Charles Isbell: Computing, Interactive AI, and Race in America

Charles Isbell (02:24.000)

follow on Spotify, support on Patreon, or connect with me on Twitter at Lex Friedman.

Lex Fridman (02:30.080)

And now, here's my conversation with Charles Isbell.

Lex Fridman (02:35.280)

You've mentioned that you love movies and TV shows.

Charles Isbell (02:39.040)

Let's ask an easy question, but you have to be definitively, objectively, conclusive.

Lex Fridman (02:44.640)

What's your top three movies of all time?

Charles Isbell (02:47.680)

So, you're asking me to be definitive and to be conclusive.

Lex Fridman (02:50.320)

That's a little hard. I'm going to tell you why.

Charles Isbell (02:51.920)

It's very simple. It's because movies is too broad of a category.

Lex Fridman (02:56.160)

I got to pick subgenres, but I will tell you that of those genres,

Charles Isbell (02:59.920)

I'll pick one or two from each of the genres, and I'll get us to three, so I'm going to cheat.

Charles Isbell (03:03.920)

So, my favorite comedy of all times, which is probably my favorite movie of all time,

Charles Isbell (03:10.000)

is His Girl Friday, which is probably a movie that you've not ever heard of,

Lex Fridman (03:14.320)

but it's based on a play called The Front Page from, I don't know, early 1900s.

Lex Fridman (03:20.400)

And the movie is a fantastic film.

Lex Fridman (03:23.840)

What's the story? What's the independent film?

Lex Fridman (03:26.080)

No, no, no. What are we talking about?

Charles Isbell (03:27.120)

This is one of the movies that would have been very popular. It's a screwball comedy.

Lex Fridman (03:31.120)

You ever see Moonlighting, the TV show? You know what I'm talking about?

Charles Isbell (03:33.920)

So, you've seen these shows where there's a man and a woman, and they clearly are in love with one another,

Lex Fridman (03:38.320)

and they're constantly fighting and always talking over each other.

Lex Fridman (03:40.960)

Banter, banter, banter, banter, banter.

Charles Isbell (03:43.280)

This was the movie that started all that, as far as I'm concerned.

Charles Isbell (03:46.880)

It's very much of its time. So, it's, I don't know, must have come out sometime between 1934 and 1939.

Charles Isbell (03:53.280)

I'm not sure exactly when the movie itself came out. It's black and white.

Lex Fridman (03:57.040)

It's just a fantastic film, and it's hilarious.

Lex Fridman (04:01.840)

So, it's mostly conversation?

Lex Fridman (04:03.840)

Not entirely, but mostly, mostly. Just a lot of back and forth.

Charles Isbell (04:07.440)

There's a story there. Someone's on death row, and they're newspaper men, including her.

Lex Fridman (04:14.480)

They're all newspaper men. They were divorced.

Charles Isbell (04:17.280)

The editor, the publisher, I guess, and the reporter, they were divorced.

Lex Fridman (04:22.400)

But, you know, they clearly, he's thinking, trying to get back together,

Lex Fridman (04:25.360)

and there's this whole other thing that's going on.

Lex Fridman (04:27.120)

But none of that matters. The plot doesn't matter.

Charles Isbell (04:28.720)

Yeah, it's just a little play in conversation.

Charles Isbell (04:31.440)

It's fantastic. And I just love everything about the conversation, because at the end of the day,

Charles Isbell (04:35.680)

sort of narrative and conversation are the sort of things that drive me.

Lex Fridman (04:38.160)

And so, I really like that movie for that reason.

Charles Isbell (04:41.280)

Similarly, I'm now going to cheat, and I'm going to give you two movies as one.

Lex Fridman (04:45.360)

And they're Crouching Tiger, Hidden Dragon, and John Wick.

Charles Isbell (04:49.760)

Both relatively modern. John Wick, of course.

Lex Fridman (04:51.440)

One, two, or three?

Charles Isbell (04:52.560)

One. It gets increasingly, I love them all for different reasons,

Lex Fridman (04:56.000)

and increasingly more ridiculous. Kind of like Loving Alien and Aliens,

Charles Isbell (04:59.440)

despite the fact they're two completely different movies.

Lex Fridman (05:01.520)

But the reason I put Crouching Tiger, Hidden Dragon, and John Wick together is because I

Charles Isbell (05:06.000)

actually think they're the same movie, or what I like about them, the same movie.

Lex Fridman (05:09.680)

Which is both of them create a world that you're coming in the middle of,

Lex Fridman (05:15.440)

and they don't explain it to you. But the story is done so well that you pick it up.

Lex Fridman (05:20.000)

So, anyone who's seen John Wick, you know, you have these little coins,

Lex Fridman (05:23.760)

and they're headed out, and there are these rules,

Lex Fridman (05:25.680)

and apparently every single person in New York City is an assassin.

Charles Isbell (05:28.960)

There's like two people who come through who aren't, but otherwise they are.

Lex Fridman (05:31.360)

But there's this complicated world, and everyone knows each other.

Charles Isbell (05:34.080)

They don't sit down and explain it to you, but you figure it out.

Lex Fridman (05:35.920)

Crouching Tiger, Hidden Dragon is a lot like that.

Charles Isbell (05:38.000)

You get the feeling that this is chapter nine of a 10 part story,

Lex Fridman (05:41.280)

and you've missed the first eight chapters, and they're not going to explain it to you,

Lex Fridman (05:44.400)

but there's this sort of rich world behind you.

Lex Fridman (05:45.920)

You get pulled in anyway, like immediately.

Charles Isbell (05:47.280)

You get pulled in anyway. So, it's just excellent storytelling in both cases,

Lex Fridman (05:50.880)

and very, very different.

Lex Fridman (05:51.840)

And also you like the outfit, I assume? The John Wick outfit?

Lex Fridman (05:54.480)

Oh yeah, of course. Well, of course. Yes. I think John Wick outfit is perfect.

Lex Fridman (05:58.000)

And so that's number two, and then…

Lex Fridman (05:59.760)

But sorry to pause on that. Martial arts? You have a long list of hobbies.

Charles Isbell (06:03.920)

Like it scrolls off the page, but I didn't see martial arts as one of them.

Lex Fridman (06:07.680)

I do not do martial arts, but I certainly watch martial arts.

Charles Isbell (06:10.240)

Oh, I appreciate it very much. Oh, we could talk about every Jackie Chan movie ever made,

Lex Fridman (06:14.000)

and I would be on board with that.

Lex Fridman (06:15.600)

The Shower, too? Like that kind of comedy of a cop?

Lex Fridman (06:18.880)

Yes, yes. By the way, my favorite Jackie Chan movie would be Drunken Master 2,

Charles Isbell (06:25.120)

known in the States usually as Legend of the Drunken Master.

Lex Fridman (06:29.200)

Actually, Drunken Master, the first one, is the first kung fu movie I ever saw,

Lex Fridman (06:33.360)

but I did not know that.

Lex Fridman (06:34.880)

First Jackie Chan movie?

Charles Isbell (06:36.000)

No, first one ever that I saw and remember, but I had no idea that that's what it was,

Lex Fridman (06:40.640)

and I didn't know that was Jackie Chan. That was like his first major movie.

Charles Isbell (06:43.200)

Yeah. I was a kid. It was done in the 70s.

Lex Fridman (06:46.320)

I only later rediscovered that that was actually.

Lex Fridman (06:49.680)

And he creates his own martial art by drinking. Was he actually drinking or was he played drinking?

Lex Fridman (06:58.000)

You mean as an actor or as a character?

Charles Isbell (06:59.600)

No. I'm sure as an actor. He was in the 70s or whatever.

Lex Fridman (07:04.240)

He was definitely drinking, and in the end, he drinks industrial grade alcohol.

Charles Isbell (07:09.440)

Ah, yeah.

Lex Fridman (07:10.480)

Yeah, and has one of the most fantastic fights ever in that subgenre.

Charles Isbell (07:15.040)

Anyway, that's my favorite one of his movies, but I'll tell you the last movie.

Lex Fridman (07:19.280)

It's actually a movie called Nothing But a Man, which is the 1960s,

Charles Isbell (07:23.760)

starred Ivan Dixon, who you'll know from Hogan's Heroes, and Abby Lincoln.

Lex Fridman (07:31.840)

It's just a really small little drama. It's a beautiful story.

Lex Fridman (07:35.040)

But my favorite scenes, I'm cheating, one of my favorite movies just for the ending is The

Charles Isbell (07:41.440)

Godfather. I think the last scene of that is just fantastic. It's the whole movie all summarized in

Charles Isbell (07:47.360)

just eight, nine seconds.

Lex Fridman (07:48.400)

Godfather Part One?

Charles Isbell (07:49.440)

Part One.

Lex Fridman (07:50.240)

How does it end? I don't think you need to worry about spoilers if you haven't seen The Godfather.

Charles Isbell (07:54.880)

Spoiler alert. It ends with the wife coming to Michael, and he says,

Charles Isbell (08:01.920)

just this once, I'll let you ask me my business. And she asks him if he did this terrible thing,

Lex Fridman (08:06.400)

and he looks her in the eye and he lies, and he says, no. And she says, thank you. And she

Charles Isbell (08:10.800)

walks out the door, and you see her going out of the door, and all these people are coming in,

Lex Fridman (08:19.920)

and they're kissing Michael's hands, and Godfather. And then the camera switches

Lex Fridman (08:24.720)

perspective. So instead of looking at him, you're looking at her, and the door

Charles Isbell (08:29.760)

closes in her face, and that's the end of the movie. And that's the whole movie right there.

Lex Fridman (08:33.840)

Do you see parallels between that and your position as Dean at Georgia Tech Chrome?

Charles Isbell (08:37.440)

Just kidding. Trick question.

Lex Fridman (08:39.520)

Sometimes, certainly. The door gets closed on me every once in a while.

Charles Isbell (08:44.160)

Okay. That was a rhetorical question. You've also mentioned that you, I think, enjoy all kinds of

Charles Isbell (08:51.120)

experiments, including on yourself. But I saw a video where you said you did an experiment where

Charles Isbell (08:56.320)

you tracked all kinds of information about yourself and a few others sort of wiring up your

Charles Isbell (09:03.440)

home. And this little idea that you mentioned in that video, which is kind of interesting, that

Charles Isbell (09:10.240)

you thought that two days worth of data is enough to capture majority of the behavior of the human

Charles Isbell (09:16.960)

being. First, can you describe what the heck you did to collect all the data? Because it's

Charles Isbell (09:23.440)

fascinating, just like little details of how you collect that data and also what your intuition

Charles Isbell (09:28.320)

behind the two days is. So first off, it has to be the right two days. But I was thinking of a

Charles Isbell (09:32.960)

very specific experiment. There's actually a suite of them that I've been a part of, and other people

Charles Isbell (09:36.480)

have done this, of course. I just sort of dabbled in that part of the world. But to be very clear,

Charles Isbell (09:40.320)

the specific thing that I was talking about had to do with recording all the IR going on in my

Charles Isbell (09:46.160)

infrared going on in my house. So this is a long time ago. So this is everything's being curled

Charles Isbell (09:50.480)

by pressing buttons on remote controls, as opposed to speaking to Alexa or Siri or someone like that.

Lex Fridman (09:56.960)

And I was just trying to figure out if you could get enough data on people to figure out what they

Charles Isbell (10:01.280)

were going to do with their TVs or their lights. My house was completely wired up at the time.

Lex Fridman (10:06.400)

But you know, what I'm about to look at a movie, I'm about to turn on the TV or whatever and just

Charles Isbell (10:10.240)

see what I could predict from it. It was kind of surprising. It shouldn't have been. But that's all

Charles Isbell (10:16.720)

very easy to do, by the way, just capturing all the little stuff. I mean, it's a bunch of computers

Charles Isbell (10:19.520)

systems. It's really easy to capture today if you know what you're looking for. At Georgia Tech,

Charles Isbell (10:22.880)

long before I got there, we had this thing called the Aware Home, where everything was wired up and

Charles Isbell (10:27.200)

you captured everything that was going on. Nothing even difficult, not with video or anything like

Charles Isbell (10:31.200)

that, just the way that the system was just capturing everything. So it turns out that,

Lex Fridman (10:37.920)

and I did this with myself and then I had students and they worked with many other people. And it

Charles Isbell (10:42.000)

turns out at the end of the day, people do the same things over and over and over again. So it

Charles Isbell (10:47.360)

has to be the right two days, like a weekend. But it turns out not only can you predict what

Charles Isbell (10:51.840)

someone's going to do next at the level of what button they're going to press next on a remote

Charles Isbell (10:55.040)

control, but you can do it with something really, really simple. You don't even need a hidden mark

Charles Isbell (11:01.360)

off model. It's like a mark, just simply, I press this, this is my prediction of the next thing.

Charles Isbell (11:05.440)

It turns out you can get 93% accuracy just by doing something very simple and stupid and just

Charles Isbell (11:11.120)

counting statistics. But what was actually more interesting is that you could use that information.

Charles Isbell (11:15.520)

This comes up again and again in my work. If you try to represent people or objects by the things

Charles Isbell (11:21.440)

they do, the things you can measure about them that have to do with action in the world. So

Charles Isbell (11:26.560)

distribution over actions, and you try to represent them by the distribution of actions that are done

Charles Isbell (11:32.400)

on them, then you do a pretty good job of sort of understanding how people are and they cluster

Charles Isbell (11:38.960)

remarkably well, in fact, irritatingly so. And so by clustering people this way,

Charles Isbell (11:44.000)

you can maybe, you know, I got the 93% accuracy of what's the next button you're going to press,

Lex Fridman (11:49.360)

but I can get 99% accuracy or somewhere there's about on the collections of things you might

Charles Isbell (11:54.240)

press. And it turns out the things that you might press are all related to number to each other and

Charles Isbell (11:58.480)

exactly what you would expect. So for example, all the key, all the numbers on a keypad, it turns out

Charles Isbell (12:04.480)

all have the same behavior with respect to you as a human being. And so you would naturally cluster

Charles Isbell (12:08.640)

them together and you discover that numbers are all related to one another in some way and all

Charles Isbell (12:14.640)

these other things. And then, and here's the part that I think is important. I mean, you can see

Charles Isbell (12:18.400)

this in all kinds of things. Every individual is different, but any given individual is remarkably

Charles Isbell (12:25.280)

predictable because you keep doing the same things over and over again. And the two things that I've

Charles Isbell (12:29.680)

learned in the long time that I've been thinking about this is people are easily predictable and

Charles Isbell (12:34.960)

people hate when you tell them that they're easily predictable, but they are. And there you go.

Charles Isbell (12:39.280)

Yeah. What about, let me play devil's advocate and philosophically speaking, is it possible to

Charles Isbell (12:46.160)

say that what defines humans is the outlier. So even though many, some large percentage of our

Charles Isbell (12:52.480)

behaviors, whatever the signal we measure is the same and it would cluster nicely, but maybe it's

Charles Isbell (12:57.760)

the special moments of when we break out of the routine is the definitive thing that we're

Charles Isbell (13:03.200)

breaking out of the routine is the definitive things. And the way we break out of that routine

Charles Isbell (13:07.920)

for each one of us might be different. It's possible. I would say that I would say it a

Charles Isbell (13:11.520)

little differently. I think I would make two things. One is a, I'm going to disagree with

Charles Isbell (13:15.920)

the premise, I think, but that's fine. I think the way I would put it is there are people who

Charles Isbell (13:22.880)

are very different from lots of other people, but they're not 0%, they're closer to 10%, right? So

Charles Isbell (13:28.240)

in fact, even if you do this kind of clustering of people, that'll turn out to be the small number

Charles Isbell (13:31.760)

they all behave like each other, even if they individually behave very differently from everyone

Charles Isbell (13:36.960)

else. So I think that's kind of important. But what you're really asking, I think, and I think

Charles Isbell (13:40.720)

this is really a question is, what do you do when you're faced with the situation you've never seen

Charles Isbell (13:46.160)

before? What do you do when you're faced with an extraordinary situation maybe you've seen others

Charles Isbell (13:49.600)

do and you're actually forced to do something and you react to that very differently. And that is

Charles Isbell (13:53.040)

the thing that makes you human. I would agree with that, at least at a philosophical level, that it's

Charles Isbell (13:56.880)

the times when you are faced with something difficult, a decision that you have to make

Charles Isbell (14:04.080)

where the answer isn't easy, even if you know what the right answer is, that's sort of what defines

Charles Isbell (14:08.560)

you as the individual. And I think what defines people broadly, it's the hard problem. It's not

Charles Isbell (14:12.800)

the easy problem. It's the thing that's going to hurt you. It's not even that it's difficult. It's

Charles Isbell (14:17.600)

just that you know that the outcome is going to be highly suboptimal for you. And I do think that

Charles Isbell (14:22.880)

that's a reasonable place to start for the question of what makes us human. So before we talk about

Charles Isbell (14:29.360)

sort of explore the different ideas underlying interactive artificial intelligence, which we

Charles Isbell (14:33.440)

are working on, let me just go along this thread to skip to kind of our world of social media,

Charles Isbell (14:39.520)

which is something that at least on the artificial intelligence side you think about.

Charles Isbell (14:44.080)

There's a popular narrative, I don't know if it's true, but that we have these silos in social

Charles Isbell (14:51.840)

media and we have these clusterings, as you're kind of mentioning. And the idea is that, you know,

Charles Isbell (14:58.400)

along that narrative is that, you know, we want to, we want to break each other out of those silos

Lex Fridman (15:06.320)

so we can be empathetic to other people, to if you're a Democrat, you'd be empathetic to the

Charles Isbell (15:12.480)

Republican. If you're Republican, you're empathetic Democrat. Those are just two silly bins that we

Charles Isbell (15:17.680)

seem to be very excited about, but there's other binnings that we can think about. Is there, from

Charles Isbell (15:24.160)

an artificial intelligence perspective, because you're just saying we cluster along the data,

Lex Fridman (15:29.840)

but then interactive artificial intelligence is referring to throwing agents into that mix,

Charles Isbell (15:35.520)

AI systems in that mix, helping us interacting with us humans and maybe getting us out of that

Charles Isbell (15:41.600)

mix, maybe getting us out of those silos. Is that something that you think is possible? Do you see

Charles Isbell (15:48.720)

a hopeful possibility for artificial intelligence systems in these large networks of people to get

Charles Isbell (15:56.160)

us outside of our habits in at least the idea space to where we can sort of be empathetic to

Lex Fridman (16:04.800)

other people's lived experiences, other people's points of view, you know, all that kind of stuff?

Charles Isbell (16:11.280)

Yes, and I actually don't think it's that hard. Well, it's not hard in this sense. So imagine that

Charles Isbell (16:16.000)

you can, let's just, let's make life simple for a minute. Let's assume that you can do a kind of

Charles Isbell (16:22.720)

partial ordering over ideas or clusterings of behavior. It doesn't even matter what I mean here,

Lex Fridman (16:28.720)

so long as there's some way that this is a cluster, this is a cluster, there's some edge

Charles Isbell (16:31.920)

between them, right? And this is kind of, they don't quite touch even, or maybe they come very

Charles Isbell (16:35.280)

close. If you can imagine that conceptually, then the way you get from here to here is not by going

Charles Isbell (16:40.480)

from here to here. The way you get from here to here is you find the edge and you move slowly

Charles Isbell (16:43.680)

together, right? And I think that machines are actually very good at that sort of thing once we

Charles Isbell (16:47.360)

can kind of define the problem, either in terms of behavior or ideas or words or whatever. So it's

Charles Isbell (16:51.840)

easy in the sense that if you already have the network and you know the relationships, you know,

Charles Isbell (16:55.840)

the edges and sort of the strengths on them and you kind of have some semantic meaning for them,

Charles Isbell (17:00.480)

the machine doesn't have to, you do as the designer, then yeah, I think you can kind of move

Charles Isbell (17:04.960)

people along and sort of expand them. But it's harder than that. And the reason it's harder than

Charles Isbell (17:08.880)

that, or sort of coming up with the network structure itself is hard, is because I'm gonna

Charles Isbell (17:13.920)

tell you a story that someone else told me and I don't, I may get some of the details a little bit

Charles Isbell (17:18.800)

wrong, but it's roughly, it roughly goes like this. You take two sets of people from the same

Charles Isbell (17:24.320)

backgrounds and you want them to solve a problem. So you separate them up, which we do all the time,

Charles Isbell (17:29.600)

right? Oh, you know, we're gonna break out in the, we're gonna break out groups. You're gonna go

Charles Isbell (17:32.400)

over there and you're gonna talk about this. You're gonna go over there and you're gonna talk

Charles Isbell (17:34.080)

about this. And then you have them sort of in this big room, but far apart from one another,

Lex Fridman (17:38.560)

and you have them sort of interact with one another. When they come back to talk about what

Charles Isbell (17:43.040)

they learn, you want to merge what they've done together. It can be extremely hard because they

Charles Isbell (17:47.440)

don't, they basically don't speak the same language anymore. Like when you create these problems and

Charles Isbell (17:51.360)

you dive into them, you create your own language. So the example this one person gave me, which I

Charles Isbell (17:56.160)

found kind of interesting because we were in the middle of that at the time, was

Charles Isbell (17:58.720)

they're sitting over there and they're talking about these rooms that you can see, but you're

Charles Isbell (18:03.520)

seeing them from different vantage points, depending on what side of the room you're on.

Charles Isbell (18:06.880)

They can see a clock very easily. And so they start referring to the room as the one with the clock.

Charles Isbell (18:12.960)

This group over here, looking at the same room, they can see the clock, but it's, you know,

Charles Isbell (18:16.720)

not in their line of sight or whatever. So they end up referring to it by some other way.

Charles Isbell (18:22.000)

When they get back together and they're talking about things, they're referring to the same room

Lex Fridman (18:26.080)

and they don't even realize they're referring to the same room. And in fact, this group doesn't

Charles Isbell (18:29.040)

even see that there's a clock there and this group doesn't see whatever's the clock on the wall is

Charles Isbell (18:32.480)

the thing that stuck with me. So if you create these different silos, the problem isn't that

Charles Isbell (18:36.160)

the ideologies disagree. It's that you're using the same words and they mean radically different

Charles Isbell (18:41.680)

things. The hard part is just getting them to agree on the, well, maybe we'd say the axioms in

Charles Isbell (18:47.680)

our world, right? But you know, just get them to agree on some basic definitions because right now

Charles Isbell (18:52.640)

they talk, they're talking past each other, just completely talking past each other. That's the

Charles Isbell (18:56.480)

hard part, getting them to meet, getting them to interact. That may not be that difficult. Getting

Charles Isbell (19:01.120)

them to see where their language is leading them to lead past one another. That's, that's the hard

Charles Isbell (19:06.400)

part. It's a really interesting question to me. It could be on the layer of language, but it feels

Charles Isbell (19:10.240)

like there's multiple layers to this. Like it could be worldview. It could be, I mean, all boils

Charles Isbell (19:14.640)

down to empathy, being able to put yourself in the shoes of the other person to learn the language, to

Charles Isbell (19:20.400)

learn like visually how they see the world, to learn like the, I mean, I experienced this now

Charles Isbell (19:28.400)

with, with trolls, the, the degree of humor in that world. For example, I talk about love a lot.

Charles Isbell (19:33.840)

I'm very like, I'm really lucky to have this amazing community of loving people. But whenever I

Charles Isbell (19:39.840)

encounter trolls, they always roll their eyes at the idea of love because it's so quote unquote

Charles Isbell (19:45.440)

cringe. So, so they, they show love by like derision, I would say. And I think about on the

Charles Isbell (19:56.080)

human level, that's a whole nother discussion. That's psychology, that's sociology, so on. But

Charles Isbell (1:00:08.000)

There's not an F you money switch where you're like, you know what? Now I'm going to be more

Charles Isbell (1:00:13.520)

bold. It doesn't, I don't see it. Well, there's a reason for that. Tenure isn't a test. It's a

Charles Isbell (1:00:18.880)

training process. It teaches you to behave in a certain way, to think in a certain way, to accept

Charles Isbell (1:00:24.240)

certain values and to react accordingly. And the better you are at that, the more likely you are to

Charles Isbell (1:00:28.880)

earn tenure. And by the way, this is not a bad thing. Most things are like that. And I think most

Charles Isbell (1:00:34.720)

of my colleagues are interested in doing great work and they're just having impact in the way that

Charles Isbell (1:00:38.880)

they want to have impact. I do think that as a field, not just as a field, as a profession,

Charles Isbell (1:00:46.480)

we have a habit of belittling those who are popular as it were, as if the word itself is a

Charles Isbell (1:00:56.000)

kind of Scarlet A, right? I think it's easy to convince yourself, and no one is immune to this,

Charles Isbell (1:01:06.720)

no one is immune to this, that the people who are better known are better known for bad reasons.

Charles Isbell (1:01:11.600)

The people who are out there dumbing it down are not being pure to whatever the values and ethos

Charles Isbell (1:01:19.520)

is for your field. And it's just very easy to do. Now, having said that, I think that ultimately,

Charles Isbell (1:01:26.480)

people who are able to be popular and out there and are touching the world and making a difference,

Charles Isbell (1:01:31.920)

you know, our colleagues do, in fact, appreciate that in the long run. It's just, you know,

Charles Isbell (1:01:36.800)

you have to be very good at it or you have to be very interested in pursuing it. And once you get

Charles Isbell (1:01:40.960)

past a certain level, I think people accept that for who it is. I mean, I don't know. I'd be really

Charles Isbell (1:01:45.200)

interested in how Rod Brooks felt about how people were interacting with him when he did Fast,

Charles Isbell (1:01:50.000)

Cheap, and Out of Control way, way, way back when.

Lex Fridman (1:01:53.120)

What's Fast, Cheap, and Out of Control?

Charles Isbell (1:01:55.200)

It was a documentary that involved four people. I remember nothing about it other than Rod Brooks

Charles Isbell (1:02:00.480)

was in it and something about naked mole rats. I can't remember what the other two things were.

Charles Isbell (1:02:05.200)

It was robots, naked mole rats, and then two others.

Charles Isbell (1:02:07.440)

By the way, Rod Brooks used to be the head of the artificial intelligence laboratory at MIT

Lex Fridman (1:02:12.480)

and then launched, I think, iRobot and then Think Robotics, Rethink Robotics.

Lex Fridman (1:02:18.160)

Yes.

Charles Isbell (1:02:18.720)

Think is in the word. And also is a little bit of a rock star personality in the AI world,

Lex Fridman (1:02:27.200)

a very opinionated, very intelligent. Anyway, sorry, mole rats and naked.

Charles Isbell (1:02:32.480)

Naked mole rats. Also, he was one of my two advisors for my PhD.

Lex Fridman (1:02:36.960)

This explains a lot.

Charles Isbell (1:02:37.840)

I don't know how to explain it. I love Rod. But I also love my other advisor, Paul. Paul,

Lex Fridman (1:02:43.840)

if you're listening, I love you too. Both very, very different people.

Charles Isbell (1:02:46.080)

Paul Viola.

Charles Isbell (1:02:46.640)

Paul Viola. Both very interesting people, very different in many ways. But I don't know what

Charles Isbell (1:02:51.360)

Rod would say to you about what the reaction was. I know that for the students at the time,

Charles Isbell (1:02:56.800)

because I was a student at the time, it was amazing. This guy was in a movie, being very

Charles Isbell (1:03:03.680)

much himself. Actually, the movie version of him is a little bit more Rod than Rod. I think they

Charles Isbell (1:03:10.720)

edited it appropriately for him. But it was very much Rod. And he did all this while doing great

Charles Isbell (1:03:15.200)

work. Was he running the iLab at that point or not? I don't know. But anyway, he was running

Charles Isbell (1:03:18.720)

the iLab, or would be soon. He was a giant in the field. He did amazing things, made a lot of his

Charles Isbell (1:03:23.280)

bones by doing the kind of counterintuitive science. And saying, no, you're doing this all

Charles Isbell (1:03:28.960)

wrong. Representation is crazy. The world is your own representation. You just react to it. I mean,

Charles Isbell (1:03:32.720)

these are amazing things. And continues to do those sorts of things as he's moved on.

Charles Isbell (1:03:37.760)

I think he might tell you, I don't know if he would tell you it was good or bad, but I know that

Charles Isbell (1:03:43.280)

for everyone else out there in the world, it was a good thing. And certainly,

Charles Isbell (1:03:46.880)

he continued to be respected. So it's not as if it destroyed his career by being popular.

Charles Isbell (1:04:21.120)

P stands for probabilistic. And what does

Charles Isbell (1:04:26.000)

FUNC stand for? So a lot of my life is about making acronyms. So if I have one quirk,

Charles Isbell (1:04:31.440)

it's that people will say words and I see if they make acronyms. And if they do, then I'm happy. And

Charles Isbell (1:04:36.880)

then if they don't, I try to change it so that they make acronyms. It's just a thing that I do.

Lex Fridman (1:04:41.440)

So PFUNC is an acronym. It has three or four different meanings. But finally, I decided that

Charles Isbell (1:04:46.960)

the P stands for probabilistic because at the end of the day, it's machine learning and it's randomness and

Charles Isbell (1:05:13.840)

it's fun.

Lex Fridman (1:05:40.720)

So there's a sense to it, which is not an acronym, like literally FUNC. You have a dark, mysterious past.

Lex Fridman (1:05:43.760)

So there's a sense to it, which is not an acronym, like literally FUNC.

Lex Fridman (1:06:13.680)

There's a sense to it, which is not an acronym, like literally FUNC.

Charles Isbell (1:06:43.600)

It's a whole set of things of which rap is a part. So tagging is a part of hip hop. I don't know why

Charles Isbell (1:06:48.480)

that's true, but people tell me it's true and I'm willing to go along with it because they get very

Charles Isbell (1:06:51.680)

angry about it. But hip hop is like graffiti. Tagging is like graffiti. And there's all these,

Charles Isbell (1:06:56.400)

including the popping and the locking and all the dancing and all those things. That's all a part of

Charles Isbell (1:06:59.600)

hip hop. It's a way of life, which I think is true. And then there's rap, which is this particular.

Lex Fridman (1:07:04.960)

It's the music part.

Charles Isbell (1:07:05.840)

Yes. A music part. I mean, you wouldn't call the stuff that DJs do the scratching. That's not rap,

Charles Isbell (1:07:12.560)

right? But it's a part of hip hop, right? So given that we understand that hip hop is this whole

Charles Isbell (1:07:16.640)

thing, what are the rap albums that best touch that for me? Well, if I were going to educate you,

Lex Fridman (1:07:21.840)

I would try to figure out what you liked and then I would work you there.

Charles Isbell (1:07:26.400)

Oh my God. I would probably start with, there's a fascinating exercise one can do by watching old

Charles Isbell (1:07:38.400)

episodes of I love the seventies. I love the eighties. I love the nineties with a bunch of

Charles Isbell (1:07:43.680)

friends and just see where people come in and out of pop culture. So if you're talking about

Charles Isbell (1:07:50.240)

those people, then I would actually start you with where I would hope to start you with anyway,

Charles Isbell (1:07:54.960)

which is public enemy. Particularly it takes a nation of millions to hold us back, which is

Charles Isbell (1:08:00.480)

clearly the best album ever produced. And certainly the best hip hop album ever produced

Charles Isbell (1:08:05.440)

in part because it was so much of what was great about the time. Fantastic lyrics is me. It's all

Charles Isbell (1:08:11.040)

about the lyrics. Amazing music that was coming from Rick Rubin was the producer of that. And he

Charles Isbell (1:08:16.160)

did a lot, very kind of heavy mental ish, at least in the eighties sense at the time. And it was

Charles Isbell (1:08:21.680)

focused on politics in the 1980s, which was what made hip hop so great. Then I would start you

Charles Isbell (1:08:28.160)

there. Then I would move you up through things that are been happening more recently. I'd probably

Charles Isbell (1:08:33.040)

get you to someone like a most deaf. I would give you a history lesson, basically. Most of them.

Charles Isbell (1:08:37.280)

That's amazing. He hosted a poetry jam thing on HBO or something like that. Probably. I don't

Charles Isbell (1:08:41.440)

think I've seen it, but I wouldn't be surprised. Yeah. Spoken poetry. That's amazing. He's amazing.

Lex Fridman (1:08:46.720)

And then I would, I, after I got you there, I'd work you back to EPMD.

Lex Fridman (1:08:51.520)

And eventually I would take you back to the last poets and particularly the first album,

Charles Isbell (1:08:57.200)

the last poets, which was 1970 to give you a sense of history and that it actually has been building

Charles Isbell (1:09:02.720)

up over a very, very long time. So we would start there because that's where your music aligns. And

Charles Isbell (1:09:08.480)

then we would cycle out and I'd move you to the present. And then I'd take you back to the past.

Charles Isbell (1:09:12.480)

Because I think a large part of people who are kind of confused about any kind of music,

Charles Isbell (1:09:16.880)

you know, the truth is this is the same thing we've always been talking about, right? It's

Charles Isbell (1:09:19.680)

about narrative and being a part of something and being immersed in something. So you understand it,

Charles Isbell (1:09:23.520)

right? Jazz, which I also like is one of the things that's cool about jazz is that you come

Lex Fridman (1:09:30.160)

and you meet someone who's talking to you about jazz and you have no idea what they're talking

Charles Isbell (1:09:32.960)

about. And then one day it all clicks and you've been so immersed in it. You go, oh yeah, that's a

Charles Isbell (1:09:37.200)

Charlie Parker. You start using words that nobody else understands, right? And it becomes part of

Charles Isbell (1:09:41.440)

hip hop's the same way. Everything's the same way. They're all cultural artifacts, but I would help

Charles Isbell (1:09:45.200)

you to see that there's a history of it and how it connects to other genres of music that you might

Charles Isbell (1:09:49.120)

like to bring you in. So that you could kind of see how it connects to what you already like,

Charles Isbell (1:09:54.320)

including some of the good work that's been done with fusions of hip hop and bluegrass.

Charles Isbell (1:10:02.480)

Oh no.

Lex Fridman (1:10:03.200)

Yes. Some of it's even good. Not all of it, but some of it is good,

Lex Fridman (1:10:09.520)

but I'd start you with, it takes a nation of millions to hold this back.

Charles Isbell (1:10:12.320)

There's an interesting tradition in like more modern hip hop of integrating almost like classic

Charles Isbell (1:10:18.800)

crock songs or whatever, like integrating into their music, into the beat, into the whatever.

Lex Fridman (1:10:25.440)

It's kind of interesting. It gives a whole new, not just classic crock, but what is it?

Charles Isbell (1:10:32.000)

Kanye with Gold Digger.

Lex Fridman (1:10:33.120)

Old R&B.

Lex Fridman (1:10:35.680)

It's taking and pulling old R&B, right?

Charles Isbell (1:10:38.080)

Well, that's been true since the beginning. I mean, in fact, that's in some ways,

Charles Isbell (1:10:41.200)

that's why the DJ used to get top billing because it was the DJ that brought all the records together

Lex Fridman (1:10:47.600)

and made it worth so that people could dance. You go back to those days, mostly in New York,

Charles Isbell (1:10:52.800)

though not exclusively, but mostly in New York where it sort of came out of,

Charles Isbell (1:10:56.400)

the DJ that brought all the music together and the beats and showed that basically music is

Charles Isbell (1:11:00.240)

itself an instrument, very meta, and you can bring it together and then you sort of wrap over it and

Lex Fridman (1:11:04.480)

so on. And it moved that way. So that's going way, way back. Now, in the period of time where I grew

Charles Isbell (1:11:10.000)

up, when I became really into it, which was mostly the 80s, it was more funk was the back for a lot

Charles Isbell (1:11:17.200)

of the stuff, Public Enemy at that time, notwithstanding. And so, which is very nice

Charles Isbell (1:11:22.800)

because it tied into what my parents listened to and what I vaguely remember listening to when I

Charles Isbell (1:11:26.720)

was very small. And by the way, complete revival of George Clinton and Parliament and Funkadelic

Lex Fridman (1:11:33.120)

and all of those things to bring it sort of back into the 80s and into the 90s. And as we go on,

Charles Isbell (1:11:38.000)

you're going to see the last decade and the decade before that being brought in.

Lex Fridman (1:11:42.240)

And when you don't think that you're hearing something you've heard, it's probably because

Charles Isbell (1:11:45.600)

it's being sampled by someone who, referring to something they remembered when they were young,

Charles Isbell (1:11:53.920)

perhaps from somewhere else altogether. And you just didn't realize what it was because it wasn't

Charles Isbell (1:11:59.040)

a popular song where you happened to grow up. So this stuff has been going on for a long time.

Charles Isbell (1:12:02.160)

It's one of the things that I think is beautiful. Run DMC, Jam Master Jay used to play, he played

Charles Isbell (1:12:08.480)

piano. He would record himself playing piano and then sample that to make it a part of what was

Charles Isbell (1:12:13.200)

going on rather than play the piano. That's how his mind can think. Well, it's pieces. You're

Lex Fridman (1:12:17.600)

putting pieces together, you're putting pieces of music together to create new music, right?

Charles Isbell (1:12:20.480)

Now that doesn't mean that the root, I mean, the roots are doing their own thing.

Charles Isbell (1:12:23.440)

Yeah. Right. Those, those are, that's a whole. Yeah. But still it's the right attitude that,

Charles Isbell (1:12:30.240)

you know, and what else is jazz, right? Jazz is about putting pieces together and then putting

Charles Isbell (1:12:33.280)

your own spin on it. It's all the same. It's all the same thing. It's all the same thing.

Charles Isbell (1:12:36.640)

Yeah. Cause you mentioned lyrics. It does make me sad. Again, this is me talking trash about

Charles Isbell (1:12:41.600)

modern hip hop. I haven't, you know, investigated. I'm sure people correct me that there's a lot of

Charles Isbell (1:12:46.880)

great artists. That's part of the reason I'm saying it is they'll leave it in the comments that you

Charles Isbell (1:12:50.640)

should listen to this person is the lyrics went away from talking about maybe not just politics,

Lex Fridman (1:12:59.680)

but life and so on. Like, you know, the kind of like protest songs, even if you look at like a

Charles Isbell (1:13:04.720)

Bob Marley or you said Public Enemy or Rage Against the Machine, More on the Rockside,

Charles Isbell (1:13:09.040)

there's, that's the place where we go to those lyrics. Like classic rock is all about like,

Charles Isbell (1:13:16.800)

my woman left me or, or I'm really happy that she's still with me or the flip side. It's like

Charles Isbell (1:13:23.360)

love songs of different kinds. It's all love, but it's less political, like less interesting. I

Charles Isbell (1:13:27.920)

would say in terms of like deep, profound knowledge. And it seems like rap is the place where you

Charles Isbell (1:13:34.320)

would find that. And it's sad that for the most part, what I see, like you look like mumble rap

Charles Isbell (1:13:40.160)

or whatever, they're moving away from lyrics and more towards the beat and the, and the musicality

Charles Isbell (1:13:45.120)

of it. I've always been a fan of the lyrics. In fact, if you go back and you read my reviews,

Charles Isbell (1:13:49.120)

which I recently was rereading, man, I wrote my last review the month I graduated. I got my PhD,

Charles Isbell (1:13:57.120)

which says something about something. I'm not sure what though. I always would always,

Lex Fridman (1:14:01.120)

but I often would start with, it's all about the lyrics. For me, it's all, it's about the lyrics.

Charles Isbell (1:14:06.560)

Someone has already written in the comments before I've even finished having this conversation

Charles Isbell (1:14:10.000)

that, you know, neither of us knows what we're talking about and it's all in the underground

Charles Isbell (1:14:13.600)

hip hop and here's who you should go listen to. And that is true. Every time I despair for popular

Charles Isbell (1:14:19.040)

rap, I get someone points me to, or I discover some underground hip hop song and I'm, I am made

Charles Isbell (1:14:25.360)

happy and whole again. So I know it's out there. I don't listen to as much as I used to because

Charles Isbell (1:14:30.720)

I'm listening to podcasts and old music from the 1980s. It's a kind of no beat at all, but you know,

Lex Fridman (1:14:37.040)

there's a little bit of sampling here and there. I'm sure. By the way, James Brown is funk or no?

Charles Isbell (1:14:42.480)

Yes. And so is junior Wells, by the way. Who's that? Oh, junior Wells, Chicago blues. He was

Charles Isbell (1:14:48.560)

James Brown before James Brown was. It's hard to imagine somebody being James Brown. Go look up

Charles Isbell (1:14:53.440)

hoodoo man blues, junior Wells, and just listen to, snatch it back and hold it and you'll see it.

Lex Fridman (1:15:01.920)

And they were contemporaries. Where do you put like little Richard or all that kind of stuff?

Charles Isbell (1:15:06.080)

Like Ray Charles, like when they get like, hit the road Jack and don't you come back. Isn't that

Charles Isbell (1:15:12.160)

like, there's a funkiness in it. Oh, that's definitely a funkiness in it. I mean, it's all,

Charles Isbell (1:15:16.160)

I mean, it's all, it's all a line. I mean, it's all, there's all a line that carries it all

Charles Isbell (1:15:20.320)

together. You know, it's, I guess I would answer your question, depending on whether I'm thinking

Charles Isbell (1:15:24.240)

about it in 2020 or I'm thinking about it in 1960. Um, I'd probably give a different answer.

Charles Isbell (1:15:28.800)

I'm just thinking in terms of, you know, that was rock, but when you look back on it, it's,

Charles Isbell (1:15:33.440)

it was funky, but we didn't use those words or maybe we did. I wasn't around. Uh, but you know,

Charles Isbell (1:15:38.240)

I don't think we used the word 1960 funk, certainly not the way we used it in the seventies

Lex Fridman (1:15:42.320)

and the eighties. Do you reject disco? I do not reject disco. I appreciate all the mistakes that

Charles Isbell (1:15:47.120)

we have made. Actually, some of the disco is actually really, really good. John Travolta.

Lex Fridman (1:15:52.320)

Oh boy. He regrets it probably. Maybe not. Well, like it's the mistakes thing.

Charles Isbell (1:15:56.720)

Yeah. And it got him to where he's going, where he is.

Charles Isbell (1:16:00.880)

Oh, well, thank you for taking that detour. You've, you've talked about computing. We've

Charles Isbell (1:16:05.920)

already talked about computing a little bit, but can you try to describe how you think about the

Charles Isbell (1:16:11.600)

world of computing or it fits into the sets of different disciplines? We mentioned College of

Charles Isbell (1:16:16.880)

Computing. What, what should people, how should they think about computing, especially from an

Charles Isbell (1:16:22.560)

educational perspective of like, what is the perfect curriculum that defines for a young mind,

Charles Isbell (1:16:30.800)

uh, what computing is? So I don't know about a perfect curriculum, although that's an important

Charles Isbell (1:16:34.960)

question because at the end of the day, without the curriculum, you don't get anywhere. Curriculum

Charles Isbell (1:16:39.280)

to me is the fundamental data structure. It's not even the classroom. I mean, the world is,

Charles Isbell (1:16:44.080)

right? I, I, I, so I think the curriculum is where I like to play. Uh, so I spend a lot of time

Charles Isbell (1:16:49.200)

thinking about this, but I will tell you, I'll answer your question by answering a slightly

Charles Isbell (1:16:52.400)

different question first and then getting back to this, which is, you know, you talked about

Charles Isbell (1:16:55.520)

disciplines and what does it mean to be a discipline? The truth is what we really educate

Charles Isbell (1:17:01.120)

people in from the beginning, but certainly through college, you sort of failed. If you

Charles Isbell (1:17:05.520)

don't think about it this way, I think is the world. People often think about tools and tool

Charles Isbell (1:17:12.640)

sets, and when you're really trying to be good, you think about skills and skill sets,

Lex Fridman (1:17:16.560)

but disciplines are about mindsets, right? They're about fundamental ways of thinking, not just the,

Charles Isbell (1:17:22.400)

the, the hammer that you pick up, whatever that is to hit the nail, um, not just the,

Charles Isbell (1:17:27.280)

the skill of learning how to hammer well or whatever. It's the mindset of like,

Charles Isbell (1:17:30.960)

what's the fundamental way to think about, to think about the world, right? And disciplines,

Charles Isbell (1:17:37.200)

different disciplines give you different mindsets to give you different ways of sort of thinking

Charles Isbell (1:17:40.960)

through. So with that in mind, I think that computing, even ask the question, whether

Charles Isbell (1:17:45.680)

it's a discipline that you have to decide, does it have a mindset? Does it have a way of thinking

Charles Isbell (1:17:48.960)

about the world that is different from, you know, the scientist who is doing a discovery and using

Charles Isbell (1:17:53.600)

the scientific method as a way of doing it, or the mathematician who builds abstractions and tries

Charles Isbell (1:17:57.680)

to find sort of steady state truth about the abstractions that may be artificial, but whatever,

Charles Isbell (1:18:03.600)

or is it the engineer who's all about, you know, building demonstrably superior technology with

Charles Isbell (1:18:08.560)

respect to some notion of trade offs, whatever that means, right? That's sort of the world that

Charles Isbell (1:18:12.640)

you, the world that you live in. What is computing? You know, how is computing different? So I've

Charles Isbell (1:18:16.880)

thought about this for a long time and I've come to a view about what computing actually is, what

Charles Isbell (1:18:22.160)

the mindset is. And, and it's, you know, it's a little abstract, but that would be appropriate

Charles Isbell (1:18:26.320)

for computing. I think that what distinguishes the computationalist from others is that he or

Charles Isbell (1:18:34.320)

she understands that models, languages and machines are equivalent. They're the same thing. And

Charles Isbell (1:18:41.600)

because it's not just a model, but it's a machine that is an executable thing that can be described

Charles Isbell (1:18:47.440)

as a language. That means that it's dynamic. So it's not the, it is mathematical in some sense,

Charles Isbell (1:18:54.000)

in the kind of sense of abstraction, but it is fundamentally dynamic and executable. The

Charles Isbell (1:18:58.320)

mathematician is not necessarily worried about either the dynamic part. In fact, whenever I tried

Charles Isbell (1:19:03.280)

to write something for mathematicians, they invariably demand that I make it static. And

Lex Fridman (1:19:08.160)

that's not a bad thing. It's just, it's a way of viewing the world that truth is a thing, right?

Charles Isbell (1:19:12.560)

It's not a process that continually runs, right? So that dynamic thing matters. That self reflection

Charles Isbell (1:19:19.440)

of the system itself matters. And that is what computing, that is what computing brought us.

Lex Fridman (1:19:23.280)

So it is a science because it, the models fundamentally represent truths in the world.

Charles Isbell (1:19:27.840)

Information is a scientific thing to discover, right? Not just a mathematical conceit that

Charles Isbell (1:19:33.040)

that gets created. But of course it's engineering because you're actually dealing with constraints

Charles Isbell (1:19:37.040)

in the world and trying to execute machines that actually run. But it's also a math because

Charles Isbell (1:19:43.120)

you're actually worrying about these languages that describe what's happening. But the fact that

Charles Isbell (1:19:51.440)

regular expressions and finite state automata, one of which feels like a machine or at least

Charles Isbell (1:19:56.400)

an abstraction machine and the other is a language that they're actually the equivalent thing. I mean,

Charles Isbell (1:19:59.760)

that is not a small thing and it permeates everything that we do, even when we're just

Charles Isbell (1:20:03.840)

trying to figure out how to, how to do debugging. So that idea I think is fundamental and we would

Charles Isbell (1:20:09.120)

do better if we made that more explicit. How my life has changed and my thinking about this

Charles Isbell (1:20:15.200)

in the 10 or 15 years it's been since I tried to put that to paper with some colleagues is

Charles Isbell (1:20:21.680)

the realization, which comes to a question you actually asked me earlier, which has to do with

Charles Isbell (1:20:28.000)

trees falling down and whether it matters, is this sort of triangle of equality.

Charles Isbell (1:20:34.560)

It only matters because there's a person inside the triangle, right? That what's changed about

Charles Isbell (1:20:40.720)

computing, computer science or whatever you want to call it, is we now have so much data

Lex Fridman (1:20:46.560)

and so much computational power. We're able to do really, really interesting, promising things.

Lex Fridman (1:20:53.040)

But the interesting and the promising kind of only matters with respect to human beings and

Charles Isbell (1:20:56.960)

their relationship to it. So the triangle exists, that is fundamentally computing.

Lex Fridman (1:21:01.680)

What makes it worthwhile and interesting and potentially world species changing is that there

Charles Isbell (1:21:08.880)

are human beings inside of it and intelligence that has to interact with it to change the data,

Charles Isbell (1:21:12.640)

the information that makes sense and gives meaning to the models, the languages and the machines.

Lex Fridman (1:21:19.120)

So if the curriculum can convey that while conveying the tools and the skills that you need

Lex Fridman (1:21:25.120)

in order to succeed, then it is a big win. That's what I think you have to do.

Lex Fridman (1:21:30.880)

Do you pull psychology, like these human things into that, into the idea,

Lex Fridman (1:21:36.960)

into this framework of computing? Do you pull in psychology and neuroscience,

Lex Fridman (1:21:41.120)

like parts of psychology, parts of neuroscience, parts of sociology?

Lex Fridman (1:21:44.560)

What about philosophy, like studies of human nature from different perspectives?

Charles Isbell (1:21:49.120)

Absolutely. And by the way, it works both ways. So let's take biology for a moment.

Charles Isbell (1:21:53.200)

It turns out a cell is basically a bunch of if, then statements, if you look at it the right way,

Charles Isbell (1:21:57.680)

which is nice because I understand if, then statements. I never really enjoyed biology,

Lex Fridman (1:22:01.680)

but I do understand if, then statements. And if you tell the biologists that and they begin to

Charles Isbell (1:22:05.360)

understand that, it actually helps them to think about a bunch of really cool things.

Charles Isbell (1:22:09.520)

There'll still be biology involved, but whatever. On the other hand, the fact of biology is,

Charles Isbell (1:22:15.040)

in fact, the cell is a bunch of if, then statements or whatever, allows the computationalist to think

Charles Isbell (1:22:19.520)

differently about the language and the way that we, well, certainly the way we would do AI machine

Charles Isbell (1:22:23.520)

learning, but there's just even the way that we think about computation. So the important thing

Charles Isbell (1:22:28.000)

to me is as my engineering colleagues who are not in computer science worry about computer science

Charles Isbell (1:22:33.680)

eating up engineering to colleges where computer science is trapped. It's not a worry. You shouldn't

Charles Isbell (1:22:40.560)

worry about that at all. Computer science computing, it's central, but it's not the

Charles Isbell (1:22:45.280)

most important thing in the world. It's not more important. It is just key to helping others do

Charles Isbell (1:22:50.960)

other cool things they're going to do. You're not going to be a historian in 2030. You're not going

Charles Isbell (1:22:54.800)

to get your PhD in history without understanding some data science and computing, because the way

Charles Isbell (1:22:58.480)

you're going to get history done in part, and I say done, the way you're going to get it done

Charles Isbell (1:23:02.960)

is you're going to look at data and you're going to let, you're going to have the system that's

Charles Isbell (1:23:06.160)

going to help you to analyze things to help you to think about a better way to describe history

Lex Fridman (1:23:09.920)

and to understand what's going on and what it tells us about where we might be going. The same

Charles Isbell (1:23:13.600)

is true for psychology. Same is true for all of these things. The reason I brought that up is

Charles Isbell (1:23:17.040)

because the philosopher has a lot to say about computing. The psychologist has a lot to say

Charles Isbell (1:23:21.920)

about the way humans interact with computing, right? And certainly a lot about intelligence, which

Charles Isbell (1:23:27.440)

at least for me, ultimately is kind of the goal of building these computational devices is to build

Charles Isbell (1:23:32.480)

something intelligent. Did you think computing will eat everything in some certain sense or almost

Charles Isbell (1:23:37.680)

like disappear because it's part of everything? It's so funny you say this. I want to say it's

Charles Isbell (1:23:41.680)

going to metastasize, but there's kind of two ways that fields destroy themselves. One is they become

Charles Isbell (1:23:47.600)

super narrow. And I think we can think of fields that might be that way. They become pure. And we

Charles Isbell (1:23:55.120)

have that instinct. We have that impulse. I'm sure you can think of several people who want computer

Charles Isbell (1:23:59.600)

science to be this pure thing. The other way is you become everywhere and you become everything

Lex Fridman (1:24:05.200)

and nothing. And so everyone says, you know, I'm going to teach Fortran for engineers or whatever.

Charles Isbell (1:24:10.240)

I'm going to do this. And then you lose the thing that makes it worth studying in and of itself.

Charles Isbell (1:24:15.440)

The thing about computing, and this is not unique to computing, though at this point in time,

Charles Isbell (1:24:19.520)

it is distinctive about computing where we happen to be in 2020 is we are both a thriving major.

Charles Isbell (1:24:26.480)

In fact, the thriving major, almost every place. And we're a service unit because people need to

Charles Isbell (1:24:34.320)

know the things we need to know. And our job, much as the mathematician's job is to help,

Charles Isbell (1:24:39.520)

you know, this person over here to think like a mathematician, much the way the point isn't the

Charles Isbell (1:24:43.440)

point of view taking chemistry as a freshman is not to learn chemistry. It's to learn to think

Charles Isbell (1:24:47.600)

like a scientist, right? Our job is to help them to think, think like a computationalist. And we

Charles Isbell (1:24:52.080)

have to take both of those things very seriously. And I'm not sure that as a field, we have

Charles Isbell (1:24:56.880)

historically certainly taken the second thing that our job is to help them to think a certain way.

Charles Isbell (1:25:01.840)

People who aren't going to be our major, I don't think we've taken that that very seriously at all.

Charles Isbell (1:25:04.960)

I don't know if you know who Dan Carlin is. He has this podcast called Hardcore History.

Charles Isbell (1:25:09.600)

Yes.

Charles Isbell (1:25:10.000)

I've just did an amazing four hour conversation with him, mostly about Hitler. But I bring him

Charles Isbell (1:25:18.160)

up because he talks about this idea that it's possible that history as a field will become,

Charles Isbell (1:25:24.160)

like currently, most people study history a little bit, kind of are aware of it. We have a

Charles Isbell (1:25:31.440)

conversation about it, different parts of it. I mean, there's a lot of criticism to say that some

Charles Isbell (1:25:35.920)

parts of history are being ignored, blah, blah, blah, so on. But most people are able to have a

Charles Isbell (1:25:41.760)

curiosity and able to learn it. His thought is it's possible given the way social media works,

Charles Isbell (1:25:49.200)

the current way we communicate, that history becomes a niche field where literally most people

Charles Isbell (1:25:55.840)

just ignore because everything is happening so fast that the history starts losing its meaning.

Lex Fridman (1:26:01.680)

And then it starts being a thing that only, you know, like the theoretical computer science,

Charles Isbell (1:26:07.840)

part of computer science, it becomes a niche thing that only like the rare holders of the

Charles Isbell (1:26:13.840)

world wars and the, you know, all the history, the founding of the United States, all those kinds of

Charles Isbell (1:26:19.920)

things, the Civil Wars. And it's a kind of profound thing to think about how these,

Lex Fridman (1:26:27.360)

how we can lose track, how we can lose these fields when they're best, like in the case of

Charles Isbell (1:26:33.360)

history, is best for that to be a pervasive thing that everybody learns and thinks about and so on.

Lex Fridman (1:26:40.480)

And I would say computing is quite obviously similar to history in the sense that it seems

Charles Isbell (1:26:46.880)

like it should be a part of everybody's life to some degree, especially like as we move into the

Charles Isbell (1:26:53.520)

later parts of the 21st century. And it's not obvious that that's the way it'll go. It might

Charles Isbell (1:26:59.840)

be in the hands of the few still. Like depending if it's machine learning, you know, it's unclear

Charles Isbell (1:27:06.640)

that computing will win out. It's currently very successful, but it's not, I would say that's

Charles Isbell (1:27:12.240)

something, I mean, you're at the leadership level of this. You're defining the future. So it's in

Charles Isbell (1:27:16.880)

your hands. No pressure. But like, it feels like there's multiple ways this can go. And there's

Charles Isbell (1:27:23.760)

this kind of conversation of everybody should learn to code, right? The changing nature of jobs

Lex Fridman (1:27:29.200)

and so on. Do you have a sense of what your role in education of computing is here? Like what's

Charles Isbell (1:27:41.200)

the hopeful path forward? There's a lot there. I will say that, well, first off, it would be an

Charles Isbell (1:27:46.000)

absolute shame if no one studied history. On the other hand, as t approaches infinity, the amount

Charles Isbell (1:27:51.360)

of history is presumably also growing at least linearly. And so you have to forget more and more

Charles Isbell (1:27:59.440)

of history, but history needs to always be there. I mean, I can imagine a world where,

Charles Isbell (1:28:02.960)

you know, if you think of your brains as being outside of your head, that you can kind of learn

Charles Isbell (1:28:07.360)

the history you need to know when you need to know it. That seems fanciful, but it's a kind of way of,

Charles Isbell (1:28:12.560)

you know, is there a sufficient statistic of history? No. And there certainly, but there may

Charles Isbell (1:28:17.440)

be for the particular thing you have to care about, but you know, those who do not remember.

Lex Fridman (1:28:20.560)

It's for our objective camera discussion, right?

Charles Isbell (1:28:23.040)

Yeah. Right. And, you know, we've already lost lots of history. And of course you have your

Charles Isbell (1:28:26.800)

own history that some of which will be, it's even lost to you, right? You don't even remember

Lex Fridman (1:28:31.120)

whatever it was you were doing 17 years ago.

Charles Isbell (1:28:33.280)

All the ex girlfriends.

Lex Fridman (1:28:34.320)

Yeah.

Charles Isbell (1:28:34.640)

Gone.

Charles Isbell (1:28:35.680)

Exactly. So, you know, history is being lost anyway, but the big lessons of history shouldn't

Charles Isbell (1:28:41.280)

be. And I think, you know, to take it to the question of computing and sort of education,

Charles Isbell (1:28:46.080)

the point is you have to get across those lessons. You have to get across the way of thinking.

Lex Fridman (1:28:50.880)

And you have to be able to go back and, you know, you don't want to lose the data,

Charles Isbell (1:28:54.480)

even if, you know, you don't necessarily have the information at your fingertips.

Charles Isbell (1:28:57.680)

With computing, I think it's somewhat different. Everyone doesn't have to learn how to code,

Lex Fridman (1:29:02.640)

but everyone needs to learn how to think in the way that you can be precise. And I mean,

Charles Isbell (1:29:07.680)

precise in the sense of repeatable, not just, you know, in the sense of not resolution in the sense

Charles Isbell (1:29:13.840)

of get the right number of bits, um, in saying what it is you want the machine to do and being

Charles Isbell (1:29:19.680)

able to describe a problem in such a way that it is executable, which we are not human beings are

Charles Isbell (1:29:26.480)

not very good at that. In fact, I think we spend much of our time talking back and forth just to

Charles Isbell (1:29:30.160)

kind of vaguely understand what the other person means and hope we get it good enough that we can,

Charles Isbell (1:29:34.080)

we can act accordingly. Um, you can't do that with machines, at least not yet. And so,

Charles Isbell (1:29:39.440)

you know, having to think that precisely about things is quite important. And that's somewhat

Charles Isbell (1:29:45.840)

different from coding. Coding is a crude means to an end. On the other hand, the idea of coding,

Lex Fridman (1:29:53.600)

what that means that it's a programming language and it has these sort of things that you fiddle

Charles Isbell (1:29:57.760)

with in these ways that you express. That is an incredibly important point. In fact, I would argue

Charles Isbell (1:30:01.920)

that one of the big holes in machine learning right now in an AI is that we forget that we are

Charles Isbell (1:30:07.120)

basically doing software engineering. We forget that we are doing, um, we're using programming,

Charles Isbell (1:30:13.280)

like we're using languages to express what we're doing. We get just so all caught up in the deep

Charles Isbell (1:30:16.720)

network or we get all caught up in whatever that we forget that, you know, we're making decisions

Charles Isbell (1:30:22.560)

based upon a set of parameters that we made up. And if we did slightly different parameters,

Charles Isbell (1:30:26.880)

we'd have completely different, different outcomes. And so the lesson of computing,

Charles Isbell (1:30:30.400)

computer science education is to be able to think like that and to be aware of it when you're doing

Charles Isbell (1:30:36.480)

it. Basically, it's, you know, at the end of the day, it's a way of, um, surfacing your assumptions.

Charles Isbell (1:30:41.120)

I mean, we call them parameters or, you know, we, we, we call them if then statements or whatever,

Lex Fridman (1:30:45.760)

but you're forced to surface those, those assumptions. That's the key, the key thing that

Charles Isbell (1:30:50.080)

you should get out of a computing education that, and that the models and languages and

Charles Isbell (1:30:53.040)

machines are equivalent, but it actually follows from that, that you have to be explicit about,

Charles Isbell (1:30:58.400)

about what it is you're trying to do because the model you're building is something you will one

Charles Isbell (1:31:02.800)

day run. So you better get it right, or at least understand it and be able to express roughly what,

Lex Fridman (1:31:08.480)

what you want to express. So I think it is key that we figure out how to educate everyone to

Charles Isbell (1:31:17.200)

think that way, because at the end, it would not only make them better at whatever it is that they

Charles Isbell (1:31:23.440)

are doing. And I emphasize doing it'll also make them better citizens. It'll help them to understand

Lex Fridman (1:31:30.480)

what others are doing to them so that they can react accordingly. Cause you're not going to

Charles Isbell (1:31:35.920)

solve the problem of social media in so far as you think of social media as a problem

Lex Fridman (1:31:40.720)

by just making slightly better code, right? It only works if people react to it

Charles Isbell (1:31:46.320)

appropriately and know what's happening and therefore take control over what they're doing.

Lex Fridman (1:31:52.160)

I mean, that's, that's my take on it.

Charles Isbell (1:31:53.920)

Okay. Let me try to proceed awkwardly into the topic of race.

Lex Fridman (1:32:00.000)

Okay.

Charles Isbell (1:32:00.560)

One is because it's a fascinating part of your story and you're just eloquent and fun about it.

Lex Fridman (1:32:05.440)

And then the second is because we're living through a pretty tense time in terms of race,

Charles Isbell (1:32:12.560)

tensions and discussions and ideas in this time in America. You grew up in Atlanta,

Lex Fridman (1:32:20.880)

not born in Atlanta. Is some Southern state, somewhere in Tennessee, something like that?

Charles Isbell (1:32:24.960)

Tennessee.

Charles Isbell (1:32:25.600)

Nice. Okay. But early on you moved, you're basically, you identify as an Atlanta native.

Charles Isbell (1:32:34.480)

Mm hmm. Yeah. And you've mentioned that you grew up in a predominantly black neighborhood,

Lex Fridman (1:32:42.320)

by the way, black African American person of color.

Charles Isbell (1:32:44.960)

I prefer black.

Lex Fridman (1:32:46.000)

Black.

Charles Isbell (1:32:46.400)

With a capital B.

Lex Fridman (1:32:47.520)

With a capital B. The other letters are...

Charles Isbell (1:32:50.720)

The rest of them, no matter.

Charles Isbell (1:32:54.640)

Okay. So the predominantly black neighborhood. And so you didn't almost see race. Maybe you

Charles Isbell (1:32:59.600)

can correct me on that. And then just in the video you talked about when you showed up to

Charles Isbell (1:33:05.200)

Georgia Tech for your undergrad, you're one of the only black folks there. And that was like,

Charles Isbell (1:33:12.640)

oh, that was a new experience. So can you take me from just a human perspective,

Lex Fridman (1:33:19.280)

but also from a race perspective, your journey growing up in Atlanta

Lex Fridman (1:33:23.120)

and then showing up at Georgia Tech?

Lex Fridman (1:33:24.800)

Okay. That's easy. And by the way, that story continues through MIT as well.

Charles Isbell (1:33:28.160)

Yeah. In fact, it was quite a bit more stark at MIT and Boston.

Lex Fridman (1:33:32.960)

So maybe just a quick pause, Georgia Tech was undergrad, MIT was graduate school.

Charles Isbell (1:33:37.920)

Mm hmm. And I went directly to grad school from undergrad. So I had no

Lex Fridman (1:33:42.240)

distractions in between my bachelor's and my master's and PhD.

Lex Fridman (1:33:45.600)

You didn't go on a backpacking trip in Europe?

Charles Isbell (1:33:47.760)

Didn't do any of that. In fact, I literally went to IBM for three months, got in a car,

Lex Fridman (1:33:52.640)

and drove straight to Boston with my mother, or Cambridge.

Lex Fridman (1:33:55.600)

Yeah.

Charles Isbell (1:33:55.840)

I moved into an apartment I'd never seen over the Royal East. Anyway, that's another story.

Lex Fridman (1:34:02.080)

So let me tell you a little bit about it.

Lex Fridman (1:34:03.440)

You miss MIT?

Lex Fridman (1:34:04.880)

Oh, I loved MIT. I don't miss Boston at all, but I loved MIT.

Charles Isbell (1:34:09.040)

That was fighting words.

Lex Fridman (1:34:11.120)

So let's back up to this. So as you said, I was born in Chattanooga, Tennessee.

Charles Isbell (1:34:14.800)

My earliest memory is arriving in Atlanta in a moving truck at the age of three and a half.

Lex Fridman (1:34:18.080)

So I think of myself as being from Atlanta, very distinct memory of that. So I grew up in Atlanta.

Charles Isbell (1:34:22.400)

It's the only place I ever knew as a kid. I loved it. Like much of the country, and certainly

Charles Isbell (1:34:28.560)

much of Atlanta in the 70s and 80s, it was deeply highly segregated, though not in a way that I

Charles Isbell (1:34:34.240)

think was obvious to you unless you were looking at it or were old enough to have noticed it.

Lex Fridman (1:34:39.680)

But you could divide up Atlanta, and Atlanta is hardly unique in this way, by highway,

Lex Fridman (1:34:43.360)

and you could get racing class that way. So I grew up not only in a predominantly

Charles Isbell (1:34:47.920)

black area, to say the very least, I grew up on the poor side of that. But I was very much aware

Charles Isbell (1:34:55.920)

of race for a bunch of reasons, one that people made certain that I was, my family did, but also

Charles Isbell (1:35:01.200)

that it would come up. So in first grade, I had a girlfriend. I say I had a girlfriend. I didn't

Charles Isbell (1:35:08.320)

have a girlfriend. I wasn't even entirely sure what girls were in the first grade. But I do remember

Charles Isbell (1:35:13.120)

she decided I was her girlfriend's little white girl named Heather. And we had a long discussion

Charles Isbell (1:35:17.680)

about how it was okay for us to be boyfriend and girlfriend, despite the fact that she was white

Lex Fridman (1:35:21.440)

and I was black. Between the two of you? Did your parents know about this?

Charles Isbell (1:35:26.400)

Yes. But being a girlfriend and boyfriend in first grade just basically meant that you spent

Charles Isbell (1:35:31.120)

slightly more time together during recess. I think we Eskimo kissed once. It didn't mean anything.

Charles Isbell (1:35:38.160)

It was. At the time, it felt very scandalous because everyone was watching. I was like,

Charles Isbell (1:35:41.440)

ah, my life is now my life has changed in first grade. No one told me elementary school would be

Charles Isbell (1:35:45.920)

like this. Did you write poetry or not in first grade? That would come later. That would come

Charles Isbell (1:35:50.400)

during puberty when I wrote lots and lots of poetry. Anyway, so I was aware of it. I didn't

Charles Isbell (1:35:56.400)

think too much about it, but I was aware of it. But I was surrounded. It wasn't that I wasn't

Charles Isbell (1:36:01.280)

aware of race. It's that I wasn't aware that I was a minority. It's different. And it's because I

Charles Isbell (1:36:07.920)

wasn't as far as my world was concerned. I mean, I'm six years old, five years old in first grade.

Charles Isbell (1:36:12.880)

The world is the seven people I see every day. So it didn't feel that way at all.

Lex Fridman (1:36:17.840)

And by the way, this being Atlanta, home of the civil rights movement and all the rest,

Charles Isbell (1:36:21.840)

it meant that when I looked at TV, which back then one did because there were only three,

Charles Isbell (1:36:25.520)

four or five channels. And I saw the news, which my mother might make me watch. Monica Kaufman was

Charles Isbell (1:36:33.520)

on TV telling me the news and they were all black and the mayor was black and always been

Charles Isbell (1:36:37.680)

black. And so it just never occurred to me. When I went to Georgia Tech, I remember the first day

Charles Isbell (1:36:43.520)

walking across campus from West campus to East campus and realizing along the way that of the

Charles Isbell (1:36:49.360)

hundreds and hundreds and hundreds and hundreds of students that I was seeing, I was the only black

Charles Isbell (1:36:53.200)

one. That was enlightening and very off putting because it occurred to me. And then of course,

Charles Isbell (1:36:59.920)

it continued that way for, well, for the rest of my, for much of the rest of my career at Georgia

Charles Isbell (1:37:05.440)

Tech. Of course, I found lots of other students and I met people cause in Atlanta, you're either

Charles Isbell (1:37:09.440)

black or you're white. There was nothing else. So I began to meet students of Asian descent and I

Charles Isbell (1:37:14.560)

met students who we would call Hispanic and so on and so forth. And you know, so my world,

Charles Isbell (1:37:18.320)

this is what college is supposed to do, right? It's supposed to open you up to people. And it

Charles Isbell (1:37:22.320)

did, but it was a very strange thing to be in the minority. When I came to Boston, I will tell you

Charles Isbell (1:37:31.200)

a story. I applied to one place as an undergrad, Georgia Tech, because I was stupid. I didn't know

Charles Isbell (1:37:38.320)

any better. I just didn't know any better, right? No one told me. When I went to grad school,

Charles Isbell (1:37:43.360)

I applied to three places, Georgia Tech, because that's where I was, MIT and CMU.

Charles Isbell (1:37:49.600)

When I got in to MIT, I got into CMU, but I had a friend who went to CMU. And so I asked him what

Charles Isbell (1:37:58.480)

he thought about it. He spent his time explaining to me about Pittsburgh, much less about CMU,

Lex Fridman (1:38:03.200)

but more about Pittsburgh, which I developed a strong opinion based upon his strong opinion,

Charles Isbell (1:38:07.760)

something about the sun coming out two days out of the year. And I didn't get a chance to go there

Charles Isbell (1:38:12.240)

because the timing was wrong. I think it was because the timing was wrong. At MIT, I asked

Charles Isbell (1:38:19.120)

20 people I knew, either when I visited or I had already known for a variety of reasons,

Charles Isbell (1:38:24.480)

whether they liked Boston. And 10 of them loved it, and 10 of them hated it. The 10 who loved it

Charles Isbell (1:38:30.160)

were all white. The 10 who hated it were all black. And they explained to me very much why

Charles Isbell (1:38:35.040)

that was the case. Both stats told me why. And the stories were remarkably the same for the

Charles Isbell (1:38:41.040)

two clusters. And I came up here, and I could see it immediately, why people would love it

Lex Fridman (1:38:46.560)

and why people would not. And why people tell you about the nice coffee shops.

Charles Isbell (1:38:50.720)

Well, it wasn't coffee shops. It was used CD places. But yeah, it was that kind of a thing.

Charles Isbell (1:38:55.840)

Nice shops. Oh, there's all these students here. Harvard Square is beautiful. You can do all these

Charles Isbell (1:39:00.000)

things, and you can walk. And something about the outdoors, which I wasn't the slightest bit

Lex Fridman (1:39:02.720)

interested in. The outdoors is for the bugs. It's not for humans.

Charles Isbell (1:39:08.400)

That should be a t shirt.

Charles Isbell (1:39:09.680)

Yeah, that's the way I feel about it. And the black folk told me completely different stories

Charles Isbell (1:39:14.800)

about which part of town you did not want to be caught in after dark. But that was nothing new.

Lex Fridman (1:39:22.000)

So I decided that MIT was a great place to be as a university. And I believed it then,

Charles Isbell (1:39:27.600)

I believe it now. And that whatever it is I wanted to do, I thought I knew what I wanted to do,

Lex Fridman (1:39:32.160)

but what if I was wrong? Someone there would know how to do it. Of course, then I would pick the

Charles Isbell (1:39:36.960)

one topic that nobody was working on at the time, but that's okay. It was great. And so I thought

Charles Isbell (1:39:42.320)

that I would be fine. And I'd only be there for like four or five years. I told myself,

Charles Isbell (1:39:46.480)

which turned out not to be true at all. But I enjoyed my time. I enjoyed my time there.

Lex Fridman (1:39:50.400)

But I did see a lot of... I ran across a lot of things that were driven by what I look like

Charles Isbell (1:39:58.400)

while I was here. I got asked a lot of questions. I ran into a lot of cops. I saw a lot about the

Charles Isbell (1:40:05.920)

city. But at the time, I mean, I haven't been here a long time. These are the things that I

Charles Isbell (1:40:09.120)

remember. So this is 1990. There was not a single black radio station. Now this is 1990. I don't

Charles Isbell (1:40:17.920)

know if there are any radio stations anymore. I'm sure there are, but I don't listen to the radio

Charles Isbell (1:40:21.920)

anymore and almost no one does, at least if you're under a certain age. But the idea is you could be

Charles Isbell (1:40:27.520)

in a major metropolitan area and there wasn't a single black radio station, by which I mean,

Charles Isbell (1:40:30.800)

a radio station to play what we would call black music then, was absurd, but somehow captured kind

Charles Isbell (1:40:37.840)

of everything about the city. I grew up in Atlanta and you've heard me tell you about Atlanta.

Charles Isbell (1:40:44.960)

Boston had no economically viable or socially cohesive black middle class.

Charles Isbell (1:40:51.840)

Insofar as it existed, it was uniformly distributed throughout large parts, not all parts,

Lex Fridman (1:40:56.480)

but large parts of the city. And where you had concentrations of black Bostonians,

Charles Isbell (1:41:02.160)

they tended to be poor. It was very different from where I grew up. I grew up on the poor side of

Charles Isbell (1:41:07.040)

town, sure. But then in high school, well, in ninth grade, we didn't have middle school. I went

Charles Isbell (1:41:13.040)

to an eighth grade school where there was a lot of, let's just say, we had a riot the year that

Charles Isbell (1:41:17.200)

I was there. There was at least one major fight every week. It was an amazing experience. But

Charles Isbell (1:41:24.800)

when I went to ninth grade, I went to Academy. Math and Science Academy, Mays High. It was a

Charles Isbell (1:41:30.640)

public school. It was a magnet school. That's why I was able to go there. It was the first high school,

Charles Isbell (1:41:36.240)

I think, in the state of Georgia to sweep the state math and science fairs. It was great. It had

Charles Isbell (1:41:44.080)

385 students, all but four of whom were black. I went to school with the daughter of the

Charles Isbell (1:41:51.200)

former mayor of Atlanta, Michael Jackson's cousin. I mean, it was an upper middle class.

Charles Isbell (1:41:56.560)

Dr. Justin Marchegiani Dropping names.

Lex Fridman (1:41:57.120)

Dr. Justin Marchegiani You know, I just drop names occasionally.

Charles Isbell (1:41:59.520)

You know, drop the mic, drop some names. Just to let you know, I used to hang out with Michael

Charles Isbell (1:42:03.120)

Jackson's cousin, 12th cousin, nine times removed. I don't know. The point is, they had money. We

Charles Isbell (1:42:07.680)

had a parking problem because the kids had cars. I did not come from a place where you had cars.

Charles Isbell (1:42:12.160)

I had my first car when I came to MIT, actually. So, it was just a very different experience for

Charles Isbell (1:42:21.280)

me. But I'd been to places where whether you were rich or whether you were poor, you know,

Charles Isbell (1:42:26.000)

you could be black and rich or black and poor. And it was there and there were places and they

Charles Isbell (1:42:29.200)

were segregated by class as well as by race. But that existed. Here, at least when I was here,

Charles Isbell (1:42:35.520)

didn't feel that way at all. And it felt like a bunch of a really interesting contradiction.

Charles Isbell (1:42:41.280)

It felt like it was the interracial dating capital of the country. It really felt that way.

Lex Fridman (1:42:49.360)

But it also felt like the most racist place I ever spent any time. You know, you couldn't go

Charles Isbell (1:42:55.120)

up the Orange Line at that time. I mean, again, that was 30 years ago. I don't know what it's

Charles Isbell (1:42:59.360)

like now. But there were places you couldn't go. And you knew it. Everybody knew it. And there were

Charles Isbell (1:43:05.520)

places you couldn't live. And everybody knew that. And that was just the greater Boston area in 1992.

Lex Fridman (1:43:12.720)

Subtle racism or explicit racism?

Charles Isbell (1:43:14.880)

Both.

Lex Fridman (1:43:16.720)

In terms of within the institutions, did you feel...

Charles Isbell (1:43:19.440)

Was there levels in which you were empowered to be first or one of the first black people in a

Charles Isbell (1:43:26.080)

particular discipline in some of these great institutions that you were a part of? You know,

Lex Fridman (1:43:31.360)

Georgia Tech or MIT? And was there a part where it felt limiting?

Charles Isbell (1:43:37.760)

I always felt empowered. Some of that was my own delusion, I think. But it worked out. So I never

Charles Isbell (1:43:45.120)

felt... In fact, quite the opposite. Not only did I not feel as if no one was trying to stop me,

Charles Isbell (1:43:52.240)

I had the distinct impression that people wanted me to succeed. By people, I meant the people in

Charles Isbell (1:43:57.760)

power. Not my fellow students. Not that they didn't want me to succeed. But I felt supported,

Charles Isbell (1:44:04.960)

or at least that people were happy to see me succeed at least as much as anyone else. But,

Charles Isbell (1:44:10.480)

you know, 1990, you're dealing with a different set of problems. You're very early, at least in

Charles Isbell (1:44:15.280)

computer science, you're very early in the Jackie Robinson period. There's this thing called the

Charles Isbell (1:44:20.720)

Jackie Robinson syndrome, which is that the first one has to be perfect or has to be sure to

Charles Isbell (1:44:27.120)

succeed because if that person fails, no one else comes after for a long time. So it was kind of in

Charles Isbell (1:44:32.960)

everyone's best interest. But I think it came from a sincere place. I'm completely sure that people

Charles Isbell (1:44:37.040)

went out of their way to try to make certain that the environment would be good. Not just for me,

Lex Fridman (1:44:43.200)

but for the other people who, of course, were around. And I was the only person in the iLab,

Lex Fridman (1:44:47.200)

but I wasn't the only person at MIT by a long shot. On the other hand, we're what?

Charles Isbell (1:44:53.120)

At that point, we would have been, what, less than 20 years away from the first black PhD to

Charles Isbell (1:44:57.760)

graduate from MIT, right? Shirley Jackson, right? 1971, something like that? Somewhere around then.

Lex Fridman (1:45:03.840)

So we weren't that far away from the first first, and we were still another eight years away from

Charles Isbell (1:45:09.680)

the first black PhD in computer science, right? So it was a sort of interesting time. But I did

Charles Isbell (1:45:16.880)

not feel as if the institutions of the university were against any of that. And furthermore, I felt

Charles Isbell (1:45:25.280)

as if there was enough of a critical mass across the institute from students and probably faculty

Charles Isbell (1:45:30.960)

that I didn't know them, who wanted to make certain that the right thing happened. It was very

Charles Isbell (1:45:35.680)

different from the institutions of the rest of the city, which I think were designed in such a way

Charles Isbell (1:45:41.600)

that they felt no need to be supportive.

Lex Fridman (1:45:44.000)

Let me ask a touchy question on that. So you kind of said that you didn't feel,

Charles Isbell (1:45:52.640)

you felt empowered. Is there some lesson, advice, in the sense that no matter what,

Charles Isbell (1:46:00.160)

you should feel empowered? You said, you used the word, I think, illusion or delusion.

Charles Isbell (1:46:05.920)

Is there a sense from the individual perspective where you should always kind of ignore, you know,

Charles Isbell (1:46:14.480)

the, ignore your own eyes, ignore the little forces that you are able to observe around you,

Charles Isbell (1:46:27.120)

that are like trying to mess with you of whether it's jealousy, whether it's hatred in its pure

Lex Fridman (1:46:33.040)

form, whether it's just hatred in its like deluded form, all that kind of stuff?

Lex Fridman (1:46:38.960)

And just kind of see yourself as empowered and confident and all those kinds of things.

Charles Isbell (1:46:44.480)

I mean, it certainly helps, but it's, there's a trade off, right? You have to be deluded enough

Charles Isbell (1:46:47.680)

to think that you can succeed. I mean, you can't get a PhD unless you're crazy enough to think you

Charles Isbell (1:46:51.360)

can invent something that no one else has come up with. I mean, that kind of massive delusion is that

Charles Isbell (1:46:56.320)

you have to be deluded enough to believe that you can succeed despite whatever odds you see

Charles Isbell (1:46:59.840)

in front of you, but you can't be so deluded that you don't think that you need to step out of

Charles Isbell (1:47:03.280)

the way of the oncoming train, right? So it's all a trade off, right? You have to kind of believe in

Charles Isbell (1:47:11.760)

yourself. It helps to have a support group around you in some way or another. I was able to find

Charles Isbell (1:47:16.800)

that, I've been able to find that wherever I've gone, even if it wasn't necessarily on the floor

Charles Isbell (1:47:21.440)

that I was in, I had lots of friends when I was here. Many of them still live here. And I've kept

Charles Isbell (1:47:26.080)

up with many of them. So I felt supported. And certainly I had my mother and my family and those

Charles Isbell (1:47:30.800)

people back home that I could always lean back on, even if it were a long distance call that cost

Charles Isbell (1:47:37.200)

money, which is not something that any of the kids today even know what I'm talking about. But

Charles Isbell (1:47:41.680)

back then it mattered, calling my mom was an expensive proposition. But you have that and

Charles Isbell (1:47:45.920)

it's fine. I think it helps. But you cannot be so deluded that you miss the obvious because it makes

Charles Isbell (1:47:51.280)

things slower and it makes you think you're doing better than you are and it will hurt you in the

Charles Isbell (1:47:55.920)

long run. You mentioned cops. You tell a story of being pulled over. Perhaps it happened more than

Charles Isbell (1:48:04.400)

once. More than once, for sure. One, could you tell that story? And in general, can you give me

Lex Fridman (1:48:11.040)

a sense of what the world looks like when the law doesn't always look at you with a blank slate?

Charles Isbell (1:48:22.640)

With a blank slate with objective eyes? I don't know how to say it more poetically.

Charles Isbell (1:48:33.840)

Well, I guess the, I don't either. I guess the answer is it looks exactly the way it looks now

Charles Isbell (1:48:39.760)

because this is the world that we happen to live in, right? It's people clustering and doing the

Charles Isbell (1:48:44.640)

things that they do and making decisions based on one or two bits of information they find

Charles Isbell (1:48:50.880)

relevant, which, by the way, are all positive feedback loops, which makes it easier for you

Charles Isbell (1:48:56.080)

to believe what you believed before because you behave in a certain way that makes it true and

Charles Isbell (1:48:59.520)

it goes on and circles and then cycles and cycles and then cycles. So it's just about being on edge.

Charles Isbell (1:49:06.960)

I do not, despite having made it over 50 now.

Lex Fridman (1:49:11.760)

Congratulations, brother.

Charles Isbell (1:49:13.520)

God, I have a few gray hairs here and there.

Lex Fridman (1:49:16.080)

You did pretty good.

Charles Isbell (1:49:16.800)

I think, I don't imagine I will ever see a police officer and not get very, very tense.

Charles Isbell (1:49:25.600)

Now, everyone gets a little tense because it probably means you're being pulled over for

Charles Isbell (1:49:30.560)

speeding or something, or you're going to get a ticket or whatever, right? I mean,

Charles Isbell (1:49:34.400)

the interesting thing about the law in general is that most human beings experience of it is

Charles Isbell (1:49:39.280)

fundamentally negative, right? You're only dealing with a lawyer if you're in trouble,

Charles Isbell (1:49:43.920)

except in a few very small circumstances, right? So that's an underlying reality.

Charles Isbell (1:49:49.680)

Now, imagine that that's also at the hands of the police officer. I remember the time when I got

Charles Isbell (1:49:55.120)

pulled over that time, halfway between Boston and Wellesley, actually. I remember thinking

Charles Isbell (1:50:04.160)

when he pulled his gun on me that if he shot me right now, he'd get away with it. That was the

Charles Isbell (1:50:11.360)

that was the worst thing that I felt about that particular moment, is that if he shoots me now,

Charles Isbell (1:50:16.480)

he will get away with it. It would be years later when I realized actually much worse than that

Charles Isbell (1:50:24.480)

is that he'd get away with it. And if it became a thing that other people knew about,

Charles Isbell (1:50:30.880)

odds would be, of course, that it wouldn't. But if it became a thing that other people knew about,

Charles Isbell (1:50:34.240)

if I was living in today's world as opposed to the world 30 years ago, that not only would

Charles Isbell (1:50:39.680)

get away with it, but that I would be painted a villain. I was probably big and scary, and I

Charles Isbell (1:50:45.360)

probably moved too fast, and if only I'd done what he said, and da, da, da, da, da, da, da,

Charles Isbell (1:50:49.040)

which is somehow worse, right? You know, that hurts not just you, you're dead, but your family,

Lex Fridman (1:50:55.760)

and the way people look at you, and look at your legacy or your history, that's terrible.

Lex Fridman (1:51:00.960)

And it would work. I absolutely believe it would have worked had he done it. Now, he didn't. I

Charles Isbell (1:51:05.360)

don't think he wanted to shoot me. I don't think he felt like killing anybody. He did not go out

Charles Isbell (1:51:08.640)

that night expecting to do that or planning on doing it, and I wouldn't be surprised if he never,

Charles Isbell (1:51:12.880)

ever did that or ever even pulled his gun again. I don't know the man's name. I don't remember

Charles Isbell (1:51:16.800)

anything about him. I do remember the gun. Guns are very big when they're in your face. I can tell

Charles Isbell (1:51:20.320)

you this much. They're much larger than they seem. But... And you're basically like speeding or

Charles Isbell (1:51:24.480)

something like that? He said I ran a light, I think. You ran a light. I don't think I ran a

Charles Isbell (1:51:28.400)

light, but you know, in fact, I may not have even gotten a ticket. I may have just gotten a warning.

Charles Isbell (1:51:33.040)

I think he was a little... But he pulled a gun. Yeah. Apparently I moved too fast or something.

Charles Isbell (1:51:37.920)

Rolled my window down before I should have. It's unclear. I think he thought I was going to do

Charles Isbell (1:51:42.240)

something, or at least that's how he behaved. So how, if we can take a little walk around your

Lex Fridman (1:51:48.640)

brain, how do you feel about that guy and how do you feel about cops after that experience?

Charles Isbell (1:51:58.880)

Well, I don't remember that guy, but my view on police officers is the same view I have about

Charles Isbell (1:52:03.680)

lots of things. Fire is an important and necessary thing in the world, but you must respect fire

Charles Isbell (1:52:12.560)

because it will burn you. Fire is a necessary evil in the sense that it can burn you. Necessary

Charles Isbell (1:52:19.360)

in the sense that, you know, heat and all the other things that we use fire for. So when I see

Charles Isbell (1:52:25.280)

a cop, I see a giant ball of flame and I just try to avoid it. And then some people might see

Lex Fridman (1:52:33.040)

a nice place, a nice thing to roast marshmallows with a family over.

Charles Isbell (1:52:37.760)

Which is fine, but I don't roast marshmallows.

Charles Isbell (1:52:40.880)

Okay. So let me go a little dark and I apologize. Just talked to Dan Carlin about

Charles Isbell (1:52:44.160)

Hitler for four hours. So sorry if I go dark here a little bit, but

Charles Isbell (1:52:50.800)

is it easy for this experience of just being careful with the fire and avoiding it to turn

Lex Fridman (1:52:57.200)

to hatred? Yeah, of course. And one might even argue that it is a logical conclusion, right?

Charles Isbell (1:53:05.840)

On the other hand, you've got to live in the world and I don't think it's helpful. Hate is something

Charles Isbell (1:53:12.560)

one should, I mean, hate is something that takes a lot of energy. So one should reserve it for

Charles Isbell (1:53:20.000)

when it is useful and not carried around with you all the time. Again, there's a big difference

Charles Isbell (1:53:25.360)

between the happy delusion that convinces you that you can actually get out of bed and

Charles Isbell (1:53:30.400)

make it to work today without getting hit by a car and the sad delusion that means you can

Charles Isbell (1:53:36.720)

not worry about this car that is barreling towards you, right? So we all have to be a

Charles Isbell (1:53:40.640)

little deluded because otherwise we're paralyzed, right? But one should not be ridiculous.

Charles Isbell (1:53:46.240)

If we go all the way back to something you said earlier about empathy,

Charles Isbell (1:53:49.360)

I think what I would ask other people to get out of this one of many, many, many stories

Charles Isbell (1:53:57.600)

is to recognize that it is real. People would ask me to empathize with the police officer.

Charles Isbell (1:54:04.560)

I would quote back statistics saying that being a police officer isn't even in the top 10 most

Charles Isbell (1:54:11.040)

dangerous jobs in the United States, you're much more likely to get killed in a taxicab.

Charles Isbell (1:54:14.960)

Half of police officers are actually killed by suicide, but that means their lives are something,

Charles Isbell (1:54:22.960)

something's going on there with them and I would more than happy to be empathetic about what it is

Charles Isbell (1:54:28.240)

they go through and how they see the world. I think though that if we step back from what I feel,

Charles Isbell (1:54:34.720)

if we step back from what an individual police officer feels, you step up a level and all this,

Charles Isbell (1:54:39.840)

because all things tie back into interactive AI. The real problem here is that we've built a

Charles Isbell (1:54:44.560)

narrative. We built a big structure that has made it easy for people to put themselves into different

Charles Isbell (1:54:50.240)

pots in the different clusters and to basically forget that the people in the other clusters are

Charles Isbell (1:54:57.440)

ultimately like them. It is useful exercise to ask yourself sometimes, I think, that if I had grown

Charles Isbell (1:55:03.840)

up in a completely different house and a completely different household as a completely different

Charles Isbell (1:55:07.760)

person, if I had been a woman, would I see the world differently? Would I believe what that crazy

Charles Isbell (1:55:12.320)

person over there believes? And the answer is probably yes, because after all, they believe it.

Lex Fridman (1:55:19.440)

And fundamentally, they're the same as you. So then what can you possibly do to fix it? How do

Charles Isbell (1:55:25.440)

you fix Twitter? If you think Twitter needs to be broken or Facebook, if you think Facebook is

Charles Isbell (1:55:29.760)

broken, how do you fix racism? How do you fix any of these things? That's all structural.

Charles Isbell (1:55:35.280)

I mean, individual conversations matter a lot, but you have to create structures that allow people

Charles Isbell (1:55:42.320)

to have those individual conversations all the time in a way that is relatively safe and that

Charles Isbell (1:55:47.760)

allows them to understand that other people have had different experiences, but that ultimately

Charles Isbell (1:55:51.600)

we're the same, which sounds very, I don't even know what the right word is. I'm trying to avoid

Lex Fridman (1:55:57.200)

a word like saccharine, but it feels very optimistic.

Lex Fridman (1:56:01.760)

But I think that's okay. I think that's a part of the delusion, is you want to be a little

Charles Isbell (1:56:06.640)

optimistic and then recognize that the hard problem is actually setting up the structures

Charles Isbell (1:56:10.160)

in the first place, because it's in almost no one's interest to change the infrastructure.

Charles Isbell (1:56:16.320)

Right. I tend to believe that leaders have a big role to that, of selling that optimistic

Charles Isbell (1:56:22.560)

delusion to everybody, and that eventually leads to the building of the structures. But that

Charles Isbell (1:56:27.600)

requires a leader that unites, sort of unites everybody on a vision as opposed to divides

Charles Isbell (1:56:33.600)

on a vision, which is, this particular moment in history feels like there's a nonzero probability,

Charles Isbell (1:56:43.600)

if we go to the P, of something akin to a violent or a nonviolent civil war. This is one of the

Charles Isbell (1:56:51.440)

most divisive periods of American history in recent, you can speak to this from a perhaps

Charles Isbell (1:56:58.800)

a more knowledgeable and deeper perspective than me, but from my naive perspective, this seems like

Charles Isbell (1:57:03.600)

a very strange time. There's a lot of anger, and it has to do with people, I mean, for many reasons.

Charles Isbell (1:57:11.440)

One, the thing that's not spoken about, I think, much is the conflict of opinion,

Charles Isbell (1:57:18.480)

much is the quiet economic pain of millions that's like growing because of COVID, because of closed

Charles Isbell (1:57:29.040)

businesses, because of like lost dreams. So that's building, whatever that tension is building.

Charles Isbell (1:57:35.440)

The other is, there seems to be an elevated level of emotion. I'm not sure if you can psychoanalyze

Charles Isbell (1:57:41.680)

where that's coming from, but this sort of, from which the protests and so on percolated. It's like,

Lex Fridman (1:57:47.440)

why now? Why this particular moment in history? Oh, because time, enough time has passed, right?

Charles Isbell (1:57:52.080)

I mean, you know, the very first race riots were in Boston, not to draw anything from that.

Charles Isbell (1:57:56.800)

Really? When? Oh, this is before like... Going way, I mean, like the 1700s or whatever,

Charles Isbell (1:58:01.200)

right? I mean, there was a massive one in New York. I mean, I'm talking way, way, way back when.

Lex Fridman (1:58:05.280)

So Boston used to be the hotbed of riots. It's just what Boston was all about,

Charles Isbell (1:58:09.920)

or so I'm told from history class. There's an interesting one in New York. I remember when

Charles Isbell (1:58:15.120)

that was. Anyway, the point is, you know, basically you got to get another generation,

Lex Fridman (1:58:22.960)

old enough to be angry, but not so old to remember what happened the last time, right?

Lex Fridman (1:58:28.160)

And that's sort of what happens. But, you know, you said like two completely, you said two things

Charles Isbell (1:58:33.120)

there that I think are worth unpacking. One has to do with this sort of moment in time.

Charles Isbell (1:58:38.320)

And, you know, why? Why is this sort of up built? And the other has to do with a kind of, you know,

Charles Isbell (1:58:43.760)

sort of the economic reality of COVID. So I'm actually, I want to separate those things because,

Charles Isbell (1:58:47.600)

for example, you know, this happened before COVID happened, right? So let's separate these two

Charles Isbell (1:58:54.160)

things for a moment. Now, let me preface all this by saying that although I am interested in history,

Charles Isbell (1:59:01.920)

one of my three minors as an undergrad was history, specifically history, the 1960s. Interesting. The

Charles Isbell (1:59:07.520)

other was Spanish. And, okay, that's a mistake. Oh, I loved that. And history of Spanish and Spanish

Charles Isbell (1:59:14.160)

history, actually, but Spanish and the other was what we would now call cognitive science. But at

Charles Isbell (1:59:17.760)

the time, that's fascinating. Interesting. I minored in Cogsci here for grad school. That was

Charles Isbell (1:59:24.960)

really, that was really fascinating. It was a very different experience. I mean, it was a very

Charles Isbell (1:59:29.440)

it was really fascinating. It was a very different experience from all the computer science classes

Charles Isbell (1:59:33.600)

I've been taking, even the Cogsci classes I was taking at an undergrad. Anyway, I'm interested

Charles Isbell (1:59:42.400)

in history, but I'm hardly a historian, right? So, you know, forgive my, I will ask the audience to

Charles Isbell (1:59:48.720)

forgive my simplification. But I think the question that's always worth asking, as opposed, it's the

Charles Isbell (1:59:58.640)

same question, but a little different. Not why now, but why not before? Right? So why the 1950s,

Lex Fridman (20:00.240)

I wonder if AI systems can help somehow and to bridge the gap of what is this person's life like?

Charles Isbell (20:10.400)

Encourage me to just ask that question, to put myself in their shoes, to experience the agitations,

Charles Isbell (20:16.800)

the fears, the hopes they have, the, to experience, you know, the, to even just to think about what

Charles Isbell (20:23.920)

was their upbringing like, like having a, a single parent home or a shitty education or all those

Charles Isbell (20:32.000)

kinds of things, just to put myself in that mind space. It feels like that's really important.

Charles Isbell (20:37.760)

For us to, to, to bring those clusters together, to find that similar language. But it's unclear

Lex Fridman (20:43.600)

how AI can help that because it seems like AI systems need to understand both parties first.

Lex Fridman (20:48.880)

So the, you know, the word understand, there's doing a lot of work, right?

Charles Isbell (20:51.760)

Yes.

Lex Fridman (20:52.240)

So do you have to understand it or do you just simply have to note that there is something

Charles Isbell (20:57.840)

similar as a point to touch, right? So, you know, you use the word empathy and I like that word,

Charles Isbell (21:04.480)

for a lot of reasons, I think you're right in the way that you're using in the ways you're describing,

Lex Fridman (21:07.600)

but let's separate it from sympathy, right? So, you know, sympathy is feeling sort of for someone,

Lex Fridman (21:13.600)

empathy is kind of understanding where they're coming from and how they, how they feel, right?

Lex Fridman (21:16.720)

And for most people, those things go hand in hand. For some people, some are very good at empathy

Lex Fridman (21:22.480)

and very, very bad at sympathy. Some people cannot experience, well, my observation would be,

Charles Isbell (21:28.160)

I'm not a psychologist, my observation would be that some people seem to be very, very,

Charles Isbell (21:32.480)

very bad at sympathy. My observation would be that some people seem incapable of feeling sympathy

Charles Isbell (21:37.680)

unless they feel empathy first. You can understand someone, understand where they're coming from and

Charles Isbell (21:42.160)

still think, no, I can't support that, right? It doesn't mean that the only way I, because if that,

Charles Isbell (21:48.160)

if that isn't the case, then what it requires is that you, you must, the only way that you can,

Lex Fridman (21:54.480)

to understand someone means you must agree with everything that they do, which isn't right, right?

Charles Isbell (21:59.760)

I can feel for someone is to completely understand them and make them like me in some way. Well,

Charles Isbell (22:06.240)

then we're lost, right? Because we're not all exactly like each other. I have to understand

Lex Fridman (22:10.480)

everything that you've gone through. It helps clearly, but they're separable ideas, right?

Charles Isbell (22:14.240)

Even though they get clearly, clearly tangled up in one another. So what I think AI could help you

Charles Isbell (22:18.480)

do actually is if, and you know, I'm, I'm being quite fanciful as it were, but if you, if you

Charles Isbell (22:23.920)

think of these as kind of, I understand how you interact, the words that you use, the, you know,

Charles Isbell (22:28.000)

the actions you take, I have some way of doing this. Let's not worry about what that is.

Lex Fridman (22:31.840)

But I can see you as a kind of distribution of experiences and actions taken upon you,

Charles Isbell (22:36.720)

things you've done and so on. And I can do this with someone else and I can find the places where

Charles Isbell (22:41.360)

there's some kind of commonality, a mapping as it were, even if it's not total, you know, the,

Charles Isbell (22:46.720)

if I think of this as distribution, right, then, you know, I can take the cosine of the angle

Charles Isbell (22:50.400)

between you and if it's, you know, if it's zero, you've got nothing in common. If it's one,

Charles Isbell (22:54.240)

you're completely the same person. Well, you know, you're probably not one. You're almost certainly

Charles Isbell (22:59.760)

not zero. I can find the place where there's the overlap, then I might be able to introduce you on

Charles Isbell (23:03.840)

that basis or connect you in that, connect you in that way and make it easier for you to take that

Charles Isbell (23:09.200)

step of that step of empathy. It's not, it's not impossible to do. Although I wonder if it requires

Charles Isbell (23:17.760)

that everyone involved is at least interested in asking the question. So maybe the hard part

Charles Isbell (23:22.000)

is just getting them interested in asking the question. In fact, maybe if you can get them to

Charles Isbell (23:24.800)

ask the question, how are we more alike than we are different, they'll solve it themselves. Maybe

Charles Isbell (23:28.480)

that's the problem that AI should be working on, not telling you how you're similar or different,

Lex Fridman (23:32.640)

but just getting you to decide that it's worthwhile asking the question. It feels like an economist's

Charles Isbell (23:37.520)

answer actually. Well, people, okay, first of all, people like would disagree. So let me disagree

Charles Isbell (23:43.040)

slightly, which is, I think everything you said is brilliant, but I tend to believe philosophically

Charles Isbell (23:49.360)

speaking that people are interested underneath it all. And I would say that AI, the, the possibility

Charles Isbell (23:57.600)

that an AI system would show the commonality is incredible. That's a really good starting point.

Charles Isbell (24:02.560)

I would say if you, if on social media, I could discover the common things deep or shallow between

Charles Isbell (24:10.400)

me and a person who there's tension with, I think that my basic human nature would take over from

Charles Isbell (24:17.920)

there. And I think enjoy that commonality. And like, there's something sticky about that, that

Charles Isbell (24:25.280)

my mind will linger on and that person in my mind will become like warmer and warmer. And like, I'll

Charles Isbell (24:31.520)

start to give a feel more and more compassion towards them. I think for majority of the

Charles Isbell (24:35.520)

population, that's true, but that might be, that's a hypothesis. Yeah. I mean, it's an empirical

Charles Isbell (24:40.800)

question, right? You'd have to figure it out. I mean, I want to believe you're right. And so I'm

Charles Isbell (24:44.480)

going to say that I think you're right. Of course, some people come to those things for the purpose

Charles Isbell (24:50.800)

of trolling, right? And it doesn't matter that they're playing a different game. Yeah. But I

Charles Isbell (24:55.520)

don't know. I, you know, my experience is it requires two things. It requires, in fact, maybe

Charles Isbell (25:00.320)

this is really at the end what you're saying. And I, and I do agree with this for sure. So

Charles Isbell (25:06.000)

you, it's hard to hold onto that kind of anger or to hold onto just the desire to humiliate someone

Charles Isbell (25:14.080)

for that long. It's just difficult to do. It takes, it takes a toll on you. But more importantly,

Charles Isbell (25:18.880)

we know this both from people having done studies on it, but also from our own experiences,

Charles Isbell (25:23.520)

that it is much easier to be dismissive of a person if they're not in front of you,

Lex Fridman (25:27.920)

if they're not real, right? So much of the history of the world is about making people other, right?

Lex Fridman (25:35.040)

So if you're social media, if you're on the web, if you're doing whatever in the internet,

Lex Fridman (25:38.880)

but being forced to deal with someone as a person, some equivalent to being in the same room,

Charles Isbell (25:45.920)

makes a huge difference. Cause then you're one, you're forced to deal with their humanity because

Charles Isbell (25:49.360)

it's in front of you. The other is of course that, you know, they might punch you in the face

Charles Isbell (25:52.800)

if you go too far. So, you know, both of those things kind of work together, I think to the,

Charles Isbell (25:56.560)

to the right end. So I think bringing people together is really a kind of substitute for

Charles Isbell (26:03.360)

forcing them to see the humanity in another person and to not be able to treat them as bits,

Charles Isbell (26:07.920)

it's hard to troll someone when you're looking them in the eye. This is very difficult to do.

Charles Isbell (26:12.720)

Agreed. Your broad set of research interests fall under interactive AI, as I mentioned,

Charles Isbell (26:18.960)

which is a fascinating set of ideas and you have some concrete things that you're

Charles Isbell (26:23.760)

particularly interested in, but maybe could you talk about how you think about the field of

Charles Isbell (26:30.000)

interactive artificial intelligence? Sure. So let me say upfront that if you look at,

Charles Isbell (26:34.880)

certainly my early work, but even if you look at most of it, I'm a machine learning guy,

Charles Isbell (26:39.600)

right? I do machine learning. First paper ever published, it was in NIPS. Back then it was

Charles Isbell (26:43.760)

NIPS. Now it's NeurIPS. It's a long story there. Anyway, that's another thing. But so,

Lex Fridman (26:48.160)

so I'm a machine learning guy, right? I believe in data, I believe in statistics and all those

Charles Isbell (26:51.280)

kinds of things. And the reason I'm bringing that up is even though I'm a newfangled statistical

Charles Isbell (26:55.360)

machine learning guy and have been for a very long time, the problem I really care about is AI,

Lex Fridman (27:00.080)

right? I care about artificial intelligence. I care about building some kind of

Charles Isbell (27:04.480)

intelligent artifact. However that gets expressed, that would be at least as intelligent as humans

Lex Fridman (27:12.160)

and as interesting as humans, perhaps in their own way. So that's the deep underlying love and

Charles Isbell (27:18.320)

dream is the bigger AI. Whatever the heck that is. Yeah. The machine learning in some ways is

Charles Isbell (27:24.640)

a means to the end. It is not the end. And I don't understand how one could be intelligent

Charles Isbell (27:29.360)

without learning. So therefore I got to figure out how to do that, right? So that's important.

Lex Fridman (27:32.720)

But machine learning, by the way, is also a tool. I said statistical, because that's what most

Charles Isbell (27:37.200)

people think of themselves as machine learning people. That's how they think. I think Pat Langley

Charles Isbell (27:40.720)

might disagree, or at least 1980s Pat Langley might disagree with what it takes to do machine

Charles Isbell (27:46.240)

learning. But I care about the AI problem, which is why it's interactive AI, not just interactive

Charles Isbell (27:50.480)

ML. I think it's important to understand that there's a longterm goal here, which I will

Charles Isbell (27:55.120)

probably never live to see, but I would love to have been a part of, which is building something

Charles Isbell (27:59.040)

truly intelligent outside of ourselves. Can we take a tiny tangent? Or am I interrupting,

Charles Isbell (28:04.800)

which is, is there something you can say concrete about the mysterious gap between

Charles Isbell (28:12.800)

the subset ML and the bigger AI? What's missing? What do you think? I mean, obviously it's

Charles Isbell (28:19.600)

totally unknown, not totally, but in part unknown at this time, but is it something like with Pat

Lex Fridman (28:25.040)

Langley, is it knowledge, like expert system reasoning type of kind of thing?

Lex Fridman (28:29.680)

So AI is bigger than ML, but ML is bigger than AI. This is kind of the real problem here,

Charles Isbell (28:35.440)

is that they're really overlapping things that are really interested in slightly different problems.

Charles Isbell (28:39.360)

I tend to think of ML, and there are many people out there who are going to be very upset at me

Charles Isbell (28:42.240)

about this, but I tend to think of ML being much more concerned with the engineering of solving a

Charles Isbell (28:45.600)

problem, and AI about the sort of more philosophical goal of true intelligence. And that's the thing

Charles Isbell (28:50.800)

that motivates me, even if I end up finding myself living in this kind of engineering ish space,

Charles Isbell (28:56.000)

I've now made Michael Jordan upset. But to me, they just feel very different. You're just

Charles Isbell (29:02.160)

measuring them differently, your goals of where you're trying to be are somewhat different.

Lex Fridman (29:07.280)

But to me, AI is about trying to build that intelligent thing. And typically, but not always,

Charles Isbell (29:13.760)

for the purpose of understanding ourselves a little bit better. Machine learning is, I think,

Charles Isbell (29:17.920)

trying to solve the problem, whatever that problem is. Now, that's my take. Others, of course,

Charles Isbell (29:22.560)

would disagree. So on that note, so with the interactive AI, do you tend to, in your mind,

Charles Isbell (29:28.000)

visualize AI as a singular system, or is it as a collective huge amount of systems interacting

Charles Isbell (29:33.760)

with each other? Like, is the social interaction of us humans and of AI systems the fundamental

Lex Fridman (29:40.640)

to intelligence? I think, well, it's certainly fundamental to our kind of intelligence, right?

Lex Fridman (29:44.800)

And I actually think it matters quite a bit. So the reason the interactive AI part matters to me

Charles Isbell (29:50.960)

is because I don't, this is going to sound simple, but I don't care whether a tree makes a sound

Charles Isbell (2:00:08.480)

60s civil rights movement as opposed to the 1930s, 1940s? Well, first off, there was a civil

Charles Isbell (2:00:12.400)

rights movement in the 30s and 40s. It just wasn't of the same character or quite as well known. Post

Charles Isbell (2:00:17.440)

World War II, lots of interesting things were happening. It's not as if a switch was turned on

Lex Fridman (2:00:22.560)

and Brown versus the Board of Education or the Montgomery bus boycott. And that's when it

Charles Isbell (2:00:27.840)

happened. These things been building up forever and go all the way back and all the way back and

Charles Isbell (2:00:30.960)

all the way back. And, you know, Harriet Tubman was not born in 1950, right? So, you know, we can

Charles Isbell (2:00:35.120)

take these things. It could have easily happened right after World War II. Yes. I think,

Lex Fridman (2:00:42.640)

and again, I'm not a scholar. I think that the big difference was TV. These things are visible.

Charles Isbell (2:00:50.800)

People can see them. It's hard to avoid, right? Why not James Farmer? Why Martin Luther King? Because

Charles Isbell (2:00:59.200)

one was born 20 years after the other, whatever. I think it turns out that, you know what King's

Charles Isbell (2:01:06.000)

biggest failure was in the early days? It was in Georgia. They were doing the usual thing,

Charles Isbell (2:01:13.120)

trying to integrate. And I forget the guy's name, but you can look this up. But he, a cop,

Charles Isbell (2:01:21.680)

he was a sheriff made a deal with the whole state of Georgia. We're going to take people and we are

Charles Isbell (2:01:26.080)

going to nonviolently put them in trucks. And then we're going to take them and put them in jails

Charles Isbell (2:01:30.800)

very far away from here. And we're going to do that. And we're not going to, there'll be no

Charles Isbell (2:01:35.760)

reason for the press to hang around. And they did that and it worked. And the press left and

Charles Isbell (2:01:41.440)

nothing changed. So next they went to Birmingham, Alabama and Bull O Connor. And you got to see on

Charles Isbell (2:01:48.320)

TV, little boys and girls being hit with fire hoses and being knocked down. And there was

Charles Isbell (2:01:53.920)

outrage and things changed, right? Part of the delusion is pretending that nothing bad is

Charles Isbell (2:01:59.840)

happening that might force you to do something big you don't want to do. But sometimes it gets

Charles Isbell (2:02:03.360)

put in your face and then you kind of can't ignore it. And a large part in my view of what happened

Charles Isbell (2:02:08.640)

right was that it was too public to ignore. Now we created other ways of ignoring it.

Charles Isbell (2:02:14.480)

Lots of change happened in the South, but part of that delusion was that it wasn't going to affect

Charles Isbell (2:02:17.680)

the West or the Northeast. And of course it did. And that caused its own set of problems, which

Charles Isbell (2:02:21.760)

went into the late sixties into the seventies. And, you know, in some ways we're living with

Lex Fridman (2:02:25.200)

that legacy now and so on. So why not what's happening now? Why didn't happen 10 years ago?

Charles Isbell (2:02:32.400)

I think it's people have more voices. There's not just more TV, there's social media. It's very easy

Charles Isbell (2:02:38.000)

for these things to kind of build on themselves and things are just quite visible. And there's

Charles Isbell (2:02:44.960)

demographic change. I mean, the world is changing rapidly, right? And so it's very difficult.

Charles Isbell (2:02:49.040)

You're now seeing people you could have avoided seeing most of your life growing up in a particular

Charles Isbell (2:02:52.560)

time. And it's happening, it's dispersing at a speed that is fast enough to cause

Charles Isbell (2:02:58.720)

concern for some people, but not so fast to cause massive negative reaction. So that's that.

Charles Isbell (2:03:06.000)

On the other hand, and again, that's a massive oversimplification, but I think there's something

Charles Isbell (2:03:10.400)

there anyway, at least something worth exploring. I'm happy to be yelled at by a real historian.

Charles Isbell (2:03:13.920)

Oh yeah. I mean, there's just the obvious thing. I mean, I guess you're implying, but not

Charles Isbell (2:03:19.520)

saying this. I mean, it seemed to have percolated the most with just a single video, for example,

Charles Isbell (2:03:24.480)

the George Floyd video. It's fascinating to think that whatever the mechanisms that put injustice

Charles Isbell (2:03:34.000)

in front of our face, not like directly in front of our face, those mechanisms are the mechanisms

Charles Isbell (2:03:42.080)

of change. Yeah. On the other hand, Rodney King. So no one remembers this. I seem to be the only

Charles Isbell (2:03:46.720)

person who remembers this, but sometime before the Rodney King incident, there was a guy who

Charles Isbell (2:03:51.360)

was a police officer who was saying that things were really bad in Southern California. And he

Charles Isbell (2:03:58.080)

was going to prove it by having some news, some camera people follow him around. And he says,

Charles Isbell (2:04:03.040)

I'm going to go into these towns and just follow me for a week. And you will see that I'll get

Charles Isbell (2:04:06.160)

harassed. And like the first night he goes out there and he crosses into the city, some cops

Charles Isbell (2:04:11.120)

pull him over and he's a police officer. Remember, they don't know that. Of course they like shove

Charles Isbell (2:04:15.840)

his face through a glass window. This was on the new, like I distinctly remember watching this as

Charles Isbell (2:04:20.640)

a kid. Actually, I guess I wasn't a kid. I was in college, I was in grad school at the time.

Lex Fridman (2:04:25.040)

So that's not enough. Well, it disappeared like a day late. It didn't go viral.

Lex Fridman (2:04:30.400)

Yeah. Whatever that is, whatever that magic thing is.

Lex Fridman (2:04:33.920)

And whatever it was in 92, it was harder to go viral in 92, right? Or 91,

Lex Fridman (2:04:38.320)

actually it must've been 90 or 91, but that happened. And like two days later,

Charles Isbell (2:04:42.000)

it's like it never happened. Again, nobody remembers this, but I'm like the only person.

Charles Isbell (2:04:45.280)

Sometimes I think I must've dreamed it. Anyway, Rodney King happens. It goes viral

Charles Isbell (2:04:50.000)

or the moral equivalent thereof at the time. And eventually we get April 29th. And I don't know

Lex Fridman (2:04:57.520)

what the difference was between the two things, other than one thing caught on and one thing

Charles Isbell (2:05:00.640)

didn't. Maybe what's happening now is two things are feeding onto one another. One is more people

Charles Isbell (2:05:07.280)

are willing to believe. And the other is there's easier and easier ways to give evidence. Cameras,

Charles Isbell (2:05:14.560)

body cams or whatever, but we're still finding ourselves telling the same story. It's the same

Charles Isbell (2:05:17.600)

thing over and over again. I would invite you to go back and read the op eds from what people were

Charles Isbell (2:05:22.640)

saying about the violence is not the right answer after Rodney King. And then go back to 1980 and

Charles Isbell (2:05:28.400)

the big riots that were happening around then and read the same op ed. It's the same words over and

Charles Isbell (2:05:33.680)

over and over again. I mean, there's your remembering history right there. I mean,

Charles Isbell (2:05:37.760)

it's like literally the same words. Like it could have just caught, but I'm surprised no one got

Charles Isbell (2:05:40.960)

flagged for plagiarism. It's interesting if you have an opinion on the question of violence

Lex Fridman (2:05:46.560)

and the popular perhaps caricature of Malcolm X versus Martin Luther King.

Charles Isbell (2:05:53.120)

You know, Malcolm X was older than Martin Luther King. People kind of have it in their head that

Charles Isbell (2:05:57.280)

he's younger. Well, he died sooner, but only by a few years. People think of MLK as the older

Charles Isbell (2:06:05.040)

statesman and they think of Malcolm X as the young, angry, whatever, but that's more of a

Charles Isbell (2:06:10.320)

narrative device. It's not true at all. I don't, I just, I reject the choice as I think it's a

Charles Isbell (2:06:18.960)

false choice. I think they're just things that happen. You just do, as I said, hatred is not,

Charles Isbell (2:06:23.360)

it takes a lot of energy, but you know, every once in a while you have to fight.

Charles Isbell (2:06:28.720)

One thing I will say without taking a moral position, which I will not take on this matter,

Lex Fridman (2:06:34.160)

violence has worked.

Charles Isbell (2:06:38.640)

Yeah, that's the annoying thing.

Lex Fridman (2:06:41.360)

That's the annoying thing.

Charles Isbell (2:06:43.600)

It seems like over the top anger works. Outrage works. So you can say like being calm and rational

Lex Fridman (2:06:53.600)

and just talking it out is going to lead to progress. But it seems like if you just look

Charles Isbell (2:06:59.600)

through history being irrationally upset is the way you make progress.

Lex Fridman (2:07:06.640)

Well, it's certainly the way that you get someone to notice you.

Charles Isbell (2:07:09.680)

Yeah.

Lex Fridman (2:07:10.320)

And if they don't notice you, I mean, what's the difference between that and what did you,

Charles Isbell (2:07:13.760)

again, without taking a moral position on this, I'm just trying to observe history here.

Charles Isbell (2:07:17.040)

If you, maybe if television didn't exist, the civil rights movement doesn't happen

Charles Isbell (2:07:22.240)

or it takes longer or it takes a very different form. Maybe if social media doesn't exist,

Charles Isbell (2:07:27.120)

a whole host of things, positive and negative don't happen. And what do any of those things

Charles Isbell (2:07:33.440)

do other than expose things to people? Violence is a way of shouting. I mean,

Charles Isbell (2:07:40.880)

many people far more talented and thoughtful than I have have said this in one form or another,

Charles Isbell (2:07:45.680)

right? That violence is the voice of the unheard. It's a thing that people do when they feel as if

Charles Isbell (2:07:54.160)

they have no other option. And sometimes we agree and sometimes we disagree. Sometimes we think

Charles Isbell (2:08:00.240)

they're justified. Sometimes we think they are not, but regardless, it is a way of shouting.

Lex Fridman (2:08:06.400)

And when you shout, people tend to hear you, even if they don't necessarily hear the words

Charles Isbell (2:08:10.240)

that you're saying, they hear that you were shouting. I see no way. So another way of putting

Charles Isbell (2:08:15.200)

it, which I think is less, let us just say provocative, but I think is true is that all

Charles Isbell (2:08:26.400)

change, particularly change that impacts power requires struggle. The struggle doesn't have to

Charles Isbell (2:08:32.000)

be violent, you know, but it's a struggle nonetheless. The powerful don't give up power

Charles Isbell (2:08:38.880)

easily. I mean, why should they? But even so, it still has to be a struggle. And by the way,

Charles Isbell (2:08:45.840)

this isn't just about, you know, violent political, whatever, nonviolent political

Charles Isbell (2:08:49.360)

change, right? This is true for understanding calculus, right? I mean, everything requires

Lex Fridman (2:08:53.280)

a struggle. We're back to talking about faculty hiring. At the end of the day,

Charles Isbell (2:08:56.560)

in the end of the day, it all comes down to faculty hiring. All a metaphor. Faculty

Charles Isbell (2:09:01.600)

hiring is a metaphor for all of life. Let me ask a strange question. Do you think everything is

Lex Fridman (2:09:10.400)

going to be okay in the next year? Do you have a hope that we're going to be okay?

Charles Isbell (2:09:16.640)

I tend to think that everything's going to be okay because I just tend to think that everything's

Charles Isbell (2:09:20.560)

going to be okay. My mother says something to me a lot and always has, and I find it quite

Charles Isbell (2:09:26.240)

comforting, which is this too shall pass and this too shall pass. Now, this too shall pass is not

Charles Isbell (2:09:32.080)

just this bad thing is going away. Everything passes. I mean, I have a 16 year old daughter

Charles Isbell (2:09:38.480)

who's going to go to college probably at about 15 minutes, given how fast she seems to be growing

Charles Isbell (2:09:43.920)

up. And you know, I get to hang out with her now, but one day I won't. She'll ignore me just as much

Charles Isbell (2:09:48.720)

as I ignored my parents when I was in college and went to grad school. This too shall pass.

Lex Fridman (2:09:52.160)

But I think that one day, if we're all lucky, you live long enough to look back on something that

Charles Isbell (2:09:57.600)

happened a while ago, even if it was painful and mostly it's a memory. So yes, I think it'll be okay.

Lex Fridman (2:10:06.080)

What about humans? Do you think we'll live into the 21st century?

Lex Fridman (2:10:11.360)

I certainly hope so.

Lex Fridman (2:10:12.560)

Are you worried that we might destroy ourselves with nuclear weapons, with AGI, with engineering?

Charles Isbell (2:10:19.920)

I'm not worried about AGI doing it, but I am worried. I mean, at any given moment, right? Also,

Lex Fridman (2:10:24.000)

but you know, at any given moment, a comet could, I mean, you know, whatever. I tend to think that

Charles Isbell (2:10:28.640)

outside of things completely beyond our control, we have a better chance than not of making it.

Charles Isbell (2:10:36.800)

You know, I talked to Alex Filipenko from Berkeley. He was talking about comets and

Charles Isbell (2:10:41.920)

that they can come out of nowhere. And that was a realization to me. Wow. We're just watching

Charles Isbell (2:10:49.040)

this darkness and they can just enter. And then we have less than a month.

Lex Fridman (2:10:53.680)

And yet you make it from day to day.

Charles Isbell (2:10:57.760)

That one shall not pass. Well, maybe for Earth they'll pass, but not for humans.

Lex Fridman (2:11:02.160)

But I'm just choosing to believe that it's going to be okay. And we're not going to get hit by

Charles Isbell (2:11:08.720)

an asteroid, at least not while I'm around. And if we are, well, there's very little I can do about

Charles Isbell (2:11:13.280)

it. So I might as well assume it's not going to happen. It makes food taste better.

Charles Isbell (2:11:17.680)

It makes food taste better.

Lex Fridman (2:11:19.680)

So you, out of the millions of things you've done in your life,

Charles Isbell (2:11:24.240)

you've also began the This Week in Black History calendar of facts.

Charles Isbell (2:11:30.640)

There's like a million questions that can ask here. You said you're not a historian,

Lex Fridman (2:11:35.760)

but is there, let's start at the big history question of, is there somebody in history,

Charles Isbell (2:11:44.240)

in black history that you draw a lot of philosophical or personal inspiration from,

Lex Fridman (2:11:50.960)

or you just find interesting or a moment in history you find interesting?

Charles Isbell (2:11:55.040)

Well, I find the entirety of the 40s to the 60s and the civil rights movement that didn't happen

Lex Fridman (2:12:01.360)

and did happen at the same time during then quite inspirational. I mean, I've read quite a bit of the

Charles Isbell (2:12:07.440)

time period, at least I did in my younger days when I had more time to read as many things as I

Charles Isbell (2:12:12.000)

wanted to. What was quirky about This Week in Black History when I started in the 80s was how

Charles Isbell (2:12:22.000)

focused it was. It was because of the sources I was stealing from. And I was very much stealing

Charles Isbell (2:12:25.280)

from sort of like, I'd take calendars, anything I could find, Google didn't exist, right? And I

Charles Isbell (2:12:29.040)

just pulled as much as I could and just put it together in one place for other people.

Lex Fridman (2:12:32.320)

What ended up being quirky about it, and I started getting people sending me information on it,

Charles Isbell (2:12:36.080)

was the inventors. People who, you know, Gerard Morgan to Benjamin Banneker, right? People who

Lex Fridman (2:12:43.120)

were inventing things. At a time when, how in the world did they manage to invent anything?

Charles Isbell (2:12:54.080)

Like, all these other things were happening, mother necessity, right? All these other things

Charles Isbell (2:12:57.600)

were happening. And, you know, there were so many terrible things happening around them. And, you

Charles Isbell (2:13:00.960)

know, they went to the wrong state at the wrong time. They may never, never come back, but they

Charles Isbell (2:13:04.160)

were inventing things we use, right? And it was always inspiring to me that people would still

Charles Isbell (2:13:10.080)

create even under those circumstances. I got a lot out of that. I also learned a few lessons. I

Charles Isbell (2:13:16.560)

think, you know, the Charles Richard Drews of the world, you know, you create things that impact

Charles Isbell (2:13:23.680)

people. You don't necessarily get credit for them. And that's not right, but it's also okay.

Lex Fridman (2:13:29.280)

TK You okay with that?

Lex Fridman (2:13:31.080)

CK Up to a point, yeah. I mean, look, in our world,

Charles Isbell (2:13:36.880)

all we really have is credit.

Lex Fridman (2:13:38.320)

TK I was always bothered by how much value credit is given.

Charles Isbell (2:13:43.520)

CK That's the only thing you got. I mean, if you're an academic in some sense,

Lex Fridman (2:13:46.960)

well, it isn't the only thing you've got, but it feels that way sometimes.

Charles Isbell (2:13:49.360)

TK But you got the actual, we're all going to be dead soon. You got the joy of having created

Charles Isbell (2:13:56.160)

the, you know, the credit with Jan. I've talked to Jorgen Schmidhuber, right? The Turing Award

Charles Isbell (2:14:05.200)

given to three people for deep learning. And you could say that a lot of other people should be on

Charles Isbell (2:14:10.880)

that list. It's the Nobel Prize question. Yeah, it's sad. It's sad. And people like talking about

Charles Isbell (2:14:16.320)

it. But I feel like in the long arc of history, the only person who will be remembered is Einstein,

Lex Fridman (2:14:22.560)

Hitler, maybe Elon Musk. And the rest of us are just like...

Charles Isbell (2:14:27.040)

CK Well, you know, someone asked me about immortality once and I said,

Lex Fridman (2:14:31.040)

and I stole this from somebody else. I don't remember who, but it was,

Charles Isbell (2:14:34.400)

you know, I asked them, what's your great grandfather's name? Any of them? Of course,

Charles Isbell (2:14:39.920)

they don't know. Most of us do not know. I mean, I'm not entirely sure. I know my grandparents,

Charles Isbell (2:14:44.960)

all my grandparents names. I know what I called them, right? I don't know their middle names,

Charles Isbell (2:14:48.560)

for example. It's within living memory, so I could find out. Actually, my grandfather

Charles Isbell (2:14:54.640)

didn't know when he was born. I had no idea how old he was, right? But I definitely don't know

Charles Isbell (2:15:00.480)

any of my great grandparents are. So in some sense, immortality is doing something preferably

Charles Isbell (2:15:06.400)

positive so that your great grandchildren know who you are, right? And that's kind of what you

Charles Isbell (2:15:11.600)

can hope for, which is very depressing in some ways. I could turn it into something uplifting

Charles Isbell (2:15:16.720)

if you need me to, but it's simple, right? It doesn't matter. I don't have to know my great

Charles Isbell (2:15:23.200)

grandfather was to know that I wouldn't be here without him. And I don't know who my great

Charles Isbell (2:15:28.400)

grandchildren are. Certainly my great, great grandchildren are, and I'll probably never meet

Charles Isbell (2:15:32.240)

them. Although I would very much like to, but hopefully I'll set the world in motion in such

Charles Isbell (2:15:38.080)

a way that their lives will be better than they would have been if I hadn't done that. Well,

Lex Fridman (2:15:41.440)

certainly they wouldn't have existed if I hadn't done the things that I did.

Lex Fridman (2:15:44.320)

So I think that's a good positive thing you live on through other people.

Lex Fridman (2:15:49.280)

Are you afraid of death?

Charles Isbell (2:15:51.200)

I don't know if I'm afraid of death, but I don't like it.

Lex Fridman (2:15:54.960)

That's another t shirt. I mean, do you ponder it? Do you think about the

Charles Isbell (2:16:02.720)

inevitability of oblivion? I do occasionally. This feels like a very rushing conversation.

Charles Isbell (2:16:07.760)

I will tell you a story, something that happened to me recently. If you look very carefully,

Charles Isbell (2:16:14.320)

you will see I have a scar, which by the way, is an interesting story of its own about why people

Charles Isbell (2:16:20.560)

have half of their thyroid taken out. Some people get scars and some don't. But anyway, I had half

Charles Isbell (2:16:26.720)

my thyroid taken out. The way I got there, by the way, is its own interesting story, but I won't go

Charles Isbell (2:16:30.320)

into it. Just suffice it to say, I did what I keep telling people you should never do, which is never

Charles Isbell (2:16:33.760)

go to the doctor unless you have to, because there's nothing good that's ever going to come

Charles Isbell (2:16:36.720)

out of a doctor's visit. So I went to the doctor to look at one thing. It's a little bump I had on

Charles Isbell (2:16:41.440)

the side that I thought might be something bad because my mother made me. And I went there and

Charles Isbell (2:16:45.600)

he's like, oh, it's nothing. But by the way, your thyroid is huge. Can you breathe? Yes,

Charles Isbell (2:16:49.120)

I can breathe. Are you sure? Because it's pushing on your windpipe. You should be dead.

Lex Fridman (2:16:52.560)

So I ended up going there. And to look at my thyroid, it was growing. I had what's called a

Charles Isbell (2:16:59.520)

goiter. And he said, we're going to have to take it out at some point. When? Sometime before you're

Charles Isbell (2:17:03.920)

85, probably. But if you wait till you're 85, that'll be really bad because you don't want to

Charles Isbell (2:17:09.280)

have surgery when you're 85 years old, if you can help it. Certainly not the kind of surgery it

Charles Isbell (2:17:14.240)

takes to take out your thyroid. So I went there and I would decide I would put it off until

Charles Isbell (2:17:21.680)

December 19th because my birthday is December 18th. And I wouldn't be able to say I made it to

Charles Isbell (2:17:26.560)

49 or whatever. So I said, I'll wait till after my birthday. In the first six months of that,

Charles Isbell (2:17:32.640)

nothing changed. Apparently in the next three months, it had grown. I hadn't noticed this at

Charles Isbell (2:17:39.120)

all. I went and had surgery. They took out half of it. The other half is still there and working

Charles Isbell (2:17:44.480)

fine, by the way. I don't have to take a pill or anything like that. It's great. I'm in the

Charles Isbell (2:17:49.520)

hospital room and the doctor comes in. I've got these things in my arm. They're going to do

Charles Isbell (2:17:56.880)

whatever. They're talking to me. And the anesthesiologist says, huh, your blood

Charles Isbell (2:18:00.880)

pressure is through the roof. Do you have high blood pressure? I said, no, but I'm terrified if

Charles Isbell (2:18:04.880)

that helps you at all. And the anesthesist, who's the nurse who supports the anesthesiologist,

Charles Isbell (2:18:11.440)

if I got that right, said, oh, don't worry about it. I've just put some stuff in your IV. You're

Charles Isbell (2:18:15.040)

going to be feeling pretty good in a couple of minutes. And I remember turning and saying,

Charles Isbell (2:18:19.280)

well, I'm going to feel pretty good in a couple of minutes. Next thing I know, there's this guy

Lex Fridman (2:18:23.680)

and he's moving my bed. And he's talking to me and I have this distinct impression that I've met

Charles Isbell (2:18:29.280)

this guy and I should know what he's talking about, but I kind of just don't remember what

Charles Isbell (2:18:35.920)

just happened. And I look up and I see the tiles going by and I'm like, oh, it's just like in the

Charles Isbell (2:18:41.040)

movies where you see the tiles go by. And then I have this brief thought that I'm in an infinitely

Charles Isbell (2:18:47.680)

long warehouse and there's someone sitting next to me. And I remember thinking, oh, she's not

Charles Isbell (2:18:53.280)

talking to me. And then I'm back in the hospital bed. And in between the time where the tiles were

Charles Isbell (2:19:00.960)

going by and I got in the hospital bed, something like five hours had passed. Apparently it had

Charles Isbell (2:19:05.520)

grown so much that it was a four and a half hour procedure instead of an hour long procedure. I

Charles Isbell (2:19:09.760)

lost a neck size and a half. It was pretty big. Apparently it was as big as my heart.

Lex Fridman (2:19:16.560)

Why am I telling you this? I'm telling you this because...

Charles Isbell (2:19:19.120)

It's a hell of a story already. Between tiles going by and me waking up in

Lex Fridman (2:19:25.920)

my hospital bed, no time passed. There was no sensation of time passing.

Charles Isbell (2:19:31.840)

When I go to sleep and I wake up in the morning, I have this feeling that time has passed. This

Charles Isbell (2:19:36.960)

feeling that something has physically changed about me. Nothing happened between the time they

Charles Isbell (2:19:42.320)

put the magic juice in me and the time that I woke up. Nothing. By the way, my wife was there

Charles Isbell (2:19:47.440)

with me talking. Apparently I was also talking. I don't remember any of this, but luckily I didn't

Charles Isbell (2:19:53.120)

say anything I wouldn't normally say. My memory of it is I would talk to her and she would teleport

Charles Isbell (2:19:58.480)

around the room. And then I accused her of witchcraft and that was the end of that.

Charles Isbell (2:20:03.760)

Her point of view is I would start talking and then I would fall asleep and then I would wake

Lex Fridman (2:20:07.760)

up and leave off where I was before. I had no notion of any time passing.

Charles Isbell (2:20:10.880)

I kind of imagine that that's death, is the lack of sensation of time passing. And on the one hand,

Charles Isbell (2:20:20.400)

I am, I don't know, soothed by the idea that I won't notice. On the other hand, I'm very unhappy

Charles Isbell (2:20:28.720)

at the idea that I won't notice. So I don't know if I'm afraid of death, but I'm completely sure

Charles Isbell (2:20:35.680)

that I don't like it and that I particularly would prefer to discover on my own whether immortality

Charles Isbell (2:20:41.520)

sucks and be able to make a decision about it. That's what I would prefer. You like to have a

Charles Isbell (2:20:47.040)

choice in the matter. I would like to have a choice in the matter. Well, again, on the Russian thing,

Charles Isbell (2:20:51.440)

I think the finiteness of it is the thing that gives it a little flavor, a little spice. Well,

Charles Isbell (2:20:57.680)

in reinforcement learning, we believe that. That's why we have discount factors. Otherwise,

Charles Isbell (2:21:00.640)

it doesn't matter what you do. Amen. Well, let me, one last question sticking on the Russian theme.

Charles Isbell (2:21:09.280)

You talked about your great grandparents not remembering their name. What do you think is the,

Lex Fridman (2:21:18.000)

in this kind of Markov chain that is life, what do you think is the meaning of it all?

Charles Isbell (2:21:26.000)

What's the meaning of life? Well, in a world where eventually you won't know who your great

Charles Isbell (2:21:32.000)

grandchildren are, I'm reminded of something I heard once or I read once that I really like,

Charles Isbell (2:21:42.000)

which is, it is well worth remembering that the entire universe, save for one trifling exception,

Charles Isbell (2:21:51.760)

is composed entirely of others. And I think that's the meaning of life.

Charles Isbell (2:22:01.040)

Charles, this is one of the best conversations I've ever had. And I get to see you tomorrow

Charles Isbell (2:22:06.240)

again to hang out with who looks to be one of the most, how should I say, interesting personalities

Charles Isbell (2:22:15.600)

that I'll ever get to meet with Michael Lippmann. So I can't wait. I'm excited to have had this

Charles Isbell (2:22:20.640)

opportunity. Thank you for traveling all the way here. It was amazing. I'm excited. I always love

Charles Isbell (2:22:25.920)

Georgia Tech. I'm excited to see with you being involved there what the future holds. So thank you

Charles Isbell (2:22:30.960)

for talking to me. Thank you for having me. I enjoyed every minute of it. Thanks for listening

Charles Isbell (2:22:34.800)

to this conversation with Charles Isbell and thank you to our sponsors, Neuro, the maker of

Charles Isbell (2:22:40.320)

functional sugar free gum and mints that I used to give my brain a quick caffeine boost, Decoding

Charles Isbell (2:22:46.880)

Digital, a podcast on tech and entrepreneurship that I listen to and enjoy, Masterclass, online

Charles Isbell (2:22:53.360)

courses that I watch from some of the most amazing humans in history, and Cash App, the app I used to

Charles Isbell (2:23:00.080)

send money to friends for food and drinks. Please check out these sponsors in the description to get

Charles Isbell (2:23:06.080)

a discount and to support this podcast. If you enjoy this thing, subscribe on YouTube, review it

Charles Isbell (2:23:11.920)

with Five Stars and Apple Podcast, follow on Spotify, support on Patreon, or connect with me

Charles Isbell (2:23:17.040)

on Twitter at Lex Friedman. And now let me leave you with some poetic words from Martin Luther

Charles Isbell (2:23:23.520)

King Jr. There comes a time when people get tired of being pushed out of the glittering sunlight

Charles Isbell (2:23:30.720)

of life's July and left standing amid the piercing chill of an alpine November.

Lex Fridman (2:23:37.280)

Thank you for listening and hope to see you next time.

Charles Isbell (30:00.000)

when it falls and there's no one around, because I don't think it matters, right? If there's no

Charles Isbell (30:04.080)

observer in some sense. And I think what's interesting about the way that we're intelligent

Charles Isbell (30:08.480)

is we're intelligent with other people, right? Or other things anyway. And we go out of our way to

Charles Isbell (30:13.760)

make other things intelligent. We're hardwired to find intention, even whether there is no intention,

Lex Fridman (30:18.960)

why we anthropomorphize everything. I think the interactive AI part is being intelligent in and

Charles Isbell (30:25.360)

of myself in isolation is a meaningless act in some sense. The correct answer is you have to

Charles Isbell (30:31.760)

be intelligent in the way that you interact with others. It's also efficient because it allows you

Charles Isbell (30:35.360)

to learn faster because you can import from past history. It also allows you to be efficient in

Charles Isbell (30:41.120)

the transmission of that. So we ask ourselves about me. Am I intelligent? Clearly, I think so.

Lex Fridman (30:47.520)

But I'm also intelligent as a part of a larger species and group of people, and we're trying to

Charles Isbell (30:51.680)

move the species forward as well. And so I think that notion of being intelligent with others is

Charles Isbell (30:56.960)

kind of the key thing because otherwise you come and you go, and then it doesn't matter. And so

Charles Isbell (31:01.920)

that's why I care about that aspect of it. And it has lots of other implications. One is not just

Charles Isbell (31:08.720)

building something intelligent with others, but understanding that you can't always communicate

Charles Isbell (31:12.240)

with those others. They have been in a room where there's a clock on the wall that you haven't seen,

Charles Isbell (31:16.240)

which means you have to spend an enormous amount of time communicating with one another constantly

Charles Isbell (31:20.720)

in order to figure out what each other wants. So this is why people project, right? You project your

Charles Isbell (31:26.400)

own intentions and your own reasons for doing things on the others as a way of understanding

Charles Isbell (31:30.800)

them so that you know how to behave. But by the way, you, completely predictable person,

Charles Isbell (31:35.840)

I don't know how you're predictable. I don't know you well enough, but you probably eat the same five

Charles Isbell (31:39.280)

things over and over again or whatever it is that you do, right? I know I do. If I'm going to a new

Charles Isbell (31:43.920)

Chinese restaurant, I will get general gals chicken because that's the thing that's easy.

Charles Isbell (31:47.600)

I will get hot and sour soup. People do the things that they do, but other people get the chicken and

Charles Isbell (31:52.560)

broccoli. I can push this analogy way too far. The chicken and broccoli. I don't know what's

Charles Isbell (31:56.960)

wrong with those people. I don't know what's wrong with them either. We have all had our trauma.

Lex Fridman (32:02.320)

So they get their chicken and broccoli and their egg drop soup or whatever. We got to communicate and

Charles Isbell (32:06.880)

it's going to change, right? So interactive AI is not just about learning to solve a problem or a

Charles Isbell (32:13.200)

task. It's about having to adapt that over time, over a very long period of time and interacting

Charles Isbell (32:18.080)

with other people who will themselves change. This is what we mean about things like adaptable

Charles Isbell (32:22.400)

models, right? That you have to have a model. That model is going to change. And by the way,

Charles Isbell (32:25.600)

it's not just the case that you're different from that person, but you're different from the person

Charles Isbell (32:28.960)

you were 15 minutes ago or certainly 15 years ago. And I have to assume that you're at least going

Charles Isbell (32:33.760)

to drift, hopefully not too many discontinuities, but you're going to drift over time. And I have

Charles Isbell (32:38.480)

to have some mechanism for adapting to that as you and an individual over time and across individuals

Charles Isbell (32:45.120)

over time. On the topic of adaptive modeling and you talk about lifelong learning, which is

Charles Isbell (32:51.200)

a, I think a topic that's understudied or maybe because nobody knows what to do with it. But like,

Charles Isbell (32:59.760)

you know, if you look at Alexa or most of our artificial intelligence systems that are primarily

Charles Isbell (33:04.880)

machine learning based systems or dialogue systems, all those kinds of things, they know very little

Charles Isbell (33:10.960)

about you in the sense of the lifelong learning sense that we learn as humans, we learn a lot about

Charles Isbell (33:21.360)

each other, not in the quantity effects, but like the temporally rich set of information that seems

Charles Isbell (33:31.280)

to like pick up the crumbs along the way that somehow seems to capture a person pretty well.

Lex Fridman (33:36.880)

Do you have any ideas how to do lifelong learning? Because it seems like most of the machine learning

Charles Isbell (33:44.000)

community does not. No, well, by the way, not only does the machine learning community not spend a

Charles Isbell (33:48.320)

lot of time on lifelong learning, I don't think they spend a lot of time on learning period in

Charles Isbell (33:52.720)

the sense that they tend to be very task focused. Everybody is overfitting to whatever problem is

Charles Isbell (33:57.520)

they happen to have. They're overengineering their solutions to the task. Even the people,

Lex Fridman (34:01.920)

and I think these people too, are trying to solve a hard problem of transfer learning, right? I'm

Charles Isbell (34:06.000)

going to learn on one task and learn another task. You still end up creating the task. It's like

Charles Isbell (34:10.000)

looking for your keys where the light is because that's where the light is, right? It's not because

Charles Isbell (34:13.280)

the keys have to be there. I mean, one could argue that we tend to do this in general. As a group,

Charles Isbell (34:20.400)

we tend to hill climb and get stuck in local optima. I think we do this in the small as well.

Charles Isbell (34:26.160)

I think it's very hard to do. Here's the hard thing about AI. The hard thing about AI is it

Charles Isbell (34:32.960)

keeps changing on us, right? What is AI? AI is the art and science of making computers act the way

Charles Isbell (34:38.480)

they do in the movies, right? That's what it is, right? But beyond that. They keep coming up with

Charles Isbell (34:44.560)

new movies. Yes. Right, exactly. We are driven by this kind of need to the ineffable quality of who

Charles Isbell (34:52.080)

we are, which means that the moment you understand something is no longer AI, right? Well, we

Charles Isbell (34:57.360)

understand this. That's just you take the derivative and you divide by two and then you average it out

Charles Isbell (35:01.360)

over time in the window. Therefore, that's no longer AI. The problem is unsolvable because

Charles Isbell (35:05.200)

it keeps kind of going away. This creates a kind of illusion, which I don't think is an entire

Charles Isbell (35:08.720)

illusion, of either there's very simple task based things you can do very well and over engineer,

Charles Isbell (35:13.600)

there's all of AI, and there's nothing in the middle. It's very hard to get from here to here,

Lex Fridman (35:18.640)

and it's very hard to see how to get from here to here. I don't think that we've done

Charles Isbell (35:23.040)

a very good job of it because we get stuck trying to solve the small problems in front of it,

Charles Isbell (35:27.360)

myself included. I'm not going to pretend that I'm better at this than anyone else. Of course,

Charles Isbell (35:31.680)

all the incentives in academia and in industry are set to make that very hard because you have

Charles Isbell (35:37.840)

to get the next paper out, you have to get the next product out, you have to solve this problem,

Lex Fridman (35:41.680)

and it's very sort of naturally incremental. None of the incentives are set up to allow you to take

Charles Isbell (35:47.600)

a huge risk unless you're already so well established you can take that big risk.

Charles Isbell (35:53.600)

If you're that well established that you can take that big risk, then you've probably spent

Charles Isbell (35:57.040)

much of your career taking these little risks, relatively speaking, and so you have got a

Charles Isbell (36:01.680)

lifetime of experience telling you not to take that particular big risk. So the whole system's

Charles Isbell (36:05.520)

set up to make progress very slow. That's fine. It's just the way it is, but it does make this

Charles Isbell (36:10.080)

gap seem really big, which is my long way of saying I don't have a great answer to it except

Charles Isbell (36:14.400)

that stop doing n equals one. At least try to get n equal two and maybe n equal seven so that you

Charles Isbell (36:21.120)

can say I'm going to, or maybe t is a better variable here. I'm going to not just solve this

Charles Isbell (36:25.120)

problem and solve this problem and another problem. I'm not going to learn just on you.

Charles Isbell (36:28.240)

I'm going to keep living out there in the world and just seeing what happens and that we'll learn

Charles Isbell (36:32.080)

something as designers and our machine learning algorithms and our AI algorithms can learn as

Charles Isbell (36:36.800)

well. But unless you're willing to build a system which you're going to have live for months at a

Charles Isbell (36:41.360)

time in an environment that is messy and chaotic, you cannot control, then you're never going to

Charles Isbell (36:47.120)

make progress in that direction. So I guess my answer to you is yes. My idea is that you should,

Charles Isbell (36:51.440)

it's not no, it's yes. You should be deploying these things and making them live for a month

Charles Isbell (36:57.760)

at a time and be okay with the fact that it's going to take you five years to do this. Not

Charles Isbell (37:02.000)

rerunning the same experiment over and over again and refining the machine so it's slightly better

Charles Isbell (37:06.320)

at whatever, but actually having it out there and living in the chaos of the world and seeing what

Charles Isbell (37:12.320)

its learning algorithm say can learn, what data structure it can build and how it can go from

Charles Isbell (37:16.800)

there. Without that, you're going to be stuck all the time. What do you think about the possibility

Charles Isbell (37:22.000)

of N equals one growing, it's probably crude approximation, but growing like if you look at

Charles Isbell (37:28.720)

language models like GPT3, if you just make it big enough, it'll swallow the world. Meaning like

Charles Isbell (37:35.040)

it'll solve all your T to infinity by just growing in size of this, taking the small overengineered

Charles Isbell (37:43.120)

solution and just pumping it full of steroids in terms of compute, in terms of size of training

Charles Isbell (37:49.040)

data and the Yann LeCun style self supervised or open AI self supervised. Just throw all of YouTube

Charles Isbell (37:56.720)

at it and it will learn how to reason, how to paint, how to create music, how to love all that

Charles Isbell (38:04.640)

by watching YouTube videos. I mean, I can't think of a more terrifying world to live in than a world

Charles Isbell (38:08.960)

that is based on YouTube videos, but yeah, I think the answer that I just kind of don't think that'll

Charles Isbell (38:14.960)

quite well, it won't work that easily. You will get somewhere and you will learn something, which

Charles Isbell (38:20.320)

means it's probably worth it, but you won't get there. You won't solve the problem. Here's the

Charles Isbell (38:25.520)

thing, we build these things and we say we want them to learn, but what actually happens, and

Charles Isbell (38:31.600)

let's say they do learn, I mean, certainly every paper I've gotten published, the things learn,

Charles Isbell (38:35.600)

I don't know about anyone else, but they actually change us, right? We react to it differently,

Charles Isbell (38:40.960)

right? So we keep redefining what it means to be successful, both in the negative in the case,

Lex Fridman (38:45.600)

but also in the positive in that, oh, well, this is an accomplishment. I'll give you an example,

Charles Isbell (38:51.840)

which is like the one you just described with YouTube. Let's get completely out of machine

Charles Isbell (38:55.040)

learning. Well, not completely, but mostly out of machine learning. Think about Google.

Charles Isbell (38:59.360)

People were trying to solve information retrieval, the ad hoc information retrieval

Charles Isbell (39:03.280)

problem forever. I mean, first major book I ever read about it was what, 71, I think it was when

Charles Isbell (39:09.280)

it came out. Anyway, we'll treat everything as a vector and we'll do these vector space models

Lex Fridman (39:14.640)

and whatever. And that was all great. And we made very little progress. I mean, we made some progress

Lex Fridman (39:20.160)

and then Google comes and makes the ad hoc problem seem pretty easy. I mean, it's not,

Charles Isbell (39:24.960)

there's lots of computers and databases involved, and there's some brilliant algorithmic stuff

Lex Fridman (39:29.920)

behind it too, and some systems building. But the problem changed, right?

Charles Isbell (39:37.520)

If you've got a world that's that connected so that you have, you know, there are 10 million

Charles Isbell (39:42.640)

answers quite literally to the question that you're asking, then the problem wasn't give me

Charles Isbell (39:48.960)

the things that are relevant. The problem is don't give me anything that's irrelevant, at least in

Charles Isbell (39:52.800)

the first page, because nothing else matters. So Google is not solving the information retrieval

Charles Isbell (39:59.120)

problem, at least not on this webpage. Google is minimizing false positives, which is not the same

Charles Isbell (40:06.560)

thing as getting an answer. It turns out it's good enough for what it is we want to use Google for,

Lex Fridman (40:11.120)

but it also changes what the problem was we thought we were trying to solve in the first place.

Charles Isbell (40:15.440)

You thought you were trying to find an answer, but you're not, or you're trying to find the answer,

Lex Fridman (40:18.960)

but it turns out you're just trying to find an answer. Now, yes, it is true. It's also very good

Charles Isbell (40:22.480)

at finding you exactly that webpage. Of course, you trained yourself to figure out what the keywords

Charles Isbell (40:26.480)

were to get you that webpage. But in the end, by having that much data, you've just changed the

Charles Isbell (40:32.640)

problem into something else. You haven't actually learned what you set out to learn. Now, the

Charles Isbell (40:35.840)

counter to that would be maybe we're not doing that either. We just think we are because, you know,

Charles Isbell (40:40.480)

we're in our own heads. Maybe we're learning the wrong problem in the first place, but I don't

Charles Isbell (40:44.720)

think that matters. I think the point is that Google has not solved information retrieval.

Charles Isbell (40:49.360)

Google has done amazing service. I have nothing bad to say about what they've done. Lord knows

Charles Isbell (40:53.040)

my entire life is better because Google exists. For Google Maps, I don't think I've ever found

Charles Isbell (40:57.440)

this place. Where is this? I see 110 and I see where did 95 go? So I'm very grateful for Google,

Lex Fridman (41:07.360)

but they just have to make certain the first five things are right.

Lex Fridman (41:11.040)

And everything after that is wrong. Look, we're going off on a totally different topic here, but

Lex Fridman (41:17.600)

think about the way we hire faculty. It's exactly the same thing.

Charles Isbell (41:21.200)

I get in controversial, getting controversial. It's exactly the same problem, right? It's

Charles Isbell (41:27.520)

minimizing false positives. We say things like we want to find the best person to be an assistant

Charles Isbell (41:34.880)

professor at MIT in the new college of computing, which I will point out was founded 30 years after

Charles Isbell (41:42.160)

the college of computing. I'm a part of both of my alma mater. I'm just saying I appreciate all

Charles Isbell (41:49.120)

that they did and all that they're doing. Anyway, so we're going to try to hire the best professor.

Lex Fridman (41:57.760)

That's what we say, the best person for this job, but that's not what we do at all, right?

Lex Fridman (42:02.320)

Do you know which percentage of faculty in the top four earn their PhDs from the top four,

Lex Fridman (42:08.720)

say in 2017, which is the most recent year for which I have data?

Charles Isbell (42:12.720)

Maybe a large percentage.

Lex Fridman (42:14.080)

About 60%.

Charles Isbell (42:15.040)

60.

Charles Isbell (42:15.600)

60% of the faculty in the top four earn their PhDs in the top four. This is computer science,

Charles Isbell (42:19.920)

for which there is no top five. There's only a top four, right? Because they're all tied for one.

Charles Isbell (42:23.280)

For people who don't know, by the way, that would be MIT Stanford, Berkeley, CMU.

Charles Isbell (42:27.040)

Yep.

Lex Fridman (42:29.520)

Georgia Tech is number eight.

Charles Isbell (42:31.440)

Number eight. You're keeping track.

Charles Isbell (42:34.080)

Oh yes. It's a large part of my job. Number five is Illinois. Number six is a tie with

Charles Isbell (42:39.280)

UW and Cornell and Princeton and Georgia Tech are tied for eight and UT Austin is number 10.

Charles Isbell (42:43.760)

Michigan is number 11, by the way. So if you look at the top 10, you know what percentage of

Charles Isbell (42:50.320)

faculty in the top 10 earn their PhDs from the top 10? 65, roughly. 65%.

Charles Isbell (42:56.960)

If you look at the top 55 ranked departments, 50% of the faculty earn their PhDs from the top 10.

Charles Isbell (43:04.160)

There is no universe in which all the best faculty, even just for R1 universities,

Charles Isbell (43:11.360)

the majority of them come from 10 places. There's no way that's true, especially when you consider

Lex Fridman (43:16.160)

how small some of those universities are in terms of the number of PhDs they produce.

Lex Fridman (43:20.000)

Yeah.

Charles Isbell (43:20.480)

Now that's not a negative. I mean, it is a negative. It also has a habit of entrenching

Charles Isbell (43:24.880)

certain historical inequities and accidents. But what it tells you is, well, ask yourself the

Charles Isbell (43:32.400)

question, why is it like that? Well, because it's easier. If we go all the way back to the 1980s,

Charles Isbell (43:40.960)

you know, there was a saying that nobody ever lost his job buying a computer from IBM,

Lex Fridman (43:45.440)

and it was true. And nobody ever lost their job hiring a PhD from MIT, right? If the person turned

Lex Fridman (43:52.480)

out to be terrible, well, you know, they came from MIT, what did you expect me to know?

Charles Isbell (43:55.920)

However, that same person coming from pick whichever is your least favorite place that

Charles Isbell (44:01.840)

produces PhDs in, say, computer science, well, you took a risk, right? So all the incentives,

Charles Isbell (44:07.280)

particularly because you're only going to hire one this year, well, now we're hiring 10,

Lex Fridman (44:10.320)

but you know, you're only going to have one or two or three this year. And by the way,

Charles Isbell (44:13.360)

when they come in, you're stuck with them for at least seven years at most places,

Charles Isbell (44:16.640)

because that's before you know whether they're getting tenure or not. And if they get tenure,

Charles Isbell (44:19.360)

stuck with them for a good 30 years, unless they decide to leave. That means the pressure to get

Charles Isbell (44:22.880)

this right is very high. So what are you going to do? You're going to minimize false positives.

Charles Isbell (44:27.680)

You don't care about saying no inappropriately. You only care about saying yes inappropriately.

Lex Fridman (44:33.120)

So all the pressure drives you into that particular direction. Google,

Charles Isbell (44:36.640)

not to put too fine a point on it, was in exactly the same situation with their search. It turns out

Charles Isbell (44:41.440)

you just don't want to give people the wrong page in the first three or four pages. And if there's

Charles Isbell (44:46.400)

10 million right answers and 100 bazillion wrong answers, just make certain the wrong answers

Charles Isbell (44:50.400)

don't get up there. And who cares if you, the right answer was actually the 13th page. A right

Lex Fridman (44:55.200)

answer, a satisficing answer is number one, two, three, or four. So who cares?

Charles Isbell (44:59.200)

Or an answer that will make you discover something beautiful, profound to your question.

Lex Fridman (45:04.160)

Well, that's a different problem, right?

Lex Fridman (45:06.800)

But isn't that the problem? Can we linger on this topic without sort of walking with grace?

Lex Fridman (45:15.040)

How do we get for hiring faculty, how do we get that 13th page with a truly special person? Like

Charles Isbell (45:25.600)

there's, I mean, it depends on the department. Computer science probably has those department,

Charles Isbell (45:29.200)

those kinds of people. Like you have the Russian guy, Grigory Perlman, like just these awkward,

Charles Isbell (45:38.800)

strange minds that don't know how to play the little game of etiquette that faculty have all

Charles Isbell (45:46.640)

agreed somehow like converged over the decades, how to play with each other. And also is not,

Charles Isbell (45:51.920)

you know, on top of that is not from the top four, top whatever numbers, the schools. And maybe

Charles Isbell (45:58.640)

actually just says a few every once in a while to the traditions of old within the computer science

Charles Isbell (46:05.520)

community. Maybe talks trash about machine learning is a total waste of time. And that's

Lex Fridman (46:11.520)

there on their resume. So like how do you allow the system to give those folks a chance?

Charles Isbell (46:19.440)

Well, you have to be willing to take a certain kind of, without taking a particular position

Charles Isbell (46:22.080)

on any particular person, you'd have to take, you have to be willing to take risk, right? A small

Lex Fridman (46:26.800)

amount of it. I mean, if we were treating this as a, well, as a machine learning problem, right?

Charles Isbell (46:31.280)

There's a search problem, which is what it is. It's a search problem. If we were treating it that

Charles Isbell (46:34.080)

way, you would say, oh, well, the main thing is you want, you know, you've got a prior,

Charles Isbell (46:37.840)

you want some data because I'm Bayesian. If you don't want to do it that way,

Charles Isbell (46:41.120)

we'll just inject some randomness in and it'll be okay. The problem is that feels very,

Charles Isbell (46:46.800)

very hard to do with people. All the incentives are wrong there. But it turns out, and let's say,

Charles Isbell (46:53.600)

let's say that's the right answer. Let's just give for the sake of argument that, you know,

Charles Isbell (46:57.040)

injecting randomness in the system at that level for who you hire is just not worth doing because

Charles Isbell (47:02.560)

the price is too high or the cost is too high. If we had infinite resources, sure, but we don't.

Lex Fridman (47:05.600)

And also you've got to teach people. So, you know, you're ruining other people's lives if you get it

Lex Fridman (47:09.120)

too wrong. But we've taken that principle, even if I grant it and pushed it all the way back, right?

Charles Isbell (47:17.120)

So, we could have a better pool than we have of people we look at and give an opportunity to.

Charles Isbell (47:25.840)

If we do that, then we have a better chance of finding that. Of course, that just pushes the

Charles Isbell (47:29.120)

problem back another level. But let me tell you something else. You know, I did a sort of study,

Charles Isbell (47:34.320)

I call it a study. I called up eight of my friends and asked them for all of their data for

Charles Isbell (47:37.360)

graduate admissions. But then someone else followed up and did an actual study. And it

Charles Isbell (47:41.440)

turns out that I can tell you how everybody gets into grad school more or less, more or less.

Charles Isbell (47:47.920)

You basically admit everyone from places higher ranked than you. You admit most people from

Charles Isbell (47:51.520)

places ranked around you. And you meant almost no one from places ranked below you, with the

Charles Isbell (47:54.800)

exception of the small liberal arts colleges that aren't ranked at all, like Harvey Mudd,

Charles Isbell (47:58.000)

because they don't have a PhD, so they aren't ranked. This is all CS. Which means the decision

Charles Isbell (48:04.000)

of whether you become a professor at Cornell was determined when you were 17, by what you knew to

Charles Isbell (48:13.520)

go to undergrad to do whatever. So if we can push these things back a little bit and just make the

Charles Isbell (48:18.400)

pool a little bit bigger, at least you raise the probability that you will be able to see someone

Charles Isbell (48:22.560)

interesting and take the risk. The other answer to that question, by the way, which you could argue

Charles Isbell (48:29.120)

is the same as you either adjust the pool so the probabilities go up, that's a way of injecting a

Charles Isbell (48:34.080)

little bit of uniform noise in the system, as it were, is you change your loss function.

Charles Isbell (48:40.080)

You just let yourself be measured by something other than whatever it is that we're measuring

Charles Isbell (48:44.640)

ourselves by now. I mean, US News and World Report, every time they change their formula

Charles Isbell (48:50.400)

for determining rankings, move entire universities to behave differently, because rankings matter.

Charles Isbell (48:55.760)

TITO Can you talk trash about those rankings for a second? No, I'm joking about talking trash.

Charles Isbell (49:01.200)

I actually, it's so funny how from my perspective, from a very shallow perspective,

Lex Fridman (49:07.280)

how dogmatic, like how much I trust those rankings. They're almost ingrained in my head.

Charles Isbell (49:13.920)

I mean, at MIT, everybody kind of, it's a propagated, mutually agreed upon

Charles Isbell (49:21.600)

TITO idea that those rankings matter. And I don't think anyone knows what they're,

Charles Isbell (49:26.960)

like, most people don't know what they're based on. And what are they exactly based on? And what

Charles Isbell (49:33.280)

are the flaws in that? TITO Well, so it depends on which rankings you're talking about. Do you

Charles Isbell (49:38.160)

want to talk about computer science or talk about universities? TITO Computer science, US News,

Charles Isbell (49:42.320)

isn't that the main one? TITO Yeah, the only one that matters is US News. Nothing else matters.

Charles Isbell (49:46.400)

Sorry, CSRankings.org, but nothing else matters but US News. So US News has formula that it uses

Charles Isbell (49:52.960)

for many things, but not for computer science because computer science is considered a science,

Charles Isbell (49:57.440)

which is absurd. So the rankings for computer science is 100% reputation. So two people

Lex Fridman (50:06.640)

at each department, it's not really department, whatever, at each department,

Charles Isbell (50:11.040)

basically rank everybody. Slightly more complicated than that, but whatever, they rank everyone. And

Lex Fridman (50:16.880)

then those things are put together and somehow. TITO So that means how do you improve reputation?

Lex Fridman (50:22.000)

How do you move up and down the space of reputation? TITO Yes, that's exactly the

Charles Isbell (50:26.960)

question. TITO Twitter? TITO It can help. I can tell you how Georgia Tech did it,

Charles Isbell (50:31.760)

or at least how I think Georgia Tech did it, because Georgia Tech is actually the case to

Charles Isbell (50:35.520)

look at. Not just because I'm at Georgia Tech, but because Georgia Tech is the only computing unit

Charles Isbell (50:40.240)

that was not in the top 20 that has made it into the top 10. It's also the only one in the last

Charles Isbell (50:45.280)

two decades, I think, that moved up in the top 10, as opposed to having someone else moved down.

Lex Fridman (50:53.040)

So we used to be number 10, and then we became number nine because UT Austin went down slightly,

Lex Fridman (50:58.320)

and now we were tied for ninth because that's how rankings work. And we moved from nine to eight

Charles Isbell (51:02.960)

because our raw score moved up a point. So something about Georgia Tech, computer science,

Charles Isbell (51:10.320)

or computing anyway. I think it's because we have shown leadership at every crisis level, right? So

Charles Isbell (51:17.120)

we created a college, first public university to do it, second college, second university to do it

Charles Isbell (51:21.040)

after CMU is number one. I also think it's no accident that CMU is the largest and we're,

Charles Isbell (51:26.400)

depending upon how you count and depending on exactly where MIT ends up with its final college

Charles Isbell (51:30.320)

of computing, second or third largest. I don't think that's an accident. We've been doing this

Charles Isbell (51:33.360)

for a long time. But in the 2000s when there was a crisis about undergraduate education,

Charles Isbell (51:39.440)

Georgia Tech took a big risk and succeeded at rethinking undergrad education and computing.

Charles Isbell (51:45.040)

I think we created these schools at a time when most public universities anyway were afraid to

Charles Isbell (51:49.920)

do it. We did the online masters and that mattered because people were trying to figure out what to

Charles Isbell (51:56.000)

do with MOOCs and so on. I think it's about being observed by your peers and having an impact. So,

Charles Isbell (52:02.400)

I mean, that is what reputation is, right? So the way you move up in the reputation rankings is by

Charles Isbell (52:07.520)

doing something that makes people turn and look at you and say, that's good. They're better than I

Charles Isbell (52:12.320)

thought. Beyond that, it's just inertia and there's huge hysteresis in the system, right? I mean,

Charles Isbell (52:17.440)

there was these, I can't remember this, this may be apocryphal, but there's a major or department

Charles Isbell (52:23.120)

that MIT was ranked number one in and they didn't have it. It's just about what you... I don't know

Charles Isbell (52:28.640)

if that's true, but someone said that to me anyway. But it's a thing, right? It's all about

Charles Isbell (52:33.360)

reputation. Of course, MIT is great because MIT is great. It's always been great. By the way,

Charles Isbell (52:37.280)

because MIT is great, the best students come, which keeps it being great. I mean,

Charles Isbell (52:42.960)

it's just a positive feedback loop. It's not surprising. I don't think it's wrong.

Charles Isbell (52:46.560)

Yeah. But it's almost like a narrative. It doesn't actually have to be backed by reality. Not to

Charles Isbell (52:53.360)

say anything about MIT, but it does feel like we're playing in the space of narratives, not the

Charles Isbell (53:01.040)

space of something grounded. One of the surprising things when I showed up at MIT and just all the

Charles Isbell (53:06.880)

students I've worked with and all the research I've done is they're the same people as I've met

Charles Isbell (53:14.880)

at other places. I mean, what MIT has going for it... Well, MIT has many things going for it. One

Lex Fridman (53:21.520)

of the things MIT has going for it is... Nice logo?

Charles Isbell (53:23.840)

Is nice logo. It's a lot better than it was when I was here. Nice colors too. Terrible,

Charles Isbell (53:30.080)

terrible name for a mascot. But the thing that MIT has going for it is it really does get the

Charles Isbell (53:35.760)

best students. It just doesn't get all of the best students. There are many more best students out

Charles Isbell (53:40.400)

there. And the best students want to be here because it's the best place to be or one of the

Charles Isbell (53:45.040)

best places to be. And it's a sort of positive feedback. But you said something earlier,

Charles Isbell (53:50.240)

which I think is worth examining for a moment. I forget the word you used. You said,

Charles Isbell (53:56.240)

we're living in the space of narrative as opposed to something objective.

Charles Isbell (54:00.000)

Narrative is objective. I mean, one could argue that the only thing that we do as humans is

Charles Isbell (54:04.320)

narrative. We just build stories to explain why we do what we do. Someone once asked me,

Lex Fridman (54:08.720)

but wait, there's nothing objective. No, it's completely an objective measure.

Charles Isbell (54:12.800)

It's an objective measure of the opinions of everybody else. Now, is that physics? I don't

Charles Isbell (54:19.600)

know. Tell me something you think is actually objective and measurable in a way that makes

Charles Isbell (54:25.280)

sense. Cameras, they don't... You're getting me off on something here, but do you know that

Charles Isbell (54:31.840)

cameras, which are just reflecting light and putting them on film, did not work for dark

Charles Isbell (54:39.440)

skin people until the 1970s? You know why? Because you were building cameras for the people who were

Charles Isbell (54:45.840)

going to buy cameras, who all, at least in the United States and Western Europe, were relatively

Charles Isbell (54:51.200)

light skin. Turns out it took terrible pictures of people who look like me. That got fixed with

Charles Isbell (54:56.560)

better film and whole processes. Do you know why? Because furniture manufacturers wanted to be able

Charles Isbell (55:03.440)

to take pictures of mahogany furniture, right? Because candy manufacturers wanted to be able

Charles Isbell (55:10.080)

to take pictures of chocolate. Now, the reason I bring that up is because you might think that

Charles Isbell (55:16.480)

cameras are objective. They're just capturing light. No, they're doing the things that they're

Charles Isbell (55:23.200)

doing based upon decisions by real human beings to privilege, if I may use that word, some physics

Charles Isbell (55:29.200)

over others, because it's an engineering problem. There are tradeoffs, right? So I can either worry

Charles Isbell (55:33.280)

about this part of the spectrum or this part of the spectrum. This costs more. That costs less.

Charles Isbell (55:37.200)

This costs the same, but I have more people paying money over here, right? And it turns out that

Charles Isbell (55:41.600)

if a giant conglomerate demands that you do something different and it's going to involve

Charles Isbell (55:46.640)

all kinds of money for you, suddenly the tradeoffs change, right? And so there you go. I actually

Charles Isbell (55:51.200)

don't know how I ended up there. Oh, it's because of this notion of objectiveness, right? So even

Charles Isbell (55:55.760)

the objective isn't objective because at the end you've got to tell a story. You've got to make

Charles Isbell (55:58.800)

decisions. You've got to make tradeoff. What else is engineering other than that? So I think that

Charles Isbell (56:03.120)

the rankings capture something. They just don't necessarily capture what people assume they

Charles Isbell (56:09.120)

capture. You know, just to linger on this idea, why is there not more people who just like play

Charles Isbell (56:19.920)

with whatever that narrative is, have fun with it, have like excite the world, whether it's in

Charles Isbell (56:24.400)

the Carl Sagan style of like that calm, sexy voice of explaining the stars and all the romantic stuff

Charles Isbell (56:31.600)

or the Elon Musk, dare I even say Donald Trump, where you're like trolling and shaking up the

Charles Isbell (56:37.040)

system and just saying controversial things. I talked to Lisa Feldman Barrett, who's a

Charles Isbell (56:43.920)

neuroscientist who just enjoys playing the controversy. Things like finds the counter

Charles Isbell (56:50.320)

intuitive ideas in the particular science and throws them out there and sees how they play in

Charles Isbell (56:56.080)

the public discourse. Like why don't we see more of that? And why doesn't academia attract an Elon

Charles Isbell (57:02.400)

Musk type? Well, tenure is a powerful thing that allows you to do whatever you want, but getting

Charles Isbell (57:08.480)

tenure typically requires you to be relatively narrow, right? Because people are judging you.

Charles Isbell (57:14.240)

Well, I think the answer is we have told ourselves a story, a narrative that is vulgar,

Lex Fridman (57:22.000)

what you just described is vulgar. It's certainly unscientific, right? And it is easy to convince

Charles Isbell (57:30.880)

yourself that in some ways you're the mathematician, right? The fewer there are in your major,

Charles Isbell (57:37.360)

the more that proves your purity, right? So once you tell yourself that story, then it is

Charles Isbell (57:46.240)

beneath you to do that kind of thing, right? I think that's wrong. I think that, and by the way,

Charles Isbell (57:51.840)

everyone doesn't have to do this. Everyone's not good at it. And everyone, even if they would be

Charles Isbell (57:54.320)

good at it, would enjoy it. So it's fine. But I do think you need some diversity in the way that

Charles Isbell (58:00.080)

people choose to relate to the world as academics, because I think the great universities are ones

Charles Isbell (58:09.440)

that engage with the rest of the world. It is a home for public intellectuals. And in 2020,

Charles Isbell (58:15.760)

being a public intellectual probably means being on Twitter. Whereas of course that wasn't true

Charles Isbell (58:21.520)

20 years ago, because Twitter wasn't around 20 years ago. And if it was, it wasn't around in a

Charles Isbell (58:25.280)

meaningful way. I don't actually know how long Twitter has been around. As I get older, I find

Lex Fridman (58:28.960)

that my notion of time has gotten worse and worse. Like Google really has been around that long?

Charles Isbell (58:33.840)

Anyway, the point is that I think that we sometimes forget that a part of our job is to

Charles Isbell (58:43.360)

impact the people who aren't in the world that we're in, and that that's the point of being at

Charles Isbell (58:47.040)

a great place and being a great person, frankly. There's an interesting force in terms of public

Charles Isbell (58:51.520)

intellectuals. Forget Twitter, we could look at just online courses that are public facing in

Charles Isbell (58:57.120)

some part. There is a kind of force that pulls you back. Let me just call it out because I don't

Charles Isbell (59:06.960)

give a damn at this point. There's a little bit of, all of us have this, but certainly faculty

Charles Isbell (59:12.000)

have this, which is jealousy. Whoever's popular at being a good communicator, exciting the world

Charles Isbell (59:20.480)

with their science. And of course, when you excite the world with the science, it's not

Charles Isbell (59:27.280)

peer reviewed, clean. It all sounds like bullshit. It's like a Ted talk and people roll their eyes

Lex Fridman (59:36.160)

and they hate that a Ted talk gets millions of views or something like that. And then everybody

Charles Isbell (59:41.360)

pulls each other back. There's this force that just kind of, it's hard to stand out unless you

Charles Isbell (59:46.400)

win a Nobel prize or whatever. It's only when you get senior enough where you just stop giving a

Charles Isbell (59:53.200)

damn. But just like you said, even when you get tenure, that was always the surprising thing to

Charles Isbell (59:58.160)

me. I have many colleagues and friends who have gotten tenure, but there's not a switch.