3 Reasons Why AI Will Never Match Human Creativity: The Complete 2025 Guide
Science-backed · Updated April 2026
GPT-4 now outperforms 72% of humans on standardized creativity tests. But the most creative humans still win. Here’s what that actually means—and what it doesn’t.
TL;DR — The Honest Snapshot
- AI beats average humans on standardized divergent creativity tests (University of Montreal, January 2026 — 100,000 participants).
- The top 10% of human creators still surpass the best AI systems tested—including GPT-4, Claude, and Gemini.
- Consumers prefer human art when they know the source—but when labels are removed, AI-generated art is chosen nearly 45% of the time.
- Three things AI still can’t replicate: intentionality (conscious choice to create), lived emotional experience, and cultural embeddedness from actually being in the world.
- The question isn’t “can AI replace human creativity?” The better question is: what kind of creativity still matters uniquely to humans—and how do you develop it?
Let’s start with the finding that made headlines in January 2026 and made a lot of creative professionals very uncomfortable.
Researchers from the University of Montreal—including deep learning pioneer Yoshua Bengio—ran the largest creativity comparison study ever conducted. They pitted GPT-4, Claude, Gemini, and other LLMs against over 100,000 human participants on validated creativity tests. The result: GPT-4 outperformed 72% of all human participants on divergent linguistic creativity tasks.
That’s not a rounding error. That’s a clear signal.
But here’s what got buried in most of the coverage: the study’s lead researcher, Professor Karim Jerbi, was explicit about the ceiling. “Even the best AI systems still fall short of the levels reached by the most creative humans,” he said. The top 10% of human participants weren’t close to being beaten. And then there’s the deeper question the study didn’t even try to address: whether divergent word association is the same as the creativity that actually moves people.
That’s what this article is actually about.
Study: “Divergent creativity in humans and large language models”
Published January 21, 2026, in Scientific Reports. Led by Professor Karim Jerbi, Université de Montréal. Co-authored by Yoshua Bengio (Mila / Google DeepMind), with collaborators from Concordia University, University of Toronto, and Google DeepMind.
Sample: 100,000+ human participants. The largest human-AI creativity comparison ever conducted.
Primary test: Divergent Association Task (DAT)—generate 10 words with maximum semantic distance from each other. Measures divergent thinking—the cognitive flexibility to make unexpected connections.
Published January 21, 2026, in Scientific Reports. Led by Professor Karim Jerbi, Université de Montréal. Co-authored by Yoshua Bengio (Mila / Google DeepMind), with collaborators from Concordia University, University of Toronto, and Google DeepMind.
Sample: 100,000+ human participants. The largest human-AI creativity comparison ever conducted.
Primary test: Divergent Association Task (DAT)—generate 10 words with maximum semantic distance from each other. Measures divergent thinking—the cognitive flexibility to make unexpected connections.
DOI: 10.1038/s41598-025-25157-3
The DAT is a legitimate, peer-validated measure of divergent thinking. It’s not a toy test. If you can generate words that are maximally semantically distant—”telescope,” “legislation,” “mud,” “grief”—you demonstrate genuine cognitive flexibility. GPT-4 excelled here.
What it didn’t measure: emotional authenticity. Cultural meaning. The intentional decision to create something. The experience of loss that makes a Johnny Cash performance different from a technically similar vocal performance. Whether the creator actually wanted to make the thing.
That gap between “performs well on divergent word association” and “creates something that matters to other humans” is where this conversation gets genuinely interesting.
72%
of humans outperformed by GPT-4 on the DAT creativity test
Top 10%
of human creators still exceed best AI tested
45%
AI art chosen by consumers when source is unknown (SCIRP study, 2025)
Higher
Monetary value assigned to human art—even when identical to AI art
The uncomfortable truth is that AI has genuinely crossed meaningful creative thresholds. Pretending otherwise is just wishful thinking. But the things it’s good at and the things humans excel at aren’t actually the same things—and conflating them produces bad analysis.
- Divergent word and concept association at scale
- Generating high volumes of creative variations rapidly
- Pattern-based creativity: remixing existing forms
- Technical execution without fatigue
- Crossing genre and medium boundaries statistically
- Initial ideation and brainstorming
- Intentional creation — the choice to make something
- Emotional authenticity from lived experience
- Cultural embeddedness and social meaning-making
- Conceptual breakthroughs that rewrite categories
- Creativity as resistance, protest, or truth-telling
- Work that moves people precisely because of who made it
Here’s the point that matters most: the things humans still lead on aren’t soft, vague, or hand-wavy. They’re the things that explain why a Johnny Cash cover of “Hurt” at age 71 is one of the most devastating recordings in popular music history, while technically accurate AI vocal approximations of the same song are interesting artifacts but not the same experience. The difference isn’t technical. It’s biographical.
Psychologist Mark Runco, director of creativity research at Southern Oregon University, puts it directly: “I believe intentionality plays an important role in human creativity.” AI systems respond to prompts. They don’t decide to make something.
This sounds like a philosophical distinction until you realize it has real creative consequences. Picasso didn’t paint “Guernica” because someone prompted him. Banksy doesn’t install pieces at 4am because of an algorithmic optimization function. The intentional human decision to create—often at personal cost, often against convention—produces work with a specific quality that AI cannot reproduce because it starts from a fundamentally different place.
Real Example
Kahlo painted this dual self-portrait during her divorce from Diego Rivera, processing the rupture of her identity—her Mexican and European selves, her capacity to love and be loved. An AI can analyze the composition, the symbolism, the color palette. It can produce stylistically similar work. What it cannot do is make the decision to externalise a specific psychic wound at a specific moment in a specific life. That gap is the painting.
AI systems simulate emotional registers. They can write sad prose, generate uplifting imagery, produce content that triggers emotional responses in humans. That’s real and it matters. But it’s different from work generated from actual emotional experience.
The research on consumer perception backs this up. A landmark study published in Cognitive Research: Principles and Implications found that identical artworks were consistently judged as more profound, more beautiful, and more valuable when labeled “human-created” versus “AI-created.” The art didn’t change. The knowledge of its origin did. What consumers are responding to isn’t the surface quality—it’s the implied experience behind it.
“When participants are provided artworks labelled either as originals or as identical forgeries, participants preferred the originals over the forgeries—indicating that humans are sensitive to an authentic process of creative production.” — Newman & Bloom, via Springer Cognitive Research, 2023
This isn’t irrational bias. It’s a reasonable response to the fact that art is communication—and what matters isn’t just the message but who’s sending it and what it cost them to say it.
3. Cultural embeddedness — being of a place and time
Hip-hop didn’t emerge from statistical analysis of music genres. It emerged from specific conditions in the South Bronx in the 1970s—economic disinvestment, specific community networks, access to specific equipment, particular social and political pressures. The form carries that history in ways that can be analyzed but not replicated from the outside.
AI can be trained on cultural data. It cannot be of a culture. The difference is the difference between reading about grief and grieving.
The embodiment problem: Human creativity is shaped by having a body—experiencing pain, pleasure, space, time, other humans at close range. This isn’t a limitation AI needs to overcome. It’s a fundamentally different mode of existence. Creative work that emerges from embodied experience has qualities that work generated from textual patterns about embodied experience doesn’t have.
What Consumers Actually Think (The Data Is More Complicated Than “Humans Win”)
This is where it gets genuinely interesting—and where the original framing of “AI can’t match human creativity” starts to need some honest qualification.
A 2025 peer-reviewed study published in Scientific Reports of Psychology tested consumers on human versus AI-generated art without disclosing the source. AI-generated art was selected nearly 45% of the time. Not a marginal fringe. Close to a coin flip on surface quality alone.
When source labels were applied—when people knew something was human or AI-created—human art was strongly preferred. Consistently higher ratings for beauty, profundity, and worth. But here’s the uncomfortable implication: a large share of that preference is about knowing, not seeing.
Meanwhile, research from Zhengzhou University published in early 2026 found a different wrinkle: for product design, consumers were actually more willing to buy AI-designed products than human-designed ones, because AI design was perceived as more novel. The preference for human creators isn’t universal—it varies dramatically by category, context, and what someone is actually evaluating the work for.
| Context | Human advantage | AI competitive / superior |
|---|---|---|
| Fine art / emotional expression | Strong — provenance and biography matter deeply to perceived value | Competitive on surface aesthetics when source is hidden |
| Product / industrial design | Valued for craft story, bespoke quality | Strong — perceived novelty advantage in consumer studies |
| Advertising copy | Strong emotional resonance, brand authenticity | Competitive on volume and variation generation |
| Music / performance | Strongest gap — biography matters enormously | Technically impressive but lacks biographical weight |
| Brainstorming / ideation | Needed for direction, judgment, selection | Strong on volume and divergent association |
The Authenticity Paradox: Why Disclosure Changes Everything
The advertising industry is living through this in real time. Questions of AI and intellectual property have moved from abstract debate to active litigation and legislation. But what’s emerging in the market is more nuanced than a blanket consumer rejection of AI content.
Clothing brand Aerie publicly committed to not featuring AI-generated bodies in their campaigns—a counter-positioning strategy explicitly betting on authenticity as a differentiator. It works. For now. Because it’s genuinely rare.
The broader truth, per Edelman UK’s Emma De La Fosse: “Creatives need to lean in and make sure they are the ones wielding the tool.” The future isn’t AI replacing humans in creative work. It’s human creative direction determining whether AI output means something.
The question isn’t whether AI can generate something beautiful. It’s whether that beauty has a source that matters to the audience.
What This Means If You Work in a Creative Field
The honest answer is: it depends on what part of your creative work people are actually paying for.
If people pay you primarily for volume—for the first 20 variations of a headline, for iterating on brief concepts quickly, for generating derivative content at scale—that part of your role has a serious competitive pressure from AI. This is not a soft threat. The Montreal study data is real.
If people pay you for judgment, cultural positioning, emotional intelligence, brand voice that reflects actual human experience, or creative work that derives value from who you are and what you’ve lived—that part is safer, and it will become more valuable as AI output floods the market and the question shifts from “can you produce this?” to “does this mean anything?”
The practical implication: lean into the things only you can bring. Not just technical things—the kinds of references, experiences, cultural position, and genuine points of view that can’t be scraped from a training dataset because they haven’t been written down yet.
What successful creative professionals are doing in 2026: Using AI for technical execution and volume generation while making their unique human perspective—their specific cultural position, their biography, their genuine aesthetic commitments—the part that makes the work worth the client’s money.
The Collaboration Model That Actually Works
The framing of “AI vs. human creativity” is already outdated for most practical applications. The question of human authorship in AI-assisted work is being actively litigated in courts and copyright offices worldwide—but in studios and agencies, the answer has largely settled into practice: humans direct, AI executes, humans judge.
The meaningful question isn’t “can AI replace human creativity?” It’s closer to: what does creative direction look like when execution is cheap and fast? The answer seems to be that it becomes more important, not less. When generating a hundred variations of an image takes seconds, the skill of knowing which one is right—and why—becomes the scarce resource.
That skill is a human one. For now. And probably for longer than the current pace of AI capability growth suggests, because it requires the kind of cultural embeddedness and experiential grounding that formal training on text and images doesn’t easily capture.
Frequently Asked Questions
Did the Montreal study prove AI is more creative than humans?
Not exactly. It proved GPT-4 outperforms the average human on divergent linguistic creativity tasks using the Divergent Association Task. The study’s lead researcher explicitly noted that the most creative humans still exceed the best AI systems tested—and that the study measures a specific type of creative thinking, not creativity in its full breadth. Professor Jerbi himself said: “We need to move beyond this misleading sense of competition.”
Why do consumers value human art more even when they can’t tell the difference?
Springer’s Cognitive Research study explains this well: humans are sensitive to the implied process of creation. Knowing that something came from human experience—from actual choices, emotions, and a specific life—changes its perceived value. This is why even identical artworks are rated higher when labeled human-created. It’s not irrational; it’s a response to what art actually communicates.
What kind of creative work is most vulnerable to AI displacement?
Volume-based, technically competent creative work without a strong human-origin value: stock imagery, generic copy, template-based design, first-draft generation, variation creation. Work that derives value from who made it, why they made it, and the cultural or emotional weight of the creator’s specific position is substantially more durable.
Can AI develop consciousness and genuine creativity eventually?
This is genuinely unresolved. Current AI systems—even frontier models—process patterns without subjective experience as cognitive science defines it. The “hard problem of consciousness” (how subjective experience arises from physical processes) remains unsolved in neuroscience. Whether AI can ever develop genuine consciousness-based creativity is a philosophical question current research can’t answer. For practical purposes in 2026, current AI systems lack it.
How should I use AI in my creative work without losing what makes it valuable?
Use AI for execution and variation where volume matters. Keep creative direction, judgment, emotional intelligence, and cultural positioning as explicitly human contributions. Be transparent about AI involvement where it affects the value proposition of your work. The artists and agencies thriving in 2026 aren’t hiding AI use—they’re being clear about what they’re bringing that AI doesn’t.
The Bottom Line
The Montreal study is real, the data is solid, and the implication matters: AI has crossed the average human creativity threshold on standardized measures. You can’t argue with 100,000 data points.
But the same study—the largest of its kind ever conducted—explicitly found that the most creative humans still exceed the best AI. And there’s a category of creativity the test didn’t even try to measure: the work that matters because of who made it, what they’d lived through, what cultural position they occupy, and what it cost them to say it.
That kind of creativity isn’t diminished by AI getting better at divergent word association. It’s arguably becoming more valuable—because as AI-generated content floods every channel, the scarcity isn’t polished output anymore. It’s genuine human perspective.
The future belongs to human creators who understand both what AI can now do and what only they can bring. That’s not a consolation prize. That’s the actual opportunity.
More reading: AI & Human Creativity · AI & IP Rights · Understanding AI Outputs · Mastering AI Art
Sources & References
- Bellemare-Pepin A, et al. — “Divergent creativity in humans and large language models.” Scientific Reports, January 21, 2026. DOI: 10.1038/s41598-025-25157-3
- University of Montreal press release — “Creative talent: Has AI knocked humans out?” January 2026
- Olson J et al. — “Humans versus AI: whether and why we prefer human-created compared to AI-created artwork.” Cognitive Research: Principles and Implications, Springer, 2023
- Scientific Reports (SCIRP) — “The Value of Creativity: Human Produced Art vs. AI-Generated Art,” 2025
- PMC / Zhengzhou University — “Perceived design source (AI vs. human) on consumers’ purchase intention,” 2025–2026
- ORF Online — “AI Advertising and the Authenticity Paradox,” April 2026
- BrainFacts.org — “Can AI Truly Match Human Creativity?” August 2025 — featuring Dr. Mark Runco, Southern Oregon University
- Cunningham CV, Radvansky GA, Brockmole JR — “Human creativity versus artificial intelligence: source attribution, observer attitudes.” Frontiers in Psychology, May 2025
https://www.bestprompt.art/ai-creativity-vs-human-creativity-2025/
https://www.bestprompt.art/common-ai-prompting-pitfalls/
https://www.bestprompt.art/best-creative-ai-prompts-for-stunning-results-in-2025/
https://www.bestprompt.art/power-of-ai-in-research-writing/
