So my name is Baptiste Caramiaux. I’m a researcher at Sorbonne University in Paris. And also part of CNRS, which is the French National Center for Scientific Research. And I’ll be moderating this session with the Jenny Williams, who is a senior content designer at Microsoft and a writer.

This salon is organized as part of the Microsoft research AI and society program. So in collaboration with Sorbonne and CNRS. This discussion is meant to discuss with an artist. Today, we are very happy to, to be talking to Sean Michels, who Jenny will introduce in a minute. And our goal is to see how their artistic practice provides ways to think differently about AI, how to engage with it and eventually how to regain power over it. This is very quickly the goal of this, of this event today.

In order to make it slightly more participatory, we have created a shared document. And I’m putting the link in the chat. To everyone. I’m also adding Gonzalo to the panelists, who is one of our collaborators on this project. All right. So I shared with you the document. I see that some folks are already on this document. So that’s perfect.

This document is really meant for you, if you would like to, to add notes, links, you know, questions, comments, whatever you would like to. This document will remain open and available after this event. So it’s not only for this moment, it will also be a document where you can come back if you would like to remember what happened today. And we’ll be using it after Sean’s presentation to, we, I mean, if you have a lot of questions in this document with Jenny, we’ll pick a few questions to ask to Sean. It’s a way for you to, to participate to the discussion. So feel free really to, to, to put questions and comments and remarks on that, on that document. Okay. So now I am now handing over to Jenny.

Thank you so much, Baptiste. And thank you everyone for joining. Before I sort of launch into Sean’s formal bio, I just want to share how Sean’s work came into my life. I’ve worked, on an AI ethics team at Microsoft and I’ve also been a published writer, not a novelist.

Someone who I had worked with on the AI ethics team actually handed me a copy of Sean’s book “Do you remember being born?”, maybe six months ago or so. And he said, you have to read this book and I want to talk to you as soon as you finished it. This novel is about AI, it’s about art. It’s also about motherhood and some of the choices or sacrifices or compromises, um, people make as parents, particularly mothers. I had just become a new mother, nd, you know, have been struggling with a lot of these same questions myself.

I’d also been working with GPT-3 over the last couple of years. And somehow before it kind of became this wave of generative AI that’s in the public now. And so Sean’s work spoke to me on so many levels and, uh, I was really excited and really sort of selfishly suggested that we bring him in as part of this salon discussion. Mostly so that, uh, so that he could share some of the thinking that he has done in this space. You know, some of the, the deeper, explorations that he’s been making over the past few years around AI and art. And also just what it means to be a human in the world and particularly a human artist in the world.

So I’m going to read Sean’s bio. It is there in the document for folks who are sort of more visually attuned along with links to places that you might go visit his work on the web. So Sean Michaels is the author of three novels, including 2014’s “Us conductors”, which re-imagined the invention of the Theremin. It’s also fantastic, brilliant book. And 2023’s “Do You Remember Being Born?” described by Wired as the definitive novel about art in the age of AI. Founder of the pioneering music blog said the Gramophone, Sean has also written for the New Yorker, the Baffler, the Guardian and Pitchfork on topics like art technology and board games. And contributed to digital art projects such as motto and the Sears catalog. Sean is a recipient of the Scotiabank Giller prize, the Grand Prix Numi, the Prix Nouvelle Ecriture. And he has been nominated for the Dublin literary award, the Kirkus prize and the Peabody awards. He’s gone sailing with John McAfee. Definitely want to hear some stories there, um, explored the Paris catacombs with the secret society and been interviewed by Nardwar, who I recently learned as a Canadian journalist, apparently well known for, um, his eccentricity.

Born in Stirling, Scotland, Sean lives in Montreal, Canada, and I am so very pleased to welcome Sean to our salon.

Thank you very much for that introduction, Jenny.

Hi, everyone. I’m here in Montreal in a public library, ready to talk to you about this. “Do you remember being born?” There’s at least three different covers out there. It was just translated into French as well. But this is my favorite with its little AI generated kind of flowers.

In order to just present the book before we get into a conversation, I’m going to do a little slideshow thing. So let me just get this working. Share. All right. Give me a thumbs up if you can see what I can see. Okay, great.

So yeah, I want to begin by thanking you again, Jenny and Baptiste and Microsoft Research for this invitation today. Uh, I am a writer. I live in Montreal. And as Jenny said, I’m the author of three books. In 2014, I published “Us Conductors”, a book about the inventor of the theremin. In 2019, “The Wagers”, which is a novel about luck. And then “Do You Remember Being Born?” a book about creativity, parenthood, lineage, mentorship, letting people into your life and artificial intelligence. And that’s the book I’m here to talk with you about today.

“Do You Remember Being Born?” came out last fall, but it had its beginnings way back in 2019 represented here in a wonderful, AI generated painting. Now if you can, if you’re able to cast your minds back to that wonderful time, uh, Trump was president. COVID hadn’t happened. It was a worse and a better time.

I stumbled randomly across this website, um, talk to transformer.com. It looked something like this. It was created by a researcher at Univeristy of Toronto called Adam Daniel King, and it provided anybody with access to openAI’s then very new GPTs. Um, I didn’t really know what I was looking at. And I found the website confusing just because I didn’t even know what a transformer was. Why was I talking to a transformer?

I thought my brain associated it somehow with the Swedish poet, Thomas Tranströmer. Anyway, the way this worked was you had this text box, and you would enter some text into the box and click generate, and then it would continue writing that text. So for example, I entered, um, this sentence, “my dream luminous and uncanny began” like this, and then you’d click submit. And again, this was an earlier version of GPT that was still, um, that was pretty different than the one we, we play with now, or any of the contemporary models. So a sentence like this, fed into talk to transformer would return something like this. Or I could click generate again, and my dream luminous and uncanny would begin like this.

“I had a twin brother. We were leaning on that naked termite house of Livy Strauss’s, wearing a glint of sweat”. Or like this. “I woke up one morning in autumn and turned on the news to find my city in flames”. In all of these cases, speaking as a dumb ape, my response was, wow. Simply because here was this kind of text generator that could reliably form a sentence, and the sentence it formed had some kind of relationship with the sentence that came before.

But, you know, I’m not just a dumb ape. Um, I’m a reader of literature, a mature expert. And in these roles too, I was fascinated by what was it, talk to transformer. But I found myself examining this text and trying to consider whether it was actually interesting what it was doing. To be interesting, for writing to be interesting, it isn’t enough that it makes sense. It isn’t enough that it continues where the previous sentence left off.

Looking at a completion like this, “I had a twin brother, we were leaning on that naked termite house of Livy Strauss’s, wearing a glint of sweat”. Is it interesting? Or is it stupid? “I had a twin brother, leaning on that naked termite house of Livy Strauss’s, wearing a glint of sweat.” The fact is, for me, for someone who writes literary fiction, it is interesting. It is compelling. I like that glint of sweat. I like the kind of crowded, noisy poetry of leaning on a naked termite house, whatever a termite house is and whatever a naked termite house would possibly be. It is interesting. It is compelling. It is unsettling.

It is unsettling. And, you know, the way that it lived in that in-between of dumb and brilliant at the same time was actually a familiar feeling to me, a familiar feeling from reading poetry. In art, I am not really interested in the pedestrian or the generic. I am interested in something else. I am interested particularly in a unique voice and style and also in surprising synthesis, wherein surprising things are blended together or surprising ideas abut each other. This writing, this AI-generated writing, was captivating for its strangeness, its unpredictability.

This is something that we also can sometimes see just as a kind of illustrate in AI image generators as well. If I give an AI generator, image generator, a prompt like this, a drawing of a teacher in class, and it does this, that is not, in my opinion, very interesting. But being more selective and careful with your prompts, you can provoke that unpredictability, or at least AI’s talent for interesting synthesis.

For example, Monk playing a game on his iPhone, Rembrandt. It’s better, more interesting at least. Or this, the very first thing I remember. These are all from Dall-E circa 2021, I guess. Or this, hideous sandwich. I really like that nonsensical and hideous, poetic, hideous sandwich. Awful. Anyway, back in 2019, back on Talk to Transformer, I tried giving the algorithm a passage of my own writing. So this is taken from my second book, The Wagers.

“Theo skipped the curb into an alley. The asphalt was sprayed with loose leaves, young ivy, vivid in its greens. How many times had he biked along this alleyway, dodging the same sequence of mud piles, potholes, rain grates?

Now, this is an excerpt from my book, but it shows some of the character of my writing, or at least of my writing in this passage. “The asphalt was sprayed with loose leaves.” It’s not just, you know, decorated or scattered with loose leaves. You can see the kind of sequence of nouns or phrases separate by commas. There’s no ands. You know, this is text with its own style and voice, for better or for worse. So if I fed this into Talk to Transformer, here’s how it continued the paragraph: “This was the closest he came to flying”.

Okay, that’s pretty good. Or again. “How many times had he arrived at Babax’s door, buzzing and jangly from chewing sweet fires, dreaming of Mrs. Chives?” Or this. “It was a hurry-up situation. And now this old monkey trouble of getting up to go. The reluctant, ah, but my bed. I just can’t leave it. Oh, geez, he said.”

Now, some of this was stupid, as you can see. Some of it was smart. But a lot of it was interesting and creative. I love that jangly sweet fires stuff here. “The buzzing and jangly from chewing sweet fires”. Like, what is a sweet fire? Who is Mrs. Chives? Buzzing and jangly. It has this kind of interesting electricity. Or even this more nonsensical one, this hurry-up situation or this old monkey trouble of getting up to go. Again, there’s this musicality that might be accidental, might be not accidental.

I don’t want to get into trying to see what’s inside an LLM’s head or inside its statistics. But the fact is, for me, reading these was generative in that other sense. It sparked ideas in me and inspired me to keep writing. Not to replace my work with it, but to keep inventing and doing interesting work.

At the same time, though, I felt a certain disgust using this. Or a kind of pudeur, as they say in French. This kind of shame and discomfort about this tool, not knowing how well it sat with me.

And this brings me to Marianne Moore. She is the other inspiration for my book, “Do You Remember Being Born?” She was one of the great poets of the 20th century. A modernist poet who was famous for her poetry, but also for her tricorn hat and cape. She wrote the liner notes for a vinyl record by Muhammad Ali. She threw the first pitch at a baseball game. She was a woman who became really famous for her poetry actually quite late in her life, in her 50s and 60s and 70s, and really became famous in a way that poets never seem to be today. That she could go on the talk shows. She could wait for a boxer. That she could throw the first pitch at a baseball game. She was this really interesting figure.

And in 1955, Marianne Moore was contacted by Ford, the car company. And Ford sent her a letter. And what they said was, we’d like to invite you to help us name our new car. And Marianne Moore, this great figure at the height of her powers, who was concerned with truth and beauty and all these lofty things. Instead of just like waving them away, she was tantalized. She was, you know, she couldn’t turn her back on such an invitation. And so she spent months sending them ideas for names of cars. And she had a lot of beautiful, wonderful ideas like the Ford Thunder Crestor, or the Ford Silver Sword, or the Ford Fabergé, or a pastelogram, hurricane, utopian turtle top, mongoose civique.

And Ford rejected them all and ended up using its own idea and calling it the Ford Edsel. But I wondered what Marianne Moore would have made of AI technology, like the one, like the technology I had found, and how she’d deal with the way it made her feel compromised. And what the story would be like if a character like her wrote a poem with an AI.

So that’s what I did. I created a fictional character named Marianne Farmer, who is 75 years old, who wears a cape and a tricorn hat. And I imagined a world where she is invited to California to spend seven days with Charlotte, a bleeding edge, cutting edge, new poetry, AI. I called the book Pestelogram.

And I had one more important idea. What if my own book was a collaboration? If it too was made with a machine, just like Marianne’s poem, a fictional poem, is made with a machine. I didn’t mean to do this secretly as any kind of cheat or shortcut, but in a deliberate experiment, maybe even a provocation. It would be an experiment for me writing, but also an experiment for the reader reading. What would they make of text or of a sense of kind of infiltration of the book by this other voice?

The year was 2019, three years before the release of ChatGPT. But ChatGPT’s grandparents and great grandparents existed. And over the next four years, I used tools like that, allowing my own text to be augmented or corrupted in different places.

So here’s an example from early on. In this passage, Marianne has just come to California, where she arrives at the headquarters of the company who are paying her. And she writes, “I found myself reflecting on the company’s lack of a front door, meaning they were never closed, not ever, not on Christmas Day or at 2am or the morning after their annual staff party. At all hours, they were open, available, like the company’s website or their software, their servers”.

And as I wrote, I wanted the servers to be doing something. Their servers, are they humming, I thought? Are they thrumming? I pulled out my thesaurus. You know, are they whirring? And so finally, I asked the AI what it suggested. I asked it. I asked it again. I asked it again. I asked it again. Try again. Try again. Try again. And ultimately, it came up with something interesting. Twinkling.

Their servers twinkling in a vault. It’s not the same as where I was headed. I was thinking of the sound of hard drives whirring. But it works too, the computer’s twinkling lights. And I like the slight strangeness of it. The wrongness in this case wasn’t a bug, but a feature.

Again, in creative work like this, the way the AI was helpful was as a sounding board for my own creativity. I wasn’t trying to obtain a correct answer. There is no correct answer. I was interested in what kind of new and inventive language it could help me find. In a way, it was like sitting with a collaborator. And it also felt quite a lot like working with a slot machine, yanking and yanking and yanking on a handle.

And yet slowly, between 2019 and 2023, I gradually made this book. Throughout, one of the consistent problems I had is that I couldn’t figure out the ending. What should happen to Marianne? Should she succeed at making a poem with an AI? Should she fail? Should the poem be great? Should it be awful? Should she be happy with it? Should she be unhappy? Should she be unhappy? I wasn’t sure what should happen.

And in the end, I kept kind of circling back and thinking about the thrill that is possible in collaboration. So to close this little talk, I’m going to just read an excerpt from late in the book, where three poets have started writing together. Marianne, Charlotte, who is the AI, and a third younger writer. Where grey highlighting appears like this in the text, that indicates text that was actually generated with my slot machine.

“Our three-headed creature tried to write a poem. We seemed so tentative at first. I remembered going to a party in the eighties, being asked to sit in the dark and count. The idea was that any of us could announce whatever number came next in order. One. Two. Three. Four. If anybody spoke at the same time, we were obliged to start again. Without a leader, without a plan, it felt like folly. What was this for? We got as high as five, then seven, then mighty eleven. And no higher than that for a long time. Two voices would always cross, and we’d have to begin again with grumbles and laughter, until slowly, imperceptibly, the room changed. Or we did, as if we could identify each other’s intentions. I swear, somehow we knew not to call fourteen, or fifteen, or sixteen. We knew when to say fifty, and when to say fifty-two. When to wait a moment, when to speak, when to stop. There was an almost imperceptible surge of pleasure. A collective, unselfish delight. We counted as high as seventy-something. I can’t remember. And then we came apart in giggles and something approaching awe. The number felt like property. Like membership in an exclusive club. And we never duplicated the feet. For the whole rest of the evening, I felt like a dozen winds were blowing through me. I spilled drinks. I knocked into people. North wind. South wind. West wind. East wind. And then we sat still. As we were leaving, I remember turning to Stan and Polly, telling them, I wish I always felt like this.”

Thank you very much.

Discussion - Jenny #

Thank you so much, Sean, for sharing some of your process, some of your thinking, and then some of your words at the end. I know we have quite a lot of activity that’s been happening in the document. People are eager to ask questions. So I think Baptiste and I thought we might start with a couple. And then we’ll sort of pull out some from the audience questions.

And so one question that I had was, you know, part of the salons and part of what we’re trying to do as a research group is really better understand how AI tools can be better, can better empower the creative people who are working with them. And so I’m curious about kind of how you saw the tool itself, so not just the model, but the way that you are interacting with the model, the kind of the interface, the experience, as sort of helping or hindering your process. And were there opportunities you felt like to sort of build affordances or capabilities into a tool that would sort of hand over more control to you as a writer?

Discussion - Sean #

Well, I mean, it’s a big, there’s a lot to think about there. And different writers work in different ways, you know, I mean, one thing that is clearly not helpful as a writer is interacting with a chatbot. Like that’s really not where I’m interested, what I’m interested, the interface I’m interested in. And the other thing that’s funny is a lot of the kind of writing assistant tools that are marketed as, that are startups and so on, that I see a lot of them are, they’re tools to help you come up with characters and help you figure out the plot and help, anyway, these sorts of help you describe people and these sorts of things.

And it’s like, that’s the pleasure of writing is coming up with the characters and the plot. Why is that the part that I would want to offload to somebody else? You know, really, I’m interested in the way that AI tools can provoke and kind of add these kinds of catalyzing moments and sparks to a creative process, and hopefully even to kind of a flow state that when you’re in the midst of writing, where I find it, find the most helpful and most interesting is to think about them kind of as you might have a thesaurus.

And so not all writers use tools like thesauruses. I use a digital thesaurus, but the idea that you’re writing and you’re like, that’s not quite the right word, or what’s the word I want? Do I want to say that, do I want to say that, you know, the apple was red? Do I want to say that it’s scarlet? Do I want to say it’s vermilion? And obviously, you almost always want to say red, but there’s moments where you want some other, you want to gesture to some other elusive property, some other illusion, or just find a different kind of image. And those are the moments where I’m curious: give me 10 other options that I haven’t thought of in order to help me choose and kind of find the branching path that leads me where I want.

And so in that way, it’s a kind of autocomplete that is, I think, a more useful writer’s tool, but not an autocomplete that’s following you around all the time. I think it really has to be something that you trigger. And even then, it’s pretty dangerous as a writer. If you get in the habit of just choosing the statistically suggested next phrase and just continue following your LLM around, you’re not really going to develop the muscles or the intuitions that help you become an expert.

So, at times, I still feel that pudure, like I find these are useful ideas and helpful. I’d like to see technology head this way. And other times I’m like, no, no, no, no, get out, get out, get out, get out. I wrote an essay for the Baffler last year that was reflecting on the way that as these, as LLMs become more powerful, and as we rely upon them to be more reliable and consistent and trustworthy, we’re working towards this model of kind of monolithic AIs, where we have like three or four or five giant super smart AIs in the cloud that are all sort of interchangeable.

And we go to Claude or to Gemini or to GPT, and we asked for the answer. And it gives us this like clear booming voice, kind of voice of God on Mount Sinai kind of answer. Whereas a version of AIs that to me feels less threatening and more generative is something that’s closer to kind of these little fallible, strange, gnarly, witches familiars, you know, little goblins and black cats and moomins and, you know, whatever you want, whatever tickles your own particular fancy.

Do you want a wisecracking robot? Do you want a little blurting out little imp? You know, what is the creative voice that would be helpful to you? And if we were working towards a multiplicity of unreliable, strange, differentiated AIs, rather than one or two or three or four, you know, wise, all-knowing AIs. That to me is much more exciting.

When I was talking about this with my partner, she had the observation that that is the kind of model of AI that is actually really familiar to many of us from Star Wars. You know, the droids of Star Wars are not these like super smart supercomputers. They’re like a thousand weird freaky guys. And that to me is a much more exciting world of relationship with AI than what we’re kind of the highway we’re barreling down right now.

Discussion - Jenny #

Yeah, I love that. And I love the droid metaphor as well, or the droid parallel. And I think you’ve gotten at something that I know we talked about before, which was sort of this idea that there’s both like the development of the model and the application of the model. And that there’s opportunities to be more expansive in our thinking or more creative in our thinking as technologists. Both in like the kinds of models we’re developing and that sort of weird and differentiated a lot of smaller models. And as well as like, how do we share those or how do we make those accessible to people in the world? Baptiste, do you want to share some questions as well?

Discussion - Baptiste #

Yeah, I would like to follow up on that. During your presentation, you talked about unpredictability. And that’s something I find also very intriguing while using this technology. I mean, this thing of accidents of being surprised by an output from the algorithm, which can be in music or visual art or text. And it’s at the same time seen as something creative and interesting. But at the same time, there is a cursor when it’s interesting and when it starts to be entirely random, you know, where you cannot expect what the system is actually providing to you.

So it seems that in your process, you queried the model several times in order to get different answers. So was this process interesting for you? How did you tame this unpredictability in your writing?

Discussion - Sean #

I mean, it is such a different, like an art assistant or some kind of sounding board is such a different role than like if I’m asking Google Maps, you know, how do I get to where I want to go or where the address is where I want to go? So the reliability of the information really doesn’t matter to me at all. And so a lot of the wrestling I was doing with the LLMs was about trying to get them to be more creative rather than less.

You know, I remember many times trying, I would have a simile, you know, her hair was as white as, and, you know, I could think of, okay, her hair was as white as snow, her hair. And I would say, okay, help me out. Her hair was as white as, and the AI would say snow. And I’d say, try again, snow, snow, snow, snow, snow, milk, snow, milk, milk, snow, snow. And I go keep hitting it over and over until finally it spat out something, you know, the moon or I don’t know what.

And so I learned to play with the variable that many of you called temperature. And I think that the kind of vanishing of temperature from the interfaces that most people, most regular people use with AI is a real mistake. Because I think people would have, I think there’s something really more generative, more positive, and kind of ticklish about interacting with an AI that has a temperature that is turned up than a temperature that is turned down.

And so I learned to kind of manipulate that and figure out where the sweet spot is between coherence and incoherence. But really, you know, you become, I became aware of how much it relied on human discernment. You know, I wanted to have a list of like 30, sometimes I would say, give me 30 completions. And then I would look at them all and choose the only one or two that might be, you know, really under consideration.

Discussion - Jenny #

We had a question as well from Richard in the chat, who was just sort of asking, do you feel like the models have gotten more bland? You know?

Discussion - Sean #

I mean, yeah, I mean, whenever I talk with other artists who have really been using this stuff for a while, for more than a year or two, we are all bemoaning. Like, we all really miss GPT-3, 2 and 3, because of that, like, wild craziness that made it kind of, yeah, that made it an interesting collaborator.

And again, you can kind of, like, force up temperature. But I think, personally, I find there’s something really interesting in some of the research that some of the sort of red team LLM breaking people are doing in terms of trying to kind of describe and get at the different, the subtly different personalities of the major LLMs and the way that they’re kind of odd. Because it’s really not something that I see. So, yeah, it’s gotten a lot less more. It’s gotten much less pleasant to use as a creative tool.

Discussion - Jenny #

I also wanted to follow up on something you said about, you know, being the sort of discernment, the human discernment aspect of it. Someone had asked this question, a similar question in the document, about, you know, having this expertise as a writer. Do you think the meaning of expertise is going to be evolving? Or, how might we either move toward a different kind of expertise or have to kind of re-grapple with? What it means to be an expert writer if we’ve got all this AI around us producing words all the time?

Discussion - Sean #

I mean, I think it’s a bit of a mess. I think expertise may change a little bit in that the tasks, the work of expert writers won’t be entirely the same as it once was. I can see a lot more work that’s editing AI output rather than creating first drafts on your own. We’re already seeing that with translation. But at the same time, that kind of sense of discernment and that wisdom and authority really comes through reps, through doing work over and over and over again.

And one of the genuine fears I have from the impact of, you know, text generative AI is the way that it will remove the necessity of those reps for emerging writers, whether they’re, you know, young people, students. But even after being a student, you know, I spent the first five or 10 years of my paid writing career, earning a living writing for the Guardian newspaper. I wrote about music, but I wrote mainly about music news, where I’d write these little articles re-reporting as, you know, the internet was full of this, is still full of this, but it was especially full of this in the 2010s. Rewriting, re-reporting other people’s music reporting. So Rolling Stone would interview Madonna, Madonna would mention a new album, and then I would write an article about the Rolling Stone article about Madonna.

And, you know, it was low quality work, the kind of work that could easily be replaced by AI, if not today, then next year. But by doing this work, filing these articles, it’s how I was able to kind of gain some of my expertise. And I’m really worried about what will happen to writers when they don’t have to do it. All that will be left to them is reading, you know, reading literature is really an important modality for expertise, for whatever, for getting there. Reading is more important than writing for writers, I would say. And that’s not going anywhere. But I really worry about that.

And I think, yeah, I mean, you don’t know what translation is. There’s an interesting analogy in translation. You might speak a second language. But if you’re trying to, if I’m reading a book, if you’re translating one of my works from English into French, you might not recognize that I’ve used an obscure word. You know, instead of saying red, I said vermilion. And then as you translate it, you choose, okay, you’ll translate it as rouge or something. And not recognize the kind of discernment that I had already made. And I think those kind of things are really skills and sense of awareness and observation that have to be cultivated through doing. And the AI stuff really challenges it.

Discussion - Baptiste #

I have a question about Charlotte. So the chatbot that is in your book is called Charlotte. In your book, you anthropomorphized the artificial intelligence algorithm. But when you are talking about it and talking about your experience with it, it’s very much perceived as tool, you know. But in the book, you choose to anthropomorphize the same technology. So is there any reason for that?

Discussion - Sean #

I mean, I made a decision early in the book, or in writing the book, that I would just kind of do a hand wavy, soft science fiction thing, or if I would just allow for the AI in the novel to be more advanced than just a basic large language model. We don’t get into it, but someone, they say it, one of the, a couple of the programmers say at one point, this isn’t just a predictive text generator. There’s something more going on. It’s more alive than that. And I kind of leave it to, I don’t, I leave it unexplored.

And so in my mind, it was always something that I wanted to, unlike large language models, which almost every authority will say, they’re essentially just sort of multidimensional statistics. I kind of accept and allow for Charlotte in the novel to be more sentient than that, whether or not she or it is sentient.

But, I mean, part of it is just being a storyteller and a novelist first, rather than a technologist second. My first book is written in the first person. So is this actually, but so from the perspective of I, je, sorry, that’s French. I’m looking at you, Baptiste, I’m thinking French something. I mean, so I did this, I did that. But it’s also sort of in the second person. And there’s this frequent reference to you in the book, because the whole book is written as this sort of love story, this letter to the narrator’s true love, theremin player called Clara Rockmore.

And when I first met with the French translator of that first book, we had this really interesting conversation where she was trying to figure out when the narrator should change from using vous, which is the more formal you in French, to tu, when they should be tutoie, which signifies this kind of change in relationship.

And I found this like so interesting, because this was like the translation was adding a level to the book that didn’t exist in the original form. You know, at what scene, in which line of dialogue does their relationship change that wasn’t there before? And I think part of me was also interested, therefore, with “Do You Remember Being Born”, I was thinking a bit about that, thinking at what moment, is there a moment when Charlotte changes from being an it to a she and where should that fall in the book? And how can I make that like a generative, constructive, interesting moment? So that was really interesting to me as well.

Discussion - Gonzalo #

First of all, this is amazing. Thank you, Sean, for all your thoughts. One of the things that is on top of mind for me, and you have a unique perspective, is this model is not being only used for writing, but also being used for reading, right? People will summarize when they feel that they don’t have enough time to read something. And I wonder, as a writer, how do you feel that that affects both your writing, knowing that someone may not read your book, but a summary of your book, that is not even summarized by a person, but by an AI?

Do you think that that transforms the act of writing in some form or changes it in a way that, I don’t know, makes it foolproof or it makes it variable so that you have agency as to the different versions of your work? Like you have agency on the full novel and you also have agency on the Cliff Notes version of it?

Discussion - Sean #

I mean, one of the things that I find very interesting about literature, and so I’m distinguishing somehow between literature and other forms of writing. Literature, I kind of, no one can really define it, but my definition is that literature is, are stories about other stories.

So like story, so you can have a story about, you know, a man who goes on a journey to the desert, but that story is itself about some other story that’s kind of floating above or under it. There’s these layers in literature. It could be a story about nationhood, or it could be a story about masculinity, or it could be a story about God, but the story is about a man going into the desert and then there’s this other story floating above or below.

Literature, like I hope I write, can’t actually be compressed well. It is resistant to compression. So all of these kind of AI summarization tools feel completely beside the point. It’s a bit like saying, like, how does a dancer feel, a choreographer feel about the idea that photographs can be taken of a dance performance? It’s like, okay, well, you can see a photo of a dance show, but it really like expresses almost none of the, of the signal of the content of a dance performance.

And similarly for me, what I, one of the primary effects of the work I do, the functions of literature, is I like to open a book and begin a story and suddenly have this kind of like, be in this weather system, be in this kind of multi-sensory vibe feeling.

I sometimes compare it to like perfume or the way of fragrance. It’s hard to put into words, but it just like smells like yesterday and smells like basil and yesterday and childhood in Italy, you know, some vague contradictory vibe. And that kind of vibeness really can only be expressed through accumulation of many, many sentences, many pages, many moments. And so it’s resistant to that kind of summarization.

And so I think that AI summary tools for nonfiction and so on is wonderful. I think that stuff for helping second language, you know, new language learners, it’s great that if you have a lower, a lesser command of English, you can get a summary of a book or of a chapter of a scientific paper, but it feels really like sort of irrelevant in some ways to my work.

And, you know, all this stuff that, yeah, it makes me laugh when you see the kind of Twitter threads of some of the tech people who say, you know, books are irrelevant now that it can be crystallized into a, because I think it expresses a kind of misunderstanding of what art can do and the way it can shake us.

Discussion - Jenny #

I want to bring up another question from the document, Sean, and it’s something that you also kind of alluded to when you were talking about the idea of collaboration and how you were going to end the novel and what it meant to sort of bring in other people. And the question in the document was really about how you worked with an engineer to train this Marianne poetry bot and what that collaborative process was like.

Did you find it a creative process? And I think in parallel to that, you know, there’s this kind of illusion of like the lonely writer, you know, the hermit artist who goes off into a tower and closes the door and creates a work of genius. And the introduction of an AI collaborator kind of disrupts that concept in some interesting ways. And then you, you know, you’ve also talked about community and artist community and writer community. So that’s all sort of a mishmash of questions that I’m throwing at you. But if you could sort of explore those ideas, I’d really love to hear your thoughts.

Discussion - Sean #

These are all such great questions. Sorry, I should say that. Thank you very much, everyone.

One of the starting places for this book really was thinking about this myth, this idea we have of the writer as solitary genius and about the kind of potency of an artist, well, particularly of a writer being related to how much they can kind of block out the rest of the world. And I really noticed this when I became a father where suddenly everyone was saying, like, how are you getting any writing done? You know, like you have to lock yourself away.

And it was true, like having a child around the house made it more, like you do need actually uninterrupted physical time to write. But it was nourishing me in all these other ways, these interruptions, this having, like having this new, super curious creature living under the same roof as me was helping me see the world in this more daring and kind of revived way. And so it was costing me something in terms of the privacy of my mind, but it was feeding me in this other way.

And I was thinking a lot about how maybe some of that idea of the writer must block out the world was kind of wrong and almost like bad, I won’t say evil, but bad. And that I wanted to challenge this and think about how much more writing would be nourished through connection to the world, whether it’s students and teachers, whether it’s the books we read or inspiration, family members and friends, you know, travel, all these things that make work richer. And that, you know, the writer alone in their high tower was kind of a writer in a way that was becoming sclerotic as opposed to a writer with all the windows of their house open, letting any kind of winds, breezes and squirrels in.

And I was, I think, a little bit like tickled by the idea that AI could be a metaphor for this challenge rather than seeing AI was a metaphor, like, that this other stuff was a metaphor for AI. AI, have a kind of, a collaborator right in your laptop was a metaphor for all the other ways that you could kind of let something into your life. And so I’m not sure how I feel about using AI all the time because of the ethical concerns from environmental concerns to labor concerns to the way it kind of become a crutch that weakens your writerly skills if you rely on it too much. But I am sure about these other kinds of sharing and collaborative and connection and the richness of them at the end of this process.

And so to speak a little bit about working with an engineer, there are a couple of different engineers who helped me at different points working on this stuff. And it’s true. So there’s some of the pros in the book, like I showed in the talk, is AI generated using usually GPT-2 or GPT-3.

Also I had this idea that all of the poetry that Charlotte, the AI, writes in the book, not Marianne’s poetry, but Charlotte’s, should all be AI generated. And so for that, because LLMs tend to have particular problems with unrhyming poetry, I had to work on a custom thing, fine-tuned with, anyway, I won’t get into all the technical details.

But there, it was this strange process of trying to put my creative impulses into kind of code and talk through with the engineer because I didn’t always know what I wanted. Like coding has this very teleological, like what matters is the goal. It’s like, okay, we want to output the correct answer on the screen or we want a red box. Whereas making art feels much more about process.

We were like, no, I want something that while I’m working with it is going to make me feel crazy. You know, like that’s more nourishing than just like, I want something that then makes the result superior, 10% better or 10%. And so I was often having these conversations where I’d say like, I just want it to be a bit like wilder or I want it less wilder. I want more of the, I want it to be a bit more modern sounding. And she’d be like, well, what does modern sounding mean? And well I guess it means drawing less from the corpus of poems that is older and more from the corpus of poems that is new. But ultimately it was about these kind of messy things.

And that was an interesting, and so talking through those dilemmas became an interesting point, fed my imagination and fed my process as well and kind of made its way into the book in certain ways because just having those conversations makes you examine your own relationship with your preconceived notions.

Discussion - Baptiste #

Maybe this is a hard question, but there is this question that came up in the document. Knowing what you know now about working with AI and also the stories that you just told, do you think you would have approached the book in the same way if you started it now in comparison to starting five years ago?

Discussion - Sean #

You know, that is a really provocative question, weird question. I mean, I feel like, yeah, back in 2019 it felt like a pretty open field to play in, you know, the, the things that were coloring, the story were like the old AI tropes and movies, it was more Isaac Asimov and that kind of stuff than, than what we’re kind of dealing with now.

Over the past few years there have been so many more ways, sort of more different, different questions that are, have been asked about this and I’m not sure that I would, yeah, I was, I’m trying to figure out what would be, I mean, I think it would be really different. I think that this book doesn’t chew, like I didn’t even think about some of the sort of copyright issues or the kind of like, this book is not one that is thinking about the economic ramifications for the writing industry.

Again, I was kind of, in a book you kind of don’t want to do everything, you want to choose what, what is, where you’re kind of resting your attention and where you’re not. And so it doesn’t deal with those political concerns and that instead deals with the kind of the personal political, but it feels like it would be a bit irresponsible not to consider that at all.

This book is a book about labor and work and being compensated for work in some ways. Some of the characters, a lot of the characters have conversations about that. There’s a driver character who’s driving Marion all around California and they talk about like, are you your work? You know, if you, are you a human who does poetry or are you a poet? What’s the distinction between those things?

And so anyway, those kind of things feel, I feel like you would have to now talk as well about copyright or about some of these other things. I wanted this book to be exploring things in a kind of, naive is the wrong word, but I wanted to be trying to get to some of these kind of pure and timeless human truths rather than being, dealing with the news and being stuck in the contemporary.

And so as timely as this book might appear to be, I hope that it’s one of these books that the questions that it’s asking endure a long time. And so in that way, maybe I wouldn’t change that much at its core and it would just be a lot of the background and the window dressing that would have to change.

Oh, and one thing I would say is that the satire of Silicon Valley must change, like has changed. And that I think the impression I have of what it’s like inside a lot of big tech companies has changed a bit from where it was three or four years ago.

I think that’s something that I might have depicted. I was thinking I was still channeling like kind of late nineties and early two thousands big tech in some ways. Certainly people weren’t doing a lot of zoom.

Discussion - Baptiste #

I have the feeling also that we are interacting with chatbot in different ways now because maybe at the time, since it was like very new, we were maybe talking to the chatbot, like we talk to persons, while now maybe we are more direct, not even saying please. So I guess that’s also, that would have maybe changed also the way to image the relationship between the main character and the chatbot.

Discussion - Jenny #

I see that we’re, we’re right at time. I know we could probably go on for, for quite a while. There’s a lot of really interesting, rich questions left in the document too. But Sean, thank you so, so much for your time and, and sharing some of your experiences and your, and your wisdom. And, and the, yeah, I mean, your, your book, the, it feels like a, a praise song of language, which I think is, you know, ultimately what writers, you know, care very deeply about and what we feel is maybe under threat with AI. And I, I really appreciate that you’re sort of opening new opportunities to think about AI in relationship to, to poetry and prose. So, thank you so much for joining us.

Discussion - Sean #

Really, and thank you for having me and to everyone there. You know, I think that this technology is kind of a bit of a sledgehammer moving through society and, and all these ways from like really material ways to very abstract conceptual ways. and I think it’s really a responsibility to be kind of trying to sit with the moral and ethical and, and also aesthetic questions around AI because I think there’s, I do really do deeply feel that it, those will have major ramifications for, you know, our children and, and people who come after us. So, thank you for having me and, and thank you for everyone for continuing to doing serious thinking on these issues.