The Women of Ill Repute with your hosts, Wendy Mesley and Maureen Holloway.
So Maureen, how's that book coming along?
My book? The one I'm- the one I'm reading, or the one I'm supposed to be writing?
Well, you said you're gonna write a book, you know, a complicated book, the one you're supposed to be writing.
I haven't started. But now I'm thinking that the easiest thing to do is just get AI to write it just like feed ChatGPT the basic plot outline, and add a few insights, and ask it to write in the style of Virginia Wolfe and bam, there's your novel baby, who's afraid of you now?
Well, I don't know, I think there'd be something missed, because I think you have a real story to tell. And so would it be- Would it really, really be as good as your story?
Well, that's the question, right? These are good questions. Well, would it be any good? What is good?
Well, our guests this week might have a few opinions on that. Stephen Marche. He's a writer. He actually used to be on my panel 1000 years ago when I did something on- Oh, he's nodding. He's here.
Only panel I'd ever drink- I've ever had alcohol on.
I know. Well, I was back at CBC. You're not supposed to be here yet.
But it's fine. It's totally fine.
Yeah, no, we had alcohol on- it was New Year's Eve. And we thought it was fine. But apparently, according to CBC board rules, alcohol was not ali- not allowed.
A crazy rule at the CBC? How could that be?
There's a couple of rules. Yeah. So anyway.
If I worked at the CBC I'd drink all the time.
Yeah. Well, yeah. You're not supposed to drink on camera. Anyway, he's written like books about everything. The Globe says he can write anything that he wants to.
Yeah, you guys. Stephen. You've written about human relationships, American politics, the British monarchy...
There's love stories. There's horror stories, there is fiction, there's nonfiction. And now there's like a lot of robot stories.
Well, this is- this brings us to artificial intelligence and creativity. So you've written a novel Stephen, or rather, you've created a novel. So even the terminologies has to change. It's actually a novella. And - and you called it Death of An Author.
Yeah, so we're gonna bring Stephen in in a second. So hi, Stephen. Anyway, he's just in this book, Death of An Author. He's used a bunch of AI programs to write it. And he he's credited, the author is credited as Aiden Marchine. I'm not sure I pronounced that right, which sort of is an amalgam of Marche and machine. I'm not sure why it's called Aiden. Well, maybe get to that. And it's arguably the first readable AI novel.
Yeah, what the New York Times said, the first readable. So the question is, is it any good?
Well, that's a good question. It's scary. It's interesting, and maybe, Maureen, it's the end of the world.
Okay, Stephen with no further ado, writer, and I'm going to say you're a literary technician.
I mean sure, whatever, no one has a word for what I did. So that sounds as good as any other.
I checked out your Twitter. I mean, you I follow you on Twitter, and you write- Oh, you- you've written so many books, and you say that you're an enemy of boredom. So you've written about marriage. I don't know what's- and we want to talk about AI. But you just you produce so much like, do you ever- do you ever turn off?
No, I mean, I'm just a really incoherent person, with- with no brand identity as a writer. I mean, basically, what happens to me is I just get sick with fascination at things. And I feel like I have to get to the end of them. And it's like, halfway to a pathology, like, it's I mean, it's not super healthy. Like the- like, I have three books coming out this year. It's not super reasonable activity to be doing that. But I just can't, I can't stop. I get obsessed with things. And I cannot- I can't turn away. Sometimes I wake up in the middle of the night. With AI for sure, I've been waking up in the middle of the night with ideas all the time. I mean, it's just so crazy and fascinating, right? I mean...
Just let me say that we're gonna concentrate on Death of an Author. But I mean, I first fell in love with your writing- I didn't even realize it was you. But I ordered your book, which was a pamphlet on writing and failure. And my takeaway from that was like, it doesn't matter about pub- being published, writers gotta write, and you're living proof of that. It's a compulsion. It's you've got to do it, whether you're successful at it or not. It's completely superfluous.
Well it's out of your control. Yeah. Right. So it's, it's, it's mostly luck. You're not going to have a writing career very long if you base it on outcomes, right? It's based on like how successful anything is, because that's just not in anyone's control. Right? So- but you know, I mean, my solution to that has been to write so much that something has to work.
Throwing it at the wall, right, eventually it'll stick. All right. Can you tell us- and I found it real- I found it fascinating reading about how you did this, but when you set out to write this novella, you must- you must be can- more than computer literate. And we have literally literate, you use three different AI platforms. Can you tell us a little bit about the process without getting too technical? Not that you did when you explained it. So it's fairly accessible.
I mean, you know, I'm doing this on a MacBook Air, right, with with a bunch of subscription to different services. So it's not like I have to be super literate. I mean, in the early days, like I wrote my first algorithmically generated story in 2017, for Wired, and a computer scientist built me a system for that, at UofT, Sci fi Q it was called. And so that was a really elaborate and expensive process to get at an algorithmically generated story. But since then, you know, the technology is sort of what's accessible. And there's a bunch of different large language models available, I use three, principally to write this thing, I use ChatGPT, particularly GPT Four. I would say, write in a buried paragraph, with simple and compound complex sentences, in the style of whatever, the following information, I would list the information, it would do that, I would then cut that and put it in a program called Pseudowrite, which is a stochastic writing instrument. And it allows you to select text, and then say, do things like shorten, add details. And then it has a Customize button where you can say you- how you want the text to be. So like, make it like Hemingway. And it does a pretty good job at it. And so I mean, it requires refinement, right, but like- and then if you do that, and then you cut and paste and say, make it sound like Raymond Chandler, then make it sound like Hemingway, then make it sound like F. Scott Fitzgerald, you might get something pretty good. So what I wanted was like, like a Raymond Chandler kind of detective story, like a driving plot driven story. But I also wanted good lines, you know, because Raymond Chandler has these lines, like it was a blonde, a blonde to make a bishop kick a hole through a stained glass window. And like I wanted, like crazy lines like that. And for that I use Cohere, which is the Canadian large language model company. And that was a different process, where I created prompts like, write a metaphor for the smell of coffee, and then I would train it on different examples of good metaphors for the smell of coffee, and then it would generate its own, and then I would just regenerate it until I found a good line that I liked. I mean, that's obviously too labor intensive a process to use for a whole book. But you know, the best lines in the book are all from Cohere. So those were the three AIs that I used to, to generate this text.
You talk about it, sort of with great affection. And there's been a lot of feedback, obviously, to the Death of an Author, a lot of people saying that, well, it's not real, it doesn't have any wit, it doesn't have any humor. And yet, it kind of does. And other people are watching you and saying, Well, you can- the most interesting part is watching where Steven, also known as Aiden, in the book is guiding the dialogue. So I just find it really interesting the comparison that you made after all this feedback. It's like hip hop. It's like the next era of creativity. So it's good, but it's also terrifying. Are you a fan? Because now obviously, everyone's come forward and said, it's the end of the world.
Certainly, one thing that I don't think is ending is creativity. I just don't believe it. I mean, instead of before, they said the camera was going to end art, right? And in some ways, the camera was the beginning of art. We think of technologies as world ending, but actually, they're just transmutation. They're just interfaces. They're just, we create a new way to do that. I mean, this work is absolutely my creation, like no one can go and do that. What I just did. The machine that I use, I do have affection for it particularly Cohere, which I have a kind of like, feelings for.
Oh, dear. Another book.
Well, it's- it's the way that that a violin player would have a relationship with a Stradivarius, or like a guitar player would have a relationship with like a Gibson flame guitar. It's a very beautiful instrument that's capable of producing really powerful experiences. But you know, when I wrote this, on the one hand, it is- I am in control. But the really interesting moments are the moments where you can feel the machine, and its relationship to language, like pushing out against the control and taking you to new places aesthetically. I'm not saying those were- that was every day, like that was probably 20 or 30 times during the entire writing process. But in those moments, I really felt like I was getting an alien to write a short story. And that alien also was the sum total of our language. It has new things in it for sure. I'm just the very beginning of this, but like someone is actually going to figure out- and it might be me, but it could also be someone else, like, what this is specifically for. I think that'll take some time. But it will be a new creative form, like it will- it will be a new art form.
Let me ask you about the actual, tangible- it isn't, there is- as far as I know, it's only available as an e book or an audiobook, but there's no actual print book.
Correct. Yeah. Well, it was because we were- like Jacob came to me with this idea in January, like you're interviewing me about it in May. I mean, you're writing a book, how long is it? Like we're talking about conception to release in three months.
Years. Years, normally. Years!
Every other book, it's been three yea- I mean, I actually don't know if a book has been this quickly produced since 18th century. I mean, we did this immediately. Right? So you know, print is kind of out of the question on those terms. You literally just can't find the presses.
But are you afraid? I mean, we've got all of these people who are huge supporters of machine intelligence for- for so many years, worked at Google, were part of promoting artificial intelligence. And they're now saying, it's out of control. There's no rules, there needs to be rules. Do you see a danger?
No. I mean, you have to understand, I've been writing about this stuff since 2017. One thing you have to remember about Doomerism, in tech, generally, is that it's a form of advertising, right? Like when they say We Work is going to end downtown cores, there's not going to be cities anymore, there's just going to be We Work. That's a form of advertising for them. When they see crypto is going to end the control of the central banks and therefore, government control of fiat money. That's, that's an ad for them. And when they- when they say the metaverse is going to end face to face interactions, that makes them feel like they're big and important. But since I've been writing about it in 2017, I've heard that the trucking industry has about a year left, I've heard that China has a trillion parameter artificial intelligence that is capable of superhuman control. I've heard that they shouldn't teach Radiology at universities anymore, because AI has already solved it. None of those things have come to pass.
But there's lots of true stuff that has happened. I mean, we see automation everywhere, and you've got like these really- I know you're really smart. But there's lots of really smart people who are going, it could mean the end of the human race, it's like a nuclear bomb. It's not like some kind of robot's going to take over certain jobs, it could mean we're not in control anymore. And as a control freak, I'm a little bit worried about that.
One thing you have to understand is, these are algorithms. Everything that you hear about natural language processing is just text prediction. That's all it is. It's not even related to truth. It's only related to probabilities. I mean, the pocket calculator is capable of doing arithmetic at a level that no human being has ever been able to do. I mean, we have chess programs that are capable of doing chess better than any human being, ever. And those are not the end of the world. And what this is, is the pocket calculator of language. One thing you have to understand, is that no computer has a will. They're all instructions, they're algorithms, I will say this, I've been around these engineers a lot. And I love being around them. God knows they're better than being around literary people. There's so much more honest and frank. And like, I mean, honestly, it's it's such a relief to be around them. But like, they don't have a super grasp of humanistic principles, right? Because they're, they're engineers. And so they don't- I mean, look at- look at Elon Musk with Twitter, like he's obviously a brilliant engineer, he also obviously understands nothing about how discourse works. And like similarily, like great geniuses, like bonafide geniuses indulge this apocalyptic thinking, it's just not very informed. To me. Also, these are the kinds of people who think that we might all be in a computer simulation. They talk these- themselves into these very... basically religious ideas about technology, because it makes them feel grander about themselves. And if you compare this to the, the atomic bomb, that means that all these people are Oppenheimer, who they are not,
Yeah, except that they're really smart people, Stephen, you know? Like, and they're people that have like, devoted their lives to developing these kinds of programs. And, and have been really proud of these kinds of programs. And now they're saying, we have rules- I used to be at CBC, I know all about rules- We have rules for everything else. There's no rules about this. And- and it's growing exponentially. So, of course, I mean, I worked in television, I remember when everyone said, oh, television, video games, whatever are going to mean the end of the world. And we're all still here. But you've got all of these people who developed the television, who developed AI, saying that it's scary and we are seeing, it's not just taking over, you know, you go to Shoppers Drug Mart or whatever, and you see the automated machine. There's real change happening.
But that's not AI. We're definitely in a technological- You know, we're in late capitalism here, right? Like things are not-
Well it's automation.
Yeah, it's automation. Facebook had to pull 2.2 billion fake profiles from its site in 2019. Right? So that's way before AI, or anything like it. There are lots of problems that we have with technology. Like I have a 10 year old daughter, she's about to go and get a phone. There's Instagram, it's like poison. We know it, and the authorities know it, and there is a solution to it. Regulating AI is more or less impossible, you have to kill it. If you want to regulate it, you're going to have to decide whether you want it at all. It's not fathomable, it's- and it cannot be made transparent. You cannot regulate the process because the process is not very well understood. You can only regulate outcomes. But we already have a lot of technological processes that we do understand, that we know are super harmful, that we do nothing about. The worry about AI, like, that it's going to create a misinformation crisis, obviously doesn't understand the point that we have already been in a misinformation crisis for a decade. It's- generation is not the problem. The political sphere, particularly in the United States, is so saturated with misinformation that I'm not sure AI can actually contribute much to it. So like the problems that we face are, first of all, they're political. Blaming the technology for them is a big mistake, because the impetus here is on what the political solutions are. And they all predate AI. The United States reached peak inequality, the highest level of inequality in its history, since 1774, in 2012. That's only a few years after the smartphone, nevermind AI. I'm not saying we don't have problems, we have tons of problems. I just don't think AI is actually going to contribute much to them. The problem is social media rather than AI.
The Women of Ill Repute.
Let's just pull back from the precipice of doomsday conversation and talk- I'm interested in the idea of cheap AI or bad AI when it comes to writing. And I'm thinking about the Writers Guild, there's a strike going on as we speak. They want the industry to either pull the plug on AI or, or not to rely on cheap AI. And this is getting back to the conversation about what is good. You know, sure, we can use AI for all sorts of things. But when we're talking about creativity and literature, in particular, how do we determine what is good? Just knowing that your novella was written by AI already made me predisposed to see it being soulless. Maybe it is, maybe it isn't. It's very- it's very hard to judge. But the experience I've had with AI and creativity is that there's something missing. There's something missing. Is it soul? Is it soul?
Well, what machines are very good at, like what AI is very good at, is imitating formulaic speech.
Boilerplate.
Boilerplate. I mean, it's incredible at it, to the point where if you say to it, write something in the style of a French semiotics professor in 1973. It's incredible. Like I would defy any writer to match it. It's better than any human writer at it. I didn't use this machine to write a plot, because I've never found an AI that could actually do a good plot. And I'm very familiar with all of them. The point of this experiment was, can it be good? On what terms can it be good, right? And there's certain things that it can't do, or can't do yet, or I'm not strong enough. I don't know enough to make it do them. I'm just not smart enough to make it happen. But there are certain things that it can't do, like, for example, that we wanted to do audio AI for this thing, right? Like, we were like, well, let's go- we'll go find an audio AI. But we were really unable to find one that anyone would listen to for two hours. Maybe six months from now, they'll have one, but there's audio AI that you could last for a minute or two and maybe fool you a bit. But like, if you're gonna use it to, to listen to for two hours, forget it. So we've got Eduardo Ballerini, who the New Yorker calls the voice of God. And you know who doesn't have to worry about a job? Eduardo Ballerini. Like they're a long way from being able to replace him.
So do you see a series? I mean, I think that's what Maureen wants to know is like- that's what the screenwriters are worried about, right? That that's gonna move in and take everybody's job.
Well, I mean, as someone who's actually used this thing for like, two months to write, the idea that you're gonna get a program where an executive goes and goes, Hey, John Wick 5, here it comes. That's not what you're gonna get. I mean, you know, John Wick's maybe a bad example, because it's so formulaic. I mean, also, we're in a period of the least originality in culture, maybe since the Industrial Revolution, right, like the last 10- the top 10 movies of the last year were all sequels. We're already in a completely derivative, cultural space. I mean, I think AI is possibly a way to lead us out of that derivative cultural space, exactly because it is an art of the archives, right? It comes from archives. But I mean, just on a purely technical level, like I just don't think this is replacing anybody. When you think of the stakes that are involved in a Hollywood screenplay, when you think of like how much money is riding on every word, you're just not going to trust a machine to do it. You might use it, it might give you insights into what work, it might be used as an analytic tool. Certainly, it'll be used in the process. Like it does amazing summaries of things, right? Like if you're an executive and you're going through 25 scripts a week, the summary function of ChatGPT has got to be useful, you know, and so it'll be stuff like that. But in terms of like, we're making a movie that we expect to earn a billion dollars worldwide, we're going to use a machine, you'd have to be out of your mind. They just will never take that kind of risk. Unless it has someone like me controlling it. Right? In which case there's a writer anyway.
Yeah, but the journalists, right, like there's so many journalists being laid off. And maybe that's a social media phenomenon, and it has nothing to do with artificial intelligence. I don't know, do- are not journalists being replaced by machines?
No, they'll never be replaced. Because, again, this is not a truth machine. They're being fired. But that's been underway since I mean, my whole life. Right? Like it's not like- it's not like- it's not like the decline in journalism is- the decline in journalism started... Pff. 1995? This is a probabilistic technology. It is not- this is not derived from the truth. So the only journalism that will value this stuff, is stuff where they do not care about the facts. The other kind of journalism that will use this stuff is stuff where they don't care about producing new information. I mean, I can absolutely see a journalist who investigates a murder for a local paper, gathers all the facts, goes out in the field gathers all the facts, goes home to ChatGPT and says, write in the style of the Baltimore Sun, a murder with the following facts and lists them, and then it generates them. And then he takes that and publishes it. Journalism is not exactly, like, the hotbed of style. It's supposed to be as frankly readable as possible. I wouldn't consider that a failure on that journalist's part. If he goes and gets the information- he or she goes and gets the information, has an opinion on it, puts that in the thing. And like, that's what- I mean, you're just fitting it into a formula anyway, why not have it be mechanized? But I don't think that's going to replace anyone. You still have to go and find it.
I think that the word opinion, the idea of opinion, I mean, AI does not have an original opinion. It can't.
No. It's completely derivative. What you can put into the machine is your will, and getting the machine to read your will, and your intention, is actually the whole trick of writing with it. Big surprise. That's the whole trick of ordinary writing as well.
Well, that's what I found so interesting when you were talking about your little love affair with one of the the bot machines. It's like- you didn't call it a love affair, but...
Cohere, yeah. I mean, I have affection for Cohere. cohere.
Cohere, yeah. And I just wonder, like, is it contributing anything new, or..? Because we've trained it, everything we've written, everything we've seen, everything- everything that's sort of available is what's trained these machines, and is there anything new coming out of it? Like, is there- or is it all human?
What you're accessing is the sum total of human language. When you get the power that comes out of that- and also on an unfathomable level, you know, I have a PhD and whatever, I've- you know, they made me do special fields, I have education and literary style, but it is accessing the sum total of human expression. So when- and you feel that, I mean, when you feel that coming out, it's an awesome experience in the, in the old sense of that term. It is an experience, that's sort of the sublime, and it's not particularly related to other kind of cultural experiences that I've ever had. You're plugging into something, that is then a new form of energy that's coming out of it, and, and what it is, is unknown. And what it is, is unfathomable. That's the future of this. That's the exciting part that's going to come out of that, how to- how to wrestle with that, how to- how to hold on to that in a meaningful way, that's meaningful to humans. Because remember, they might use ChatGPT, like a lot of the time that I'm using it, it just produces junk that I just throw out. And the reason that my book is better than other people's AI books, which it is, is that I know what a good sentence is. And I know what a good paragraph is.
So you're driving the boat?
Well, not only am I driving the boat, I know where I'm going. The machine doesn't know any of that. But the machine can get you places you wouldn't have gone before. And that's where the excitement comes in. But you know, as I said, that would only be a- maybe two or three dozen moments of that, in the entire experience of writing it. The rest is my control.
Death of an Author by Aidan- Why Aidan? I have a son named Aidan. Did you just like the name?
No, I mean, literally, we couldn't think of a pseudonym. So we went to ChatGPT and said, What's a pseudonym for a collaboration between Stephen Marche and a bunch of AIS? And it said, Aidan-
Ah, AI. AI and Aidan.
So it's a fusion of Marche and machine, which I didn't think of.
Yeah, that's clever.
So I mean, it was clever. It was a relatively clever answer by the machine. Do you see what I mean? I didn't think of Aidan Marchine, but it's- it's perfect. Right?
It is perfect.
So that's like a very odd thing.
So Death of an Author by Aiden Marchine is available as an ebook, or you can get it through Amazon and also as an audiobook narrated by Eduardo
Ballerini.
Ballerini. The voice of God.
He's a big deal Maureen, we should know this.
I know.
That's going to be our future.
The voiceover world.
Voice of God.
The voice of God. What- You've got three more books coming out.
This is number two.
Whe- anything else you want to tell us about that you're working on? No, you
No, you could have everyone buy On Writing and Failure. Tell them it's good and tell them that they should go and buy it.
Yeah. I bought it. I have bought it three times to give to friends of mine who write or are trying to write. Yeah, yeah. Well, it's easy, right? It's easy. I'm a huge fan. And I feel strangely optimistic after this conversation.
You know, no one's ever accused me of being an optimist in my life, right? I mean, my last book was On Writing and Failure, the book before that was about the possibility of a civil war in the United States. Like, it's not like I'm not prone to catastrophic thinking. I just don't see it. From all of my usage of it, from my seeing it develop. It's not going to replace anyone. It's going to replace low level clerks. In law firms. It's going to replace ad copywriters, Junior ad copywriters. It's going to replace maybe some basic research people, but almost all of it needs to be supervised. I don't think it'll even be used for mass translation, it just has that little factor of unreliability in it. What use is a journalist if they're totally hallucinating, wrong, one every 100 times? Now, you know, we've known some drunks in journalism.
On Wendy's show!
Just once a year.
Yeah, just once a year, they go completely off. But like, so I mean, maybe it's not as bad as them. But they're the ones get fired, right? Because it's like one out of 100 times, the drunken stupor is too much.
Well, as long as we remain in control, I'm okay. As long as there's like a role for us, and we're still feeding them, and they're not feeding us. That's, you know, I mean, some of the stuff you read sounds a little scary, but it's- it's nice to hear that there is- because you've argued that this is sort of the new sort of creative element. And, and we just keep going on. So yeah.
It's a tool. It's a tool.
A tool.
Yeah.
Like two turntables and a microphone. It's a tool.
Anyway, it's been lovely to see you and talk to you again, and-
Yes. Lovely to see you, too.
Yeah. And I can't believe that your daughter is 10. How old- How old is the other one?
17.
Oh, wow. Okay, well, good luck.
Yeah, right. Thank you.
Lovely to talk to you, Stephen.
Nice to have met you, Steven.
Lovely to talk to you.
Talk soon.
You know, I just feel like he's the smartest boy in the class.
I know. I know. It's kind of unbelievable. I mean, I tried to get a list of all of his books. I think the first one was, was something about Shakespeare, and then he wrote about The Unmade Beds. It was all about marriage. So I wanted to ask him about what's scarier, Civil War, artificial intelligence, or kids?
Or marriage.
Or marriage.
But he's also I mean, his essays is- we know we didn't give him the full resume off the top. But he's write- writes for The Guardian, The Atlantic, the Esquire. Every- he's just, he's, he's, he's a machine.
He is, and he's even written a great piece about the Royals. Yeah, so yeah, he just sort of pumps it out. And it's kind of amazing. He doesn't need AI. And yet, he's fascinated by it. So it's, it's kind of nice because we are in a period where it seems that a lot of really smart people who know a lot, a lot more than I will ever know about machine intelligence, are going, this is completely out of control. And he's like, No, this is great. I'm learning so much.
You've been obsessed- well, we're all obsessing about it. But I mean, you got snowed in the middle of last winter and you've just- you've spent too much time in your own brain, and you got really worried, and you're not alone. I mean, again to your point earlier, there are a lot of very smart people who see AI as a serious threat, particularly you know, within the military and so on. But when it comes to creativity- my takeaways is, is strangely optimistic after talking to Stephen, because he so clearly sees it as a tool, as much as a camera is, or as much as- you know, we will never lose our creativity and things like AI may- things like. There is nothing like- other than AI. AI can help us become even more creative, and...
So your book, are you going to ChatGPT it?
Now I'm thinking I might look into it, because the story is mine, but the meat and potatoes of actually, you know, putting pen to paper. And I don't know, it's interesting. I might- you might- we might all become better editors of our own stories if we, if we use AI to help us write them.
Well, that was what I found to be the most interesting thing was, was someone saying it was like watching somebody play hip hop. You could just see his fingers, you could see- see him like leaving at some- whoo, I'm sorry I keep touching the microphone. I shouldn't do that. I'm still learning still learning. Maybe I'll call ChatGPT get a little few lessons. Anyway, lovely to see you.
Lovely to see you.
Women of Ill Repute was written and produced by Maureen Holloway and Wendy Mesley, with the help from the team at The Sound Off Media Company and producer Jet Belgraver.
