - I think someone from MIT told me that term.

01:37:14.340 | [laughing]

01:37:15.420 | Wasn't Lex.

01:37:18.460 | - So first of all, there's power dynamics in bed

01:37:22.040 | and power dynamics in relationships and power dynamics

01:37:25.140 | on the street and in the work environment,

01:37:26.880 | those are all very different.

01:37:28.280 | I think power dynamics can make human relationships,

01:37:34.280 | especially romantic relationships, fascinating and rich

01:37:39.740 | and fulfilling and exciting and all those kinds of things.

01:37:42.740 | So I don't think in themselves they're bad.

01:37:48.740 | And the same goes with robots.

01:37:50.820 | I really love the idea that a robot would be a top

01:37:54.220 | or a bottom in terms of like power dynamics.

01:37:57.500 | And I think everybody should be aware of that.

01:37:59.980 | And the manipulation is not so much manipulation,

01:38:02.460 | but a dance of like pulling away, a push and pull

01:38:06.700 | and all those kinds of things.

01:38:08.340 | In terms of control, I think we're very, very, very far away

01:38:12.860 | from AI systems that are able to lock us up.

01:38:16.340 | They, to lock us up in, like to have so much control

01:38:21.340 | that we basically cannot live our lives

01:38:25.140 | in the way that we want.

01:38:26.920 | I think there's a, in terms of dangers of AI systems,

01:38:29.680 | there's much more dangers that have to do

01:38:31.260 | with autonomous weapon systems and all those kinds of things.

01:38:34.240 | So the power dynamics as exercised in the struggle

01:38:37.900 | between nations and war and all those kinds of things.

01:38:40.660 | But in terms of personal relationships,

01:38:42.740 | I think power dynamics are a beautiful thing.

01:38:45.900 | Now there is, of course, going to be all those kinds

01:38:48.340 | of discussions about consent and rights

01:38:52.460 | and all those kinds of things.

01:38:53.300 | - Well, here we're talking, I always say, you know,

01:38:54.820 | in any discussion around this,

01:38:56.420 | if we need to define really the context,

01:38:59.180 | it's always, always should be consensual, age appropriate,

01:39:03.280 | context appropriate, species appropriate.

01:39:06.060 | But now we're talking about human robot interactions.

01:39:09.200 | And so I guess that-

01:39:10.780 | - No, I actually was trying to make a different point,

01:39:13.620 | which is I do believe that robots will have rights

01:39:17.020 | down the line.

01:39:17.860 | And I think in order for us to have deep meaningful

01:39:21.780 | relationship with robots,

01:39:23.060 | we would have to consider them as entities in themselves

01:39:26.580 | that deserve respect.

01:39:28.360 | And that's a really interesting concept

01:39:31.660 | that I think people are starting to talk about

01:39:34.360 | a little bit more, but it's very difficult

01:39:36.340 | for us to understand how entities that are other than human.

01:39:39.700 | I mean, the same as with dogs and other animals

01:39:42.460 | can have rights on a level as humans.

01:39:44.900 | - Well, yeah, I mean, we can't and nor should we

01:39:48.220 | do whatever we want with animals.

01:39:49.940 | We have a USDA, we have departments of agriculture

01:39:54.700 | that deal with, you know, animal care and use committees

01:39:58.260 | for research, for farming and ranching and all that.

01:40:02.060 | So I, while when you first said it, I thought, wait,

01:40:05.220 | why would there be a bill of robotic rights?

01:40:07.380 | But it absolutely makes sense in the context of everything

01:40:11.240 | we've been talking about up until now.

01:40:13.860 | Let's, if you're willing, I'd love to talk about dogs

01:40:18.700 | because you've mentioned dogs a couple of times,

01:40:21.260 | a robot dog, you had a biological dog, yeah.

01:40:26.260 | - Yeah, I had a Newfoundland named Homer

01:40:31.300 | for many years growing up.

01:40:34.220 | - In Russia or in the US?

01:40:35.620 | - In the United States.

01:40:37.300 | And he was about, he was over 200 pounds.

01:40:40.140 | That's a big dog. - That's a big dog.

01:40:41.840 | - If people know, people know Newfoundland,

01:40:43.880 | so he's this black dog that's a really long hair

01:40:48.580 | and just a kind soul.

01:40:50.460 | I think perhaps that's true for a lot of large dogs,

01:40:53.220 | but he thought he was a small dog.

01:40:55.200 | So he moved like that and--

01:40:56.500 | - Was he your dog?

01:40:57.380 | - Yeah, yeah.

01:40:58.340 | - So you had him since he was fairly young?

01:41:00.480 | - Oh, since, yeah, since the very, very beginning

01:41:02.720 | to the very, very end.

01:41:03.760 | And one of the things, I mean, he had this kind of,

01:41:08.620 | we mentioned like the Roombas,

01:41:11.300 | he had a kind hearted dumbness about him

01:41:15.540 | that was just overwhelming.

01:41:17.120 | Part of the reason I named him Homer

01:41:20.020 | because it's after Homer Simpson,

01:41:22.740 | in case people are wondering which Homer I'm referring to.

01:41:25.260 | I'm not, you know.

01:41:26.420 | So there's a-- - Not "The Odyssey."

01:41:29.580 | - Yeah, exactly.

01:41:30.880 | There's a clumsiness that was just something

01:41:35.120 | that immediately led to a deep love for each other.

01:41:37.960 | And one of the, I mean, he was always,

01:41:42.300 | it's the shared moments.

01:41:43.320 | He was always there for so many nights together.

01:41:46.500 | That's a powerful thing about a dog that he was there

01:41:51.080 | through all the loneliness, through all the tough times,

01:41:53.540 | through the successes and all those kinds of things.

01:41:55.820 | And I remember, I mean,

01:41:57.740 | that was a really moving moment for me.

01:42:00.100 | I still miss him to this day.

01:42:02.140 | - How long ago did he die?

01:42:03.500 | - Maybe 15 years ago.

01:42:07.060 | So it's been a while.

01:42:09.140 | But it was the first time I've really experienced

01:42:13.920 | like the feeling of death.

01:42:15.680 | So what happened is he got cancer

01:42:21.980 | and so he was dying slowly.

01:42:26.020 | And then at a certain point he couldn't get up anymore.

01:42:29.680 | There's a lot of things I could say here

01:42:31.780 | you know, that I struggle with.

01:42:33.960 | That maybe he suffered much longer than he needed to.

01:42:38.960 | That's something I really think about a lot.

01:42:42.060 | But I remember I had to take him to the hospital

01:42:47.060 | and the nurses couldn't carry him, right?

01:42:52.160 | So you talk about 200 pound dog.

01:42:54.420 | I was really into powerlifting at the time.

01:42:56.680 | I remember like they tried to figure out

01:42:59.340 | all these kinds of ways to,

01:43:01.740 | so in order to put them to sleep,

01:43:03.580 | they had to take them into a room.

01:43:07.220 | And so I had to carry him everywhere.

01:43:09.500 | And here's this dying friend of mine

01:43:13.540 | that I just had to,

01:43:15.220 | first of all, it's really difficult to carry

01:43:16.980 | somebody that heavy when they're not helping you out.

01:43:20.100 | And yeah, so I remember it was the first time

01:43:25.100 | seeing a friend laying there

01:43:28.460 | and seeing wife drained from his body.

01:43:34.040 | And that realization that we're here for a short time

01:43:38.360 | was made so real.

01:43:40.580 | That here's a friend that was there for me

01:43:42.760 | the week before, the day before, and now he's gone.

01:43:46.240 | And that was, I don't know,

01:43:49.180 | that spoke to the fact that he could be deeply connected

01:43:52.400 | with the dog.

01:43:53.240 | Also spoke to the fact that the shared moments together

01:44:00.800 | that led to that deep friendship will make life so amazing,

01:44:05.800 | but also spoke to the fact that death is a motherfucker.

01:44:11.700 | So I know you've lost Costello recently

01:44:16.120 | and you've been going. - Yeah,

01:44:16.960 | and as you're saying this,

01:44:17.780 | I'm definitely fighting back the tears.

01:44:20.180 | Thank you for sharing that,

01:44:23.320 | that I guess we're about to both cry over our dead dogs.

01:44:28.680 | It was bound to happen just given when this is happening.

01:44:32.620 | Yeah, it's-

01:44:35.060 | - How long did you know that Costello was not doing well?

01:44:38.480 | - Well, let's see.

01:44:41.000 | A year ago, during the start of,

01:44:44.280 | about six months into the pandemic,

01:44:46.440 | he started getting abscesses and he was not,

01:44:49.060 | his behavior changed and something really changed.

01:44:51.560 | And then I put him on testosterone

01:44:55.760 | because, which helped a lot of things.

01:44:58.100 | It certainly didn't cure everything,

01:44:59.300 | but it helped a lot of things.

01:45:00.600 | He was dealing with joint pain, sleep issues,

01:45:03.880 | and then it just became a very slow decline

01:45:07.740 | to the point where two, three weeks ago,

01:45:11.360 | he had a closet full of medication.

01:45:15.320 | I mean, this dog was, yeah, it was like a pharmacy.

01:45:17.360 | It's amazing to me when I looked at it the other day,

01:45:19.880 | still haven't cleaned up and removed all his things

01:45:22.080 | 'cause I can't quite bring myself to do it, but...

01:45:24.480 | - Do you think he was suffering?

01:45:27.480 | - Well, so what happened was about a week ago,

01:45:30.300 | it was really just about a week ago, it's amazing.

01:45:32.480 | He was going up the stairs, I saw him slip,

01:45:35.020 | and he was a big dog.

01:45:35.860 | He wasn't 200 pounds, but he was about 90 pounds,

01:45:37.740 | but he's a bulldog, that's pretty big, and he was fit.

01:45:41.200 | And then I noticed that he wasn't carrying a foot

01:45:43.920 | in the back like it was injured.

01:45:45.160 | It had no feeling at all.

01:45:46.400 | He never liked me to touch his hind paws and I could do,

01:45:48.860 | that thing was just flopping there.

01:45:50.480 | And then the vet found some spinal degeneration

01:45:53.840 | and I was told that the next one would go.

01:45:55.720 | Did he suffer?

01:45:57.300 | I sure hope not, but something changed in his eyes.

01:46:01.420 | - Yeah. - Yeah, it's the eyes again.

01:46:02.840 | I know you and I spend long hours on the phone

01:46:05.120 | and talking about like the eyes and what they convey

01:46:07.320 | and what they mean about internal states

01:46:09.180 | and forsaken robots and biology of other kinds, but-

01:46:12.800 | - Do you think something about him was gone in his eyes?

01:46:17.800 | - I think he was real, here I am anthropomorphizing.

01:46:22.860 | I think he was realizing that one of his great joys in life,

01:46:26.840 | which was to walk and sniff and pee on things.

01:46:31.840 | [laughter]

01:46:33.120 | This dog loved to pee on things, it was amazing.

01:46:37.240 | I've wondered where he put it.

01:46:38.900 | He was like a reservoir of urine, it was incredible.

01:46:42.280 | I think, oh, that's it, he'd put like one drop

01:46:44.840 | on the 50 millionth plant,

01:46:46.560 | and then we get to the 50 millionth in one plant

01:46:49.120 | and he'd just have, you know, leave a puddle.

01:46:51.220 | And here I am talking about Costello peeing.

01:46:54.520 | He was losing that ability to stand up and do that.

01:46:57.020 | He was falling down while he was doing that.

01:46:58.880 | And I do think he started to realize,

01:47:01.560 | and the passage was easy and peaceful,

01:47:04.560 | but you know, I'll say this, I'm not ashamed to say it.

01:47:09.140 | I mean, I wake up every morning since then just,

01:47:11.480 | I don't even make the conscious decision

01:47:13.240 | to allow myself to cry, I wake up crying.

01:47:16.080 | And I'm fortunately able to make it through the day

01:47:18.160 | thanks to the great support of my friends

01:47:19.960 | and you and my family, but I miss him, man.

01:47:24.080 | - You miss him?

01:47:24.920 | - Yeah, I miss him.

01:47:25.760 | And I feel like he, you know, Homer, Costello,

01:47:29.180 | you know, the relationship to one's dog is so specific, but.

01:47:33.000 | - So that party is gone.

01:47:36.060 | That's the hard thing.

01:47:39.520 | - You know, what I think is different

01:47:41.880 | is that I made the mistake, I think.

01:47:50.600 | I hope it was a good decision,

01:47:51.800 | but sometimes I think I made the mistake

01:47:53.260 | of I brought Costello a little bit to the world

01:47:56.480 | through the podcast, through posting about him.

01:47:58.400 | I gave, I anthropomorphized about him in public.

01:48:01.560 | Let's be honest, I have no idea what his mental life was

01:48:03.700 | or his relationship to me.

01:48:04.840 | And I'm just exploring all this for the first time

01:48:06.800 | because he was my first dog,

01:48:07.760 | but I raised him since he was seven weeks.

01:48:09.680 | - Yeah, you got to hold it together.

01:48:11.020 | I noticed the episode you released on Monday,

01:48:14.960 | you mentioned Costello.

01:48:16.560 | Like you brought him back to life for me

01:48:18.900 | for that brief moment.

01:48:20.340 | - Yeah, but he's gone.

01:48:22.080 | - Well, that's the,

01:48:23.440 | he's going to be gone for a lot of people too.

01:48:26.920 | - Well, this is what I'm struggling with.

01:48:29.820 | I think that maybe, you're pretty good at this, Lila.

01:48:33.840 | Have you done this before?

01:48:36.180 | This is the challenge is I actually, part of me,

01:48:40.560 | I know how to take care of myself pretty well.

01:48:43.080 | Not perfectly, but pretty well.

01:48:44.700 | And I have good support.

01:48:46.000 | I do worry a little bit about how it's going to land

01:48:48.780 | and how people will feel.

01:48:50.200 | I'm concerned about their internalization.

01:48:54.040 | So that's something I'm still iterating on.

01:48:56.280 | - And they have to watch you struggle, which is fascinating.

01:48:59.360 | - Right, and I've mostly been shielding them from this,

01:49:01.620 | but what would make me happiest

01:49:04.000 | is if people would internalize

01:49:06.380 | some of Costello's best traits.

01:49:08.200 | And his best traits were that he was incredibly tough.

01:49:13.200 | I mean, he was a 22 inch neck, bulldog, the whole thing.

01:49:17.680 | He was just born that way.

01:49:18.960 | But what was so beautiful is that his toughness

01:49:22.320 | is never what he rolled forward.

01:49:24.040 | It was just how sweet and kind he was.

01:49:27.100 | And so if people can take that,

01:49:29.140 | then there's a win in there someplace.

01:49:32.400 | So.

01:49:33.240 | - I think there's some ways in which

01:49:34.880 | you should probably live on in your podcast too.

01:49:37.800 | You should, I mean, it's such a,

01:49:40.560 | one of the things I loved about his role in your podcast

01:49:45.800 | is that he brought so much joy to you.

01:49:48.740 | I'll mention the robots, right?

01:49:51.920 | I think that's such a powerful thing to bring that joy into,

01:49:56.300 | like allowing yourself to experience that joy,

01:49:59.380 | to bring that joy to others, to share it with others.

01:50:02.520 | That's really powerful.

01:50:03.820 | And I mean, not to, this is like the Russian thing is,

01:50:07.520 | it touched me when Louis CK had that moment

01:50:14.040 | that I keep thinking about in his show, Louis,

01:50:17.960 | where like an old man was criticizing Louis

01:50:20.240 | for whining about breaking up with his girlfriend.

01:50:22.840 | And he was saying like the most beautiful thing

01:50:27.100 | about love, I mean, a song that's catchy now,

01:50:32.480 | that's now making me feel horrible saying it,

01:50:35.660 | but like is the loss.

01:50:37.700 | The loss really also is making you realize

01:50:41.220 | how much that person, that dog meant to you.

01:50:46.340 | And like allowing yourself to feel that loss

01:50:48.700 | and not run away from that loss is really powerful.

01:50:51.420 | And in some ways that's also sweet,

01:50:55.060 | just like the love was, the loss is also sweet

01:50:58.080 | because you know that you felt a lot for that,

01:51:01.440 | for your friend.

01:51:03.760 | So I, and I continue bringing that joy.

01:51:07.140 | I think it would be amazing to the podcast.

01:51:09.400 | I hope to do the same with robots

01:51:13.680 | or whatever else is the source of joy, right?

01:51:16.920 | And maybe you think about one day getting another dog.

01:51:21.920 | - Yeah, in time, you're hitting on all the key buttons here.

01:51:27.460 | I want that to, we're thinking about, you know,

01:51:32.700 | ways to kind of immortalize Costello in a way that's real,

01:51:35.620 | not just, you know, creating some little logo

01:51:38.780 | or something silly.

01:51:39.740 | You know, Costello, much like David Goggins is a person,

01:51:43.840 | but Goggins also has grown into kind of a verb.

01:51:47.140 | You're going to Goggins this or you're going to,

01:51:48.820 | and there's an adjective like that's extreme.

01:51:51.020 | Like I think that for me, Costello was all those things.

01:51:54.420 | He was a being, he was his own being.

01:51:56.820 | He was a noun, a verb and an adjective.

01:52:00.340 | So, and he had this amazing superpower

01:52:02.420 | that I wish I could get,

01:52:03.300 | which is this ability to get everyone else

01:52:05.060 | to do things for you without doing a damn thing.

01:52:07.800 | The Costello effect, as I call it.

01:52:10.260 | - So as an idea, I hope he lives on.

01:52:12.260 | - Yes, thank you for that.

01:52:14.440 | This actually has been very therapeutic for me,

01:52:16.740 | which actually brings me to a question.

01:52:22.360 | We're friends, we're not just co-scientists colleagues

01:52:26.720 | working on a project together and in the world

01:52:31.200 | that's somewhat similar.

01:52:34.760 | - Just two dogs.

01:52:37.200 | - Just two dogs, basically.

01:52:40.540 | But let's talk about friendship

01:52:42.700 | because I think that I certainly know as a scientist

01:52:49.140 | that there are elements that are very lonely

01:52:51.100 | of the scientific pursuit.

01:52:52.740 | There are elements of many pursuits that are lonely.

01:52:56.980 | Music, math always seemed to me

01:53:00.100 | like they're like the loneliest people.

01:53:02.400 | Who knows if that's true or not.

01:53:04.060 | Also people work in teams

01:53:05.380 | and sometimes people are surrounded by people

01:53:06.980 | interacting with people and they feel very lonely.

01:53:09.300 | But for me, and I think as well for you,

01:53:14.300 | friendship is an incredibly strong force

01:53:17.940 | in making one feel like certain things are possible

01:53:22.940 | or worth reaching for.

01:53:25.780 | Maybe even making us compulsively reach for them.

01:53:28.200 | So when you were growing up,

01:53:30.700 | you grew up in Russia until what age?

01:53:32.780 | - 13.

01:53:33.700 | - Okay, and then you moved directly to Philadelphia?

01:53:38.260 | - To Chicago.

01:53:39.820 | And then Philadelphia and San Francisco and Boston and so on

01:53:44.820 | but really to Chicago, that's when I went to high school.

01:53:48.260 | - Do you have siblings?

01:53:49.840 | - Older brother.

01:53:51.080 | - Most people don't know that.

01:53:52.540 | - Yeah, he is a very different person

01:53:57.880 | but somebody I definitely look up to.

01:53:59.580 | So he's a wild man, he's extrovert.

01:54:01.820 | He was into, I mean, so he's also a scientist,

01:54:06.820 | a bioengineer, but when we were growing up,

01:54:10.260 | he was the person who did drink and did every drug

01:54:15.260 | but also is the life of the party.

01:54:19.380 | And I just thought he was the,

01:54:21.100 | when you're the older brother, five years older,

01:54:23.420 | he was the coolest person that I always wanted to be him.

01:54:28.300 | So to that, he definitely had a big influence.

01:54:31.180 | But I think for me, in terms of friendship growing up,

01:54:35.620 | I had one really close friend.

01:54:40.060 | And then when I came here, I had another close friend,

01:54:42.280 | but I'm very, I believe, I don't know if I believe,

01:54:47.280 | but I draw a lot of strength from deep connections

01:54:52.520 | with other people and just a small number of people,

01:54:57.740 | just a really small number of people.

01:54:59.060 | That's when I moved to this country,

01:55:00.500 | I was really surprised how like there would be

01:55:04.140 | these large groups of friends, quote unquote,

01:55:08.280 | but the depth of connection was not there at all

01:55:12.260 | from my sort of perspective.

01:55:14.260 | Now I moved to the suburb of Chicago, was Naperville.

01:55:17.700 | It's more like a middle class, maybe upper middle class.

01:55:20.980 | So it's like people that cared more

01:55:23.500 | about material possessions than deep human connection.

01:55:26.460 | So that added to the thing.

01:55:28.420 | But I drew more meaning than almost anything else

01:55:34.020 | was from friendship early on.

01:55:35.620 | I had a best friend, his name was, his name is Yura.

01:55:40.180 | I don't know how to say it in English.

01:55:43.500 | - How do you say it in Russian?

01:55:44.700 | - Yura.

01:55:45.520 | - What's his last name?

01:55:46.420 | Do you remember his last name?

01:55:47.620 | - Mirkulov, Yura Mirkulov.

01:55:51.840 | So we just spent all our time together.

01:55:56.640 | There's also a group of friends, like, I don't know,

01:56:00.140 | it's like eight guys in Russia.

01:56:02.980 | Growing up, it's like parents didn't care

01:56:07.500 | if you're coming back at a certain hour.

01:56:09.660 | So we spent all day, all night just playing soccer,

01:56:13.340 | usually called football and just talking about life

01:56:18.060 | and all those kinds of things.

01:56:18.900 | Even at that young age.

01:56:20.780 | I think people in Russia and the Soviet Union

01:56:23.860 | grow up much quicker.

01:56:25.120 | I think the education system at the university level

01:56:30.900 | is world-class in the United States

01:56:33.740 | in terms of creating really big, powerful minds,

01:56:38.740 | at least it used to be, but I think that they aspire to that.

01:56:42.020 | But the education system for younger kids

01:56:46.380 | in the Soviet Union was incredible.

01:56:49.300 | They did not treat us as kids.

01:56:51.220 | The level of literature, Tolstoy, Dostoyevsky--

01:56:54.580 | - When you were just a small child?

01:56:55.980 | - Yeah.

01:56:56.820 | - Amazing.

01:56:57.660 | - And the level of mathematics.

01:57:00.300 | And you're made to feel like shit

01:57:02.060 | if you're not good at mathematics.

01:57:03.780 | Like we, I think in this country, there's more,

01:57:07.180 | like especially young kids 'cause they're so cute,

01:57:09.700 | like they're being babied.

01:57:11.500 | We only start to really push adults later in life.

01:57:15.260 | Like, so if you want to be the best in the world at this,

01:57:17.860 | then you get to be pushed.

01:57:19.460 | But we were pushed at a young age, everybody was pushed.

01:57:22.900 | And that brought out the best in people.

01:57:25.020 | I think they really forced people to discover,

01:57:29.020 | like discover themselves in the Goggin style,

01:57:31.820 | but also discover what they're actually passionate about,

01:57:35.140 | what they're not.

01:57:35.980 | - Was this true for boys and girls,

01:57:37.420 | were they pushed equally there?

01:57:38.940 | - Yeah, they were pushed equally, I would say.

01:57:41.940 | There was a, obviously there was more, not obviously,

01:57:45.540 | but at least from my memories, more of,

01:57:49.340 | what's the right way to put it?

01:57:52.020 | But there was like gender roles,

01:57:53.980 | but not in a negative connotation.

01:57:56.260 | It was the red dress versus the suit and tie

01:57:59.460 | kind of connotation, which is like,

01:58:01.560 | there's, you know, like guys like lifting heavy things

01:58:06.560 | and girls like creating beautiful art and, you know,

01:58:13.020 | like there's-

01:58:14.180 | - A more traditional view of gender, more 1950s, '60s.

01:58:18.200 | - But we didn't think in terms of, at least at that age,

01:58:20.380 | in terms of like roles and then like a homemaker

01:58:23.780 | or something like that, or no,

01:58:25.420 | it was more about what people care about.

01:58:28.220 | Like girls cared about this set of things

01:58:31.360 | and guys cared about this set of things.

01:58:33.240 | I think mathematics and engineering was something

01:58:36.500 | that guys cared about and sort of,

01:58:38.580 | at least my perception of that time.

01:58:40.460 | And then girls cared about beauty.

01:58:43.980 | So like guys want to create machines,

01:58:46.060 | girls want to create beautiful stuff.

01:58:47.740 | And now of course that I don't take that forward

01:58:52.740 | in some kind of philosophy of life,

01:58:54.820 | but it's just the way I grew up and the way I remember it.

01:58:57.500 | But all, everyone worked hard.

01:59:01.580 | The value of hard work was instilled in everybody.

01:59:06.500 | And through that, I think it's a little bit of hardship.

01:59:11.500 | Of course, also economically, everybody was poor,

01:59:14.940 | especially with the collapse of the Soviet Union.

01:59:17.100 | There's poverty everywhere.

01:59:18.620 | You didn't notice it as much,

01:59:19.820 | but there was a, because there's not much

01:59:22.660 | material possessions, there was a huge value

01:59:25.740 | placed on human connection.

01:59:28.020 | Just meeting with neighbors, everybody knew each other.

01:59:31.220 | We lived in an apartment building,

01:59:33.600 | very different than you have

01:59:34.820 | in the United States these days.

01:59:36.460 | Everybody knew each other.

01:59:37.780 | You would get together, drink vodka,

01:59:40.700 | smoke cigarettes and play guitar

01:59:42.380 | and sing sad songs about life.

01:59:47.260 | - What's with the sad songs and the Russian thing?

01:59:50.580 | I mean, Russians do express joy from time to time.

01:59:54.700 | - Yeah, they do.

01:59:55.900 | - Certainly you do.

01:59:57.540 | But what do you think that's about?

02:00:00.220 | Is it 'cause it's cold there?

02:00:01.320 | But it's cold other places too.

02:00:02.820 | - I think, so first of all, the Soviet Union,

02:00:08.340 | the echoes of World War II

02:00:12.100 | and the millions and millions and millions of people there,

02:00:14.660 | civilians that were slaughtered

02:00:17.020 | and also starvation is there, right?

02:00:20.180 | So like the echoes of that, of the ideas,

02:00:24.040 | the literature, the art is there.

02:00:25.900 | Like that's grandparents, that's parents, that's all there.

02:00:29.320 | So that contributes to it,

02:00:30.580 | that life can be absurdly, unexplainably cruel.

02:00:35.580 | At any moment, everything can change.

02:00:37.380 | So that's in there.

02:00:38.600 | Then I think there's an empowering aspect

02:00:40.920 | to finding beauty and suffering,

02:00:43.220 | but then everything else is beautiful too.

02:00:45.480 | Like if you just linger or it's like

02:00:47.660 | why you meditate on death is like if you just think about

02:00:50.800 | the worst possible case and find beauty in that,

02:00:53.100 | then everything else is beautiful too.

02:00:54.660 | And so you write songs about the dark stuff

02:00:57.420 | and that somehow helps you deal with whatever comes.

02:01:02.420 | There's a hopelessness to the Soviet Union

02:01:07.220 | that like inflation, all those kinds of things

02:01:10.280 | where people were sold dreams and never delivered.

02:01:15.420 | And so like, if you don't sing songs about sad things,

02:01:20.420 | you're going to become cynical about this world.

02:01:24.300 | - Interesting.

02:01:25.140 | - So they don't want to give into cynicism.

02:01:27.340 | Now, a lot of people did, one of the,

02:01:30.360 | but it's the battle against cynicism.

02:01:34.240 | One of the things that may be common in Russia

02:01:38.440 | is the kind of cynicism about,

02:01:41.040 | like if I told you the thing I said earlier

02:01:43.180 | about dreaming about robots, it's very common for people

02:01:46.460 | to dismiss that dream of saying, no, that's not,

02:01:51.100 | that's too wild.

02:01:52.180 | Like who else do you know that did that?

02:01:54.320 | Or you want to start a podcast, like who else?

02:01:57.000 | Like nobody's making money on podcasts.

02:01:58.780 | Like why do you want to start a podcast?

02:02:00.020 | That kind of mindset I think is quite common,

02:02:03.360 | which is why I would say entrepreneurship in Russia

02:02:07.520 | is still not very good, which to be a business,

02:02:11.300 | to be an entrepreneur, you have to dream big

02:02:13.460 | and you have to have others around you,

02:02:14.980 | like friends and support group that make you dream big.

02:02:19.080 | But if you don't give into cynicism

02:02:21.380 | and appreciate the beauty and the unfairness of life,

02:02:27.340 | the absurd unfairness of life,

02:02:29.940 | then I think it just makes you appreciative of everything.

02:02:34.500 | It's like a, it's a prerequisite for gratitude.

02:02:38.020 | And so, yeah, I think that instilled in me

02:02:42.060 | ability to appreciate everything,

02:02:43.980 | just like everything, everything's amazing.

02:02:46.740 | And then also there's a culture

02:02:48.900 | of like romanticizing everything.

02:02:56.100 | Like it's almost like romantic relationships

02:03:01.100 | were very like soap opera like.

02:03:04.280 | It's very like over the top dramatic.

02:03:07.680 | And I think that that was instilled in me too.

02:03:10.900 | Not only do I appreciate everything about life,

02:03:13.640 | but I get like emotional about it.

02:03:15.940 | In a sense, like I get like a visceral feeling of joy

02:03:20.600 | for everything and the same with friends

02:03:24.500 | or people of the opposite sex.

02:03:26.080 | Like there's a deep, like emotional connection there

02:03:30.700 | that like, that's like way too dramatic to like,

02:03:35.300 | I guess, relative to what the actual moment is,

02:03:38.760 | but I derive so much deep, like dramatic joy

02:03:43.760 | from so many things in life.

02:03:46.620 | And I think I would attribute that to bringing in Russia.

02:03:50.340 | But the thing that sticks most of all is the friendship.

02:03:54.000 | And I've now since then had one other friend like that

02:03:59.000 | in the United States, he lives in Chicago.

02:04:02.620 | His name is Matt.

02:04:04.300 | And slowly here and there accumulating

02:04:08.280 | really fascinating people,

02:04:09.840 | but I'm very selective with that.

02:04:11.540 | Funny enough, the few times, it's not few,

02:04:16.540 | it's a lot of times now interacting with Joe Rogan.

02:04:19.900 | It sounds surreal to say,

02:04:21.840 | but there was a kindred spirit there.

02:04:24.180 | Like I've connected with him.

02:04:25.620 | And there's been people like that also

02:04:28.300 | in the grappling sports that are really connected with.

02:04:31.600 | I've actually struggled, which is why I'm so glad

02:04:35.360 | to be your friend, is I've struggled

02:04:37.740 | to connect with scientists.

02:04:40.440 | - They can be a little bit wooden sometimes.

02:04:43.200 | Even the biologists, I mean, one thing that I'm--

02:04:45.940 | - Even the biologists.

02:04:47.000 | - Well, I'm so struck by the fact that you work with robots,

02:04:50.080 | you're an engineer, AI, science, technology,

02:04:52.940 | and that all sounds like hardware, right?

02:04:55.320 | But what you're describing, and I know is true about you,

02:04:58.260 | is this deep emotional life and this resonance.

02:05:01.640 | And it's really wonderful.

02:05:02.700 | I actually think it's one of the reasons

02:05:04.120 | why so many people, scientists and otherwise,

02:05:07.700 | have gravitated towards you and your podcast

02:05:09.900 | is because you hold both elements.

02:05:12.860 | In Hermann Hesse's book, I don't know if you read,

02:05:14.680 | "Narcissus and Goldman," right?

02:05:16.100 | It's about these elements of the logical, rational mind

02:05:19.400 | and the emotional mind and how those are woven together.

02:05:22.940 | And if people haven't read it, they should.

02:05:24.860 | And you embody the full picture.

02:05:27.300 | And I think that's so much of what draws people to you.

02:05:30.020 | - I've read every Hermann Hesse book, by the way.

02:05:31.940 | - As usual, as usual, I've done about 9% of what Lex is.

02:05:36.940 | No, it's true.

02:05:38.560 | Well, you mentioned Joe, who is a phenomenal human being,

02:05:42.360 | not just for his amazing accomplishments,

02:05:44.260 | but for how he shows up to the world one-on-one.

02:05:48.500 | I think I heard him say the other day on an interview,

02:05:51.620 | he said there is no public or private version of him.

02:05:55.660 | He's like, "This is me."

02:05:56.980 | And he said the word, it was beautiful.

02:05:58.340 | He said, "I'm like the fish that got through the net."

02:06:00.940 | There is no on-stage, off-stage version.

02:06:03.580 | You're absolutely right.

02:06:04.580 | And I, so, but well, you guys,

02:06:08.100 | I have a question actually about-

02:06:09.820 | - But that's a really good point

02:06:10.860 | about public and private life.

02:06:12.620 | He was a huge, if I could just comment real quick.

02:06:15.620 | Like that, he was, I've been a fan of Joe for a long time,

02:06:18.560 | but he's been an inspiration

02:06:20.620 | to not have any difference between public and private life.

02:06:25.060 | I actually had a conversation with Naval about this.

02:06:28.180 | And he said that you can't have a rich life,

02:06:33.180 | like an exciting life

02:06:36.620 | if you're the same person publicly and privately.

02:06:39.340 | And I think I understand that idea,

02:06:42.420 | but I don't agree with it.

02:06:45.380 | I think it's really fulfilling and exciting

02:06:49.100 | to be the same person privately and publicly

02:06:51.340 | with very few exceptions.

02:06:52.500 | Now, that said, I don't have any really strange sex kinks.

02:06:57.500 | So like, I feel like I can be open with basically everything.

02:07:00.420 | I don't have anything I'm ashamed of.

02:07:02.340 | There's some things that could be perceived poorly,

02:07:05.660 | like the screaming Arumbas, but I'm not ashamed of them.

02:07:09.040 | I just have to present them in the right context.

02:07:11.420 | But there is a freedom

02:07:14.020 | to being the same person in private as in public.

02:07:16.620 | And that, Joe made me realize that you can be that

02:07:22.460 | and also to be kind to others.

02:07:25.220 | It sounds kind of absurd,

02:07:28.620 | but I really always enjoyed being good to others.

02:07:33.620 | Just being kind towards others.

02:07:41.100 | But I always felt like the world didn't want me to be.

02:07:45.380 | Like there's so much negativity when I was growing up,

02:07:48.500 | like just around people.

02:07:49.700 | If you actually just notice how people talk,

02:07:52.520 | from like complaining about the weather,

02:07:57.460 | this could be just like the big cities that I visited in,

02:07:59.620 | but there's a general negativity

02:08:02.020 | and positivity is kind of suppressed.

02:08:05.820 | One, you're not seen as very intelligent.

02:08:08.660 | And two, you're seen as like a little bit of a weirdo.

02:08:13.220 | And so I always felt like I had to hide that.

02:08:15.780 | And what Joe made me realize, one,

02:08:17.780 | I can be fully just the same person, private and public.

02:08:22.260 | And two, I can embrace being kind

02:08:25.260 | and just in the way that I like,

02:08:28.260 | in the way I know how to do.

02:08:31.240 | And sort of for me on like on Twitter

02:08:33.900 | or like publicly, whenever I say stuff,

02:08:37.520 | that means saying stuff simply,

02:08:39.620 | almost to the point of cliche.

02:08:41.220 | And like, I have the strength now to say it,

02:08:44.540 | even if I'm being mocked, you know what I mean?

02:08:47.140 | Like just, it's okay.

02:08:48.680 | If everything's going to be okay, okay.

02:08:50.700 | Some people will think you're dumb.

02:08:52.500 | They're probably right.

02:08:53.520 | The point is like, it's just enjoy being yourself.

02:08:56.260 | And that Joe, more than almost anybody else,

02:08:58.300 | because he's so successful at it, inspired me to do that.

02:09:03.060 | Be kind and be the same person, private and public.

02:09:06.200 | - I love it.

02:09:07.040 | And I love the idea that authenticity

02:09:08.660 | doesn't have to be oversharing, right?

02:09:11.520 | That it doesn't mean you reveal every detail of your life.

02:09:14.460 | It's a way of being true to an essence of oneself.

02:09:19.000 | - Right, there's never a feeling when you deeply think

02:09:23.620 | and introspect that you're hiding something from the world

02:09:26.100 | or you're being dishonest in some fundamental way.

02:09:28.740 | So yeah, that's truly liberating.

02:09:33.480 | It allows you to think, it allows you to like think freely,

02:09:37.040 | to speak freely, just to be freely.

02:09:42.100 | That said, it's not like there's not still a responsibility

02:09:47.100 | to be the best version of yourself.

02:09:52.120 | So, I'm very careful with the way I say something.

02:09:57.120 | So the whole point, it's not so simple

02:10:00.680 | to express the spirit that's inside you with words.

02:10:04.660 | Some people are much better than others.

02:10:08.740 | I struggle, like oftentimes when I say something

02:10:12.740 | and I hear myself say it, it sounds really dumb

02:10:15.240 | and not at all what I meant.

02:10:16.700 | So that's the responsibility you have.

02:10:18.420 | It's not just like being the same person publicly

02:10:21.340 | and privately means you can just say whatever the hell.

02:10:24.460 | It means there's still a responsibility to try to be,

02:10:27.640 | to express who you truly are.

02:10:29.300 | And that's hard.

02:10:32.740 | - It is hard.

02:10:33.580 | And I think that we have this pressure,

02:10:37.640 | all people, when I say we, I mean all humans,

02:10:40.900 | and maybe robots too, feel this pressure

02:10:44.480 | to be able to express ourselves in that one moment,

02:10:47.560 | in that one form.

02:10:48.900 | And it is beautiful when somebody, for instance,

02:10:51.260 | can capture some essence of love or sadness

02:10:53.740 | or anger or something in a song or in a poem

02:10:57.160 | or in a short quote.

02:10:58.740 | But perhaps it's also possible to do it in aggregate,

02:11:02.740 | you know, all the things, you know, how you show up.

02:11:06.080 | For instance, one of the things that initially drew me

02:11:08.660 | to want to get to know you as a human being and a scientist,

02:11:11.420 | and eventually we became friends,

02:11:13.700 | was the level of respect that you brought

02:11:16.600 | to your podcast listeners by wearing a suit.

02:11:19.620 | I'm being serious here.

02:11:20.620 | You know, I was raised thinking

02:11:22.380 | that if you overdress a little bit,

02:11:24.340 | overdressed by American, certainly by American standards,

02:11:27.260 | you're overdressed for a podcast, but this is,

02:11:29.740 | but it's genuine, you're not doing it for any reason,

02:11:31.820 | except I have to assume, and I assumed at the time,

02:11:35.060 | that it was because you have a respect for your audience.

02:11:37.940 | You respect them enough to show up a certain way for them.

02:11:42.400 | It's for you also, but it's for them.

02:11:44.440 | And I think between that and your commitment

02:11:47.080 | to your friendships, the way that you talk about friendships

02:11:49.560 | and love and the way you hold up these higher ideals,

02:11:52.820 | I think at least as a consumer of your content

02:11:56.620 | and as your friend, what I find is that in aggregate,

02:12:01.760 | you're communicating who you are.

02:12:03.360 | It doesn't have to be one quote or something.

02:12:05.860 | And I think that we were sort of obsessed

02:12:08.140 | by like the one Einstein quote

02:12:10.120 | or the one line of poetry or something,

02:12:13.020 | but I think you so embody the way that, and Joe as well,

02:12:18.020 | it's about how you live your life and how you show up

02:12:21.080 | as a collection of things and said and done.

02:12:24.880 | - Yeah, that's fascinating.

02:12:25.840 | So the aggregate is the goal, the tricky thing,

02:12:30.380 | and Jordan Peterson talks about this

02:12:32.220 | because he's under attack way more than you and I

02:12:34.400 | will ever be, but that- - For now.

02:12:37.200 | - For now, right?

02:12:38.500 | This is very true for now.

02:12:40.420 | That the people who attack on the internet,

02:12:45.420 | this is one of the problems with Twitter,

02:12:49.080 | is they don't consider the aggregate.

02:12:53.160 | They take a single statements.

02:12:55.300 | And so one of the defense mechanisms,

02:12:58.420 | like again, why Joe has been an inspiration

02:13:01.140 | is that when you in aggregate are a good person,

02:13:05.540 | a lot of people will know that.

02:13:07.060 | And so that makes you much more immune

02:13:08.940 | to the attacks of people

02:13:10.020 | that bring out an individual statement

02:13:11.900 | that might be a misstatement of some kind

02:13:14.220 | or doesn't express who you are.

02:13:16.820 | And so that, I like that idea is the aggregate

02:13:20.940 | and the power of the podcast

02:13:23.780 | is you have hundreds of hours out there

02:13:27.180 | and being yourself and people get to know who you are.

02:13:30.180 | And once they do and you post pictures of screaming Roombas

02:13:34.740 | as you kick them, they will understand

02:13:37.320 | that you don't mean well.

02:13:38.160 | By the way, as a side comment,

02:13:40.120 | I don't know if I want to release this

02:13:42.940 | because it's not just the Roombas.

02:13:45.520 | - You have a whole dungeon of robots.

02:13:48.460 | - Okay, so this is a problem.

02:13:51.580 | The Boston Dynamics came up against this problem,

02:13:54.420 | but let me just, let me work this out,

02:13:57.260 | like workshop this out with you.

02:13:59.860 | And maybe because we'll post this, people will let me know.

02:14:03.700 | So there's legged robots.

02:14:07.580 | They look like a dog.

02:14:08.820 | They have a very, I'm trying to create

02:14:10.420 | a very real human robot connection,

02:14:13.980 | but like they're also incredible

02:14:15.420 | because you can throw them like off of a building

02:14:19.500 | and they'll land fine.

02:14:21.420 | And it's beautiful.

02:14:22.380 | - That's amazing.

02:14:23.220 | I've seen the Instagram videos of like cats

02:14:25.300 | jumping off of like fifth story buildings

02:14:27.260 | and then walking away.

02:14:29.100 | No one should throw their cat out of a window.

02:14:30.980 | - This is the problem I'm experiencing.

02:14:32.540 | I'll certainly kicking the robots.

02:14:34.700 | It's really fascinating how they recover from those kicks,

02:14:38.000 | but like just seeing myself do it

02:14:40.440 | and also seeing others do it, it just does not look good.

02:14:43.340 | And I don't know what to do with that.

02:14:44.920 | 'Cause it's such a-

02:14:46.260 | - I'll do it.

02:14:47.100 | See, but you don't, 'cause you-

02:14:52.860 | - Robot.

02:14:53.700 | No, I'm kidding.

02:14:54.540 | Now I'm, you know what's interesting?

02:14:55.960 | - Yeah.

02:14:56.800 | - Before today's conversation,

02:14:58.660 | I probably could do it.

02:14:59.980 | And now I think I'm thinking about robots,

02:15:03.020 | bills of rights and things.

02:15:04.140 | I'm actually, and not for any,

02:15:05.700 | not to satisfy you or to satisfy anything,

02:15:08.580 | except that if they have some sentient aspect

02:15:13.020 | to their being, then I would loathe to kick it.

02:15:16.380 | - I don't think you'd be able to kick it.

02:15:17.660 | You might be able to kick the first time,

02:15:18.980 | but not the second.

02:15:20.120 | This is the problem I've experienced.

02:15:21.820 | One of the cool things is one of the robots I'm working with,

02:15:26.580 | you can pick it up by one leg and it's dangling.

02:15:29.340 | You can throw it in any kind of way

02:15:31.540 | and it'll land correctly.

02:15:33.660 | So it's really-

02:15:34.500 | - I had a friend who had a cat like that.

02:15:35.860 | [laughter]

02:15:37.580 | - Oh man, we look forward to the letters from the cat.

02:15:40.560 | - Oh no, I'm not suggesting anyone did that.

02:15:42.100 | But he had this cat and the cat,

02:15:44.540 | he would just throw it onto the bed from across the room

02:15:46.820 | and then it would run back for more.

02:15:48.820 | Somehow they had, that was the nature of the relationship.

02:15:51.580 | I think no one should do that to an animal,

02:15:54.100 | but apparently this cat seemed to return for it

02:15:57.380 | for whatever reason.

02:15:58.220 | - The robot is a robot and it's fascinating to me

02:16:00.100 | how hard it is for me to do that.

02:16:02.240 | So it's unfortunate,

02:16:04.000 | but I don't think I can do that to a robot.

02:16:06.260 | Like I struggle with that.

02:16:08.240 | So for me to be able to do that with a robot,

02:16:13.520 | I have to almost get like into the state

02:16:15.820 | that I imagine like doctors get into

02:16:17.700 | when they're doing surgery.

02:16:19.340 | Like I have to start,

02:16:20.740 | I have to do what robotics colleagues of mine do,

02:16:22.980 | which is like start seeing it as an object.

02:16:25.140 | - Dissociate.

02:16:25.980 | - Like dissociate.

02:16:26.880 | So it was just fascinating that I have to do that

02:16:28.820 | in order to do that with a robot.

02:16:30.700 | I just wanted to take that a little bit of a tangent.

02:16:33.620 | - No, I think it's an important thing.

02:16:34.980 | I mean, I am not shy about the fact that for many years,

02:16:39.980 | I've worked on experimental animals

02:16:42.420 | and that's been a very challenging aspect

02:16:44.700 | to being a biologist, mostly mice,

02:16:46.880 | but in the past no longer, thank goodness,

02:16:49.220 | 'cause I just don't like doing it, larger animals as well.

02:16:52.700 | And now I work on humans,

02:16:53.740 | which I can give consent, verbal consent.

02:16:56.260 | So I think that it's extremely important

02:17:00.820 | to have an understanding of what the guidelines are

02:17:03.880 | and where one's own boundaries are around this.

02:17:06.180 | It's not just an important question.

02:17:09.020 | It might be the most important question

02:17:11.040 | before any work can progress.

02:17:12.860 | - So you asked me about friendship.

02:17:14.940 | I know you have a lot of thoughts about friendship.

02:17:18.180 | What do you think is the value of friendship in life?

02:17:22.320 | - Well, for me personally,

02:17:24.820 | just because of my life trajectory and arc of friendship,

02:17:29.820 | and I should say I do have some female friends

02:17:34.460 | that are just friends,

02:17:35.620 | they're completely platonic relationships,

02:17:37.100 | but it's been mostly male friendship to me has been-

02:17:39.660 | - Has been all male friendships to me actually.

02:17:41.920 | - Interesting, yeah.

02:17:43.020 | It's been an absolute lifeline.

02:17:45.680 | They are my family.

02:17:47.020 | I have a biological family

02:17:48.260 | and I have great respect and love for them

02:17:50.040 | and an appreciation for them,

02:17:51.260 | but it's provided me the,

02:17:55.300 | I wouldn't even say confidence

02:17:58.880 | because there's always an anxiety in taking any good risk

02:18:02.460 | or any risk worth taking.

02:18:04.220 | It's given me the sense that I should go for certain things

02:18:08.860 | and try certain things to take risks,

02:18:10.720 | to weather that anxiety.

02:18:12.140 | And I don't consider myself

02:18:14.640 | a particularly competitive person,

02:18:16.780 | but I would sooner die than disappoint

02:18:21.780 | or let down one of my friends.

02:18:24.620 | I can think of nothing worse actually

02:18:26.920 | than disappointing one of my friends.

02:18:29.120 | Everything else is secondary to me.

02:18:31.260 | - Well, disappointment-

02:18:33.100 | - Disappointing meaning not, I mean,

02:18:36.140 | certainly I strive always to show up

02:18:39.580 | as best I can for the friendship.

02:18:41.580 | And that can be in small ways.

02:18:43.020 | That can mean making sure the phone is away.

02:18:45.000 | Sometimes it's about, I'm terrible with punctuality

02:18:49.920 | 'cause I'm an academic and so I just get lost in time

02:18:52.380 | and I don't mean anything by it,

02:18:53.420 | but it's striving to listen, to enjoy good times

02:18:57.620 | and to make time.

02:18:59.100 | It kind of goes back to this first variable we talked about,

02:19:01.860 | to make sure that I spend time

02:19:03.980 | and to get time in person and check in.

02:19:06.720 | And I think there's so many ways

02:19:10.640 | in which friendship is vital to me.

02:19:12.580 | It's actually, to me, what makes life worth living.

02:19:14.820 | - Yeah, well, I am surprised with the high school friends

02:19:19.820 | how we don't actually talk that often these days

02:19:22.640 | in terms of time, but every time we see each other,

02:19:24.940 | it's immediately right back to where we started.

02:19:27.060 | So I struggle with that, how much time you really allocate

02:19:30.360 | for the friendship to be deeply meaningful

02:19:34.400 | because they're always there with me,

02:19:36.740 | even if we don't talk often.

02:19:38.280 | So there's a kind of loyalty.

02:19:40.980 | I think maybe it's a different style,

02:19:44.000 | but I think to me, friendship is being there

02:19:48.440 | in the hard times, I think.

02:19:51.120 | I'm much more reliable when you're going through shit

02:19:56.460 | than like-

02:19:57.580 | - You're pretty reliable anyway.

02:19:59.560 | - No, but if you're like a wedding or something like that,

02:20:03.120 | or like, I don't know, like you want an award of some kind,

02:20:08.120 | like, yeah, I'll congratulate the shit out of you,

02:20:12.200 | but like that's not, and I'll be there,

02:20:14.460 | but that's not as important to me as being there

02:20:16.700 | when like nobody else is, like just being there

02:20:20.240 | when shit hits the fan or something's tough

02:20:23.820 | where the world turns their back on you,

02:20:25.900 | all those kinds of things.

02:20:26.880 | That to me, that's where friendship is meaningful.

02:20:29.340 | - Well, I know that to be true about you,

02:20:30.920 | and that's a felt thing and a real thing with you.

02:20:33.420 | Let me ask one more thing about that actually,

02:20:35.560 | because I'm not a practitioner of jiu-jitsu.

02:20:38.420 | I know you are, Joe is, but years ago I read a book

02:20:41.420 | that I really enjoyed,

02:20:42.400 | which is Sam Sheridan's book, "A Fighter's Heart."

02:20:44.780 | He talks about all these different forms of martial arts,

02:20:46.920 | and maybe it was in the book, maybe it was in an interview,

02:20:50.640 | but he said that fighting or being in physical battle

02:20:55.420 | with somebody, jiu-jitsu, boxing,

02:20:57.340 | or some other form of direct physical contact

02:21:00.220 | between two individuals creates this bond unlike any other,

02:21:04.300 | because he said, it's like a one-night stand.

02:21:06.580 | You're sharing bodily fluids with somebody

02:21:08.480 | that you barely know.

02:21:09.800 | And I chuckled about it 'cause it's kind of funny

02:21:13.100 | and kind of tongue-in-cheek.

02:21:14.820 | But at the same time, I think this is a fundamental way

02:21:18.960 | in which members of a species bond

02:21:22.280 | is through physical contact.

02:21:24.300 | And certainly there are other forms.

02:21:25.880 | There's cuddling and there's hand-holding

02:21:27.360 | and there's sexual intercourse

02:21:29.680 | and there's all sorts of things.

02:21:30.520 | - What's cuddling?

02:21:31.720 | I haven't heard of it.

02:21:32.720 | - I heard this recently, I didn't know this term,

02:21:34.840 | but there's a term, they've turned the noun cupcake

02:21:38.180 | into a verb, cupcaking, it turns out.

02:21:40.540 | I just learned about this.

02:21:41.740 | Cupcaking is when you spend time just cuddling.

02:21:45.060 | I didn't know about this.

02:21:46.300 | You heard it here first,

02:21:47.300 | although I heard it first just the other day,

02:21:48.800 | cupcaking is actually-

02:21:49.980 | - So cuddling is everything, it's not just like,

02:21:51.780 | is it in bed or is it on the couch?

02:21:53.740 | Like what's cuddling?

02:21:55.380 | I need to look up what cuddling is.

02:21:56.380 | - We need to look this up

02:21:57.220 | and we need to define the variables.

02:21:58.980 | I think it definitely has to do with physical contact,

02:22:02.860 | I am told.

02:22:04.180 | But in terms of battle, competition and the Sheridan quote,

02:22:09.180 | I'm just curious, so do you get close

02:22:15.060 | or feel a bond with people that, for instance,

02:22:18.820 | you rolled jujitsu with,

02:22:20.660 | even though you don't know anything else about them?

02:22:23.940 | Was he right about this?

02:22:25.500 | - Yeah, I mean, on many levels.

02:22:27.140 | He also has the book, "A Fighter's Mind" and "A Fighter's Art."

02:22:30.900 | - He's actually an excellent writer.

02:22:32.140 | What's interesting about him, just briefly about Sheridan,

02:22:34.500 | I don't know, but I did a little bit of research.

02:22:36.140 | He went to Harvard, he was an art major at Harvard.

02:22:40.100 | He claims all he did was smoke cigarettes and do art.

02:22:43.420 | I don't know if his art was any good.

02:22:45.100 | And I think his father was in the SEAL teams.

02:22:48.980 | And then when he got out of Harvard, graduated,

02:22:51.740 | he took off around the world,

02:22:52.900 | learning all the forms of martial arts

02:22:54.420 | and was early to the kind of ultimate fighting

02:22:56.640 | to kind of mix martial arts and things.

02:22:59.020 | Great, great book.

02:22:59.960 | - Yeah, it's amazing.

02:23:01.260 | I don't actually remember it, but I read it

02:23:03.140 | and I remember thinking there was an amazing encapsulation

02:23:06.980 | of what makes fighting, like what makes it compelling.

02:23:11.980 | I would say that there's so many ways that jiu-jitsu,

02:23:17.100 | grappling, wrestling, combat sports in general,

02:23:21.100 | is like one of the most intimate things you could do.

02:23:24.240 | I don't know if I would describe it

02:23:25.620 | in terms of bodily liquids and all those kinds of things.

02:23:27.900 | - I think he was more or less joking.

02:23:29.460 | - But I think there's a few ways that it does that.

02:23:34.460 | So one, because you're so vulnerable.

02:23:39.380 | So that the honesty of stepping on the mat

02:23:44.200 | and often all of us have ego thinking we're better

02:23:49.580 | than we are at this particular art.

02:23:52.380 | And then the honesty of being submitted

02:23:55.900 | or being worse than you thought you are

02:23:58.700 | and just sitting with that knowledge.

02:24:01.000 | That kind of honesty, we don't get to experience it

02:24:03.700 | in most of daily life.

02:24:06.040 | We can continue living somewhat of an illusion

02:24:08.540 | of our conceptions of ourselves

02:24:10.740 | 'cause people are not going to hit us with the reality.

02:24:13.740 | The mat speaks only the truth,

02:24:15.780 | that the reality just hits you.

02:24:17.820 | And that vulnerability is the same

02:24:19.940 | as like the loss of a loved one.

02:24:22.500 | It's the loss of a reality that you knew before.

02:24:26.300 | You now have to deal with this new reality.

02:24:28.180 | And when you're sitting there in that vulnerability

02:24:30.460 | and there's these other people

02:24:32.100 | that are also sitting in that vulnerability,

02:24:34.320 | you get to really connect like, fuck.

02:24:36.920 | Like I'm not as special as I thought I was

02:24:40.220 | and life is like not,

02:24:42.760 | life is harsher than I thought I was

02:24:46.020 | and we're just sitting there with that reality.

02:24:47.660 | Some of us can put words to them, some of them can't.

02:24:50.020 | So I think that definitely is the thing

02:24:51.660 | that leads to intimacy.

02:24:54.620 | The other thing is the human contact.

02:24:58.580 | There is something about, I mean, like a big hug.

02:25:03.580 | Like during COVID, very few people hugged me

02:25:06.780 | and I hugged them and I always felt good when they did.

02:25:10.240 | Like we're all tested and especially now we're vaccinated

02:25:13.860 | but there's still people, this is true in San Francisco,

02:25:16.360 | this is true in Boston.

02:25:17.340 | They want to keep not only six feet away

02:25:19.340 | but stay at home and never touch you.

02:25:21.720 | That loss of basic humanity is the opposite

02:25:26.720 | of what I feel in jiu-jitsu where it was like

02:25:30.480 | that contact where you're like, I don't give a shit

02:25:34.900 | about whatever rules we're supposed to have in society

02:25:37.400 | where you're not, you have to keep a distance

02:25:39.380 | and all that kind of stuff.

02:25:40.440 | Just the hug, like the intimacy of a hug

02:25:45.000 | that's like a good bear hug

02:25:46.440 | and you're like just controlling another person.

02:25:49.720 | And also there is some kind of love communicated

02:25:52.100 | through just trying to break each other's arms.

02:25:54.660 | I don't exactly understand why violence

02:25:57.600 | is such a close neighbor to love but it is.

02:26:02.600 | - Well, in the hypothalamus,

02:26:04.740 | the neurons that control sexual behavior

02:26:08.220 | but also non-sexual contact are not just nearby the neurons

02:26:13.220 | that control aggression and fighting,

02:26:15.780 | they are salt and pepper with those neurons.

02:26:19.540 | It's a very interesting and it almost sounds

02:26:23.120 | kind of risque and controversial and stuff.

02:26:25.020 | I'm not anthropomorphizing about what this means

02:26:27.820 | but in the brain, those structures are interdigitated.

02:26:32.020 | You can't separate them except at a very fine level.

02:26:36.020 | And here the way you describe it is the same

02:26:38.780 | as a real thing.

02:26:39.700 | - I do want to make an interesting comment.

02:26:42.860 | Again, these are the things

02:26:43.740 | that could be taken out of context

02:26:45.660 | but one of the amazing things about jiu-jitsu

02:26:50.540 | is both guys and girls train it.

02:26:52.240 | And I was surprised.

02:26:54.740 | So I'm a big fan of yoga pants at the gym kind of thing.

02:26:59.740 | It reveals the beauty of the female form.

02:27:04.780 | But the thing is girls are dressed in skintight clothes

02:27:08.740 | in jiu-jitsu often.

02:27:10.180 | And I found myself not at all thinking like that at all

02:27:13.900 | when training with girls.

02:27:14.980 | - Well, the context is very non-sexual.

02:27:17.140 | - But I was surprised to learn that.

02:27:20.040 | When I first started jiu-jitsu,

02:27:21.180 | I thought wouldn't that be kind of weird

02:27:22.660 | to train with the opposites in something so intimate?

02:27:26.340 | - Boys and girls, men and women,

02:27:28.500 | they rolled jiu-jitsu together completely.

02:27:30.940 | - Interesting.

02:27:31.780 | - And the only times girls kind of try to stay away

02:27:34.500 | from guys, I mean, there's two contexts.

02:27:36.380 | Of course, there's always going to be creeps in this world.

02:27:38.860 | So everyone knows who to stay away from.

02:27:42.100 | And the other is like, there's a size disparity.

02:27:44.280 | So girls will often try to roll with people

02:27:46.260 | a little bit closer weight-wise.

02:27:48.260 | But no, that's one of the things

02:27:51.100 | that are empowering to women.

02:27:53.100 | That's what they fall in love with

02:27:54.260 | when they started doing jiu-jitsu is I can,

02:27:56.320 | first of all, they gain an awareness

02:27:58.700 | and a pride over their body, which is great.

02:28:00.980 | And then second, they get, especially later on,

02:28:04.360 | start submitting big dudes,

02:28:05.940 | like these like bros that come in who are all shredded

02:28:10.100 | and like muscular, and they get to technique to exercise

02:28:14.760 | dominance over them.

02:28:16.100 | And that's a powerful feeling.

02:28:17.720 | - You've seen women force a larger guy to tap

02:28:21.560 | or even choke him out.

02:28:22.400 | - Well, I was deadlifting for,

02:28:26.080 | oh boy, I think it's 495.

02:28:31.520 | So I was really into powerlifting when I started jiu-jitsu.

02:28:35.000 | And I remember being submitted by,

02:28:37.400 | I thought I walked in feeling like I'm going to be,

02:28:40.660 | if not the greatest fighter ever, at least top three.

02:28:43.220 | And so as a white belt, you roll in like all happy.

02:28:47.420 | And then you realize that as long as you're not applying

02:28:50.740 | too much force that you're having,

02:28:52.940 | I remember being submitted many times by like 130,

02:28:55.540 | 120 pound girls at a balance studios in Philadelphia

02:28:59.860 | that a lot of incredible female jiu-jitsu players.

02:29:02.460 | And that's really humbling too.

02:29:04.360 | The technique can overpower in combat pure strength.

02:29:09.360 | And that's the other thing that there is something

02:29:15.540 | about combat that's primal.

02:29:18.140 | Like it just feels, it feels like we were born to do this.

02:29:23.140 | - Like that we have circuits in our brain

02:29:29.040 | that are dedicated to this kind of interaction.

02:29:32.380 | There's no question.

02:29:34.060 | And like, that's what it felt like.

02:29:35.700 | It wasn't that I'm learning a new skill.

02:29:38.940 | It was like somehow I am remembering echoes

02:29:42.780 | of something I've learned in the past.

02:29:44.380 | - Well, it's like hitting puberty.

02:29:45.660 | A child before puberty has no concept of boys and girls

02:29:49.260 | having this attraction regardless of whether or not

02:29:51.840 | they're attracted to boys or girls.

02:29:53.140 | It doesn't matter at some point.

02:29:54.600 | Most people, not all, but certainly,

02:29:56.380 | but most people when they hit puberty,

02:29:58.340 | suddenly people appear differently.

02:30:01.300 | And certain people take on a romantic or sexual interest

02:30:05.540 | for the very first time.

02:30:07.560 | And so it's like, it's revealing a circuitry in the brain.

02:30:11.380 | It's not like they learn that, it's innate.

02:30:14.320 | And I think when I hear the way you describe jiu-jitsu

02:30:18.100 | and rolling jiu-jitsu, it reminds me a little bit,

02:30:21.060 | Joe was telling me recently about the first time

02:30:23.300 | he went hunting and he felt like it revealed a circuit

02:30:26.660 | that was in him all along,

02:30:28.540 | but he hadn't experienced before.

02:30:30.900 | - Yeah, that's definitely there.

02:30:32.320 | And of course there's the physical activity.

02:30:34.920 | One of the interesting things about jiu-jitsu

02:30:37.220 | is it's one of the really strenuous exercises

02:30:40.860 | that you can do late into your adult life,

02:30:43.860 | like into your 50s, 60s, 70s, 80s.

02:30:48.000 | When I came up, there's a few people in their 80s

02:30:50.500 | that were training.

02:30:51.700 | And as long as you're smart,

02:30:53.080 | as long as you practice techniques

02:30:54.840 | and pick your partners correctly,

02:30:55.940 | you can do that kind of art.

02:30:57.320 | That's late into life and so you're getting exercise.

02:30:59.860 | There's not many activities I find

02:31:01.900 | that are amenable to that.

02:31:04.500 | So because it's such a thinking game,

02:31:07.740 | the jiu-jitsu in particular is an art

02:31:10.860 | where technique pays off a lot.

02:31:13.340 | So you can still maintain, first of all,

02:31:17.220 | remain injury-free if you use good technique

02:31:20.820 | and also through good technique,

02:31:22.960 | be able to go be active with people

02:31:26.920 | that are much, much younger.

02:31:28.460 | And so that was, to me, that and running

02:31:31.560 | are the two activities you can kind of do late in life

02:31:33.820 | because to me, a healthy life has exercises

02:31:38.260 | as a piece of the puzzle.

02:31:39.380 | - No, absolutely.

02:31:40.420 | And I'm glad that we're on the physical component

02:31:42.720 | because I know that there's, for you,

02:31:47.720 | you've talked before about the crossover

02:31:49.680 | between the physical and the intellectual and the mental.

02:31:52.480 | Are you still running at ridiculous hours of the night

02:31:57.680 | for ridiculously long?

02:31:59.600 | - Yeah, so definitely.

02:32:01.520 | I've been running late at night here in Austin.

02:32:03.520 | People tell me, the area we're in now,

02:32:05.800 | people say is a dangerous area,

02:32:07.240 | which I find laughable coming from the bigger cities.

02:32:10.580 | No, I run late at night.

02:32:12.780 | There's something.

02:32:14.180 | - If you see a guy running through Austin at 2 a.m.

02:32:17.580 | in a suit and tie, it's probably nice.

02:32:20.140 | [laughing]

02:32:22.200 | - Well, yeah, I mean, I do think about that

02:32:24.080 | 'cause I get recognized more and more in Austin.

02:32:26.640 | I worry that, not really,

02:32:29.040 | that I get recognized late at night.

02:32:30.800 | But there is something about the night

02:32:34.960 | that brings out those deep philosophical thoughts

02:32:38.960 | and self-reflection that I really enjoy.

02:32:40.720 | But recently, I started getting back to the grind.

02:32:44.560 | So I'm gonna be competing or hoping to be compete

02:32:47.720 | in September and October.

02:32:49.880 | - In jiu-jitsu. - In jiu-jitsu, yeah,

02:32:51.240 | to get back to competition.

02:32:53.120 | And so that requires getting back into great cardio shape.

02:32:58.120 | I've been getting running as part of my daily routine.

02:33:02.280 | - Got it.

02:33:03.640 | Well, I always know I can reach you,

02:33:05.140 | regardless of time zone in the middle of the night,

02:33:08.000 | wherever that happens.

02:33:09.280 | - Well, part of that has to be just being single

02:33:11.120 | and being a programmer.

02:33:13.780 | Those two things just don't work well

02:33:16.040 | in terms of a steady sleep schedule.

02:33:18.040 | - It's not banker's hours kind of work, nine to five.

02:33:21.440 | I want to, you mentioned single.

02:33:23.280 | I want to ask you a little bit

02:33:24.840 | about the other form of relationship,

02:33:26.420 | which is romantic love.

02:33:29.700 | So your parents are still married?

02:33:32.360 | - Still married, still happily married.

02:33:34.120 | - That's impressive.

02:33:34.960 | - Yeah. - A rare thing nowadays.

02:33:36.620 | - Yeah.

02:33:37.460 | - So you grew up with that example.

02:33:38.960 | - Yeah, I guess that's a powerful thing, right?

02:33:40.920 | If there's an example that I think can work.

02:33:43.120 | - Yeah, I didn't have that in my own family,

02:33:46.480 | but when I see it, it's inspiring

02:33:50.880 | and it's beautiful, the fact that they have that

02:33:53.280 | and that was the norm for you, I think is really wonderful.

02:33:57.000 | - In the case of my parents, it was interesting to watch

02:34:00.060 | 'cause there's obviously tension.

02:34:03.240 | Like there'll be times where they fought

02:34:04.920 | and all those kinds of things.

02:34:06.680 | They obviously get frustrated with each other

02:34:09.920 | and they like, but they find mechanisms

02:34:13.320 | how to communicate that to each other,

02:34:15.000 | like to make fun of each other a little bit,

02:34:16.520 | like to tease, to get some of that frustration out

02:34:19.400 | and then ultimately to reunite

02:34:21.040 | and find their joyful moments and be that the energy.

02:34:25.280 | I think it's clear 'cause I got together in there,

02:34:27.600 | I think early twenties, like very, very young.

02:34:30.280 | I think you grow together as people.

02:34:32.440 | - Yeah, you're still in the critical period

02:34:34.320 | of brain plasticity.

02:34:35.560 | - And also, I mean, it's just like divorce

02:34:40.200 | was so frowned upon that you stick it out.

02:34:43.840 | And I think a lot of couples,

02:34:44.960 | especially from that time in the Soviet Union,

02:34:46.640 | that's probably applies to a lot of cultures.

02:34:48.600 | You stick it out and you put in the work,

02:34:50.640 | you learn how to put in the work.

02:34:52.280 | And once you do, you start to get to some of those

02:34:54.920 | rewarding aspects of being like through time,

02:34:59.060 | sharing so many moments together.

02:35:00.760 | That's definitely something that was an inspiration to me,

02:35:07.820 | but maybe that's where I have,

02:35:09.720 | so I have a similar kind of longing

02:35:11.340 | to have a lifelong partner that have that kind of view

02:35:14.880 | where same with friendship,

02:35:16.760 | lifelong friendship is the most meaningful kind,

02:35:20.600 | that there is something with that time

02:35:22.040 | of sharing all that time together,

02:35:24.120 | like till death do us part as a powerful thing,

02:35:26.800 | not by force, not because the religion said it

02:35:29.200 | or the government said it or your culture said it,

02:35:31.580 | but because you want to.

02:35:33.240 | - Do you want children?

02:35:34.520 | - Definitely, yeah.

02:35:35.920 | Definitely want children.

02:35:37.320 | It's common.

02:35:39.200 | - How many room was, do you have?

02:35:41.280 | - Oh, I thought you should know human children.

02:35:43.240 | - No, human children.

02:35:44.440 | - 'Cause I already have the children.

02:35:45.560 | - Exactly, well, I was saying you probably need

02:35:46.880 | at least as many human children as you do room with.

02:35:49.520 | Big family, small family.

02:35:50.980 | In your mind's eye, is there a big,

02:35:55.160 | are there a bunch of freedman's running around?

02:35:59.160 | - So I'll tell you like realistically,

02:36:01.240 | I can explain exactly my thinking.

02:36:04.080 | And this is similar to the robotics work

02:36:06.160 | is if I'm like purely logical right now,

02:36:10.080 | my answer would be, I don't want kids

02:36:12.340 | because I just don't have enough time.

02:36:15.720 | I have so much going on.

02:36:17.280 | But when I'm using the same kind of vision

02:36:19.240 | I use for the robots is I know my life

02:36:22.620 | will be transformed with the first.

02:36:25.000 | Like I know I would love being a father.

02:36:27.640 | And so the question of how many,

02:36:30.000 | that's on the other side of that hill.

02:36:33.140 | It could be some ridiculous number.

02:36:35.880 | So I just know that--

02:36:36.800 | - I have a feeling and I don't have a crystal ball,

02:36:40.820 | but I don't know.

02:36:42.600 | I see an upwards of certainly three or more comes to mind.

02:36:47.600 | - So much of that has to do with the partner you're with too.

02:36:51.920 | So like that's such an open question,

02:36:55.560 | especially in this society of what the right partnership is.

02:36:58.880 | 'Cause I'm deeply empathetic.

02:37:02.800 | I want to see, like to me,

02:37:05.100 | what I look for in a relationship is for me

02:37:08.520 | to be really excited about the passions of another person,

02:37:12.080 | like whatever they're into.

02:37:13.020 | It doesn't have to be a career success, any kind of success,

02:37:16.860 | just to be excited for them

02:37:18.440 | and for them to be excited for me

02:37:20.540 | and they can share in that excitement

02:37:21.840 | and build and build and build.

02:37:23.720 | But there was also practical aspects of like,

02:37:25.800 | what kind of shit do you enjoy doing together?

02:37:28.760 | And I think family is a real serious undertaking.

02:37:32.340 | - Oh, it certainly is.

02:37:34.040 | I mean, I think that I have a friend who said it,

02:37:37.480 | I think best, which is that you first have,

02:37:41.240 | he's in a very successful relationship and has a family.

02:37:44.440 | And he said, you first have to define the role

02:37:47.480 | and then you have to cast the right person for the role.

02:37:50.180 | - Well, yeah, there's some deep aspects to that,

02:37:53.880 | but there's also an aspect to which you're not smart enough

02:37:58.280 | from this side of it to define the role.

02:38:03.000 | There's part of it that has to be a leap

02:38:04.840 | that you have to take.

02:38:06.520 | And I see having kids that way,

02:38:11.520 | you just have to go with it and figure it out also.

02:38:16.120 | As long as there's love there,

02:38:17.640 | like what the hell is life for even?

02:38:20.580 | So there's so many incredibly successful people that I know,

02:38:25.360 | that I've gotten to know, that all have kids.

02:38:28.960 | And the presence of kids for the most part

02:38:32.800 | has only been something that energized them,

02:38:36.620 | something that gave them meaning,

02:38:37.900 | something that made them the best version of themselves,

02:38:40.200 | like made them more productive, not less,

02:38:42.380 | which is fascinating to me.

02:38:43.680 | - It is fascinating.

02:38:44.520 | I mean, you can imagine if the way that you felt about Homer,

02:38:47.260 | the way that I feel and felt about Costello

02:38:49.760 | is at all a glimpse of what that must be like then.

02:38:54.600 | - Exactly.

02:38:55.840 | The downside, the thing I worry more about

02:39:00.020 | is the partner side of that.

02:39:04.560 | I've seen the kids are almost universally

02:39:07.840 | a source of increased productivity and joy and happiness.

02:39:10.840 | Like, yeah, they're a pain in the ass.

02:39:13.280 | Yeah, it's complicated.

02:39:14.120 | Yeah, so on and so forth.

02:39:15.640 | People like to complain about kids.

02:39:17.440 | But when you actually look past

02:39:19.620 | that little shallow layer of complaint, kids are great.

02:39:22.940 | The source of pain for a lot of people

02:39:24.760 | is when the relationship doesn't work.

02:39:27.600 | And so I'm very kind of concerned about,

02:39:30.940 | dating is very difficult and I'm a complicated person.

02:39:36.560 | And so it's been very difficult

02:39:38.220 | to find the right kind of person.

02:39:41.980 | But that statement doesn't even make sense

02:39:45.040 | because I'm not on dating apps.

02:39:46.880 | I don't see people.

02:39:48.260 | You're like the first person I saw in a while.

02:39:50.380 | It's like you, Michael Malice and like Joe.

02:39:53.040 | So like, I don't think I've seen like a female.

02:39:58.040 | What is it?

02:40:00.840 | An element of the female species in quite a while.

02:40:03.680 | So I think you have to put yourself out there.

02:40:06.520 | What is it?

02:40:07.440 | Daniel Johnson says, true love will find you,

02:40:10.240 | but only if you're looking.

02:40:11.760 | So there's some element of really taking the leap

02:40:13.680 | and putting yourself out there

02:40:14.780 | in kind of different situations.

02:40:17.000 | And I don't know how to do that

02:40:18.400 | when you're behind a computer all the time.

02:40:20.460 | - Well, you're a builder and you're a problem solver

02:40:25.200 | and you find solutions and I'm confident this solution is,

02:40:30.200 | the solution is out there and-

02:40:34.260 | - I think you're implying

02:40:35.100 | that I'm going to build the girlfriend, which I think-

02:40:38.500 | - Or that you, well,

02:40:39.780 | and maybe we shouldn't separate this friendship,

02:40:43.360 | the notion of friendship and community.

02:40:45.460 | And if we go back to this concept of the aggregate,

02:40:48.900 | maybe you'll meet this woman through a friend

02:40:52.420 | or maybe or something of that sort.

02:40:53.960 | - So one of the things,

02:40:54.800 | I don't know if you feel the same way,

02:40:56.920 | I definitely one of those people

02:40:59.240 | that just falls in love and that's it.

02:41:02.480 | - Yeah, I can't say I'm like that.

02:41:04.000 | With Costello, it was instantaneous.

02:41:06.560 | - Yeah. - It really was.

02:41:07.760 | I mean, I know it's not romantic love,

02:41:09.680 | but it was instantaneous.

02:41:10.760 | No, I, but that's me.

02:41:12.320 | You know, and I think that if you know, you know,

02:41:14.680 | because that's a good thing that you have that.

02:41:18.260 | - Well, it's, I'm very careful with that

02:41:20.860 | because you don't want to fall in love with the wrong person.

02:41:24.980 | So I try to be very kind of careful with,

02:41:27.340 | I've noticed this because I fall in love with everything,

02:41:29.460 | like this mug, everything.

02:41:31.040 | I fall in love with things in this world.

02:41:34.020 | So like, you have to be really careful

02:41:35.500 | because a girl comes up to you and says,

02:41:40.500 | "She loves Dostoevsky."

02:41:42.700 | That doesn't necessarily mean you need to marry her tonight.

02:41:46.700 | - Yes, and I like the way you said that out loud

02:41:49.060 | so that you heard it.

02:41:49.980 | It doesn't mean you need to marry her tonight.

02:41:52.540 | - Exactly. - Right, exactly.

02:41:53.600 | - But I mean, but people are amazing

02:41:56.120 | and people are beautiful.

02:41:57.140 | And that's, so I'm fully embraced that,

02:42:00.540 | but I also have to be careful with relationships.

02:42:02.940 | And at the same time, like I mentioned to you offline,

02:42:05.820 | I don't, there's something about me

02:42:08.880 | that appreciates swinging for the fences and not dating,

02:42:13.640 | like doing serial dating or dating around.

02:42:15.860 | - Yeah, you're a one guy, one girl kind of guy.

02:42:17.580 | - Yeah. - You said that.

02:42:18.940 | - And it's tricky because you want to be careful

02:42:23.100 | with that kind of stuff,

02:42:24.180 | especially now there's a growing platform

02:42:26.380 | that have a ridiculous amount of female interest

02:42:29.140 | of a certain kind, but I'm looking for deep connection

02:42:33.580 | and I'm looking by sitting home alone.

02:42:36.580 | And every once in a while, talking to Stanford professors.

02:42:40.180 | - Perfect solution. - On a podcast.

02:42:42.380 | - Perfect solution. - It's going to work out great.

02:42:44.260 | - It's well incorporated, it's part of,

02:42:47.260 | that constitutes machine learning of sorts.

02:42:49.540 | - Yeah, of sorts.

02:42:50.580 | - You mentioned what has now become a quite extensive

02:42:55.860 | and expansive public platform, which is incredible.

02:42:59.260 | I mean, the number of people,

02:43:01.260 | first time I saw your podcast, I noticed the suit.

02:43:03.540 | I was like, he respects his audience, which was great.

02:43:05.460 | But I also thought, this is amazing.

02:43:08.480 | People are showing up for science and engineering

02:43:10.740 | and technology information and those discussions

02:43:12.820 | and other sorts of discussions.

02:43:14.100 | Now, I do want to talk for a moment about the podcast.

02:43:18.100 | So my two questions about the podcast are,

02:43:21.720 | when you started it, did you have a plan?

02:43:24.300 | And regardless of what that answer is,

02:43:27.500 | do you know where you're taking it?

02:43:29.820 | Or would you like to leave us?

02:43:31.960 | I do believe in an element of surprise is always fun.

02:43:35.220 | But what about the podcast?

02:43:36.380 | Do you enjoy the podcast?

02:43:37.680 | I mean, your audience certainly includes me,

02:43:40.340 | really enjoys the podcast, it's incredible.

02:43:42.660 | So I love talking to people.

02:43:46.540 | And there's something about microphones

02:43:50.200 | that really bring out the best in people.

02:43:52.300 | Like you don't get a chance to talk like this.

02:43:54.500 | If you and I were just hanging out,

02:43:56.100 | we would have a very different conversation

02:43:58.500 | in the amount of focus we allocate to each other.

02:44:02.100 | We would be having fun talking about other stuff

02:44:04.380 | and doing other things.

02:44:06.000 | There'd be a lot of distraction.

02:44:07.480 | There would be some phone use and all that kind of stuff.

02:44:11.040 | But here we're 100% focused on each other

02:44:13.980 | and focused on the idea.

02:44:16.080 | And sometimes playing with ideas

02:44:18.200 | that we both don't know the answer to,

02:44:21.080 | like a question we don't know the answer to.

02:44:23.080 | We're both like fumbling with it, trying to figure out,

02:44:25.480 | trying to get some insights.

02:44:27.120 | That's something we haven't really figured out before

02:44:29.520 | and together arriving at that.

02:44:31.280 | I think that's magical.

02:44:32.400 | I don't know why we need microphones for that,

02:44:34.200 | but we somehow do.

02:44:35.240 | - It feels like doing science.

02:44:36.720 | - It feels like doing science for me, definitely.

02:44:38.720 | That's exactly it.

02:44:39.900 | Then, and I'm really glad you said that

02:44:42.220 | because I don't actually often say this,

02:44:46.300 | but that's exactly what I felt like.

02:44:48.900 | I wanted to talk to friends and colleagues at MIT

02:44:52.780 | to do real science together.

02:44:56.420 | That's how I felt about it.

02:44:57.780 | Like to really talk through problems

02:45:00.180 | that are actually interesting.

02:45:01.860 | As opposed to like incremental work

02:45:06.060 | that we're currently working for a particular conference.

02:45:10.840 | So really asking questions like, what are we doing?

02:45:14.140 | Like, where's this headed to?

02:45:16.000 | Like, what are the big,

02:45:17.200 | is this really going to help us solve,

02:45:19.960 | in the case of AI, solve intelligence?

02:45:22.800 | Like, is this even working on intelligence?

02:45:24.700 | There's a certain sense,

02:45:26.360 | which is why I initially called it artificial intelligence,

02:45:29.980 | is like most of us are not working

02:45:32.900 | on artificial intelligence.

02:45:34.760 | You're working on some very specific problem

02:45:37.680 | and a set of techniques.

02:45:39.480 | At the time, it's machine learning

02:45:41.260 | to solve this particular problem.

02:45:42.980 | This is not going to take us to a system

02:45:45.560 | that is anywhere close to the generalizability

02:45:49.240 | of the human mind.

02:45:51.320 | Like the kind of stuff the human mind can do

02:45:52.940 | in terms of memory, in terms of cognition,

02:45:54.620 | in terms of reasoning, common sense reasoning.

02:45:56.940 | This doesn't seem to take us there.

02:45:58.700 | So the initial impulse was,

02:46:00.540 | can I talk to these folks, do science together,

02:46:04.040 | through conversation?

02:46:05.620 | And I also thought that there was not enough,

02:46:08.620 | I didn't think there was enough good conversations

02:46:13.860 | with world-class minds that I got to meet.

02:46:17.660 | And not the ones with the book,

02:46:19.820 | or this was the thing.

02:46:21.740 | Oftentimes you go on this tour when you have a book,

02:46:24.260 | but there's a lot of minds that don't write books.

02:46:26.640 | - And the books constrain the conversation too,

02:46:28.860 | 'cause then you're talking about this thing, this book.

02:46:31.720 | But I've noticed that with people

02:46:34.460 | that haven't written a book who are brilliant,

02:46:37.260 | we get to talk about ideas in a new way.

02:46:40.060 | We both haven't actually, when we raise a question,

02:46:43.780 | we don't know the answer to it when the question is raised.

02:46:47.140 | And we try to arrive there.

02:46:48.700 | Like, I don't know, I remember asking questions

02:46:52.220 | of world-class researchers in deep learning

02:46:57.140 | of why do neural networks work as well as they do?

02:47:02.580 | That question is often loosely asked,

02:47:06.600 | but like when you have microphones

02:47:09.220 | and you have to think through it

02:47:11.000 | and you have 30 minutes to an hour

02:47:12.460 | to think through it together, I think that's science.

02:47:16.120 | I think that's really powerful.

02:47:17.600 | So that was the one goal.

02:47:19.680 | The other one is,

02:47:20.560 | I again don't usually talk about this,

02:47:25.460 | but there's some sense in which I wanted

02:47:28.500 | to have dangerous conversations.

02:47:30.320 | Part of the reasons I wanted to wear a suit

02:47:35.320 | is like, I want it to be fearless.

02:47:38.700 | The reason I don't usually talk about it

02:47:40.180 | is because I feel like I'm not good at conversation.

02:47:43.000 | So it looks like it doesn't match the current skill level,

02:47:48.000 | but I want it to have really dangerous conversations

02:47:53.240 | that I uniquely would be able to do.

02:47:58.160 | Not completely uniquely, but like I'm a huge fan

02:48:01.620 | of Joe Rogan and I had to ask myself

02:48:04.540 | what conversations can I do that Joe Rogan can't?

02:48:08.240 | For me, I know I bring this up,

02:48:10.560 | but for me, that person I thought about

02:48:13.760 | at the time was Putin.

02:48:15.640 | Like that's why I bring him up.

02:48:17.580 | He's just like with Costello, he's not just a person.

02:48:22.080 | He's also an idea to me for what I strive for,

02:48:25.580 | just to have those dangerous conversations.

02:48:27.860 | And the reason I'm uniquely qualified is both the Russian,

02:48:31.440 | but also there's the judo and the martial arts.

02:48:34.060 | There's a lot of elements that make me have a conversation

02:48:37.820 | he hasn't had before.

02:48:39.460 | And there's a few other people that I kept in mind,

02:48:44.460 | like Don Knuth, he's a computer scientist from Stanford,

02:48:49.340 | that I thought is one of the most beautiful minds ever.

02:48:54.060 | And nobody really talked to him, like really talked to him.

02:48:59.060 | He's did a few lectures, which people love,

02:49:01.380 | but really just have a conversation with him.

02:49:03.720 | There's a few people like that.

02:49:04.820 | One of them passed away, John Conway, that I never got,

02:49:07.300 | we agreed to talk, but he died before we did.

02:49:10.660 | There's a few people like that, that I thought like,

02:49:13.460 | it's such a crime to not hear those folks.

02:49:19.660 | And I have the unique ability to know how to purchase

02:49:24.660 | a microphone on Amazon and plug it into a device

02:49:28.000 | that records audio and then publish it,

02:49:30.420 | which seems relatively unique.

02:49:32.060 | Like that's not easy in the scientific community,

02:49:34.700 | people knowing how to plug in a microphone.

02:49:36.700 | - No, they can build Faraday cages and two photon

02:49:39.700 | microscopes and bioengineer, all sorts of things.

02:49:43.040 | But the idea that you could take ideas and export them

02:49:47.100 | into a structure or a pseudo structure that people would

02:49:49.740 | benefit from seems like a cosmic achievement to them.

02:49:54.060 | - I don't know if it's a fear or just basically

02:49:57.460 | they haven't tried it, so they haven't learned

02:49:59.340 | the skill level.

02:50:00.300 | - But I think they're not trained.

02:50:01.580 | I mean, we could riff on this for a while,

02:50:03.380 | but I think that, but it's important and maybe we should,

02:50:08.040 | which is that it's, they're not trained to do it.

02:50:11.060 | They're trained to think in specific aims

02:50:12.820 | and specific hypotheses and many of them don't care to.

02:50:17.340 | They became scientists because that's where they felt safe.

02:50:22.660 | And so why would they leave that haven of safety?

02:50:25.880 | - Well, they also don't necessarily always see

02:50:27.860 | the value in it.

02:50:29.300 | We're all together learning.

02:50:30.580 | You and I are learning the value of this.

02:50:33.740 | I think you're probably, you have an exceptionally successful

02:50:38.340 | and amazing podcast that you started just recently.

02:50:40.900 | - Thanks to your encouragement.

02:50:42.420 | - Well, but there's a raw skill there that you're definitely

02:50:47.420 | an inspiration to me in how you do the podcast

02:50:50.400 | in the level of excellence you reach.

02:50:52.860 | But I think you've discovered that that's also

02:50:55.240 | an impactful way to do science, that podcast.

02:50:58.020 | And I think a lot of scientists have not yet discovered

02:51:01.620 | that this is, if they apply same kind of rigor

02:51:06.620 | as they do to academic publication

02:51:09.340 | or to even conference presentations,

02:51:11.780 | and they do that rigor and effort to podcast,

02:51:16.100 | whatever that is, that could be a five minute podcast,

02:51:18.500 | a two hour podcast, it could be conversational,

02:51:20.980 | or it can be more like lecture like.

02:51:22.920 | If they apply that effort, you have the potential to reach

02:51:26.260 | over time, tens of thousands, hundreds of thousands,

02:51:28.600 | millions of people.

02:51:29.800 | And that's really, really powerful.

02:51:32.460 | But yeah, for me, giving a platform to a few of those folks,

02:51:37.460 | especially for me personally,

02:51:40.940 | so maybe you can speak to what fields you're drawn to,

02:51:45.940 | but I thought computer scientists

02:51:48.940 | were especially bad at this.

02:51:53.420 | So there's brilliant computer scientists

02:51:56.300 | that I thought it would be amazing to explore their mind,

02:52:00.300 | explore their thinking.

02:52:02.060 | And so I took that almost on as an effort.

02:52:06.540 | And at the same time, I had other guests in mind

02:52:11.140 | or people that connect to my own interests.

02:52:13.900 | So the wrestling,

02:52:15.500 | wrestling, music, football,

02:52:18.900 | both American football and soccer,

02:52:21.260 | I have a few particular people

02:52:22.620 | that I'm really interested in.

02:52:24.180 | Bovisar Satiev, the Satiev brothers,

02:52:27.980 | even Khabib for wrestling, just to talk to them.

02:52:30.660 | - Oh, 'cause you guys can communicate.

02:52:32.980 | - In Russian and in wrestling, right?

02:52:36.380 | As wrestlers and as Russians.

02:52:38.840 | And so that little,

02:52:41.740 | it's like an opportunity to explore a mind

02:52:44.060 | that I'm able to bring to the world.

02:52:47.680 | And also, I feel like it makes me a better person,

02:52:52.680 | just that being that vulnerable

02:52:55.620 | and exploring ideas together.

02:52:57.380 | I don't know, like good conversation.

02:52:59.820 | I don't know how often you have really good conversation

02:53:01.780 | with friends, but like podcasts are like that.

02:53:04.460 | And it's deeply moving.

02:53:07.100 | - It's the best.

02:53:08.500 | And what you brought through,

02:53:09.920 | I mean, when I saw you sit down with Penrose,

02:53:12.180 | Nobel prize-winning physicists and these other folks,

02:53:15.020 | it's not just 'cause he has a Nobel,

02:53:16.300 | it's what comes out of his mouth is incredible.

02:53:18.020 | And what you were able to hold in that conversation

02:53:22.900 | was so much better.

02:53:24.400 | Light years beyond what he had any other interviewer,

02:53:28.880 | I don't want to even call you an interviewer

02:53:30.100 | 'cause it's really about conversation.

02:53:31.620 | Light years beyond what anyone else had been able

02:53:34.460 | to engage with him was such a beacon of what's possible.

02:53:39.460 | And I know that, I think that's what people are drawn to.

02:53:42.440 | And there's a certain intimacy that,

02:53:45.540 | certainly if two people are friends as we are

02:53:47.540 | and they know each other, that there's more of that,

02:53:49.860 | but there's an intimacy in those kinds

02:53:51.940 | of private conversations that are made public.

02:53:54.300 | - Well, that's the, with you,

02:53:57.900 | you're probably starting to realize, and Costello is like,

02:54:01.940 | part of it, because you're authentic

02:54:04.540 | and you're putting yourself out there completely,

02:54:07.000 | people are almost not just consuming

02:54:10.860 | the words you're saying.

02:54:13.300 | They also enjoy watching you, Andrew, struggle

02:54:18.180 | with these ideas or try to communicate these ideas.

02:54:20.700 | They like the flaws.

02:54:21.700 | They like a human being exploring ideas.

02:54:24.860 | - Well, that's good 'cause I got plenty of those.

02:54:26.500 | - Well, they like the self-critical aspects,

02:54:28.660 | like where you're very careful,

02:54:30.300 | where you're very self-critical about your flaws.

02:54:33.060 | I mean, in that same way, it's interesting, I think,

02:54:35.720 | for people to watch me talk to Penrose,

02:54:37.940 | not just because Penrose is communicating ideas,

02:54:42.300 | but here's this like silly kid trying to explore ideas.

02:54:46.980 | Like they know this kid that there's a human connection

02:54:50.100 | that is really powerful.

02:54:51.240 | Same, I think, with Putin, right?

02:54:53.580 | Like it's not just a good interview with Putin.

02:54:57.440 | It's also, here's this kid struggling

02:55:00.900 | to talk with one of the most powerful

02:55:03.780 | and some would argue dangerous people in the world.

02:55:08.260 | They love that, the authenticity that led up to that.

02:55:11.620 | And in return, I get to connect everybody I run to

02:55:15.940 | in the street and all those kinds of things.

02:55:18.140 | There's a depth of connection there

02:55:21.020 | almost within like a minute or two.

02:55:22.940 | That's unlike any other.

02:55:24.300 | - Yeah, there's an intimacy that you've formed with them.

02:55:26.840 | - Yeah, we've been on this like journey together.

02:55:29.740 | I mean, I have the same thing with Joe Rogan

02:55:31.300 | before I ever met him, right?

02:55:32.740 | Like I was, because I was a fan of Joe for so many years,

02:55:36.620 | there's something, there's a kind of friendship

02:55:40.900 | as absurd as it might be to say in podcasting

02:55:44.460 | and listening to podcasts.

02:55:45.960 | - Yeah, maybe it fills in a little bit of that

02:55:48.860 | or solves a little bit of that loneliness

02:55:50.640 | that you were talking about earlier.

02:55:51.480 | - Until the robots are here.

02:55:53.120 | [both laughing]

02:55:54.540 | - I have just a couple more questions,

02:55:56.540 | but one of them is on behalf of your audience,

02:55:59.560 | which is, I'm not going to ask you

02:56:02.720 | the meaning of the hedgehog,

02:56:04.460 | but I just want to know, does it have a name?

02:56:08.460 | And you don't have to tell us the name,

02:56:09.920 | but just does it have a name, yes or no?

02:56:12.400 | - Well, there's a name he likes to be referred to as,

02:56:17.400 | and then there's a private name

02:56:19.280 | in the privacy-owned company that we call each other.

02:56:21.280 | No, I'm not that insane.

02:56:24.380 | No, his name is Hedgy.

02:56:25.680 | He's a hedgehog.

02:56:28.680 | I don't like stuffed animals,

02:56:30.280 | but his story is one of minimalism.

02:56:35.620 | So I gave away everything I own now three times in my life.

02:56:40.620 | By everything, I mean, almost everything,

02:56:43.000 | kept jeans and shirt and a laptop.

02:56:45.260 | And recently it's also been guitar, things like that.

02:56:50.940 | But he survived because he was always in the,

02:56:55.280 | at least in the first two times, was in the laptop bag,

02:56:58.540 | and he just got lucky.

02:57:00.240 | And so I just liked the perseverance of that.

02:57:02.960 | And I first saw him in the,

02:57:05.960 | the reason I got a stuffed animal

02:57:07.360 | and I don't have other stuffed animals

02:57:09.600 | is it was in a thrift store

02:57:11.560 | in this like giant pile of stuffed animals.

02:57:16.000 | And he jumped out at me

02:57:18.080 | because unlike all the rest of them,

02:57:20.000 | he has this intense mean look about him,

02:57:25.000 | that he's just, he's upset at life,

02:57:28.640 | at the cruelty of life.

02:57:31.000 | And just, especially in the contrast

02:57:32.680 | of the other stuffed animals,

02:57:33.680 | they have this dumb smile on their face.

02:57:35.840 | If you look at most stuffed animals,

02:57:37.100 | they have this dumb look on their face

02:57:38.700 | and they're just happy.

02:57:39.640 | It's like Pleasantville.

02:57:40.560 | - That's what we say in neuroscience.

02:57:41.640 | They have a smooth cortex, not many fold.

02:57:44.440 | - Exactly.

02:57:45.280 | And this, like Hedgy like saw through all of it.

02:57:48.000 | He was like Dostoevsky's man from underground.

02:57:52.000 | I mean, there's a sense that he saw the darkness

02:57:54.800 | of the world and persevered.

02:57:56.560 | So I got, and there's also a famous Russian cartoon,

02:58:00.700 | Hedgehog in the Fog, that I grew up with,

02:58:03.920 | I connected with.

02:58:04.760 | There's people who know of that cartoon.

02:58:07.600 | You can see it on YouTube.

02:58:09.420 | It's like-- - Hedgehog in the Fog?

02:58:10.780 | - Yeah.

02:58:11.620 | He, it's just as you would expect,

02:58:15.300 | especially from like early Soviet cartoons.

02:58:17.900 | It's a hedgehog, like sad, walking through the fog,

02:58:22.900 | exploring like loneliness and sadness.

02:58:25.300 | It's like, but it's beautiful.

02:58:26.740 | It's like a piece of art.

02:58:27.820 | People should, even if you don't speak Russian,

02:58:29.620 | you'll see, you'll understand.

02:58:31.580 | - Oh, it's, the moment you said that I was gonna ask,

02:58:33.900 | so it's in Russian, but of course it's in Russian.

02:58:35.380 | - It's in Russian, but it's more,

02:58:37.220 | there's very little speaking in it.

02:58:39.020 | It's almost, there's an interesting exploration

02:58:43.020 | of how you make sense of the world

02:58:47.360 | when you see it only vaguely through the fog.

02:58:52.140 | So he's trying to understand the world.

02:58:54.100 | - Here we have Mickey Mouse.

02:58:56.660 | - Yeah. - We have Bugs Bunny.

02:58:58.740 | We have all these crazy animals,

02:59:01.540 | and you have the hedgehog in the fog.

02:59:03.580 | - So there's a certain period, and this is, again,

02:59:07.860 | I don't know what it's attributed to,

02:59:09.260 | but it was really powerful,

02:59:10.820 | which there's a period in Soviet history,

02:59:12.940 | I think probably '70s and '80s,

02:59:16.020 | where like, especially kids were treated very seriously.

02:59:21.020 | Like they were treated like they're able to deal

02:59:24.300 | with the weightiness of life.

02:59:27.820 | And that was reflected in the cartoons.

02:59:29.960 | And there was, it was allowed to have

02:59:33.180 | like really artistic content,

02:59:37.180 | not like dumb cartoons that are trying to get you

02:59:39.500 | to be like smile and run around, but like create art.

02:59:42.320 | Like stuff that, you know how like short cartoons

02:59:45.300 | or short films can win Oscars?

02:59:47.040 | Like that's what they're swinging for.

02:59:48.700 | - So what strikes me about this is a little bit

02:59:51.100 | how we were talking about the suit earlier.

02:59:52.780 | It's almost like they treat kids with respect.

02:59:55.140 | - Yeah. - Like that they have

02:59:57.060 | an intelligence and they honor that intelligence.

02:59:59.740 | - Yeah, they're really just adult in a small body.

03:00:02.360 | Like you want to protect them

03:00:04.620 | from the true cruelty of the world.

03:00:06.140 | - Sure. - But in terms

03:00:06.980 | of their intellectual capacity

03:00:08.460 | or like philosophical capacity, they're right there with you.

03:00:11.540 | And so the cartoons reflected that,

03:00:14.220 | the art that they consumed, education reflected that.

03:00:17.720 | So he represents that.

03:00:19.100 | I mean, there's a sense of, because he survived so long

03:00:24.100 | and because I don't like stuffed animals,

03:00:27.320 | that it's like we've been through all of this together

03:00:30.900 | and it's the same, sharing the moments together.

03:00:32.980 | It's the friendship.

03:00:34.340 | And there's a sense in which, you know,

03:00:36.220 | if all the world turns on you and goes to hell,

03:00:39.000 | at least we got each other.

03:00:41.180 | And he doesn't die because he's an inanimate object, so.

03:00:45.460 | - Until you animate him.

03:00:47.380 | - Until you animate him.

03:00:49.060 | And then I probably wouldn't want to know

03:00:50.420 | what he was thinking about this whole time.

03:00:52.500 | He's probably really into Taylor Swift

03:00:55.200 | or something like that.

03:00:56.040 | It's like that I wouldn't even want to know.

03:00:57.700 | Anyway.

03:00:58.540 | - Well, I now feel a connection to Hedgy the Hedgehog

03:01:02.460 | that I certainly didn't have before.

03:01:04.100 | And I think that encapsulates the kind of possibility

03:01:07.340 | of connection that is possible between human

03:01:12.340 | and other object and through robotics, certainly.

03:01:17.620 | There's a saying that I heard when I was a graduate student

03:01:19.580 | that's just been ringing in my mind

03:01:22.520 | throughout this conversation in such a,

03:01:25.060 | I think, appropriate way, which is that,

03:01:27.180 | Lex, you are in a minority of one.

03:01:31.300 | You are truly extraordinary in your ability

03:01:35.140 | to encapsulate so many aspects of science, engineering,

03:01:40.140 | public communication about so many topics,

03:01:43.740 | martial arts, and the emotional depth that you bring to it,

03:01:47.120 | and just the purposefulness.

03:01:49.060 | And I think if it's not clear to people,

03:01:51.620 | it absolutely should be stated,

03:01:53.880 | but I think it's abundantly clear

03:01:55.460 | that just the amount of time and thinking

03:01:58.400 | that you put into things is,

03:02:01.360 | it is the ultimate mark of respect.

03:02:04.800 | So I'm just extraordinarily grateful for your friendship

03:02:08.040 | and for this conversation.

03:02:09.120 | - I'm proud to be your friend.

03:02:11.140 | And I just wish you showed me the same kind of respect

03:02:13.420 | by wearing a suit and make your father proud

03:02:15.820 | and maybe next time.

03:02:17.480 | - Next time, indeed.

03:02:19.080 | Thanks so much, my friend.

03:02:20.440 | - Thank you.

03:02:21.280 | Thank you, Andrew.

03:02:22.360 | - Thank you for joining me for my discussion

03:02:24.440 | with Dr. Lex Friedman.

03:02:26.280 | If you're enjoying this podcast and learning from it,

03:02:29.140 | please consider subscribing on YouTube.

03:02:31.380 | As well, you can subscribe to us on Spotify or Apple.

03:02:35.480 | Please leave any questions and comments and suggestions

03:02:38.200 | that you have for future podcast episodes and guests

03:02:40.840 | in the comment section on YouTube.

03:02:43.240 | At Apple, you can also leave us up to a five-star review.

03:02:46.800 | If you'd like to support this podcast, we have a Patreon.

03:02:49.640 | That's patreon.com/andrewhuberman.

03:02:52.840 | And there you can support us at any level that you like.

03:02:56.200 | Also, please check out our sponsors mentioned

03:02:58.520 | at the beginning of the podcast episode.

03:03:00.840 | That's the best way to support this podcast.

03:03:03.200 | Links to our sponsors can be found in the show notes.

03:03:06.740 | And finally, thank you for your interest in science.

03:03:09.760 | [upbeat music]

03:03:13.480 | (upbeat music)

03:03:16.060 | (upbeat music)

Dr. Lex Fridman: Machines, Creativity & Love | Huberman Lab Podcast #29

Chapters