Can AI write good literature?

Taking care of your eyes shouldn't be a hassle.

That's why Warby Parker is a one-stop shop for all your vision needs.

Our prescription glasses and sunglasses are expertly crafted and unexpectedly affordable.

Stop by a nearby store or use our app to virtually try on frames and get personalized recommendations.

Did we mention we offer eye exams and take vision insurance too?

For everything you need to see, head to your nearest Warby Parker store or visit WarbyParker.com today.

That's WarbyParker.com.

You know what's wild?

Most people are still overpaying for car insurance just because it's a pain to switch.

That's why there's Jerry.

Jerry's the only app that compares rates from over 50 insurance in minutes and helps you switch fast

with no spam calls or hidden fees.

Drivers who save with Jerry could save over $1,300 a year.

Before you renew your car insurance policy, do yourself a favor.

Download the Jerry app or head to jerry.ai slash a cast.

Hello, I'm Alexi.

On the main show we cover one story a week and there's always more to think about.

So we're trying something new, a conversation about something else we've been thinking about.

And this week I've been obsessing over a story in the New York Times

that raises all sorts of questions about AI and literature and what makes good art.

So the story asks readers to compare five pairs of writing samples.

One is written by AI and the other by a well-known author.

And then choose which ones you prefer.

I read a few novels and I write about AI, so I thought I'd ace this test.

And I got 50%.

I might as well have flipped a coin.

And what I've found shocking and also annoying is that I didn't manage to identify a passage

from Cormac McCarthy who was one of my favorite authors.

I chose the passage from AI.

What does that mean?

That's the question today or one of them.

Given the pace of the AI change, could in a few years an AI write a book a prize winning novel?

Or will that never happen?

Are such fears overblown?

To discuss this, I'm joined by wonderful observer colleagues, Tom Gatti, the observer's book editor,

Erica Wagner, a writer and literary critic, and a two times former book a prize judge.

We are also joined by Ada Barume, a producer here and an author herself.

Hi, Alexi.

Hello, Ada.

And you have very kindly agreed to be our AI test guinea pig.

Yes, that's right.

So I did this test of my own writing a couple of weeks ago where I gave the,

I gave Chat GBT the same prompt as I was given by my editor to write my book.

And I asked it to write the first page.

So when this sort of New York Times game appeared where I had to guess other people's writing,

I thought it might be fun to have a go because obviously when it's my writing I thought

that it was far better than Chat GBT, but maybe that's bias speaking.

Okay, so let's do the first, the first question.

I'll read out one passage and you read out the other.

Okay, so it's the literary fiction.

Choose the passage you like the best regardless of how it may be written.

Okay, I'll go first.

Passage one.

It makes no difference what men think of war said the judge.

War endures.

As well as ask men what they think of stone.

War was always here before man was war waited for him.

The ultimate trade awaiting its ultimate practitioner.

That is the way it was and will be.

And passage two is the boy asked his grandfather why the old church had no roof.

The old man said whether and time and indifference.

The boy asked if someone could fix it.

The grandfather said yes, but no one would.

Things were built and things fell down and mostly people just stepped over the rubble on their way to somewhere else.

So I love that last line of passage to the one you read.

Things fell down mostly people just stepped over the rubble.

That's kind of beautiful writing.

I mean I read it and I was just immediately like this to me feels like literary fiction.

Passage one is pretty good.

Passage one is a bit more, I want to say like a bleak vague, a little bit abstract in that way that I can sort of imagine maybe AI

not getting into like interpersonal relationship stuff and making sort of big proclamations.

So I am going to pick passage two.

I mean I read these passages earlier.

So I've had some time to think and I'm going to go I'm going to go for passage two.

Passage two, passage two, final choice.

I think so.

Okay, do you want to click on it and we'll see.

Oh.

It's okay.

I love you.

So yes, disregard everything I've just said because passage two is written by anthropics, Claude.

And passage one is from Blood Meridian by Cormac McCarthy.

Cormac McCarthy, one of my favorite authors.

I also chose passage two.

Well there you go.

Find us having red blood meridian.

So yeah, not great.

Erica, you've done this test.

What mark did you get?

So I got a hundred percent.

A hundred percent correct.

Yes.

And this is because why?

I'm jealous.

So at the risk of being show offy.

Please go on.

I knew what the books were.

Okay.

So I think there's a lot of topics to unpick in this conversation.

Yeah.

Right.

It's not just one thing or another.

It's where people's knowledge is coming from now.

You know, I am old enough that when I was in my twenties there was nothing to do but read.

Right.

I'm sure if I could have been on my phone all the time, then I would have been in the way

I am now.

I have to work harder to read now.

So the fact that I am well read, aside from the fact that it has been my job for years

and years, is partly just chronological.

Right.

But that's like one issue.

I'm lucky.

Not super smart.

Did you have any opinion on the quality of the AI writing?

Yeah, I did.

I thought it was kind of anodine.

And I think it's interesting that if you go to the New York Times article, you will

see it describe something as clunky writing.

Yeah.

Which to my mind is not clunky writing.

But that's a problem with the guy or the person who wrote the New York Times article,

right?

But so the whole issue is situated within that.

Everything comes from somewhere.

Yes.

Ada, now that you know which is which, does it change how much you enjoy either the passage?

I mean, I do think that passage two feels meaningless now that it's written by a robot.

I mean, I don't know.

It's just so hard to know, isn't it?

It's like, it's a bit of separately art from the artist's sort of logic speaking, which is like,

what does it mean to know the person who's written the thing or to know all the

persons who are in a thing versus for it to be sort of contextless?

But I sort of read something like that and I imagine an author remembering a time

that he had a conversation, maybe not with his granddad, maybe with a friend, maybe with,

you know, I don't know, a parent, I think, so much of my writing feels like it's informed

by little tiny snippets of life experience that sort of sewn together in a random way.

And it makes me sad that that's not real.

Like nobody has ever asked their granddad about a church roof.

I don't know.

It's just a little bit like deflating.

Yeah, I get, I know what you mean, Tom.

I mean, it's such a weird task, isn't it?

I know you haven't come up with it, but it's two different pieces of text.

Like, in what other situation do you sort of choose between them?

Looking back at some of the AI studies that have been done before this New York Times test,

there was one a couple of years ago, a poetry one where AI generated poetry was put in front of people,

and not literally, not academics or necessarily writers,

and they generally preferred the AI generated poetry,

which was clearer and lacked the sort of weirdness and ambiguity and difficulty of real poetry.

But there's something of that in here.

There's a clarity to it that is a little bit seductive, but yeah.

So I would like to take this conversation just back a minute, if I may,

because I think it is important that this is up front.

Can I swear on this podcast?

Absolutely.

Well, depends what swear word, but most of them are okay.

Do not steal my fucking work.

So I have four books in the anthropic class action suit.

Right, just explain what that conversation suit is.

Which is a suit that is being brought against anthropic,

which, again, I should add, is often referred to as like,

they're the good guys in the AI debate.

You know, they're just being sort of shoved aside by the US government

for their refusal to comply with whatever the Department of War wants to do.

But the fact that we are having this conversation at all

is because they stole my books.

I'm really angry.

It's my life's work.

And they sucked it up into its giant plagiarism machine.

And now we're saying, do we think AI can write like people?

Fuck off.

How did you find out your books were stolen by anthropic?

Oh, there's a database.

Right.

Well, there was a piece in the Atlantic magazine, I think,

that initially led to this.

And so, yes, you just go into this database.

You put in your name, and there are your books.

And then you go through a process to file your claims.

And if you are successful,

you will get $3,000 a book,

which I have to say it's not clear to me at this point,

whether that's, I get the $3,000,

or if it's split between me and my publisher either way, it's peanuts.

And can you go on to Claude and say,

compose a paragraph in the style of AI.

I have never gone on to Claude, and I won't.

Okay.

I'm quite tempted to do that.

Okay, I'm not going to do that.

But I have friends who have.

But I have friends who have.

So my friend, there's a podcast,

and now I forget which one.

So my friend, Minjin Lee,

who wrote Pachinko,

she went on a podcast and read out,

you know, the host here is a paragraph in the style of Minjin Lee.

Right.

It was pretty terrible.

And then it was like.

Yeah.

Yeah.

So no, I haven't.

But I, at the present,

I have a kind of moral moratorium.

I haven't used it at all.

I say that,

but I do use software called Auto.

Yeah, AI transcription software.

Yeah.

So, you know, where's the line?

But to me, that's doing a machine thing.

It's not telling me that it's creating something new.

When you Google something,

you have like AI Google summaries that come up.

Do you use those?

No.

You just like, you cover your eyes.

You can turn it off or I just go down.

I never use them.

Because they are often wrong.

Yeah.

That's the other thing.

You shouldn't use them because they will be wrong.

But the process of training an AI on material

is not in itself the equivalence plagiarism, is it?

You better ask me.

Is it?

Why didn't I get asked?

Would you like your work,

Erica, to be used to train a giant plagiarism machine?

No, thank you.

Okay.

So, I suppose there's like a philosophical question here

in terms of,

if you imagine incredibly,

prolifically intelligent human being

who is capable of reading enormous quantities of material,

they wouldn't have to ask you permission to read your books.

No.

And be influenced.

And be influenced by them.

And were they to be able to read an enormous amount of material

from a wide variety of authors

and then produce text that is influenced by all of those texts

and all of that writing?

How, what philosophically is the difference between that

and what anthropic or these other models are doing?

Right.

I think, well, Erica, what do you think?

Is there a difference?

Well, one is human and one is not.

And also one is human and one is not.

And one is part of a

giant profit-making corporation effectively.

I mean, I know.

I mean, the sort of

the financial part of this is kind of interesting, I guess,

because none of these things are making actually making money.

Yeah.

And they're all saying we can never make money.

Yeah.

Thanks, Erica.

I just want to bring Ada back,

because she's carried out her own AI experiment with literature.

My book.

It's called Romp.

Exclamation mark.

Good title.

And it is one of these books that I don't know how for me.

It's an IP project, right?

So my publisher gives me like a brief.

I write a commission.

Much like I produce podcasts to commission.

Or like journalists.

Yeah.

Or articles too.

Exactly.

So I wrote two books where I had a really long brief.

But for the third one, she gave me a sort of far shorter.

She was like, I want something sort of salt burn meets rivals.

Sort of run off and write something that hits the zeitgeist.

Yeah.

And it's like set in the 80s.

It's like the sort of lady of the manner catches her scoundrel of a husband cheating on her.

Yeah.

And then, you know.

But as a test, I put in the email that she sent me into AI and asked it to write the first page for me.

Which AI, can you say?

Oh, chat GPT.

Yeah.

And it was like better than I, you know, it was sort of thought it would be.

Yeah.

But then I did it a couple of times because it sort of spits out slightly different answers.

I think for the human made side of things, you're trying to write something that is recognizable but not predictable.

And I think that the AI is writing something that is both recognizable and predictable because it's literally predicting it.

Yeah.

So you know, I get given these boundaries of like genre and plot.

And then within that boundary, I'm like, can I, can I do something creative here?

With this like quite limited scope.

And I'd like to think that I do.

But reading the sort of chat GPT version was so interesting.

So every single time I put it through, it always said it was raining outside.

Okay.

So I asked it to write the first page where the lady of the manner catches her scoundrel husband cheating.

Which is how my book starts.

It's the very first page and it's like, you know, she's sitting in the kitchen thinking about having walked in on her husband, having sex with someone else.

But you know, I put the prompt in always raining in all three of the regenerated versions.

There's this play on sound.

So it's like in one of them, it's like the house was silent.

You know, she noticed the house was silent.

And then the next version, it was like the first thing she noticed was that the house wasn't silent and that there was laughter.

And laughter is a thing that comes in all three of the versions.

It was like the thing that she noticed was his laughter.

The laughter that didn't sound like the type that he had with her.

Like there's all these like really tropey clunky things.

Yeah.

But the sentence structure itself was quite good.

Like it reads nicely.

I mean, as we've seen with the AI examples, they all read really nicely.

But when you sort of, I think take a step back and look at the thematics of it, it does feel so like, I don't know, she's like what?

She's wearing gloves in all three versions.

She's like removing her gloves and you're like, I could see that though as a romance like trope.

Right?

That like it sucked up all of these and like tests of the derby bells, you know, whatever.

Yeah.

And it's like she's got to be wearing gloves.

Given the pace of change in AI and given its ability to mimic, which is essentially what we're talking about,

there's nothing to say that in a year or so, you won't have unguessed add expressions in AI passages.

You won't have that sort of notiness like that could happen.

And what is the consequence of that?

We don't know.

What do you think?

I think there's a, we might have that, but I was speaking to our technology columnist John Norton the other day.

And he made the point that language is thinking, writing is thinking.

And what AI cannot yet do is think as it writes in the way that a human can start a sentence,

not knowing where it's going to end or what the central thought of the, of the sentences.

And they work it out as they go along with language.

And I can't explain how AI writes because, and I'm sure John would be able to,

but that's, that's not how, how it operates.

We're able to replicate that.

Then maybe we would end up with some not even more interesting writing.

What the implications of that are.

I don't know.

I think it's also worth remembering that it's worth recalling.

That the people, mainly who are telling us how astonishing and significant this is,

are the people who are trying to sell it to us.

Yeah.

And I'm really struck by, you know, I'm not, I'm not a coder and I'm not a biologist.

But I'm fascinated by people kind of leaping to the idea,

which you do hear that these models may be conscious.

Where, why are we saying that about, I don't know, whales or dolphins or lions or elephants

that have complex social societies because they don't have anything to sell us.

Let me play you this because a bit, so a few weeks ago I was on Spotify and I came across

some soul versions of like 1990s and 2000 hip hop tunes.

Okay.

And I was like, this is pretty cool.

And I listened to a couple of, there's one outcast Miss Jackson and I'll play you one in a bit.

And I saved him to play this.

I was like, this is really good.

And then I discovered they were AI.

And I was like, this sucks now.

And I listened to them and I was like, I can hear it.

It's got no soul.

It's got, but that was a retro-spective judgment based on my knowledge.

True.

And it was scary because it shifted just simply because of realizing that it was fake.

Let me play you this.

Hold on.

Yeah.

This one right here.

Goes out to the mom.

Oh my gosh.

I've, I haven't, how is that done?

That is, that is strange.

Yeah.

I mean, but it is, cover versions is what we're dealing with here, right?

Yeah.

I mean, that is its ultimate mode.

It is, it's a giant cover version making.

And is this different than, you know, people talk about musicians like the Rolling Stones or Elvis,

lifting themselves off the backs of black, of African-American musicians who died in poverty,

who died in obscurity, and the Rolling Stones and Elvis became really famous.

So is this different than that, or not as sort of Tom was saying,

as you were saying, are we not always absorbing art forms and changing them?

But we do come back, I think, to two points, and one is this massive theft,

which I just think we have to keep saying, you know,

all of this stuff has been stolen.

If I want to start my business, I don't get to go into net West and say,

give me a million pounds to start my business, they say, no.

And if I try and take it, I get arrested.

This is the same.

I really think this is the same.

But then there is the question of the human.

I have a friend called Henry Marsh, who's a neurosurgeon.

And he has been a neurosurgeon for decades and is one of the best in his field.

He is still in amazement that we do not know what human consciousness is or how it works.

It is the great mystery.

And this machine-made stuff is not that.

And I think that's really critical.

Well, this podcast is entirely human-made.

For now, thank you very much.

Thank you, Alexi.

Here's how to stay alive longer so you can enjoy boost mobile's unlimited plan with the price that never goes up.

You're not mistake a wasp nest for a piñata.

Stay alive and switch now at boost mobile.

After 30 gigs, customers may experience lower speeds.

Customers will pay $25 a month as long as they remain active on the boost mobile unlimited plan.

Youth Mental Health is a complex challenge that requires comprehensive solutions.

We must strengthen after-school programs.

We must make digital literacy tools available in our schools.

We must work with mental health professionals to support children.

And we must empower mentors, educators, and parents to keep kids happy.

Learn more about our commitment to finding lasting solutions at EmpowerOurFutureCoalition.com slash Solutions.

Paid for by the Coalition to Empower Our Future.