Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:15
Pushkin.
0:24
A few weeks ago, I went to Chicago to interview
0:26
two people on stage about
0:28
creative work they've done using artificial
0:31
intelligence. One of the people
0:33
was Stephen marsh He's a writer. He's
0:36
done nonfiction books, novels, magazine
0:38
articles, and earlier this year,
0:41
he used AI to help him write a
0:43
short novel called Death of an Author. That
0:46
book, by the way, was published in audio
0:48
form by Pushkin Industries, the same company
0:51
that publishes this podcast. The
0:53
other person on stage with us was Lucas
0:56
Contor. Lucas is a composer.
0:59
Among other things, He's won a couple Emmys
1:01
for his work scoring.
1:02
The Olympics for NBC.
1:04
He co produced a Lord song that was
1:06
in one of the Hunger Games movies. And
1:09
the reason he was there talking with us Lucas
1:12
used AI to help him finish Schubert's
1:14
unfinished symphony. It was
1:16
a really interesting conversation and I
1:18
thought it would make a great episode of What's Your Problem?
1:21
So here it is now,
1:23
please join me and welcome our panel.
1:31
So let's do a
1:33
thing you're never supposed to do in narrative.
1:36
Let's answer the question right
1:38
at the beginning. So,
1:41
uh, the sort of headline question
1:44
for this panel is will
1:46
AI kill creativity?
1:49
I want to ask both of you. I
1:51
want you to answer in one word at
1:53
the same time, on the count of three.
1:56
It's gonna be it's not one two three go, You're gonna
1:58
go on what it is? One two three
2:00
go?
2:02
Yeah?
2:02
Okay, will AI
2:04
kill creativity? One two three
2:07
No? No, great done,
2:10
let's go. Yeah, thank you, Thank you very
2:12
much.
2:15
So I'm delighted to be here with
2:17
both of you, in particular
2:20
because you have made things with
2:22
AI. Right there have been countless
2:24
panels of people sort of waving their hands.
2:26
About the theory of AI or the
2:28
future of AI.
2:30
But I love that we're here talking
2:32
about things that you have made, creative
2:35
work that you've made. And so what I want to do
2:37
is, I want to start by talking a
2:39
little bit about process. I love talking
2:42
about how people make creative things.
2:45
And we'll just do that in order, frankly, just because
2:47
I want to get into first of the book and then
2:50
the symphony, and then we can talk more
2:52
generally about AI and creativity and
2:55
humanity, and then we can wave our hands in that classic
2:57
can'd wave away. So Stephen,
2:59
let's start with you. I
3:02
want to read an excerpt from your book,
3:05
in part because this book that
3:07
was written with Ai has a very
3:09
particular, I don't know
3:11
quality to the pros. There's a really
3:14
interesting feel to the pros, and I
3:16
don't know if you'll quite get it from a paragraph,
3:18
but I want to give you something to hold
3:20
on to as we're talking about the book.
3:22
So I
3:24
think I have this right.
3:25
This passage I'm going to read, it's in the first person,
3:27
and it's actually in the book, spoken by
3:30
a digital avatar, an AI
3:32
avatar in the book, who is an avatar
3:35
of a dead author whose
3:38
death is the title of the book. So
3:40
the passage in her voice goes like this,
3:45
I learned the limits of machines when they wanted
3:47
me to fly bombers. They were going
3:49
to force me to push a button that would end
3:51
the world.
3:53
I hope you can.
3:53
Understand that my stance as a pacifist
3:56
wasn't cowardice or principle, but
3:58
a confession. I could never
4:01
bring myself to press that button.
4:03
Human beings cannot stop making buttons,
4:06
and once we've made them, we can't
4:08
stop pushing them.
4:10
Pretty good for a machine, it really pretty good for
4:12
a machine.
4:13
Yeah, I'm gonna read that last sentence again because
4:15
I like it, and because it comes up a couple times in the
4:17
book. Human beings cannot stop
4:19
making buttons, and once we've
4:22
made them, we can't stop pushing
4:24
them. So maybe Stephen,
4:27
we should actually start with that sentence.
4:29
Right.
4:29
It's a great sentence, I think, I or really
4:32
interesting sentence. Sounds like a sentence
4:34
a human being would write. It ends up being important
4:36
in the book sort of thematically. How
4:38
did the machine write
4:41
that sentence?
4:42
Okay, let me see if I can get it exactly
4:44
right. So that
4:47
was the first person from the
4:49
death of an author. So Jacob came to
4:51
me in February and said, we need to release this
4:53
thing.
4:53
This is Jacob Weisberg, actually the
4:56
person who runs Yeah, pushed It. Let's actually
4:58
start at the beginning of the sure and then we'll get to that sentence.
5:00
So right, So Jacob
5:03
Weisberg, who runs Pushkin, which is the company
5:05
where I make a podcast, came to you in February.
5:08
And said, can you write a book
5:10
that's AI that's generated by AI?
5:12
In fact, he said, can you create an AI author and
5:15
then have that author create a book. Now I'd
5:17
been working on this for a while i'd
5:19
been working. I'd wroteen my first you
5:21
know, algorithmically generated story for Wired
5:24
in twenty seventeen, which was before the Transformer,
5:27
so the Dark Ages of AI really,
5:31
and so I said, yes, I can definitely do that.
5:34
It'll be about ninety five percent computer
5:36
generated. I don't want to if I want to change
5:39
heat to the character's name or something like
5:41
that, I want to be able to do that without forcing
5:43
all these iterations and so on. And
5:46
basically I used GBT
5:49
four and I
5:51
would use it to generate texts.
5:54
I knew from having done AI
5:57
AI text before that A is very
5:59
poor at generating plots, okay, and
6:01
it's very poor at certain other tasks.
6:03
It's incredibly good at style, okay.
6:05
Right, So I would, you
6:07
know, have very clear ideas of where the narrative
6:09
what's going. I'd give very specific grammatical
6:11
and syntactical commands write
6:14
a paragraph
6:16
with high variability, like very
6:18
very specific commands like wait, do the whole.
6:21
Give me an example in its entirety
6:23
of a command.
6:24
It would be almost impossible to do because it's exactly like
6:26
do it when you've seen them for visual
6:28
stuff, where it's like they'll just to get really
6:31
interesting AI generated pictures. You often
6:33
have like one hundred different references.
6:36
Like it almost impossible,
6:38
but just give me something.
6:39
Give me something.
6:42
Write a hard boiled detective
6:44
story paragraph
6:47
with a variability between short and long
6:49
sentences and
6:53
clear, elegant syntax,
6:56
containing the following information, and then you write
6:59
out information it would generate that.
7:01
Then you would take that and I would put it into a program
7:04
called pseudo.
7:04
Right, and wait just before we go to the next program,
7:07
when you say containing the following information, like
7:10
that.
7:11
One would be it would be like in this one.
7:13
The author says, well,
7:15
that would be slightly different because with characters,
7:17
I would use a whole different set of commands.
7:20
So you know the author and here
7:22
was basically a combination of Margaret
7:24
Outwood and my dead father. Because I was writing
7:27
this thing fast, so I needed to know something that
7:29
I needed to have a character that I would automatically
7:31
be interested in, and.
7:32
I should say, you're a Canadian, like
7:35
basically the next closest thing after your.
7:38
If women were alive for you, yes,
7:40
right, and so uh, and
7:43
so that I would say write something
7:45
like Sylvia Plath meets Philip
7:48
Roth and meets a bunch of different other things
7:50
and get hurt it.
7:51
So you're doing a very specific character. And
7:53
then do you do all of the sort of exposition
7:56
or plot points, like what what in terms of substance,
7:58
what is an example of what you might put?
8:00
Well, I would that would probably actually be mostly
8:03
the machine, but for plot details would be like she
8:05
walks to a bridge.
8:07
And but this this paragraph about like the
8:10
you know, the buttons. I wouldn't
8:12
press the button, and it's like, how do you It would
8:14
be something like you
8:16
know it to be something like the character.
8:19
Reminisces about her times as a
8:22
UH and and expounds
8:24
philosophically on the
8:27
difference between AI and being a
8:30
fighter pilot.
8:31
And or the character expounds on being a pacifist
8:34
in the military. Exactly okay, right, and
8:36
so sometimes more, sometimes
8:38
less. Tried to get as little as possible,
8:41
but you know you want specificity here,
8:43
like you're the more precise the command,
8:45
the better information, the
8:47
precise the command, the more it's just you writing
8:49
it with the weird kind of intermediation is.
8:52
My creation, right, this is a tool
8:54
which you will I will say the same thing.
8:57
So just the same as if like this is the thing people
8:59
don't understand, right, it's like, of
9:02
course this is a creative act. It's just a different creative
9:04
act, right, Like it's this is one hundred
9:06
percent me. It's just I didn't
9:08
write the words right
9:10
like like like so that's like
9:13
like that's weird, Like I am, yeah,
9:15
it's very weird, like I am.
9:16
Don't you didn't write the words that ended up in the
9:18
book word you weren't
9:21
the words that were the instructions to the machine
9:23
to write the words that.
9:23
Well, so good as any computer that's true, any
9:26
computer program.
9:26
So so okay, so I want to get back to the specific
9:29
sort of process narratives. So you put this very
9:31
specific prompt into GPD
9:34
four, which is basically chat GPT.
9:36
I would say, it's actually better fine,
9:38
and chat EPT four is now it was
9:40
better than what chatchat is now fine
9:42
for creative stuff.
9:44
Uh. Then you get some output,
9:46
you get the paragraph for it, and
9:48
then what and then.
9:49
It usually it's very bad, right, And
9:51
then you take that and you put it in a program called pseudo
9:54
right. Okay, and pseudo right is a stochastic
9:56
writing instrument. So you could you then select
9:58
the text and you say shorten lengthen
10:01
you say and then it has another button, which is a customized
10:04
feature, which is make it sound
10:06
like X. So, make
10:08
it sound like Ernest Hemingway, make
10:11
it sound like f Scott Fitzgerald, and
10:13
and and you know, the of course,
10:15
the thing I figured out very quickly is that if you
10:17
want something to sound like Margaret out With, the very
10:20
last thing you should do is put in make
10:22
it sound like markered out.
10:23
That's not enough course to me.
10:25
Well, of course, because markered
10:27
Outwood is in trying to sound like Margaret Out would she's trying
10:29
to sound like Sylvia Plathmas Philip
10:32
Roth meets, it meets
10:34
a bunch of other things.
10:35
Right, then you ultimately always
10:37
get back.
10:38
Yeah, And so that when you the way you get interesting
10:40
things in this text is by essentially folding
10:43
these layers of style onto
10:45
each other.
10:46
Now I also use and then so
10:48
pseudo right has some output. Yeah,
10:50
and then is that output what we're reading in the
10:52
book?
10:52
Correct?
10:54
Or you know, if I don't like it, I
10:56
just try again, just refresh,
10:58
refresh, refresh until I guess something that I like.
11:00
And so so this is very much a creative
11:02
act.
11:03
And you're doing that basically a paragraph
11:05
at a time.
11:06
Yeah, Well, with dialogue, it would
11:08
go like die would be a lot longer, right,
11:11
like, because you want flow and
11:13
you want so I could do up
11:15
to maybe five hundred words of dialogue
11:17
at a time. Uh huh, So that would have been part of
11:19
a much longer series of instructions.
11:21
So this sentence human beings cannot
11:23
stop making buttons, and once we've made
11:25
them, we can't stop pushing them. A nice sentence,
11:28
you know, big idea. I certainly didn't think
11:30
of that.
11:30
You didn't. It just came out of some refreshment,
11:33
yeah, fresh, and.
11:34
It was in some I mean, obviously
11:36
I made it, and I authorized
11:38
it too. You know, I've compared it in the
11:40
Atlantic to doing hip hop
11:42
in the sense that you're you're folding
11:44
things on top of each other, right,
11:47
You're folding styles and
11:49
metrics and effects on top
11:51
of each other until you get
11:53
something new and weird.
11:55
Right.
11:56
And I would say about twenty
11:58
times during the course of writing it, I felt
12:00
like I was, you know, putting
12:02
my hand up against something new
12:05
and weird.
12:05
That's fun, right, like something.
12:08
But you know this is for most
12:10
of the process, it's just a writing tool,
12:13
right, Like, it writes it for you. You decide
12:15
if it works, right, and you tell
12:17
it's you tell it what to write
12:20
in.
12:20
A very granular way.
12:21
The more granular, just like writing
12:23
normally, the more you know about the
12:26
bigger planning. The more planning you have for
12:28
any essay, the better the essay is going
12:30
to be right. And in this case,
12:32
so you have a plan and then you have the editing
12:34
process, and in between there's this machine.
12:37
But how much of that, how much does
12:39
that matter? Is actually I
12:42
don't know if it's like twenty times it did
12:44
matter where it was like, oh that's not something I would
12:47
have written, but.
12:47
It's very beautiful.
12:48
Yeah, and it's very strange, and
12:50
it's you know, there's a there's
12:52
a Danish journalist who deals with go players who
12:54
play ai go against
12:56
each other, and they say it's like listening to an
12:58
alien make music right, because
13:00
it's like it's not how they would play go, it's not
13:02
how a human could play go, but it's
13:05
obviously makes sense on some level. Similarly,
13:08
that's how I felt like most
13:10
of the time, it's just a writing machine that does what
13:12
I tell it and then I correct it. But
13:14
then maybe twenty times you feel this
13:17
new presence. That's what's
13:19
exciting.
13:22
We'll be back in a minute to hear how Lucas
13:24
Contour used AI to help
13:26
him finish Schubert's unfinished symphony.
13:44
Okay, back to the conversation in Chicago
13:46
with Stephen Marsh and Lucas Contour.
13:49
Lucas's story of using AI
13:51
to finish Schubert's unfinished symphony
13:53
goes back to twenty nineteen. He
13:56
was approached by a Chinese tech company called
13:58
Huawei. They said, we
14:00
want our phone, which runs AI,
14:03
to finish Schubert's on Finnish symphony. And
14:06
they didn't know what that meant. They had a
14:08
tech team in place that was running the
14:10
AI and I knew those people.
14:12
That's why they, I think brought me in. I was told
14:14
that. So my friend, the technologist who
14:16
brought me in on this project, told me that he thought
14:19
that I would be a good fit because I have a
14:21
corporate friendly bio where they could say,
14:23
oh, he can do it. And
14:25
he said, I know they you don't
14:27
have to say that part. You don't have to say that part.
14:29
He said, uh, But he said I. He said
14:32
that you, I know you can command an orchestra,
14:35
but I don't think you'll be precious about the project,
14:37
meaning that I won't be. He didn't think I would
14:39
say like, oh, well, this is heresy. We
14:41
shouldn't take Schubert's perfect work that
14:44
was so perfect that he didn't even finish it and
14:47
do something with it. And
14:49
uh yeah. So I think
14:51
they thought they would just that I would press
14:53
a button on the phone and a symphony would come
14:55
out and somehow a bunch of musicians would
14:57
play.
14:58
So they need you for it. They
15:01
just pushed the button.
15:02
So this is the conversation we had, and eventually
15:04
I had to I was on a call with them and I said, look,
15:07
this is this is not I mean, what you're asking for in
15:09
principle doesn't exist, like you can't And
15:12
I mean, what do you even want the machine to do? Do you want it to generate
15:14
audio for you? Do you want it to generate a score? Do
15:16
you want it to perform the score? So, I mean,
15:18
right off the bat, this was a
15:20
fascinating project because I had to think about
15:23
the very nature of music to even really
15:25
get started. I don't know if that answers the
15:27
question about I think it does.
15:29
I mean, I just wanted you to set yourself
15:31
up, and I think you've done it.
15:32
You want to I think I'm set up, so I'm
15:34
gonna try something new for you today. So on the on
15:37
the prep call for this event, we
15:40
discussed I said something
15:42
that I don't often say out loud, but I realized as a hallmark
15:44
of my presence on stage, is that I like to
15:46
do things that might spectacularly
15:49
fail in the hopes that they will
15:51
be entertaining to an audience. So I'm going to
15:54
do one of them for you.
15:54
Now.
15:55
I'm going to I wrote a little thing about the
15:57
Unfinished Symphony. I'm going to
15:59
explain it while I'm playing some music in the background
16:01
and basically scoring it as
16:04
i'm talking. So you know, wish me
16:06
luck and hopefully it'll be interesting. This
16:12
is how the Unfinished Symphony starts.
16:29
A symphony has four movements,
16:31
but Schubert only wrote two and sketched
16:34
a third of his eighth Symphony, the Unfinished
16:36
Symphony. No one knows why
16:38
he abandoned the Unfinished Symphony, but
16:40
he did, and now it's probably his most
16:43
famous work, along with his greatest hit, Ave
16:45
Maria. Some
16:50
scholars believe that Schubert couldn't find a
16:52
way to fit the Eighth Symphony into the orthodoxy
16:55
of the time. Which forbade three movements
16:57
in a row in triple meter meters
16:59
like three, four and sixty eight.
17:01
But I don't believe this.
17:03
Schubert showed little reverence for orthodoxy
17:05
during his short life, and the AI
17:08
that I used to finish Ubert's on Finnish Symphony
17:10
didn't believe it either. At
17:19
first, we trained the AI on
17:21
recordings of Schubert's entire catalog, then
17:25
prompted it with the first two movements of the unfinished
17:27
symphony. Seems like a reasonable strategy,
17:29
right, This was the result
17:37
sounds like Kat's walking on a piano,
17:42
But this was actually pretty logical.
17:45
Recorded music has almost no mathematically
17:47
discernible patterns to it, so
17:49
from the AI's perspective, the input
17:52
was nonsense, so more nonsense was
17:54
a logical output. Music
18:04
as an abstraction is math, but
18:06
music in practice is convention. Music
18:10
is understood by groups of humans, and like
18:12
any language, music doesn't have objective
18:14
meaning. Music is emotionally
18:17
inert left
18:22
myself. A water break is symphony. A
18:25
symphony is like a skyscraper. It's
18:27
enormous, but every inch of it is designed
18:29
in meticulous detail. It's
18:31
beautiful on the outside, but the inside
18:34
is filled with utilitarian solutions to simple
18:36
problems. A
18:38
skyscraper has electrical
18:40
columns to distribute power throughout the building,
18:43
It has plumbing, it has elevators, but
18:45
you don't see any of this essential detail when you
18:47
admire the building from outside. A
18:50
symphony is like a skyscraper, but
18:53
a recording of a symphony is
18:55
like a skyscraper's facade.
19:03
There is no way to tell from photos
19:05
of even a million facades that skyscrapers
19:08
should have electricity, bathrooms and a
19:10
way for humans to move from one floor to another.
19:13
Similarly, there is no way to tell
19:15
from the morass of frequencies that is a piece of
19:17
recorded music which frequencies are the most
19:19
important.
19:30
There we go.
19:33
So analyzing recorded
19:35
music got us nowhere, and
19:38
I thought that the best way to proceed
19:41
was to simplify the task and
19:43
just train the AI on
19:46
the blueprints of music rather than a finished
19:48
building. So train the AI on a blueprint
19:50
rather than a finished building. So what you just heard, what you're
19:52
hearing now is the main theme from the unfinished
19:55
symphony. Here it is again, just
19:57
really listen and try to listen for the melody.
20:10
And here is that same theme reduced
20:12
to its blueprint. This
20:22
structure, this blueprint in music,
20:25
is just a simple melody. So
20:27
my team and I went to work extracting just
20:29
the melodies from as much of Schubert's music as
20:31
we could get our hands on. These
20:34
are some examples of the melodies we extracted. These
20:39
sound robotic because they are.
20:42
They sound emotionally inert. But
20:44
these are Schubert's melodies reduced to their simplest
20:46
forms, the forms that human
20:48
composition students would use when beginning a study
20:50
of Schubert. Your ear knows how
20:53
to pick a melody out of a dense arrangement, but
20:55
an untrained AI cannot
20:57
do this. The
21:02
reason that, since the results we wanted were simple, we
21:04
needed to train the AI on simple data.
21:08
We trained on hours of these simple melodies
21:10
and then prompted again. We prompted
21:12
it with the unfinished symphony reduced to
21:15
its blueprint, and these were some of the results.
21:23
So this is what it suggested might be
21:25
something that Subert would have written.
21:29
These are simple, but much more musical
21:31
than the cats walking on a piano that came from the audio
21:33
only training data.
21:39
This one, for some reason, caught my attention. Let's
21:41
hear it again. I
21:49
liked it, so I selected it for embellishment.
21:52
I decided to use this. I decided
21:54
to use this blueprint. This
22:05
melody is a bit more modern sounding than
22:07
any of Schubert's work. If
22:09
Schubert lived to old age, these sonorities
22:11
would have been available to him.
22:17
The orthodoxy around triple meters
22:19
and other constraints of form would have given
22:22
way to the exploration of
22:24
the Romantic period. Providing
22:33
simple singable melodies is
22:35
perhaps not how most people would imagine
22:38
that an AI would be useful in writing
22:40
a symphony. But what is a symphony?
22:44
Typically people think about a symphony as something
22:46
that you hear, while the score is just a byproduct
22:49
of the notated sounds.
22:53
But to me, the
22:55
sound is a byproduct, and the symphony
22:58
is something that you see. It's something
23:00
that you read. It's a collection
23:02
of abstract ideas in
23:04
abstract notation. It's
23:07
markings on a page that serve as instructions
23:10
for how to create sounds. A
23:12
symphony itself is a blueprint,
23:15
and those instructions that blueprint
23:18
will be executed differently at every performance.
23:32
Let me just check out this music. It's pretty cool. The
23:37
sounds are a byproduct of the abstractions
23:39
that are expressed in the notation, and that byproduct
23:42
is what the audience experiences as a symphony.
23:46
The byproduct is what you hear.
23:48
I didn't know that I thought about music in this way
23:51
until I had to explain how I think about music
23:53
to a machine. This
23:55
project taught me to
23:57
question the assumptions I make when thinking about
24:00
my own craft. I
24:04
think this is the job of the AI assisted composer
24:06
today to think about
24:08
what we know and to guide our audience
24:11
to rethink what happens inside their own minds.
24:23
I think it's our job to question orthodoxy.
24:27
I think it's our job to use new tools to
24:29
make new art. Today's
24:33
artists are not on the verge of being replaced.
24:36
On the contrary, we are possessed of powers
24:38
so great that we will expose more truth
24:40
about the human mind and the human soul
24:43
than any generation before us. We
24:47
stand on the shoulders of giants. They
24:50
have given us the language, they
24:52
have given us the blueprints, they
24:55
have given us the technology. What
24:58
we build with these tools will be more powerful,
25:00
and more beautiful, and more profound than
25:02
anything we can now imagine.
25:07
Artificial intelligence is nothing like us than
25:09
a prosthetic for the human mind. It
25:16
will enhance art the way writing enhanced memory,
25:18
the way printing enhanced literature, the
25:20
way the steam engine enhanced travel. Artificial
25:24
intelligence is an automobile. We're
25:26
only beginning to emerge from the age of horse and
25:28
buggy. Artificial
25:31
intelligence helped me write the music that you're hearing
25:34
right now. So
25:37
will AI kill creativity?
25:42
No, that's
25:49
really rather Good's that
25:51
more or less worked? I think that's really rather good.
25:54
Thanks.
25:57
We'll be back in a minute to wave
26:00
our hands a little bit about the future of
26:02
AI and creativity.
26:14
That's the end of the ads.
26:15
Now we're going back to the show.
26:17
The reason I knew AI was going to take off
26:20
was when I was writing a piece for The New Yorker about
26:22
GPT three and I
26:25
got it to finish off Coleridge's
26:29
Kubla Khan is great unfinished
26:31
poem, and it did
26:33
it perfectly well. Like I mean,
26:35
if somebody told me, yeah, this is how it ended, I would have been
26:37
like, great, right and so,
26:40
And it did it like that like one second.
26:42
I mean, it was just so incredible to me.
26:43
Just to sort of close this part of the conversation,
26:47
I'm curious. I mean, both of these projects.
26:49
We were very AI forward,
26:51
right, They were like high concept,
26:54
you know, sort of let's explicitly
26:56
wrap this thing in AI.
26:57
Fine.
26:58
Interesting, But presumably
27:01
the real action comes in the things that are
27:03
just what you guys are working on that
27:05
just happens to have AI as a tool, the same
27:07
way say a Google search, which by the way, is
27:09
a kind of AI, is also a tool, right,
27:12
And so I'm curious in your work
27:14
now on other projects that
27:16
are not like, hey, look this was made with AI
27:19
kind of projects. Are you guys using AI? And
27:21
if so, how what do you
27:23
want to go first?
27:24
Yeah? Yeah, first, so
27:26
yeah, obviously of course, like it's in everybody's
27:28
pockets, you use it all the time. And AI
27:31
has done nothing so far
27:33
other than help my career. And I don't mean
27:35
just by doing this, which was fantastic. But when
27:38
I write a piece of music and put it on Spotify, the
27:40
reason you hear it is because an AI recommended
27:43
it to you. You know, that's the only reason you're going to
27:45
find it. And so and these types
27:47
of algorithms that are generating that are keeping
27:49
people out on apps longer and keeping
27:51
people on Netflix and on Spotify longer,
27:54
are putting money not enough money, and
27:56
that's another panel discussion, but putting
27:58
money in our pockets directly?
28:00
Let me let me ask a more precise
28:02
version of the question in response to that clever
28:05
answer.
28:05
Do you use generative AI?
28:08
Yes?
28:09
And also this is a terminology
28:11
problem.
28:12
But you know what do you
28:14
use music?
28:15
Do you use AI to generate musical
28:17
ideas for you?
28:19
Yes?
28:19
But also like what is a musical idea?
28:21
I use a parametric eque that I mean they
28:23
were using a they were using this was there
28:25
was probably good. I'm
28:27
trying, well the answer the answer is yes.
28:30
I know what you're saying. But I feel like you know
28:32
what I'm saying.
28:33
Well, yes, I'm The reason I'm trying to drill
28:35
down here is because this there tell
28:37
me how to ask the question I want to asking
28:39
doesn't have the answer that you want?
28:42
Right, So fair, what's the what?
28:45
What's the smarter version of the question?
28:47
I'm not well enough equipped to ask.
28:50
I don't know if I can.
28:51
I don't know if I can help you with that.
28:53
I don't
28:58
let.
28:58
Me ask the question to you.
29:00
Thank you for your Stephen.
29:02
Do you you use generative AI when
29:04
you're writing with other things?
29:06
Okay, here's the thing, and I think this is sort of
29:08
where we're going. Like I would when
29:10
I write something for a magazine or newspaper
29:12
or novel that I'm working on, I would never use chatchipt.
29:15
Even to get an idea because here
29:17
or whatever they because I'm
29:19
so much smarter than chat GPT.
29:22
Right, And I'm like when you
29:24
and what you have to also have to understand is chatchypt.
29:26
The reason it's so successful is exactly that it has
29:28
been banalified, like when
29:30
you use other generative ais that
29:32
we have access to, because you realize
29:35
that like these are the ones
29:37
that the public uses are very poor
29:39
creatively, like they're actually.
29:41
But you have access to the good ones, to
29:43
the good stuff.
29:43
Here's the thing you can't get on when when you use
29:46
the good stuff.
29:46
What the good stuff is going to be used
29:49
for stuff that doesn't exist yet.
29:51
What we're seeing here is the birth of a
29:53
new medium, right and what
29:56
and so when it comes to write an essay,
29:59
what people want when they write, when they
30:01
read an essay, is a human being communicating
30:04
their thoughts and feelings, right, they
30:07
don't want like they don't That's why they go
30:09
to it. And a generative AI cannot
30:11
do that generative Like it's
30:13
sort of like asking, like do you use film
30:16
to make theater? Like at
30:18
first, you know, when you when film was invented,
30:20
all they did was cannibalized theater and they were putting
30:22
on weird shows or they were recreating news
30:24
events and things like this. That's where
30:27
we're at right now. This is going to be used for
30:29
new art forms that don't exist,
30:32
and that's that's the exciting stuff.
30:34
And it's also why it's almost impossible to do.
30:36
You mean, like the book that is never done, the book
30:38
where it can or like what like.
30:40
I'm written that I have written a short story
30:42
that is infinite art
30:44
forms?
30:45
Like what do you have in your mind when you say it?
30:47
Well, like, for example, I'm working
30:49
with cohere to recreate the
30:51
Oracle at Delphi. Right there's a
30:53
large amount of information that you can glean from
30:55
that, and there's also pretty interesting historical
30:57
record.
30:58
And so you'll ask it a question and it will
31:00
answer, yes.
31:01
We're try and recreate the experience
31:03
of going to the Oracle at Delphia as closely as we
31:05
can use effects.
31:07
Yeah, it's a perfect use of AI and so oracles.
31:09
This is one of the things that has come up in my research is
31:11
that we use oracles because we're bad
31:13
at doing things randomly. So if
31:16
we're out in the wilderness, we'll just
31:18
go hunt in the same place over and over and over
31:20
again, right, And eventually animals figure
31:22
it out and they say, just don't hang out there.
31:23
And you won't get eaten by the humans.
31:25
And so when we like consult an oracle,
31:27
or roll some dice, or like ask the sacred
31:29
chickens if we should go to war, they're basically giving
31:32
us a random answer.
31:33
That's right. There are randomization engines, see,
31:35
and it's.
31:35
Things of this nature that I think will
31:38
be that I'm excited about
31:40
to use it. We're cannibalizing forms. That's what
31:42
I do writing short stories to It's very interesting.
31:44
But the truth is that what this can be used for
31:47
we don't know yet, and what it's going to be used
31:49
for is some weird and the
31:51
problem is there's absolutely no institutions
31:53
to do it with, right, Like.
31:55
Nobody will buy your oracle of Veli's.
31:58
Supposed to take oracle of Hi.
32:00
My name Stephen, I'd like to recreate the
32:02
oracle at DELFI using generative
32:05
AI. I'm sorry, sir, this
32:07
is a key mark, you
32:09
know what I mean, like like it like it's
32:12
not that's that's not like there's no one to
32:14
go to. So that's that's
32:16
where we're at. To me, Like, I think
32:18
the the the thing that I think is very obvious
32:20
is that when you use generative
32:23
AI, what it is very good at is
32:25
the most stock answer, right.
32:28
And that's why it's so such a
32:30
threat to like the undergraduate essay, right,
32:32
because that there you're basically looking for the
32:34
fulfillment of a stylistic you
32:37
know, set pattern that
32:39
it can do.
32:40
Right.
32:41
But people respond to human
32:43
like there's this weird idea
32:46
that art is something external to our experience
32:48
of it. It isn't. It's just
32:50
we we have we create tools. As
32:53
the moment we find tools, all we're thinking of is
32:55
can we do something weird with it? And I think,
32:57
I mean, one thing that I've really learned doing this is
32:59
that creativity is instructible.
33:03
Like it it doesn't matter what comes
33:05
down technologically, what comes down politically,
33:08
what Like, we are creative
33:11
animals and we have to understand
33:13
that that's just our nature and
33:15
nothing is gonna kill it, nothing, not
33:17
certainly not chat gept.
33:19
Great I can I can sum up the history
33:21
of music from the year sixty
33:24
thousand before present to now with
33:26
one sentence, and maybe you'll agree that this sums up the history
33:28
of art already. It's the search for new sounds.
33:31
Yeah, that's it. That's all there is to
33:33
it. If something exists, nobody cares and
33:35
chat geept I will chat. Chapet doesn't do music.
33:37
But there are many music generative ais, and
33:40
they generate music that, like
33:42
charitably would call insipid. Yeah
33:44
it's fine, like it's music. You would
33:46
recognize it as music, but nobody. You wouldn't listen
33:49
to it. It'll get bad music. It
33:51
won't, so it'll sound better,
33:53
it'll sound better. So this is the but,
33:55
but nobody cares about that. So as soon as like,
33:57
as soon as you can have so Jacob
34:00
for your podcast, as soon as you can have beautiful
34:02
sounding orchestral music like this for free,
34:05
you're gonna want something else because this is
34:07
available and it's everywhere, and so what you're
34:09
gonna what you're gonna want is like the thing where
34:11
like Lucas plays a guitar with a really nice
34:13
sounding reverb. That's gonna be the style and
34:16
you can trace and we have a we have a composer
34:18
in the audience who could, hopefully will agree with me on this, and a
34:20
professor of this kind of thing. But you can
34:22
trace musical styles in media, and it's
34:24
like whatever is ubiquitous just falls out of fashion
34:27
and then that whatever the opposite of it is becomes
34:30
becomes fashionable. So yeah,
34:33
that's my that's my two cents the search for new
34:35
sounds.
34:36
Thanks you guys. This is closure. Yeah.
34:48
My conversation with Lucas Contour
34:50
and Stephen Marsh was organized.
34:52
By Chicago Humanities.
34:56
Today's show was edited by Karen
34:58
Chakerji, produced by Edith Russolo,
35:01
and engineered by Amanda k Wong.
35:04
You can email us.
35:05
At Problem at pushkin dot fm.
35:07
We are always, always, always
35:09
trying to find interesting new guests
35:11
for the show, So if there's somebody who think we should book,
35:13
please let us know. I'm Jacob Goldstein
35:16
and we'll be back next week with another episode
35:18
of What's Your Problem.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More