Latest / What's Your Problem? / Using AI for Creative Work

Transcript

Using AI for Creative Work · What's Your Problem?

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