August 29, 2024

"Oh, what if it was this contest between A.I. and me? We would both get the same prompts."

"We’ll give a list of possible ideas that readers can choose from, a combination of very summery stuff, like pop songs and sand and sunscreen, and also mixed in are things that I like to write about, like ennui and ambivalence and self-consciousness and privilege.... [We asked A.I. to w]rite a thousand-word short story in the style of Curtis Sittenfeld that includes these elements: lust, kissing, flip-flops, regret and middle age.... To see the speed at which it spits out sentences was the most unsettling part because, in a very biased way, I’m like, my story’s better. But it wrote its story — I don’t know — a million times faster? And my story is probably not a million times better. The speed is unsettling...."

Said the fiction writer Curtis Sittenfeld, in the podcast "Can You Tell Which Short Story ChatGPT Wrote?" (YouTube for NYT podcast).

You'll hear the beginning of the human-written story and the machine-written story and get a chance to notice the distinctive differences. I know that I myself want to learn how to recognize A.I.-written material so that I can maintain a healthy disgust for what is artificial. The uncanny valley must continue to feel uncanny... or what are we living for?

50 comments:

Lawnerd said...

It seems that half the articles on the msn website are AI, even though they list a named author. They seem AI to me because they state some fact then run with garbage platitudes regarding said fact.

ron winkleheimer said...

no "uncanny valley" tag?

Lawnerd said...

AI may have the potential of making law firm associates obsolete.

RideSpaceMountain said...

"...or what are we living for?"

The scary part is we might just have created something that in a few short years could answer that question. Do I want it to give us the answer? Do we?

chuck said...

AI material tends to be generic, i.e., lacking specificity. The details an expert might offer seem to be missing. In fiction, this might show up in handling emotion, or tricky details in the plot. That said, I have no confidence that I could reliably recognize AI, especially in short pieces. Might be interesting to see how well AI writes in the style of Saki.

Yancey Ward said...

Inside of 5 years it will be impossible to tell the difference.

ron winkleheimer said...

There was speculation that the first season of "Rings of Power" was at least partially written by AI because of idiotic actions taken by the lead character and her tone deaf emotional responses. But I think it was the result of the writers being hired due to identity politics and they were just that bad. So, how can you tell the difference between AI and bad writing?

Yancey Ward said...

The really creative human writers of today will be the ones that can more successfully use AI to write their stories, novels, and non-fiction. There is still a place for good human writers in the future, but just not as many such jobs as today because the actual market is always going to be human readers/watchers.

RideSpaceMountain said...

Old and busted: The Turing Test.

The new hotness: A General AI test that measures capacitance by its ability to crack jokes and insult us generally and personally.

Think about it. If AI can get to the point where it can make bj jokes about Kamala Harris or make orange man sad, I think you'd finally have something that would be indistinguishable from a David Burge, The Babylon Bee, or George Carlin.

That would be quite the feat.

Wince said...

From what I heard:

The author's story was told heavily first person POV: I, me, I'd...

The AI was third person narration.

Lem Vibe Bandit said...

Resist promises of utopia in all its incantations and incarnations.

Paddy O said...

With weekly responses and research papers in my undergraduate and graduate courses I teach, AI is definitely an increasing challenge. For the weeklies, I have the students connect the material with their life and context. That does a lot of the work to keep AI out, though I have had 1 or 2 students this term that I am fairly sure were using AI. Those generalities and lack of specifics. The trouble is that while there are online detectors, some better than others, they're like lie detectors, helping for confirming a suspicion but not accepted by schools as actual proof.

I tend to then just grade very strictly for those I suspect of using AI. Research papers are harder, but using Turn It In and I think that's getting better at recognizing. I have a student who I am fairly sure has been using AI for assignments tell me on Monday (when the research paper was due) that he wouldn't get it in until Friday, accepting all late penalties. I allow late submissions, but in this case I'm leaning toward the fact he did an AI research paper but it got heavily flagged by Turn It In so he had to start it from scratch, having done little original work over the quarter. He'll pass the class, I don't have institutional support to keep that from happening, but it'll not help his GPA.

And this in a theology class...

Paddy O said...

Of course, maybe I'm wrong about all this, and he just writes like that, and it'll be a great research paper. I keep an open mind.

Ann Althouse said...

"no "uncanny valley" tag?"

Thanks! I forgot I had one.

CJinPA said...

Does no one want to make a fortune developing an app to flag AI-created content, text, video, images, etc.? It would instantly reduce the potential downside of AI by approximately 1,000%.

Do I have to make that money myself?

Wa St Blogger said...

I cringed during the first story. I was convinced it was generic and had to be AI. After the discussion about which was AI, I knew better what the distinctives might be, but I was quite impressed with the AI as I expected AI to be even more cliche and generic than it was. I must say that I was not drawn in to the first story. I could not relate to it and thought that it was not very good fiction. I read a lot of fiction (not beach romance, to be sure), and I thought the author's story failed to draw me in, and that was what I was looking for as my delineator between the stories. I guess I should have used other criteria. Still I was impressed on how much detail was in the chatgpt story.

Ann Althouse said...

I get an "uncanny valley" feeling from the Kamala Harris that appears in public.

RideSpaceMountain said...

It already exists. Many are using AI to combat AI. They're call AI/IAs (Artificial Intelligence Investigation Apps). The problem is many researchers are beginning to suggest that all using them does is help train the AI further.

Kate said...

"So, how can you tell the difference between AI and bad writing?" An AI will have been programmed to know proper screenwriting structure. It can still churn out dreck, but the structural markers will be noticeable. Bad writing by a human is usually the result of no structure. They get carried away by their own genius and forget how to write.

Ann Althouse said...

"I cringed during the first story. I was convinced it was generic and had to be AI. After the discussion about which was AI, I knew better what the distinctives might be, but I was quite impressed with the AI as I expected AI to be even more cliche and generic than it was. I must say that I was not drawn in to the first story. I could not relate to it and thought that it was not very good fiction."

I didn't think it was good either, but when the other story came on, I could clearly see the differences. It wasn't even really a story. It was more of a summary or an outline of characters who could appear in a story to be developed later. We were just told generalities about fictional characters personalities. So and so is nervous or bold or some damned thing. Who cares?

I could have believed the first one was fake... if only the second one hadn't been faker.

Most fiction written by humans is bad. Now there will be much more, with the machines whizzing away. I'll just read old things, from the era before it was possible to take shortcuts to the kind of junk I wouldn't read anyway.


Kate said...

I love what people are creating visually with AI. (See the YouTube by @demonflyingfox called "Harry Potter - Redneck Wizard", for instance.) I'm curious if the ironic tone influences AI to learn to be more playful. Are we teaching it to be fun?

Lazarus said...

In the future you will be able to buy new novels by Dickens, Twain, Fitzgerald and others, all written by AI and probably more appealing than what living novelists will write. Some people will always read novels, but increasingly the novel seems like a lonely waystation between the factual material (reportage, essays, memoirs, anecdotes, journalism, history, biography, autobiography) and the eventual film or television series. Mass audiences won't take the time to read "literary novels." We get what we want from the original material, the audiovisual experience, and the criticism and journalism about both and don't have time for the novelization. "Reality Hunger" by (fiction writer) David Shields may have more to say about this.

Wa St Blogger said...

Most fiction written by humans is bad.

I am amazed at what get published, but then I think about just how much reading I do and can see why. More demand than there is quality (same for movies and TV.

However, writing is amazingly difficult. I write periodically (not able to dedicate much time as I over commit myself), but there is some short story styles that I can crank out and not cringe when I re-read it 6 months later. But when I attempt a novel, I just can't get traction. Usually I get so bogged down in the details (names of places, and people, etc. I just can't get the narrative flow to start. I am thinking of using ChatGPT to do all the details, and then re-write it with my own voice using the generated facts and details that I get lost in. I will let it provide the skeleton, and I put in the flesh and blood. At least that is my thought if I ever stop wasting my time on blogs and try writing again.... ;)

Big Mike said...

For the time being, I would tell any aspiring author that if they can’t do better than AI they either should try harder, or look for some other endeavor where they really have talent. But just as IBM’s “Deep Blue” eventually beat chess world champion Gary Kasparov*, the day will come when AI dominates even the best writers. Deal with it.

_____________
* In 1997! I didn’t realize it was that long ago.

Big Mike said...

I agree that most human writers can’t write worth a damn. Luckily a lot of readers can’t tell the difference.

RideSpaceMountain said...

I don't about the rest of you, but I certainly am.

Tim said...

My first thought is that the human written work does not have to be a million times better. The difference between mid-tier authors and best-selling authors is rather small. But when one sells millions of books, and the other sells 50K books, then you see how much effect a small difference can have. I do not expect AI works to ever gain true popularity, at least not in my lifetime. True AI might be a different story, but that may not even be possible.

n.n said...

More Human Than Human

White Zombie... the end is near.

ron winkleheimer said...

"It was more of a summary or an outline of characters who could appear in a story to be developed later. We were just told generalities about fictional characters personalities. So and so is nervous or bold or some damned thing. Who cares?"

Which brings up the question, how do you tell the difference between AI and bad fanfiction?

Howard said...

The first story sounded like it was written in vocal fry. The second one sounded like rhhardin.

imTay said...

Sounds to me like the guy is mailing it in anyways.

narciso said...

a whole row of minitrue offerings,

Nancy said...

I had to read quite a way in to the human-written story to understand that the narrator was female.

Iman said...

funny, but I get a Valley of Gwangi feel, as in something is terribly wrong with this picture.

Lawnerd said...

A new Turing test, AI writes a story and we can’t tell if it is fake or real.

Jaq said...

I am a novelist by avocation, rather than vocation, I use AI quite a bit to review my stuff, and it has taught me a lot about writing. For instance I asked it to review a paragraph that I just didn't like, but couldn't understand why, and it pointed out that I had a POV shift in there, which was pretty subtle, and it was right. It also pointed out that little flaws like that in a story break immersion, and the main idea of writing an entertaining novel is to maintain immersion in the world you have created. It points out issues with dialogue tags, punctuation, grammar, etc. But when it "helpfully" rewrites my stuff for me, while once in a while there is something good in there that I wouldn't have thought of, my usual response is "don't quit your day job," but ngl, it would cost me a lot of time and money to have a human editor pore over my work with me, available to me at any time, and endlessly patient. And its knowledge of the mechanics of fiction writing is up there with any professor I ever had in college.

Jaq said...

Based on my experience using AI, and I have been using it for months, here is my suggestion to weed out AI from a "real writer," look at the passages of description, writers love doing those. AI doesn't have the experience to describe a room, for example, with specifics that ring true and seem original.

mccullough said...

Louis L’Amour weeps

mccullough said...

Louis L’Amour weeps

Jaq said...

When AI can write a passage like this from the Shooting of Dan McGrew, we can all hang up our keyboards, because what would be the point?

Were you ever out in the Great Alone, when the moon was awful clear,
And the icy mountains hemmed you in with a silence you most could hear;
With only the howl of a timber wolf, and you camped there in the cold,
A half-dead thing in a stark, dead world, clean mad for the muck called gold;
While high overhead, green, yellow and red, the North Lights swept in bars? Then you've a hunch what the music meant. . . hunger and night and the stars.

Jaq said...

Only when AI learns to fake authenticity would a writer like Louis L'Amour have anything to worry about. What probably has him rolling in his grave are the unfinished and unpublished novels of his that his estate puts out as finished work after a little effort at polishing them up, diluting the overall quality of his oeuvre, even if they are of value to writers regarding what a work in progress of a master looks like.

Steven said...
This comment has been removed by the author.
Paddy O said...

As I think about this, there really is an overlap between bad fiction and AI that might be hard to discern, but really good fiction writers are unique even among OIs (organic intelligence). AI could mimic a style but not the heart of really good fiction writers because such writers are often doing something that breaks the mold.

But bad fiction is formulaic to a fault, so AI could probably do better. I think of the Star Wars sequels, three movies that took the Star Wars universe and had a lot of Star Wars elements too them, but in a way that turned out very unsatisfying, like the Nutrimatic Drink Dispenser's version of tea. It was written by humans, at least that is what they tell us, but it really turned out like someone put a description of Star Wars movies and books from before and asked AI to churn out new versions. A lot of the elements, none of the heart or consistency.

I think that's how a lot of fan fiction can be described, so maybe AI can really just be better fan fiction, but not ever quite be real art that hits our hearts.

Iman said...

Nothing you can do that can’t be faked…

https://x.com/ClownWorld_/status/1828834736527265932?s=19

traditionalguy said...

Elon warned us.

rhhardin said...

Unless you're working an engrossing problem, you're as fake as the text it's generating.

Jaq said...

The first one sounds obviously AI to me, and the second one sounds obviously human. I don't care for the genre, but if the first one was written by a human, OMG, it was a human programmed by algorithms. Plus, if I am not mistaken, the first one doesn't mention flip flops, which was part of the prompt. The second one was much more in touch with the physical world and its impact on the senses. That "woodpecker riding on the back of a deer" image comes right out of one of those bizzaro world AI images that AI likes to generate.

I don't know if my vocal fry tolerance is strong enough to see if I am right, but if I am wrong, then AI is a much better writer than this person writing high class smut for ladies for glossy magazines. Anyway, story number two was significantly better than story number one.

Leland said...

Combining topics of this and the previous post, here is a YouTube channel that uses AI to create 1950ish movie trailers out of modern movies. What surprised me was how well the AI pulled off the film noir style and the stop motion animation of sci-fi movies of the time. The one replicating "The Fifth Element" (the first one I came across) actually fooled me into looking into whether the 1997 movie was a remake.

rhhardin said...

Is AI female?

Women impersonate themselves as having an orgasm even at the time of orgasm. Within the historical understanding of women as incapable of orgasm, Nietzsche is arguing that impersonation is woman's only sexual pleasure. At the time of greatest self-possession-cum-ecstasy, the woman is self-possessed enough to organize a self-representation without an actual presence of sexual pleasure to represent. This is an originary displacement. The virulence of Nietzsche's misogyny occludes an unacknowledged envy: a man cannot fake an orgasm. His pen must write of prove impotent.

Displacement and the Discourse of Women, Gayatri Spivak

Leland said...

How about this Kamala Harris X post in which she claims a handwritten letter from Tucker [Carlson] supporting her "commonsense gun safety legislation" (sic).