Intelligence and Creativity

Everyone Gets Better. Nothing Gets Weirder.

What three quietly alarming research papers tell us about AI, creativity, and the future of human originality.

December 2025

Here’s a question that should bother you more than it probably does: what happens to culture when everyone’s creative output improves, but starts sounding the same?

We’re having the wrong argument about AI and creativity. Most of it is either breathless: “AI is going to replace artists!” or dismissive: “It’s just autocomplete, relax.” Both camps are missing the more interesting, weirder, and frankly more unsettling story that’s starting to emerge from the actual research.

Three papers published in the last two years point toward something nobody really prepared for. Not a robot uprising. Not creative obsolescence. Something quieter and stranger: a slow, voluntary narrowing of the human imagination.

The tragedy of the commons, but make it art

Doshi and Hauser ran a clean, clever experiment. They gave 300 writers a story-writing task, splitting them into three groups: no AI help, one AI idea to work from, or up to five AI ideas to choose between. Then they had 600 separate people judge the stories.

The results look like good news at first glance. Stories written with AI assistance were rated as more creative, better written, and more enjoyable. Especially among writers who weren’t naturally gifted. AI was a rising tide. Everyone’s boat went up.

Then you look at the other half of the data and your stomach drops a little.

The AI-assisted stories were measurably more similar to each other than the human-only stories. Give everyone the same AI idea machine, and you get convergence. The individual outputs improve, but the collective output gets blander. Less varied. Less surprising.

“With generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced.”

Doshi & Hauser, Science Advances

The researchers actually used the phrase “tragedy of the commons” to describe it, a term from economics that describes situations where individually rational behavior produces collectively bad outcomes. Everybody overfishes the lake. Everybody uses AI for their story ideas. Everybody’s story gets better. Nobody writes anything genuinely weird anymore.

Think about what this means at scale. Music. Advertising. Novels. Scientific papers. Marketing copy. All of it getting slightly more polished, slightly more optimized, and slowly converging toward whatever the latent average of human creativity looked like when the training data was assembled. We’d be collectively sanding down the sharp edges of culture: the very edges that tend to produce the next big thing.

AI didn’t democratize creativity. It just gave everyone a better pencil.

There’s a particular story Silicon Valley loves to tell about AI. The great equalizer. The tool that lets anyone write like a novelist, code like an engineer, design like a professional. It’s a genuinely appealing vision. Talent gap? Closed. Creative inequality? Solved.

This preprint, tracking 442 participants through creative tasks both with and without AI, is a direct challenge to that story. And it’s worth sitting with, because the findings are not what anyone was hoping for.

People who were more creative before AI got their hands on it were also more creative with AI. Baseline creativity predicted AI-assisted performance at β = .42. General intelligence added more on top. Together, those two baseline factors explained 40% of the variance in how well people created with AI assistance.

The gap between high and low performers didn’t close. Some of the data hints it may have quietly widened.

Here’s the thing that makes this finding particularly hard to wave away: the researchers were able to demonstrate that AI-assisted creativity is its own distinct cognitive construct, separable from both your general intelligence and your raw creative ability. Performance on one AI-assisted task predicted performance on a totally different one. Writing a sharp social media post correlated with generating a persuasive pitch. There’s a skill in working with AI well, and that skill itself is partly downstream of the cognitive advantages you brought to the table before you ever opened a chat window.

Which means if you were already good at thinking, AI made you better at thinking with AI. And if you weren’t, AI gave you a modest boost, but didn’t close the gap. The tide rose, but some boats rose faster.

“People who wrote more original stories without AI also wrote more original stories with AI. The gap didn’t close. It may have widened.”

OSF preprint, 2025

There’s something deeply counterintuitive here. We tend to imagine that access to the same tool should reduce inequality. Give everyone a calculator and math ability matters less. But creativity, real, original, surprising creativity, turns out to be more like a multiplier than a floor. AI doesn’t pull people up to some baseline. It amplifies what’s already there.

Creativity is a feeling, not a product

The first two papers are about outputs and effects. This one is about something harder to measure, and maybe more important: what creativity actually is, and whether AI is doing it at all.

Jaan Aru, a neuroscientist at the University of Tartu, makes an argument that sounds almost obvious once you hear it, but has surprisingly deep implications. He points out that when we evaluate whether AI is “creative,” we’re almost exclusively judging the product: the poem, the image, the story. We look at the output and ask: is this novel? Is this good? And sometimes the answer is yes.

But creativity isn’t just a product. It’s a process with an experience baked into it. The frustration of being stuck. The weird, itchy feeling when something’s almost right but not quite. The specific joy (which anyone who’s ever cracked a hard problem knows is genuinely unlike anything else) of suddenly seeing the solution. Aru argues that these experiences aren’t side effects of creativity. They’re part of what creativity is.

And AI systems don’t have them. Not even close. The scientific consensus among consciousness researchers is clear: current AI models are not conscious. Which means they can’t feel the thrill. They can’t be frustrated. They can’t be satisfied by their own work. They’re not intrinsically motivated to do anything. They generate output because that’s what they do when prompted, not because something in them cares about the outcome.

“Even if the creative output of AI systems and humans might be similar, the process and experience might be vastly different.”

Jaan Aru, arXiv:2412.04366

Now, you might say: who cares? If the poem is good, the poem is good. Does the inner life of the poet matter?

Aru’s answer, and I think he’s right, is: maybe more than we think. Because the process shapes the product in ways that aren’t always visible. Real creativity involves building skills over time: the basal ganglia grinding through thousands of practice reps so that mastery becomes intuitive. It involves integrating knowledge across domains in ways that only happen when you’ve actually lived with that knowledge, turned it over, wondered about it at 2am. It involves the kind of deep personal investment that makes creative work strange and specific and irreducibly human.

When we outsource the process, we might be outsourcing more than we realize. Not just the labor. The growth.

What you’re left with

Put these three papers next to each other and a picture emerges that’s harder to dismiss than either the techno-optimist or techno-pessimist narratives.

AI is making individual creative outputs measurably better, especially for people at the lower end of the ability spectrum. That’s real, and it’s genuinely good. The Doshi and Hauser study isn’t pessimistic about this part. Neither is the gap study.

But the collective creative ecosystem is narrowing at the same time that individual outputs are improving. The people who were already more creative are pulling further ahead in their ability to use these tools. And the internal experience of creating, the struggle, the growth, the felt sense of having made something, is being quietly bypassed.

None of this is inevitable. The research suggests that how you use AI matters enormously. Being a co-creator rather than an editor. Treating it as a thinking partner rather than an answer machine. Using it to expand your search space rather than narrow it to the first plausible-sounding idea it spits out.

But the default behavior, just asking for the thing, taking the first good answer, shipping it, is probably the worst possible way to interact with these tools if you care about your own creative development over time. It’s the equivalent of always using GPS and then wondering why you can’t navigate without it.

The uncomfortable truth is that the future of human creativity might depend less on what AI can do, and more on what we’re willing to keep doing ourselves, even when it’s harder and slower and less immediately impressive than just asking a language model to do it for us.

The frustration of being stuck is not a bug. It might be the whole point.

Sources: Doshi & Hauser (2024), Science Advances · Aru (2024), arXiv:2412.04366 · anonymous preprint (2025), OSF. All three are open access or available as preprints (links above and in Reading & watching).