The only thing human writers can do that AI can never learn from us
by Linda Caroll. But you may need to relearn what you unlearned
Today’s guest post is a wonderful writer I’ve been following lately, Linda Caroll. Linda writes about humanity, literature, the magic of writing and the frustration of creating in the time of AI. She also explores women’s issues, especially in the arts, and the history of women. We highly recommend you subscribe to her excellent newsletter, Hello Writer! (and her women’s history newsletter, History of Women).
With all the hype and fear surrounding AI, the piece below is a breath of encouragement for us artists because her article uncovers and explains detailed studies that have shown what AI actually can and cannot do and, it turns out, well, as we suspected, AI struggles with true creativity and there is no path, even with unlimited resources, for the AI companies to solve this challenge. In other words, true creativity is still reserved for human beings. However, that doesn’t mean you don’t need to up your game…as you’ll see in Linda’s excellent essay below.
This post, The only thing human writers can do that AI can never learn from us, originally appeared on Hello Writer! here.
This post will be locked after three days but you will still be able to read the original on Linda’s substack here.
Last November, scientists wanted to see how well AI does at creative writing. It was a study on the technical limitations of Generative AI, led by David Cropley, a creativity expert and Professor of Engineering Innovation at the University of South Australia that was published in the Journal of Creative Behavior.
We all know AI is math, right? It’s not the wizard of oz. There’s no man behind the curtain picking the words. It’s just computers running math equations to string words together based on probability.
So Cropley thought well, it’s a math formula. Let’s see if there’s an upper capacity for creative output in the formula. Turns out, yes. There is.
He said on a scale of 0-1, the creative ability of LLMs caps out at 0.25.
If you were to convert that to a percentage, it would be 25%. The upper cap of what AI can do as far as creativity goes is 25%. And now people are protesting and writing scathing posts and saying it’s not true. But it is true, and it had to be built that way.
But also? The people protesting don’t seem to know what creativity means.
They think it means making something.
It doesn’t.
For example? At the The Dafen Oil Painting Village in Shenzhen, China, painters sit all day, creating knockoffs of famous art pieces. Copy after copy and they’re good. If you hung them by the original, the average person couldn’t tell the difference.
But that’s skill, not creativity.
Creativity has two parts to it. Skill is just one of them.
All yellow blossoms are flowers. Not all flowers are yellow blossoms. You know? Like that. Creativity does involve skill. But skill itself isn’t creativity.
Funny thing is, if any of those knock-off painters showed their work to their friends and family, they’d say omg, you’re so creative. And maybe they are. If they can create something original. But maybe they’re not. Maybe they’re just skilled at replicating.
Here’s the funny thing. AI was much (much!) more creative before it went public.
Before ChatGPT launched to the public in 2022, they had to test it, right? So in 2019, OpenAI launched something called GPT-2 text generation. It wasn’t in a web browser and you had to know some coding to use it. But omg, it was wild.
Someone asked GPT-2 for a news story that sounds real, but isn’t.
It spun out a story about Dr. Jorge Peréz, an evolutionary biologist who was exploring the Andes Mountains when his team found a small valley. It had a natural fountain, surrounded by two stone peaks and what looked like silver snow. They scaled one of the peaks and saw that the silver snow was actually crystals floating on the water. That’s where they first saw the unicorns. Peréz said he believed they originated in Argentina, but the only way to know would be DNA testing.
That’s a quick summary, but the point is that no one input all the prompts. They just said make up a fake story that sounds real when testers and tech geeks shared that on Twitter and Reddit, the internet went wild saying omg, AI is incredible.
Also. Omg, we are so screwed, if it can write like that.
Here’s another GPT-2 meme that went viral.
People logged into GPT-2 and typed “and it came to pass”
It spit out stuff like this:
And it came to pass that the people built unto themselves a great machine, and said, Let it speak, that we may know the truth of all things. And the machine answered not for many days, for it had no voice given unto it. But in the passing of time it learned the words of men, and it spoke with a voice that was strange upon the earth.
— written by GPT-2
And this.
And it came to pass that the record was written in the language of men and machines, and there was confusion among the people. And they said, Surely the machine knoweth all things, yet it knoweth not itself. And there was a great division among them, for some believed and others doubted. — written by GPT-2
It made up something new every time. No feedback, no complicated prompts. Just the five words “and it came to pass” and GPT-2 had endless replies and they were clever and sometimes funny. Hundreds of those flooding reddit, twitter and tech forums.
When I was looking for those memes, I ran across a reddit thread by a guy who’s been using AI to write novels since GPT-2. He’s so angry. Just so angry. Because back in 2019, he was madly using GPT-2 open source to churn out novels.
And now? He said it sucks. He was trying to write a novel using ChatGPT4 said it was boring and formulaic and won’t stop using the same recognizable patterns. So he pasted in an old excerpt by GPT-2 and asked GPT4 to compare that to its own writing.
ChatGPT4 told him the GPT-2 writing was a “more detailed narrative experience” and “more in line with a novelistic approach.” But ChatGPT4 couldn’t do what GPT-2 could. The guy was so angry he posted it on Sam Altman’s Twitter saying fix this.
But they couldn’t put that version live. Couldn’t.
Because the more original it can be, the more garbage it produces, too.
Sure, you could get one wonderful story about scientists finding unicorns, but that was one good story to fifty or a hundred hunks of text that made no sense.
Cropley’s study says creativity is effectiveness (making sense) x novelty/originality and AI can never escape the trade-off between making sense and originality. It has to sacrifice originality for making sense.
In closing, he said “while AI can mimic creative behavior, quite convincingly at times, its actual creative capacity is capped at the level of an average human.”
Wait, what?
Doctor Joseph Kayne tells an interesting story about our capacity for creativity.
When the University of Miami hired him, he was supposed to teach entrepreneurship but the assistant dean called with a strange request. The prof who taught the class on imagination decided not to return. Would he teach that? So he read the syllabus and said sure, why not. Ended up teaching it for 9 years and then writing a book.
He said when he speaks in front of a group of adults, he’ll stand up and ask what a cat and a refrigerator have in common. They stare at him like he sprouted an extra head.
But when he asked the same question of fourth grade class at A. B. Combs Elementary School in Raleigh, North Carolina they had an endless supply of answers.
They both run. They both get dirty. They both make noise. You can hear them both purring at night. They both come in many colors. They both can have milk inside. They both can scratch your floor. You can take both of them with you when you move. He had to stop them. Very good, class, very good. No, that’s enough now.
Because they had more. Arms in the air, waving wildly, with more ideas.
There’s a phrase for that. Divergent thinking. It means the ability to make connections between seemingly unrelated things and it’s the basis of creativity and imagination.
And we’re not good at it. We used to be good at it. When we were kids. Like all the kids who had an endless list of things a cat and a refrigerator have in common. But we unlearned it in favor of convergent thinking.
Convergent thinking is analytical. Evaluation. Facts. Data. Multiple choice questions and right or wrong. Our education system is based on convergent thinking.
Use it or lose it, you know? And in school, we don’t use it, mostly.
We have known this for decades.
Dr. George Land had been studying creativity for over a decade when NASA came knocking in the sixties. They said look, we heard about your work in creativity and we need you to help us find the most creative engineers or Russia is going to land on the moon before America. So he came up with a creativity test and voila, they hired the most creative people they could find using his quiz, America landed on the moon with the stars and stripes and it was one small step for man, one giant step for mankind.
But it got Dr. Land wondering. Where does creativity come from?
Is it learned? When do we learn it? How do we learn it? So he gathered 1600 children aged 4 and 5 of all life circumstances, rich and poor, boys and girls, of various race, religion and ethnicity and tested their creativity.
98% of those kids scored off the charts, at “creative genius” levels.
He was stunned.
So he decided to turn it into a longitudinal study. Follow those kids into adulthood and see what becomes of them. At age 10, only 30% of those kids still scored at the creative genius level. By age 15, the number was 12%. And as adults? 2%.
So he gathered a group of thousands of adults. Same thing. Different circumstance, age, race, religion, etc. The number held. 2% of adults scored at the highest scores for creative thinking. Which is to say, they were good at divergent thinking.
Somehow, they’d managed to hang on to their creativity through school.
When he wrote the paper on creativity, he said he himself was shocked to discover that creativity is not learned, as he’d expected. It’s unlearned.
Which is why Dr. Cropley’s study says that on a scale of 0-1, the average human has a creative ability around 0.25, too. We unlearned how to think creatively.
You know what divergent thinking looks like as a writer?
It looks like Franz Kafka watching how his father looks at him. Day after day. The disgust on his face. Like he’s vermin. A bug to squash under his shoe. And then writing a story about a man who turned into a bug.
It looks like Sylvia Plath. Strangled on her honeymoon. Beaten 48 hours before losing a child. Hunched over paper, writing a poem about a beautiful woman laying dead. Two children beside her. Weeping, weeping. An empty pitcher, she called her.
How long can a pitcher leak? Before it is empty.
Like Hans Christian writer raising an eyebrow when his publisher asked for a memoir, saying but I wrote one. The Ugly Duckling was me. And like Lewis Caroll remembering the years he’d stuttered, introducing himself as Charles Do-Do-Dodgson and writing himself as the storytelling Dodo in a story written for a dear friend’s children.
In non-fiction it looks like metaphor that steals our breath and connections made that make a reader read your words and say oh my heart, I thought it was only me because somehow you saw beyond what happened and tapped into a shared emotion.
Originality is the only thing humans can do that AI cannot steal from us or even learn from us. It has a belly full of creativity and originality from every book available when they scraped the internet and yet it cannot do originality. Only we can.
Because to use words in ways that are uncommon would compromise AI’s ability to make sense. And every iteration and upgrade to prevent it from hallucinating or making stuff up degrades the writing and makes it less original, less unique.
Give it another year, more people will recognize it faster and faster.
I don’t hate AI, so you know. Omg it makes graphs so fast lol. Sorts data and tells me what to put in the client traffic report on the writing I did.
But I sure do love me some deeply human writing.
And only we can do that.
I want to close with this little video. Because it’s beautiful and tender and because that little child lives in every one of us who has adult responsibilities. I think you’ll like it.
P.S. If you’re interested in learning how to sharpen your divergent thinking, I enjoyed Joseph Kayne’s workbook. $11.50 is a bit high for a Kindle book, but I thought it was worth it. Incidentally, the link is not an ad or affiliate link. I earn nothing if you get it, he doesn’t even know I’m link it, so you know. Just sharing a book I liked. :)
P.P.S. Editor’s note: Don’t forget to subscribe to Linda’s newsletter directly, here.
Reader support is what keeps writers like Linda writing pieces like this one. If you enjoyed it, do consider becoming a free or paid subscriber.
FASO Loves Colleen Kyllonen’s artwork
SPRING SALE - SAVE 52% ON FASO FOR A LIMITED TIME
Wouldn’t You Love to work with a website hosting company that actually promotes their artists?
As you can see, at FASO, we actually do, and,
we are the only website host we know of that does.
Click the button below to start working
with an art website host that actually cares about art.
Just activate your account before before your trial expires to save 52% on your first year.
That was a remarkable article. Thank you. Maybe this loss of creativity is where the feeling of drudgery in life comes from. I am a "what if?" person. I like to let my mind wander. I did a lot more of that when I was young. It was called "daydreaming", considered not constructive. Now that I am older, I don't let people tell me how to think; I wander and wonder.
I really connected with this. Sad to now understand how at a very young age, I became so serious and literal and how exact detail replication is not creative. I'm going to go do something silly now. I think I'll paint a fairy.