Music. Books. Film. Artificial intelligence is changing how we express ourselves. Is AI a tool for creativity — or a barrier?

Guest

Coral Hart, romance novelist. Founder of AI writing course company, PlotProse.

Adam Green, cognitive neuroscientist and professor at Georgetown University. Founder of The Society for the Neuroscience of Creativity. Editor-in-chief at Creativity Research Journal.

Also Featured

Judith E. Fan, assistant professor of psychology at Stanford University, courtesy of Education and Computer Science departments.

Lennart Meincke, researcher at Wharton Generative AI Lab.

The version of our broadcast available at the top of this page and via podcast apps is a condensed version of the full show. You can listen to the full, unedited broadcast here:

Part I  

AMORY SIVERTSON: In February 2025, the writer Mia Ballard self-published her second book, Shy Girl. Pretty soon, the horror novel had the book community on Instagram shivering, in a good way.

INSTAGRAM USER #1: The way Shy Girl by Mia Ballard has taken over all of my thoughts lately.

INSTAGRAM USER #2: Have I mentioned Shy Girl enough this month? I don’t know. You guys, this is the most unhinged female rage type of book I have ever read in the best way possible. It shook me to my core.

INSTAGRAM USER #3: This is a four-star read. I really much enjoyed that.

SIVERTSON: That was Instagrammers madisoncanread, ardently.cad, and selamseifu. Shy Girl tells the story of a young woman down on her luck, experiencing money trouble, and soon to be evicted, who agrees to meet with a mysterious and affluent man she encounters on a sugar daddy website.

The man asks her to pretend to be his pet in exchange for wiping out her debts, and from there, terror ensues of both a mental and physical nature. A lot of horror fans praised Ballard’s presentation of the protagonist’s emotional landscape. Here’s Instagrammer ardently.cad again.

INSTAGRAM USER: Mia Ballard, the author, top tier. Girlfriend, you did your thing. I don’t know what she was going through in her life, like who hurt her, but dang, whatever she was going through in her brain, whatever she was going through in that moment, it helped her create this piece of literature that is just so good. So good.

SIVERTSON: With online endorsements rolling in and more than 5,000 reviews of the book on the website Goodreads, publishers came a-knockin’. Shy Girl was picked up by major publishing house Hachette Book Group, and it hit stores in the UK in November of 2025. The U.S.-based imprint Orbit Books was scheduled to release the novel here this past spring. Carina Joe on Instagram and many others told folks to mark their calendars.

I give this five stars. This was so good. If you like horror, go ahead and put it on your watch list. You’re going to want to watch out for this when it comes out April 6th.

SIVERTSON: But shortly before publication, Mia Ballard experienced an author’s real-life horror story.

This is the first time I think I’ve seen a traditionally published book be pulled from shelves and not reprinted because of suspected AI use.

SIVERTSON: That was Instagrammer Amy Danielle Books.

Following the release of a two-and-a-half-hour commentary on the popular YouTube channel Frankie’s Shelf titled I’m Pretty Sure This Book Is AI Slop, Orbit Books canceled the publication of Shy Girl in the U.S. In January, people on Reddit started discussing characteristics of the writing that are frequently associated with AI-generated text.

And prior to that Frankie’s Shelf review, Max Spero, founder of the AI detection company Pangram, posted on X that much of the book appeared to have been written by AI. For her part, the author, Mia Ballard, has said that while she did not use AI in the creation of her book, a friend whom she asked to edit the book did use some AI tools in the editing process.

Ballard told The New York Times, quote, “This controversy has changed my life in many ways, and my mental health is at an all-time low, and my name is ruined for something I didn’t even personally do,” end quote. This publishing scandal has gotten critics, authors, and artists more broadly wondering, how is AI being used in creative work?

How is it changing how we think of the creative act itself? And is it altering our ability to be creative? I’m Amory Sivertson in for Meghna Chakrabarti. This is On Point. And joining me now from Cape Town, South Africa, is Coral Hart. She’s a romance author who says that last year she wrote more than 200 novels under more than 20 pseudonyms with assistance from artificial intelligence.

Coral Hart, welcome to On Point.

CORAL HART: Hi, Amory. Thank you for having me.

SIVERTSON: It’s great to have you. And I guess I’d love to start with a quick reaction from you about what we just heard there with the writer Mia Ballard and her novel Shy Girl getting pulled after it was discovered that she used AI in the process.

HART: I did make a statement about this when it happened and my statement went something like, “Shame on that publisher and shame on that agent because they’re supposed to stand up for their authors,” and there’s no way it got where it got in the process without anybody reading it. So they just reacted.

It’s a mob mentality. So for me, I’ve read Shy Girl. Horror’s like my first love and I thought it was brilliant. So I think I understand how Mia feels because the hate that comes when you get pointed out or say you’re using AI is not funny. But I think she was really let down by her team.

SIVERTSON: So that’s interesting, and I want to get a lot more to your perspective, but I feel let’s just lay a foundation here for your own work, because you’ve been writing novels for a long time and before, yeah, before you started using AI in your own work, what did your writing process look like?

HART: So I’ve been writing 17 years this year. Published 17 years this year. So my writing process used to be outline, concept, market research, write. I could write, I’m one of the lucky few who could write 10, 12,000 words a day, and that’s what I did Monday through Saturday, six days a week, every day for about 13 years straight.

I published 94 books before AI. Ghostwritten many more than that in the same time. I won awards. I’ve been an Amazon bestseller, been on the USA Today. I did it the old way for a very long time. I did it the way we’re taught in every MFA, in every writing class, in every way followed the creative path that is set up for writers when you start, and I did that, and I loved doing it every single day for years.

HART:  So in, I want to say about mid-2020, when ChatGPT first came out and it could generate three sentences at a time, I started to look at it, and my first thought was, “This is going to change my industry. If this gets good, it’s going to change things.” So I looked at it. I didn’t really use it. The first heavy use for me of AI was a developmental editing AI program that came out quite early on called Marlowe. I used that first. That was the first shift I made, was from human dev edits to machine dev edits, because it was just so much faster.

And then around end of 2024, I actively started seeing that AI probably was getting good enough to write, and that ghostwriting jobs were falling away in their droves, being replaced by AI first drafts. And I decided early in 2025, January 2025, I took off. I took three months off completely to learn.

And I spent those three months learning.

SIVERTSON: And was any part of you hesitant to experiment with AI as someone who is a writer by trade? Did this feel like potentially toothpaste you couldn’t put back in the tube?

HART: Yeah, so I had, so I knew it wasn’t going back. This cat’s out the bag.

It’s never going back in the bag. So it’s either going to get better, it’s not going to go away. But there was always, in the beginning, there was this part of me that was like it was trained on my work. And I sat down with people in the field and spoke to them, and I expressed that feeling, and it was silly, I don’t wanna call them silly. It was statements like, “Google has had our data since the dawn of the internet. They just didn’t know what to do with it.” “Now they’re using it.”

SIVERTSON: So I want to get specific about how you use AI, because I know AI is this blanket term for a lot of different tools.

What tools specifically are you using, and can you break down what the writing process looks like now?

HART: Yes. Initially I was using ChatGPT, Claude, Gemini, a bit of Grok, all of the different interfaces, and it got very clunky and frustrating. And AI has become so phenomenal that now you can build what you want.

So I’ve built my own tools. So I start every project with data. I have a data mining set that mines sales data, reader data. What are readers talking about? What do they want? What do the reviews say? What do good reviews say? What do bad reviews say? And I make a conscious decision then, as far as writing to market, what the reader is looking for.

From that data, I plan my story. So my stories still come from the millions of squirrely ideas inside my head that will never run out as long as I live. I take that market data, I take my ideas, and I build what I call my story seed or my story concept, and from that story concept, I move into working.

I’ve built my own, we’ve built my own writing dashboard, that first draft. So when I go into that dashboard, it starts with craft. It starts with everything I would have done before. My style sheet, my beat sheet, my character cards, outlines.

SIVERTSON: So what is the, is it a prompt that you are feeding into AI?

What information are you giving it, and what is it giving you back?

HART: Okay. So when I very first start I’m going to start with prompts. AI’s changed now, so now we work with skills and things like that, where it’s more trained and it’s more specific and it remembers. But prompts, yes, so I would initially start with my basic story idea and say, “This is my story idea.

These are my characters. This is the idea I have for them in my head. Let’s flesh out the character cards,” and I want my character card. I am a craft writer, so my character card has to have conflict, wounds, internal, external conflict, backstory, physical features, family history, relationship dynamics, things like, exactly as I would’ve built it before, but I build it now in conversation with my AI tool.

SIVERTSON: And we have to take a break here in just a minute but very quickly, how quickly can AI produce a draft for you?

HART: I can do a five-book series in a day.

SIVERTSON: One day. Five books in a day? Five books. Oh my gosh. I could do more.

Part II

SIVERTSON: We started talking about prompts before the break. Do you have an example of a specific prompt? How does this actually work? A specific thing that you would feed into an AI service like ChatGPT or Claude to get it started?

HART: Yes. So good authors are going to make great books with AI, a plumber is probably not going to make a good book with AI. I still need to know my craft, so my start with AI is always my style sheet. This is my voice, this is my style, this is the syntax I use, the point of view, the tense, the narration.

Good authors are going to make great books with AI, a plumber is probably not going to make a good book with AI.

Coral Hart

I build that first with my AI before we begin writing, and then we develop the story. So I always go in with a story idea, and then I would say, “Let’s generate the character cards or build the characters for this story.” And again, it’s back and forth. And then once we actually get to the drafting of the actual book, it becomes, it already knows what I want.

With Claude, we build skills now in Claude Cowork. So the skill is built from my writing, from my process, from how I do it. If I’m working in a system like Authorism, where it’s already built for authors, it does all the hard work in the background, but you still have to understand the craft to get a good book out.

So if I don’t know that what it’s given me is good, it could be absolute nonsense. So you still have to understand how to, I have to tell you, the biggest one I have to tell you, the biggest fight I have with AI is story structure, because it wants to write every book on Save the Cat, and Save the Cat doesn’t work for romance.

Romance authors use Romancing the Beats, and Save the Cat doesn’t work for thrillers. We use The Fichtean Curve.

SIVERTSON: You’re pulling out some terminology that I am not familiar with, but I’m learning. I’m learning about romance novels along the way.

HART: That’s the scaffolding of the story.

So that’s where things happen to deliver what a reader expects. So Claude wants to just write everything one way, or Chat wants to just write it all one way, and you have to know that’s not going to work to get it to actually write a good book.

SIVERTSON: Okay. So I hear you on this point of, kind of garbage in, garbage out but more finely tuned, more creative ideas in, you maybe get better material out.

But you’ve also hinted at some of the limitations of AI in the writing process. So what are these AI tools currently not so good at?

HART: They’re actually really good at almost everything, but you still need a human to read the book. … So naming conventions and then small continuity, things that we as humans would pick up, but because it’s a machine, it’s not gonna pick it up.

Little things like it will say the room next door in this paragraph, and then the next paragraph it’s gonna go, “I went upstairs to the room,” and we’re like, “But the room’s next door.” Those types of continuity, we’re still seeing no matter which model we use, no matter how we prompt it. It’s going to pull little things like that into it.

It’s not very good at telling time so you have to watch timelines and things like that.

SIVERTSON: When you read back the work that you’ve created with AI does it feel like your writing? Does it feel the same as your writing before you got AI involved or is it fundamentally different in some way?

HART: Yes, because I’ve taught it to be me.

So I’ve given it hundreds and hundreds of thousands of words that I’ve written, and from that we’ve built a style sheet. From that we’ve built an instructions. From that we’ve built the rules. So it’s very close to my writing.  It still throws in the AI-isms, as we call them, but it is very close to my writing.

SIVERTSON: Okay. And have readers noticed? Do you know what your readers think of the books that you’ve written with AI?

HART: So there’s, after I was in The New York Times, there’s a lady that spent an hour on YouTube berating me and telling me that readers can tell the difference, and her profile had five-starred about 10 of my AI books under one of my pen names.

So what readers don’t know, readers don’t know. I have very few one-star reviews on the pen names that people don’t know are me saying, “This is AI,” or, “I think this is AI.” It’s down to how you work with it, and it’s down to how good you tell a story. Readers want good stories, and if you can tell a phenomenal story, whether you tell it with a 2H pencil or you tell it with Claude, the reader will be invested.

Readers want good stories, and if you can tell a phenomenal story, whether you tell it with a 2H pencil or you tell it with Claude, the reader will be invested.

Coral Hart

SIVERTSON: And so do you disclose to your readers that you’ve created a book using AI, or are they typically none the wiser?

HART: Yes and no. Amazon requires us to disclose when we upload, so I do disclose that I use AI on there. It doesn’t disclose it further than that. I have two pen names that are clearly marked, “This work is created with artificial intelligence through the Authorithm guided system with the author’s input,” et cetera, on it.

I don’t think people even read the copyright page of books anymore, to be totally honest. I haven’t seen any difference.

SIVERTSON: So this is a theme that we’ll probably be coming back to over the course of the hour, but part of what makes the creative process rewarding for some people is that it’s hard.

Sometimes the work might just come tumbling out of you, but a lot more of the time there is some struggle involved, what some have called productive friction. It takes time. It takes revisions. Is there still a certain amount of struggle in your work or have you, would you say that this has removed that friction?

HART: It’s removed a lot of friction, but I also, who decided art has to be hard? Who decided writing has to be hard? As a rule, I want to wake up every day and love my job and love what I do and be excited to create stories. And AI has given me that freedom. Because I can sit down and be excited and not smash 10,000 words a day on my keyboard with carpal tunnel and shoulder pain and fatigue.

I’ve taken the pain points out of my job. I still do everything I did before, except the first draft.

SIVERTSON: I hear you on that. I hear that the work should be fun, and yet I would, I don’t write romance novels. I do other kinds of creative work. And I would say that for me, the struggle is maybe not fun, but it is what makes some of the work rewarding.

Like pushing through the fear of the blank page, so to speak, and landing on something that you really arrived at through a process. I can appreciate that that feels rewarding, but it sounds like for you, you want something different from the process, and I can understand that too.

HART: I think creative process for any creative is so unique to the person. And I think that’s what makes creativity so beautiful. Is my creative process, if I’m writing a book, and your creative process if you’re writing a book, even if we were given the same tools, we wouldn’t follow the same path.

SIVERTSON: Okay, that’s a really interesting point. And I want to bring another voice into this conversation now because I want to look at creativity itself. And to do that, we have Adam Green. He’s a professor of psychology and neuroscience at Georgetown University. He’s the founder of the Society for the Neuroscience of Creativity and editor-in-chief at Creativity Research Journal.

Adam Green, welcome to On Point.

ADAM GREEN: Great to be with you.

SIVERTSON: So maybe let’s start just right there with what is creativity or what does the act of being creative mean?

GREEN: It’s a great question. I think it’s useful to start with the way that we’ve been defining creativity for a long time. There’s what’s called the standard definition of creativity, and it’s really focused on creative products, so things like a book or a solution to a problem.

And it’s used a couple of criteria, novel and useful or sometimes novel and valuable or meaningful if we’re verging more toward the arts. But these are criteria that are applied to the product. And if we want to understand the creative process, which I love that you guys are already engaging that question, and I think it’s the question when we’re interested in how humans differ from AI and creativity and how AI affects human creativity.

I think focusing on process is really essential, and a definition based on products doesn’t get us there. So a definition based on products leaves us just with, okay, so your creative process was whatever it was that you did to create a product that was novel and useful. Really not very informative about what that process is made of.

And so a group of us who study the neuroscience of creativity, sparked really by an initiative at the Society for the Neuroscience of Creativity, have been looking at what we know and what we’ve learned over the last twenty or so years about how brains create and building a definition of what creativity is made of, what creativity as a process means in terms of the components that make it up.

And one of the things that turns out to be really important across lots of different kinds of creativity is what we do with our attention. So specifically, when brains are creating, we see them shift their attention internally. And that might be counterintuitive. Because we often think of creativity as being inspired by the world around us or people around us, and it very often is.

But in that moment when we’re creating, if we are in fact the creator, then it has to come from in us. It has to come from something internal, what we’ve made of our learned and stored representations of all those inspirational things around us. And so when we think about creativity as a process, it has a lot to do with focusing internally, and what we focus on, what brains are focused on when they shift that attention inward is a lot of the story of how brains and AI differ in creativity and how AI affects the human creative process.

SIVERTSON: Okay. That’s what we’re going to be getting to here, and I want to point out something that you said at the top there because you said when we talk about creativity in the arts, and so I think sometimes our brains hear creativity, and we jump to, okay, we’re talking about artists.

But we’re actually really thinking about creativity that is an essential ingredient of being human. Whether you consider yourself an artist or not, we all possess some degree of creativity and perhaps need some degree of creativity. But I know you also describe creativity not as a trait or as something that you either possess or don’t, but as a state of mind.

Can you say more about that?

GREEN: Yeah. That’s always been interesting to me as a researcher, as a scientist. Another thing about the story of the science of creativity is that for decades and decades, it looked a lot like the science of intelligence, which has really focused on a few tests like IQ measures, and the way IQ is more or less construed is that it doesn’t change.

You take a test, you get your number, you go live your life. There are a number of creativity tests that are built in a very similar way and actually inspired directly by the intelligence psychometric work. And the idea there is, okay, so I might take the test and get my number and Amory, you might take the test and get your number, and then we just go about the rest of our lives.

It’s never been clear to me exactly what we do with that information. So it’s always been much more interesting to me how we can boost our creativity, how our creativity might be dynamic. So it’s not quite true to say that creativity is not a trait, because it is so many other things.

But almost all of our traits are also malleable, depending on how we live our lives, how we exercise, how we use our brains to become better at remembering. All of these things may, we may have some band that’s our sort of minimum and maximum, but where we end up in that is really largely up to us.

And so what’s really been exciting to me in studying how brains create is to look at how brains go from thinking less creatively to thinking more creatively, and how we can help brains do that in schools and through direct boosting of brain function through neurostimulation. And also, now in the context of AI, how we can how we can help people focus their attention on what makes their ideas unique and what allows them to escape some of this homogenization that Coral was talking about that really is a consistent effect of AI on creativity.

SIVERTSON: Okay, and so you mentioned tests there that you’re able to do. How is creativity measured in the brain?

What tools are you using for that?

GREEN: It really does depend on the kind of creativity that you’re interested in. So there are measures for divergent thinking. A very common task is the alternate uses task, where you come up with funny ways to use something like a paperclip or a brick or a shoe.

And that’s intended to assess more divergent, different, weird kind of stuff that we can generate or ideate. Other kinds of creativity include seeing different kinds of connections between things. So a kind of creativity that’s fascinating to me is analogical connections, making connections between concepts that seem far apart on the surface, but there’s something new and interesting in the connection.

Classic examples like the connection between the structure of the solar system and the structure of the atom that were important in developing early theories of how the atom is composed. Then other kinds of creativity would include assessment of short stories by experts or by AI, and that’s something that we could tackle.

Or even, how do you find a common associate between a set of words?

Part III

SIVERTSON: Lennart Meincke is a researcher at the Wharton School of Business of the University of Pennsylvania, and in late 2023, he and his advisor decided that they would study how humans respond to creative concepts developed by AI.

LENNART MEINCKE: One of the things we were curious about was creativity has always been seen as this human frontier. I think most people, when they thought about AI at least five years ago, would say, “Oh, the last thing to be automated would be human creativity or cognition or maybe even ethics,” like these things are all like super, super difficult and we’ll always have a human.

And so we wanted to put it to a test empirically.

SIVERTSON: And so they put a few different large language models, you know the ones, Claude, ChatGPT, et cetera, to the test.

For about a decade, Wharton MBA students have participated in an innovation tournament where competitors are challenged to come up with creative physical products that would cost a consumer less than $50.

Those concepts are then subjected to a review by a panel of humans, and a winner is chosen. Lennart and his collaborators provided the LLMs with the same prompts given to students and then presented both the winning human ideas from the previous 10 years and the AI-generated ideas to a panel of human judges.

MEINCKE: Of the top 10% of all our ideas, 35 out of 40 ended up being AI-generated ideas. So it was a seven-to-one advantage in favor of the AI to humans, and the top funnel or the top piece is especially interesting since in innovation you usually only care about the very few best ideas.

SIVERTSON: A seven-to-one advantage for the AI.

So given that result, Lennart and his collaborators concluded that to some extent AI had the potential to aid human creativity, to build on the ideas from which it was built and refine them into more desirable concepts.

MEINCKE: I think AI can really help you to push your thinking. It makes it very cheap to iterate on ideas.

Something I always like to do is, and recommend to people, start with a few ideas that you have yourself, like plot the map with some of the things that you would like to do, and then use the AI to explore the region around that idea, right? Give me five smaller variants of this, and you can come with some stupid constraints to see where it ends up.

SIVERTSON: But Lennart felt that this didn’t tell the whole story, so he and his advisor looked more closely at the types of ideas AI put forth, and they found that although AI-generated ideas received high marks from the judges, those ideas were often very similar.

MEINCKE: That’s a phenomenon we see to this day, that AI lacks the diversity when it comes to ideas.

AI lacks the diversity when it comes to ideas.

Lennart Meincke

It often gives you very similar ideas. So I like soccer/football. You can have 11 great goalkeepers, but that’s not what you need, right? You need a full team, and so it’s a similar thing here. If all your ideas are great but they’re all the same idea, then probably that’s not good enough.

SIVERTSON: And so the takeaway for Lennart was that AI is a tool that can be used creatively, but it’s one that still needs human guidance.

MEINCKE: I always try to encourage people to use your own brain first. Exercise that muscle.

SIVERTSON: That’s Lennart Meincke, researcher at the Wharton School of Business. I’m joined today by Adam Green, a neuroscientist and professor at Georgetown University, and Coral Hart, who’s a romance author who uses AI to write novels for her, to help her in the creative process.

And Adam Green I first just want to hear some, some thoughts from you on what we just heard from Lennart Meincke, because it echoes some of what you started to talk about in terms of how AI affects the ideas that we are able to come up with. Can you say more about that?

GREEN: Yeah, I think everything that we just heard is very much in line with where all the data are pointing us right now, and that is that AI can generate ideas that at least sound good, and in many cases are good, but those ideas end up being really similar.

And the effect, if we frame that in terms of what brains do when brains create, I mentioned that we need to focus our attention internally, and what we’re focusing on when we do that is a network of meanings that we’ve built up over the course of our lives, and each person’s life and each person’s experience being different, each person’s semantic network is different.

So when we focus our attention internally during that creative process, we’re following along and leaping the different links of that semantic network that’s based on our own experience. I might connect a bicycle to a watermelon because of an experience I had. You might connect a waterfall to a butterfly.

I’m just riffing. … But that’s different. Yours would be different from mine. And so when you’re searching internally, you’re gonna generate words and ideas that are different from the words and ideas that I’m gonna generate, and that becomes very important when we start to think about what AI is doing and how humans are interacting with AI to create.

Because AI’s starting from a place of gathering all the things it can find that all of us have written and glomming them into one training set, and then it’s also not searching a network the same way that we do. So it’s doing a probabilistic token prediction based on what it thinks is a good response for any number of reasons and based on how you’ve prompted it, but that good response probability is always going to be a narrow band of what’s out there.

So in a certain sense, depending on how we use the tools, they can be very much homogenization machines, and the homogenization comes from several other factors as well. It’s the single clearest result that has come out of all the work on AI and creativity, I think, at this point, is this homogenization.

SIVERTSON: And Coral … Oh, sorry, Adam. … I just want to, I want to get a quick response from Coral Hart on that. Does this resonate with you, this homogenization of ideas? Have you found that?

HART: So if I don’t put the idea in, if it’s not a, a story concept out of my head, if I go, “Hey, Claude, let’s write a cozy mystery today,” I can almost list off the exact 10 ideas that it’ll come back with.

Cookie cutter, formulaic ideas that it’s going to come back with. Whereas if I go and say, “Hey, Claude, I had this idea for a cozy mystery about a haunted house that gets up and moves on its own every night. Let’s do that,” it’s going to work with my idea and expand my idea. But if I let it come up with the idea on its own, I’m gonna get a very narrow set of ideas.

SIVERTSON: Adam Green, I wanna understand better what the brain is doing when we engage in creative work and maybe what we know about what the brain is or isn’t doing when we bring an AI tool into that creative work.

GREEN: Yeah. So on our own, we are swinging the branches of our own semantic network and exploring these very life-specific, experience-specific pathways that lead to different words and ideas that are ours.

When we work with AI, a couple of things happen. One is that we focus instead of in focusing our attention internally, which is this big brain signal that we’re doing, that we’re in the act of creating. Instead, our attention is drawn outward to these ideas that are coming from outside, and they’re not just benign.

They’re coming from outside, and they’re coming from essentially a homogenization engine, right? And so when you combine that sort of lack of internal direction of attention with the input that’s coming in from outside being homogenized, it’s fundamentally changing the capacity for originality.

But I do think it’s always important to say it depends on how you use the tools. I think if you surrender your creativity entirely to the LLM, then that creativity will become highly homogenized. If you start with your own ideas, which I think everyone in the conversation has mentioned some variety of, I think that’s the most powerful thing you can do.

But I also think it’s very hard for us to know, and the research supports this, where our ideas end and the AI influence begins when we interact. And so I think mapping that originality is the next big step, and it’s one that a lot of people are interested in taking.

SIVERTSON: Okay, so how do you do that?

Because that’s exactly what I was thinking is as long as you have original ideas, you can keep feeding them into the AI and getting better material out, as Coral Hart is saying. But if you start using, the more we use AI, do we lose or do we become more and more detached from the originality that maybe made that early human AI collaboration feel more human than it might, in years to come as we use this technology more and more?

GREEN: Yes. And the thing is, we’re talking about this as a problem for grownups. It’s so much more of a concern and of a threat for kids because we’re talking about essentially turning on and off a capability that we’ve already developed over our many 10,000 hours or however many hours we’ve invested in creating things and thinking, writing, drawing, whatever it is that we do.

We have it, we’ve developed it, and it might atrophy if we’re not careful, and that’s a real concern. But if you are a kid, I’ve got a four-year-old, if you’re a kid and you have an alternative to developing this kind of desirable difficulty or creative friction, I think you called it, through that really tough challenge of the blank page and learning and expressing and owning your creativity, that practice, I don’t think there’s a replacement for it, especially during those critical times when the brain is developing.

SIVERTSON: And Coral, do you experience this yourself where the more you use AI, the harder it feels to come up with ideas that feel fresh and different from your previous ones?

HART: I don’t think so. I’ve always worried that I’m gonna die one day with ideas still in my head. I think I’m a creative, but I also have ADHD, so I have a very divergent, crazy brain that’s always got ideas happening.

So I’ve experienced it as a freedom to explore every one of those creative ideas instead of only being allowed to sit down and work on the ones that are commercially viable because that’s what I have to do, writing my job. So before, if I sat down to write a book, I had to know I was writing the book that was gonna make the most money, not the most exciting book, not the most fun book, not the most wild story.

And I think for me, I don’t have a shortage of ideas, but what I’m saying, I spent 16 years learning my craft first. I didn’t have the shortcut 16 years ago. I spent 16 years getting there, and I think for me it’s a tool like a typewriter or a printer or a pencil now, but it’s just allowed me the speed and the freedom to explore my creativity actually on a much, much bigger scale.

SIVERTSON: And that’s a really important point there, Coral, because Adam, I’m thinking of students today, people today who are growing up in a world where they can ask AI to write a paper for them, to analyze a book for them, whatever the insert any various homework assignment on and on. And kids are busy.

They don’t want to struggle with creative work that comes in the form of homework when they have a million other things to do, and yet they’re missing out on these semantic connections that you were talking about, the networks that they’re building that they won’t have if they do take this AI shortcut.

So how do we impress upon young people the importance and the value of creative work, of developing their own voice before bringing in these kinds of tools?

GREEN: It’s such an important point. And so my lab works with a lot of high schools, and high schoolers are feeling really this big sort of cloud hanging over them, this thing that’s smarter than them, this thing that makes maybe learning and trying meaningless, because it’ll always be better than me at everything.

And that’s incredibly demoralizing also for teachers. But when we work with these schools, and now we’re actually starting a collaboration, we’re gonna look at a neural signal of internally directed attention and help kids keep that attention internal even when they’re working with AI, which I’m excited about.

But overall, we’ve been mapping originality and mapping homogenization. And what you’re able to do when you can show anyone really, but especially young people, when they think for themselves, their ideas grow beyond those AI homogenized black holes or clusters. And that’s a superpower, that being them, the uniqueness of their own ideas contributes to idea diversity in a way that AI cannot do.

And that’s nice and it sounds pretty, but it has real market value because in a landscape, in a creative landscape where everybody has these same models and the models are homogenizing ideas, the ideas that actually lead to differentiation and innovation will be coming from the people who continue to think for themselves and can push outside of that homogenization.

And a good idea that everybody has is a lot less valuable than an equally good idea, not necessarily better idea, that you’re the only one that has it.

SIVERTSON: Creative work is often thought of as the thing that makes us feel the most human and what teaches other human beings what it means to be human.

And yet, Coral, do you ever feel like we’re being too purist or too precious about creativity, creating new things of our own minds as opposed to collaborating with this technology that is here whether we like it or not?

HART: So I think for me, creativity has always been a collaboration.

I’ve co-written with other authors. I’ve worked with artists. I’ve worked with musicians. So for me I’m just collaborating in a different way now. I don’t have that precious feeling about it. I love making amazing stories, and this just helps me do that.

The first draft of this transcript was created by Descript, an AI transcription tool. An On Point producer then thoroughly reviewed, corrected, and reformatted the transcript before publication. The use of this AI tool creates the capacity to provide these transcripts.