Scott Draves Interview

Scott Draves Interview

Transcript

S1 Speaker 1 0

Yeah. Could you talk a little bit about how you first got into art with A I, I'm

assuming this was in your phd at Carnegie Mellon or what was it in undergrad

for you?

S2 Speaker 2 0

I would say I started work, you know, the A I algorithm stuff that I did like the,

the early generative art.

S2 Speaker 2 0

T Home Your work Projects Community Upgrade · 7 days left J

Really the art of, you know, the thing that became the electric sheep, it sort of

started with something called the flame algorithm and that was something I

I was working as an undergrad at Brown University already.

S2 Speaker 2 0

And it's just, it's been a lifelong pursuit of mine.

S2 Speaker 2 0

And so, yeah, it goes, it goes way back. So I guess it was originally sort of more

on, on the graphical side. And then as time went by, I sort of like, like kind of

like a rendering algorithm. And then as, as time went by, I got more interested

sort of and, you know, did more work on the kind of like the, the design of

the, of, of the images, which was more where things more like originally, like

just the genetic algorithm.

S2 Speaker 2 1

And then now, you know, neural networks got involved. So, yeah, it go, it go, it

go, it goes way back. My, my phd research was sort of related and inspired

by kind of that pursuit actually.

S2 Speaker 2 1

So my phd thesis was an attempt to sort of solve the, the technical problems

or the language problems which I was having as I was like, I implementing the

essentially the artistic generative algorithms. And was it using this like the,

the fusion model or was that, is that around or how are you? I mean d

diffusion models didn't exist? So the, the, the flame algorithm was sort of

more like a particle system and it was kind of like a combination between a

particle system and a fractal.

S2 Speaker 2 2

And then, so that was one of, you know what one of my one of my generative

algorithms, the one that was actually sort of most similar to the diffusion in

the sense that it was like a kind of like a an I an image model and that you

would like give it images and it would like essentially reproduce new images

based on the style and the inputs.

S2 Speaker 2 2

I did one of those that was called fuse.

S2 Speaker 2 2

And you know, it basically, it's sort of at, at that same time as you know, like

this was stuff that I was doing when I was a grad student but, you know, it

was kind of like AAA side pursuit or it was like, that was like, my passion, like,

you know, like I was like a regular, you know, you know, department of

defence funded phd student, you know, like, they, they weren't funding

computer art, you know what I mean? But as, you know, so the, the, that, you

know, that was the, the res my research was in, you know, sort of special

compilers that would allow this kind of artistic exploration. But really what

allows, you know, research in general and, you know, one of kind of the things

that I don't know, the touch points of my career are sort of is that, you know,

sort of like artistic creation and sort of, you know, scientific or like research

are actually extremely sort of sim similar processes and, you know, sim

similar tools and languages, you know, facilitate both of them when you say

there's they have similar processes is that more of the fact that these tools

are utilised by both communities.

S1 Speaker 1 3

So they're, they're both just require like rapid experimentation, you like, yeah,

rapid iteration, just kind of like messing around, you don't necessarily know

what you're trying to do, which is in sort of, in contrast to kind of like

engineering where you can like make a plan and then like build the whole

thing, you know, like re research and art, it's more like, you know, you don't

really know what you're looking for until you see it.

S2 Speaker 2 4

So you're just trying to like mess around and so you need like rap, both rapid

experimentation, but you also need like, you know, high performance, you

know, at least if you're working with, you know, media, like video, audio, you

know, la latency and total throughput are, are, are important. So when it

comes to like, you know, and that's true whether you're doing, you know,

scientific research or artistic research, definitely.

S2 Speaker 2 4

Yeah, so, so yeah, so my ph sort of my phd research at, at CMU was involved

in sort of the, the language and compiler technology to enable this kind of

like me. So essentially I was using meta programming to enable the kind of cr

creative research both you know, ar artistic or which it was, was, you know,

which was my sort of nighttime passion inspiration. But it's, it's, these are

general fundamental problems in, in computer science and what, what's

medic or creation or excuse me, meta programming, meta programming.

Yeah. Yeah, that's, so that's programs that create programs. So there's like

just a whole you know, compilers, interpreters, macro programming.

S2 Speaker 2 5

And you know, there are all sorts of different forms of meta programming.

S1 Speaker 1 5

OK, great. And so can you talk a little bit about Electric Sheep and kind of the

genesis of that? Yeah. And inspiration because I mean, I I was reading that

you use your previous algorithms for that program.

S2 Speaker 2 6

That's right. So that was, that was the flame. So the previous algorithm that

was the flame algorithm and then that the evolution algorithm is that the

flame algorithm is the flame algorithm. The evolution was the thing that was

sort of added I see to make it to make the electric sheep. Well, I mean, there

there were, you know, several things added. The base. So the flame

algorithm is the sort of particle system fractal thing. So it's really just a

visual, it's like a rendering, a visual, it's an image generator. So that's like a

traditional, you know, you know, read, it's actually in some, in some sense, it

is like the stable diffusion because it reads a text file and then it writes out an

image, you know what I mean? And the text file and it's not a prompt with like

a bunch of English words. It's like its own mathematical language. It's like a

bunch of numbers which are like parameters to equations, you know.

S2 Speaker 2 6

So, but, but it, you know, so, but that was and it, it has some properties that

match up well with the sort of the, the sort of the with neural networks and

the, and the way they work today in the sense that it was a sort of

connections like a high compute kind of vector language, you know, which is

so it worked in a sort of a fun fundamentally similar way to, to neural

networks. But it was really sort of mathematically, it was like a kind of a kind

of a fractal that makes sense. Anyway, that was the flame algorithm. And

that was something that, like I said before I had actually been working on, like

for, for many years back at, at Brown University. I was really fortunate to, you

know, get into these places that sort of supported this kind of work. And then

I really, you know, I owe, owe my sort of my, my teachers or my professors or

whatever, the guys, Andy Van Dam, Peter Lee, you know, Andy Witkin, all, all

those guys Thomas ban off, you know, II, I owe them a lot for sort of giving

me a lot of rope to, to, to do this kind of stuff. But anyway, so I had this

rendering algorithm. And then I had, I had gotten my degree, I had moved to

the Bay area like, you know, San Francisco. And I think in the late nineties,

essentially to participate in the, in the.com boom, you know, like I, I finished

my degree which was, yeah, and you know, went to do start ups. I really was,

I was already super into open source and I was somewhat disillusioned with

academia, with the bureaucracy of it or, just with more like the usual stuff like

pub publishing. Kind of the mid, mid. Well, I think what's called now is the

least publishable unit kind of approach to like career advancement.

S2 Speaker 2 9

People, like, sort of fighting over kind of CRE credit.

S2 Speaker 2 9

I had some incidents where, like, I felt like my ideas were kind of pub, you

know, sort of ripped off. And so I thought that, you know, I would, I, like,

thought open source was a, could be a better system because like, it's all in

the code, you know, there's like, it's, it's online and you can, you know, you,

you know, who did, what contributions essentially speak for themselves and

can be reproduced, you know, science claims to be reproducible, but then

people, like, publish papers and you can't, they don't give you the code so

you can't actually run it. So I thought like open source was like a better and

everything was going digital anyway. So, like, let's just do it, you know, like I

I just had this. So I was like, that's part of why I didn't, like, go become a

professor somewhere, you know what I mean? Right. So I had this, I'm, you

I know, and I'm still, I'm still really into open source. Maybe I'm, I'm a, maybe

less idealistic than I, than I was back then or so any anyway, so II I, but you

know, that was, you know, part, part of what made the electric sheep, you

know what, what, what it is. And that's part of my belief in this kind of, you

know, remix culture comes from that, you know, from, from, from open

source.

S1 Speaker 1 10

Yeah, I mean, it, it really is and then in the sense, like remix everything

creative is a remix in a sense because you're pulling inspiration and kind of

splicing together inspirations and different pieces.

S2 Speaker 2 10

No, I mean, it's, I mean, I was just, it's, it's just, it's just how the human brain

works. I mean, I was just, you know, talking to my daughter and she was like,

you know, drawing something and we were talking about how it was like,

similar to something that she had seen, you know, and we were talking about

like, just, you know, s singing songs and when you just make up a melody, it's

maybe inspired by something you heard, you know, last month or, or

something about your brain.

S1 Speaker 1 11

But, yeah, exactly.

S2 Speaker 2 11

So, so I was, yeah, so I, I'm, you know, I'm, I'm a, I'm a believer in this system.

And that's, and also there was a lot of, you know, in the, in the Bay area,

some, some of the other ins inspiration for Electric Sheep was definitely like

the sort of like, just like the burning man culture.

S2 Speaker 2 11

That was, that was sort of part of it.

S2 Speaker 2 11

But, yeah, so I had the and I, I learned a lot from, from making stuff open

source. So it's, it's very, it's very interesting, like, and, you know, today and

back then it was considered kind of radical.

S2 Speaker 2 11

And I don't know what the word is, con contrarian or, or whatever, you know,

it was like, essentially, you know, like Berkeley Hippies or whatever. And, you

know, like the free software foundation, like the mit nerd squad kind of thing.

So it was, it was definitely a counterculture, you know what I mean, like a

rebellious thing and the big battle was with like, you know, Microsoft or

whatever, you know, and Apple, like sort of the corporate version. So I've, you

know, so this was I was definitely on the sort of the side of the, you know, the,

the free software thing. You know, we Bill Gates, whatever said, open sources

of cancer, you know what I mean?

S2 Speaker 2 12

Yeah, I'd, I'd honestly like, and now of course open source runs the world,

you know, like basically every web application runs on Linux on, you know,

the whole cloud is open source, even Microsoft capitulated completely and

runs Azure, you know. Right.

S2 Speaker 2 13

So, you know, Android phones are open source. Apple is kind of half in, half

out, right? So they, they're but anyway, it was I mean, so I've, my experience

is that it's a real, a really, you know, beneficial and effective way of advancing

culture. So, we haven't even gotten into machine learning yet.

S1 Speaker 1 13

I know. Yeah, I realise we only have eight minutes left. But yeah, I, I would love

maybe in our next game to dive into open source a little bit more. And then

also, obviously, I mean, there's because there's, you know, like I said, like I

said, unless maybe, every everything, everything can be taken too far and I

don't want to, you know, I'm, I'm definitely not an, not an, like an absolutist.

S2 Speaker 2 14

Right. You know, there are people who say like everything should be open

source or like, you know, copyright shouldn't exist.

S2 Speaker 2 14

And, you know, I'm not really, I'm not really in that camp.

S2 Speaker 2 14

You can rate a bunch of music back in the day. I don't know.

S2 Speaker 2 14

I, I certainly, I certainly under information must be free or something like that.

Who said that information wants to be free, John Perry Barlow.

S2 Speaker 2 14

I think, you know, I'm a, I'm a fan of him and the Grateful Dead.

S2 Speaker 2 14

But, you know, ultimately you also have to, pay the bills and so, you know,

part of, part of, you know, getting old and having kids is facing some of that,

some of those questions, which, you know, you know, and, but even, even,

even, even before that, you know, the, the essential thing, I, I found one of the

things I found, you know, from the electric sheep was, you know, just giving

stuff away, just doesn't, doesn't really work and actually sort of just, and

actually is, leaves unrealized potential because without any kind of returns,

essentially it's hard to justify the investment and it takes real work to make

great things, you know, and yeah, so is that so electricity was all open

sourced?

S1 Speaker 1 15

And then so it was not funded because it's, that's right, it's free and it was

actually, it was super, it became super negative in some, in some ways

because, you know, the great thing about the Electric Sheep was it had this

sort of distributed renderer, which was a, a way of kind of reducing the costs,

right?

S2 Speaker 2 15

Of creating these graphics because, you know, they were, were you need a

supercomputer to render all these animations.

S2 Speaker 2 15

And so we had, you know, but that was so it was great to have like a, you

know, 100,000, a strong, you know, army of like distributed renderers like my,

My free supercomputer was great. But, but that also meant I had 100,000

people like downloading videos from my web server, you know, which was

very expensive.

S2 Speaker 2 16

And of course, I mean, I sort of freeloaded off of whatever I could, you know,

like I had like a free web, a free website from my, you know, from grad school

or whatever, you know. But as, as soon as it went viral, they, like, saw the load

go through the roof and they said, you know, sorry, you can't, you can't do

this, you, you're shut, you know, you're saturating our server. So I had to you

know, find there was like a long series of techniques to like find like

essentially a free bandwidth that was like the limiting factor for the Electric

Sheep was, was bandwidth.

S2 Speaker 2 17

But yeah, so there was, it was a, it was a cost sink and I did eventually figure

out how to monetize it. But like how, how you, you know, the, the sort of the

strategy is super tricky, you know, like my first idea for how to monetize this

stuff was to be a DJ. So I was like a performing artist, right? Interesting.

S1 Speaker 1 17

The idea was performance, live performance.

S2 Speaker 2 17

That's right. Like, you know, with a DJ plays the music and the VJ does the

visuals and it's a dance party. So, and that, so that's the idea is that like, I'm,

I'm writing generative software, whatever the Electric Sheep Internet or also,

you know, I had like real time interactive, like audio visual stuff that was open

The source also said that you should look for bombs. That was actually it even

predates Electric Sheep and very, was very much sort of more specifically

oriented towards remix kind of like culture and ideas.

S2 Speaker 2 18

And that, you know, so I was like performing live with that software. And so

The idea was that my software is open source. But if you want to hire me to

perform with it, like if you want the, you know, like the best show, obviously,

I'm the expert and so you can, you can hire me to perform at your party. So

that was a business model free in the paid option, excuse me, three in the

paid option. That's right. So like, you know, the I'm the premium, whatever

you, you pay for real life, right?

S2 Speaker 2 18

And it turns out that was a good way to get invited to a lot of parties but not a

good, not a good way to, to make a lot of money.

S2 Speaker 2 18

And yeah, it really sounds like the difference between open source for on the

smaller scale versus like the big tech corporation.

S1 Speaker 1 18

You know, it seems like it's just on the, this was, this was, this was really

small, nowadays, of course, and back then there weren't, there were no, I

mean, there were no corporations which were on board with it.

S2 Speaker 2 19

Like literally none, I think, at that time, you know, in the, in the nineties, the,

I think that maybe the first corporate open source was Mozilla.

I'm not, I'm not really sure if you, if you go back, there's probably some

antecedents. It's a good question. But, yeah, so this, this was, well, it

depends on when you, depends on when you count it anyway. So that was,

That was one business model.

S2 Speaker 2 19

This, the second one I did was, see what the other problem with the, the VJ

thing was essentially in, in, in this, in that culture. It's really the VJ was always

second fiddle to the DJ.

S2 Speaker 2 20

Yes, I mean, it's a dance party. So it's, it's a, it's all about the music.

S2 Speaker 2 20

And it's just not so I always try to like, and it's just a gro it's just a gross

business. I mean, I like a good party but like, you know, clubs with alcohol,

drugs and like sexual assault. I mean, it's, it's, it's not good, right? So, I,

so I was, I was like, I had asked myself the question like, you know, where,

where are the, where are, where the visuals really respected, you know what I

mean? And so, you know, the answer was you know, in a museum or in, in like

a gallery, like, I could be a, a fine artist like, but like a com a commercial, fine

artist because I had already done, like, academic art, you know, you know

What do I mean? Like, sort of like the, the European, like, prizes and like, kind of

like the, the, the, the shows, where the, where they, you know, talk about the

ideas of, of the art. Now, that stuff I had sort of gotten into in, in the nineties

and had some success there but that's different from, like, the artists who,

like, you know, make things and sell them for $10,000 or 50 thou. You know

What do I mean? Like, and really make a living as professional artists as opposed

to like, academic artists who are like, you know, usually the professor, they

write some theories, they make, they make artworks stuff like that.

S2 Speaker 2 21

So I thought, like, I'll do, I'll do the, I, you know, I need to make money. I'm

gonna do the money thing that a moma and all the, and that's how you

started showcasing in museums. Yeah. So I moved, I moved to New York.

That, that's where this, you know, that, that's where the business is and, you

know, tried, tried to sell stuff and had, you know, had some, had some

success better, better than a VJ.

S2 Speaker 2 21

You know, I found that like, financial tech kind of companies would like to buy

a screen to, like, go behind the, you know, the receptionist like that kind of art

in the lobby of a, of a, of a fancy company. So I, I sold those things for like,

you know, 20,000 bucks.

S2 Speaker 2 22

And I felt great. I mean, it's like, oh, it's open source. It's funny, you know

What do I mean? Those high res works were not open source but like the, the,

The electric Sheep system was open source. And what I figured out was how

to use that to create some, in addition to creating all the free content, I was

also creating some, you know, some non free content that was more curated

by me, my, you know, like my aesthetic as opposed to like the publicly A I

driven aesthetic, right? So limited edition higher quality stuff made with the

same factory essentially.

S2 Speaker 2 22

And so that was like the sort of the, the second business model which had,

which had some success but essentially not, not enough.

S2 Speaker 2 23

Especially once 2008 hit. Basically, the, you know, the financial crisis people

were like, you know, fancy art was kind of like first thing to be cut, not a

priority, not a priority. I mean, I, you know, of course, and then, but the real

killer was I met my future wife and we decided to have kids and, you know,

That kind of makes you want to have a job.

S2 Speaker 2 23

So I had to like, I get like a regular job and I've kind of built, back my, like,

regular tech career as opposed to like my, my sort of art, art career.

S2 Speaker 2 23

But, and now they're sort of like, you know, rein intersecting with stability and

with my new company. Oh, I have to go. It's like, oh, yeah.

S1 Speaker 1 24

Yeah, thanks so much for meeting. I'll schedule another meeting for us.

S2 Speaker 2 24

All right. Thanks Sam.

S1 Speaker 1 24

Yeah. Thank you. Have a great day.