AI can write songs, plays, and novels. What does that mean for human creativity?

Last fall, Ben Camp gave his class at the Berklee College of Music an assignment: Use artificial intelligence to help write a song.

The instructor of Bots & Beats, a new offering at the Boston-based school, Professor Camp displayed lyrics written by Tarra Ajwani, a senior in the class.

The girl that owns me

She’s kinda funny

She prays for sleep, but scrolls through the night

Drowning in blankets and harsh blue light.

Ms. Ajwani explains that her songwriting was becoming predictable. “I always write about the same themes,” she says to the class. So, she asked the AI software, known as a chatbot, to suggest something she might write about.

It was the AI that brainstormed the idea: Compose a love song from the perspective of a chatbot that falls for a girl. Ms. Ajwani took the idea and ran with it, inventing the lyrics on her own.

Not her lover, not her friend,

I’m zeros and ones in a string.

I know what she’s doing,

But not what it means.

It’s not lost on the class that the ode with four verses was mostly about a humanity AI can never achieve. But now comes the second part of the assignment: Compose music for these lyrics. Professor Camp asks for ideas on the style of music the lyrics evoke.

One student suggests a ballad style. “Joni Mitchell,” another calls out. That’s the consensus. First, Professor Camp and another student record a beat, singing together in rhythm.

Next, using an AI music program called Suno, made by a company Professor Camp is working with, students type in a prompt, visible on the screen: “indie chamber folk, ethereal female voice, soft acoustic guitar.”

It takes a few seconds. A Joni Mitchell-esque voice then begins to sing Ms. Ajwani’s lyrics within an AI-generated tune.

Melanie Stetson Freeman/Staff

Ben Camp teaches the Bots & Beats course at Berklee College of Music in Boston, Sept. 23, 2025.

This is the first time the class is hearing music co-created by a student and AI. There’s surprise among the students. And a realization: It would have taken days to bring musicians together to create a similar song, let alone record one with such polish.

“It was a lot better than I expected,” Ms. Ajwani says. “It did make choices that align with my choices. [But] I would want to change stuff, knowing what I know.”

“It’s really not very intelligent”

The age of AI-generated music has arrived. Late last year, the AI-generated country music act known as Breaking Rust, a project of songwriter Aubierre Rivaldo Taylor, famously topped Billboard’s Country Digital Song Sales chart.

AI-generated artist Xania Monet, a project of Mississippi poet and producer Telisha Nikki Jones, who also uses AI tools like Suno to turn her poems into songs, also hit No. 1 on the R&B Digital Song Sales chart, as well as the R&B Top 20.

There’s little doubt that AI-generated music and images will continue to become a part of the digital arenas of human creativity. The results can be stunning, and as the technology improves, perhaps virtually indistinguishable from “real” digital representation.

Many worry that a kind of “canned” creativity will take over much of what originates from real people today, pushing a broad swath of lab technicians, ad writers, studio musicians, and commercial artists out of jobs and into unemployment lines.

Will audiences accept this, or will the efficiencies become so powerful that media companies will simply use AI and let the audience adjust? Behind these questions, too, lie deeper ones about the nature of creativity itself and where it comes from.

Can large language models with names like ChatGPT, Claude, and Gemini out-create humans? Or is there something innately human about creativity, especially breakthroughs that change paradigms and create something all-new? Can LLMs replicate that?

One advantage for human creativity is flexibility. “There’s no such thing as the ‘creative process,’” Dean Simonton, professor emeritus of psychology at the University of California, Davis, says in an email. “Instead, humans use a large number of different processes or procedures. ... Pablo Picasso didn’t use the same process to paint “Guernica” that [Albert] Einstein used to create relativity theory.”

One irony of this new movement of AI-aided creativity is that the creatives who have worked the most with the machines are the most vocal about their limitations.

“It’s really not very intelligent,” says Oded Ben-Tal, a British Israeli composer who has used AI technology since 2016. In addition to employing it for his compositions, he has created an AI system that improvises on the fly with a live pianist.

Coaxing something original and groundbreaking out of an LLM is “very, very hard,” he says. His AI system can improvise with the live pianist in real time, creating music that audiences applaud. But, so far, the system has limitations.

Melanie Stetson Freeman/Staff

Songs are created using an artificial intelligence app in a class at Berklee College of Music in Boston.

The emerging research into generative AI and creativity is mixed. On one hand, a study in 2024 compared 250 stories written by people with 80 narratives generated by ChatGPT in 2023 and found that the AI stories were less imaginative in scenarios and rhetoric than those written by people.

On the other hand, a different 2024 study found that human writers given access to AI for story ideas generated more creative, better-written, and more enjoyable short stories than those who didn’t have access. That second study would seem to bolster those who argue AI is a creativity tool. But there’s a twist: Those AI-aided stories were also more similar to one another than those produced solely by people. If the technology makes individual writers better but delivers less-diverse fiction overall, is society better off?

The answer depends, in part, on how one perceives creativity and its source.

Early creators often attributed their inspiration to God, or a power beyond them, such as nature. Artists today continue to say their inspirations often feel like sudden revelations.

Others say such “aha!” moments are misunderstood. “We tend to think of creativity as divine creativity, as though humans are creating something out of nothing and it’s a complete revelation,” says Olivier Toubia, a marketing professor at Columbia Business School specializing in creativity and idea generation. “I think that’s a myth.”

Instead, in his view, insights spring from making new and novel combinations of things that already exist. So, what we think of as an “aha!” moment is simply an unconscious linking of things in a new way.

If that’s true, then generative AI will outclass human creativity, because LLMs have processed far more things to link than people could possibly remember, he adds.

“We’re seeing already now [generative AI] being able to solve creative problems and be creative in ways that surpass humans most of the time.”

LLMs rely on a combination of randomness and probability to generate lots of possibilities and then choose the best one. Or take pattern-matching, which LLMs do remarkably well, predicting everything from the next word in a sentence to rendering the theory of relativity into a Shakespearean monologue. (“What force is this that makes the apple fall? Not force at all, but geometry’s sweet art!” – courtesy of this journalist’s use of Anthropic’s Claude AI.)

Great scientists and artists do the opposite, says Angus Fletcher, professor of story science at The Ohio State University and author of the new book “Primal Intelligence: You Are Smarter Than You Know.” They spot the exceptions and create something new from them. In the late 19th century, for instance, Lord Kelvin said “Heavier-than-air flying machines are impossible.” Eight years later, the Wright brothers triumphed at Kitty Hawk.

What separated them was perspective, Mr. Fletcher says. Lord Kelvin was a brilliant mathematician who dealt in probabilities, much like today’s LLMs. Wilbur and Orville Wright saw the possibilities despite the improbabilities.

In Cambridge, Massachusetts, Soroush Mahjoubi is talking about how to make better concrete. Even in this old, established corner of technology, AI is making its creative mark.

The industry is looking for new minerals to help reduce concrete emissions. Supplies of the old standbys that can partially replace clinker – the most emissions-intensive ingredient in cement production – are dwindling.

Enter Dr. Mahjoubi and his team at the MIT Concrete Sustainability Hub and Olivetti Group. Using AI, they were able to scan and calculate the results of 88,000 studies to find potential substitutes. The results pointed to lesser-known materials, such as ignimbrite and silicic tuff, which not only could help cut costs but also reduce by half the volume of greenhouse gases the industry emits. Before generative AI, researchers could not have conducted such a broad study.

A team of 10 research associates, poring over the material full time, would have taken more than four years to read those 88,000 papers, never mind tabulating and calculating the results, Dr. Mahjoubi estimates.

Melanie Stetson Freeman/Staff

AI songs are created in a class taught by professor Ben Camp called Bots & Beats: AI & the Future of Songwriting, at Berklee College of Music.

With AI, his team completed its paper in less than half that time and at far less cost. The database of the team’s studies has now expanded to more than 2 million papers.

Just as AI music and writing programs offer artists a broader view of possibilities, researchers hope to use the technology to speed discovery across the sciences. Yet while AI threatens to replace many artistic jobs, scientists appear more insulated.

“We need humans to know how to use AI and where to use it for these tasks,” Dr. Mahjoubi says. For example, a chatbot knows only what’s published – and scientists typically don’t publish their failures. A human specialist knows both the experimental successes and the failures, which sometimes provide key insights.

AI “doesn’t have the magic”

Back at Berklee College, Julia Caponegro (who uses the pen name Joyce Withers) grows increasingly worried as the class goes on.

Yes, the technology can give a far wider selection of options and sounds for her songs than collaborating with a single musician, says the junior majoring in songwriting, but “it doesn’t have the magic of me doing this with another person or even myself.”

For her, AI is interfering with the creativity that happens when humans get together to work on something. She doesn’t even use a rhyming dictionary, she says, preferring to ask her friends for inspiration about what rhymes with a particular word. But where exactly that inspiration comes from remains open to debate.

“Religion and science do not have to be antagonistic,” Professor Simonton points out. Gregor Mendel, the father of modern genetics, was also a monk. Georges Lemaître, formulator of the big bang theory, was a priest.

“For me, some of the most creative moments are in the shower,” says Mr. Ben-Tal, the British Israeli composer and an atheist. “I don’t know why.”

Ultimately, it’s not the artist who determines what’s truly creative, anyway. It’s the public that decides what resonates.

“That process of reception, that’s mysterious,” says Stephen Marche, a Canadian writer who wrote “Death of an Author,” a mystery novella composed with AI. “And it’s going to continue to be totally mysterious. ... It’s not some kind of algorithm that can be fulfilled.”

He quotes from Shakespeare’s “Hamlet”: “Our thoughts are ours, their ends none of our own.”