A few weeks ago I overheard something in my kitchen. My son, who writes comics, was showing my husband a page and asking what he thought of it. Somewhere in the conversation my husband asked, almost as an aside, whether he'd used AI on any of it. My son didn't even pause. No, he said. AI takes the soul out of it.
I've been turning that over ever since, partly because I agree with him and partly because I've had a harder and harder time explaining why. If you hand a generative AI system a blank page it will give you back a poem that didn't exist a minute ago, or a picture nobody has seen, or a metaphor that catches you off guard. Some of it is bad. Some of it is quite good, better than I'd like to admit.
So can AI create? The answer I keep landing on is that it depends what you mean by the word, which sounds like a dodge until you notice that the word is the whole problem.
The Easy Answer
The answer I hear most often is that AI can't really create because it only rearranges what already exists. It has read an enormous amount of human work, learned the patterns, and now produces new arrangements of them. That's a fair description of how the technology works. I just don't think it settles anything, because it describes human creativity almost as well. Nobody writes without carrying everything they've read into the sentence, and musicians play inside traditions somebody else built. A designer's new form is usually two or three old ones that hadn't been put next to each other yet. None of us creates from nothing. Only God does that. If our definition of creativity requires complete independence from what came before, we've written a definition that excludes most of what people do too.
The research doesn't rescue the easy answer either. A 2024 study in Scientific Reports put GPT-4 up against human participants on standard measures of divergent thinking and found the AI's responses rated more original and more elaborate (Hubert et al., 2024). A much larger study published this year in Nature Human Behaviour, drawing on more than nine thousand people and over two hundred thousand model outputs, found the opposite on average: humans came out slightly ahead, and the gap widened among the most creative people (Wang et al., 2026). On the tests we've built to measure creativity, AI does surprisingly well, and the best humans still do better. We don't have to deny either half of that to say something important is missing.
What's Missing
Margaret Boden has spent most of her career on the question of whether machines can be creative, and her work gave me the distinction I'd been fumbling toward. She describes different kinds of creativity, combining familiar ideas in unfamiliar ways, say, or pushing out to the edges of a space that already exists, and she's frank that computers can do these well. But she also says that once you ask whether a machine is really creative, you've left the question of novelty behind and walked into much harder territory: intention, autonomy, judgment about value, consciousness (Boden, 1998).
That separates two questions I had been treating as one. Can AI produce something creative? By a lot of the standards we actually use, yes. Is AI a creator, in the way a person is? That's a bigger claim, and a different one.
When a person makes something there is a reason underneath it. Somebody writes because there's something she wants another person to understand. Somebody builds a solution because a problem got under his skin and wouldn't leave. What we make comes out of what we've lived through and what we've slowly learned to care about, and the finished thing is only a piece of the act. I think that's what my son meant, even if he'd never put it in those terms. The AI can produce the page. It can't bring a life to the page, no childhood behind the story, nobody in particular it's hoping to reach. That's harder to measure than originality. I don't think it's any less real.
Copyright law, of all things, has ended up drawing a similar line. In 2025 the U.S. Copyright Office concluded that using AI somewhere in the process doesn't disqualify a work from protection, but that material generated entirely by AI, with no human expressive contribution, can't be copyrighted on its own. The test is whether a human authored enough of it (U.S. Copyright Office, 2025). Using a tool and authoring a work are not automatically the same act, and the law has had to say so out loud.
Created to Create
For Christians, there’s another reason the question matters. Scripture begins with God creating a world and placing people within it. We make things too, but never ex nihilo. We work with what we’ve been given: materials, language, ideas, abilities, experiences. From the beginning, though, people are given real work to do with those gifts. Adam cultivates and names. Bezalel designs and builds. Human making happens inside a world we received, but it still asks something of us.
That is why creativity matters for more than novelty. Making involves judgment about what is worth saying, building, or bringing into the world, and responsibility for what we do with what we’ve been given. AI can help with that process, but the more it can do for us, the more important it becomes to ask which parts of creating were never just about the finished product.
When Help Becomes Substitution
There's a study I keep coming back to. In an experiment published in Science Advances, people who used generative AI to brainstorm wrote stories that readers judged more creative and more enjoyable than stories written without help. Each writer got better. But the AI-assisted stories also started to resemble one another. The individual work improved while the collective range shrank (Doshi & Hauser, 2024). Later work on other creative tasks has found the same thing: AI can lift what one person produces while nudging a lot of people toward the same few ideas.
I don't worry much that AI will become creative. I worry that we'll mistake generating for creating and slowly give away the parts of the process that were ours to do. Sitting with a half-formed idea, developing taste, deciding what stays and what has to go, learning to tell the difference between something impressive and something worth saying. All of that takes longer than generation does, and I suspect it's the part that shapes us.
So, Can It?
It depends on what you mean. AI can generate novel combinations. It can produce work that people rate as creative, and it can make a person more creative at certain tasks than she'd have been on her own. I'm not going to pretend otherwise, and I don't think believers should either.
But creativity, for people, has always been more than producing something new. It carries a judgment about what deserves to exist, an intention about why it's being made, and responsibility for what we put into the world. AI can take part in that, and sometimes it makes the process better. That doesn't mean every piece of it should be handed over.
So the question I've started asking myself, whenever AI helps me make something, is not whether the machine created it. I already know it didn't, not in the sense that matters. The questions that stay with me are harder ones. Which part of making this was mine to do? Did I do it? And what am I becoming if I keep handing it over?
References
Boden, M. A. (1998). Creativity and artificial intelligence. Artificial Intelligence, 103(1–2), 347–356. https://doi.org/10.1016/S0004-3702(98)00055-1
Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), Article eadn5290. https://doi.org/10.1126/sciadv.adn5290
Hubert, K. F., Awa, K. N., & Zabelina, D. L. (2024). The current state of artificial intelligence generative language models is more creative than humans on divergent thinking tasks. Scientific Reports, 14, Article 3440. https://doi.org/10.1038/s41598-024-53303-w
U.S. Copyright Office. (2025). Copyright and artificial intelligence, part 2: Copyrightability.
Wang, D., Huang, D., Shen, H., & Uzzi, B. (2026). A large-scale comparison of divergent creativity in humans and large language models. Nature Human Behaviour, 10(3), 531–540. https://doi.org/10.1038/s41562-025-02331-1