GPT-4 ranks in the 99th percentile for originality on the Torrance test, a global benchmark for creative thinking—a score surpassing almost all human participants. Yet a study published in Frontiers in Psychology in 2025 shows that this same model cannot distinguish an original idea from an ordinary one. This paradox captures the debate now engaging businesses, research laboratories and product teams: can generative AI, built on statistical models, truly create, or does it merely recombine what already exists?
This article examines the creative mechanics of AI systems, compares recent research with practical realities, and offers a framework for decision-makers who want to benefit from AI without confusing productivity with innovation.
TL;DR — Generative AI excels at divergent thinking—idea volume and rapid variations—but fails at critical evaluation and conceptual breakthroughs. Companies gaining the most value use it to amplify human creativity rather than replace it. The key is combining the machine's combinatorial power with human judgment, intention and strategic vision.
What Is Creativity, and Why Does the Question Arise for AI?
Creativity Is About More Than Novelty
In psychology, creativity rests on two pillars: novelty—producing something that did not previously exist in that form—and relevance—that something solves a problem or creates value in a given context. An original but useless idea is not creative; it is merely random.
This distinction is essential to understanding AI performance. A language model can generate thousands of unprecedented combinations of words or concepts. But assessing whether one of those combinations meets a real need, positively surprises a user or opens a market requires human judgment rather than statistical calculation.
How a Generative Model Actually Works
A Large Language Model (LLM) such as GPT-4 or Claude does not think. It predicts the most likely next token in a sequence, drawing on statistical patterns learned from billions of documents. A diffusion model such as Midjourney or DALL-E progressively removes noise from a random image, guided by a text prompt.
In both cases, the fundamental mechanism is the same: statistical recombination of existing patterns. The term “stochastic parrot,” coined in 2021 by Emily Bender and Timnit Gebru, describes this reality: a system that assembles sequences of linguistic forms using probabilistic information about how they combine, without any reference to meaning.
The scientific debate has evolved, however. A study published in WIREs Computational Statistics in 2025 shows that the most advanced models demonstrate composition and reasoning capabilities beyond simple memorization. Reality lies somewhere between parrot and genius, and this middle ground is precisely what interests businesses.
Why This Question Directly Concerns Decision-Makers
“Is AI creative?” is not a philosophical question for a CIO selecting tools, a CEO investing in innovation or a product manager designing a product. The answer determines:
- Which processes can be delegated to AI, and which cannot
- Which roles remain irreplaceable within teams
- What real added value to expect from investment in generative AI
According to McKinsey, generative AI could unlock USD 2.6–4.4 trillion in additional global value. But that value materializes only when organizations understand where and how AI actually creates, and where it merely gives the illusion of creation.
What Research Shows: AI's Creative Performance and Limits
Impressive Divergent-Thinking Scores
The raw results are spectacular. According to a study published in the Journal of Creativity in 2023, GPT-4 reaches the 99th percentile for originality and fluency on the Torrance Tests of Creative Thinking (TTCT), an international benchmark for divergent thinking. Its flexibility ranks between the 93rd and 99th percentiles.
A more recent 2025 study comparing ChatGPT-4o, DeepSeek-V3 and Gemini 2.0 with human participants on the Alternative Uses Test (AUT) and Remote Associates Test (RAT) confirms the trend: generative models produce a significantly greater volume of ideas and demonstrate comparable or superior ability to generate original responses.
These figures feed an appealing narrative that AI is more creative than humans. The reality is more nuanced.
The Paradox of Generative Creativity
The 2025 Frontiers in Psychology study reveals a fundamental paradox. Using the egg task, researchers observed that ChatGPT-4o:
- Produces more ideas than the panel's 47 human participants
- Shows fixation bias comparable to humans: about 80% of generated ideas remain within dominant conceptual categories
- Fails to distinguish original from conventional ideas, whereas humans naturally make that distinction
The last point is decisive. Human creativity goes beyond producing ideas: it includes the ability to evaluate, filter and select ideas worth pursuing. This metacognitive ability—knowing an idea is good—remains absent from current models.
AI Does Not Make Fundamental Discoveries
A study published in Scientific Reports (Nature, 2025) reinforces the point: current generative AI can make incremental discoveries but cannot produce fundamental discoveries from nothing. The authors describe the current approach as world-taking—drawing from the existing world—rather than world-making, or creating a new one.
This distinction clarifies a frequently misunderstood point: AI excels at exploring possibilities within a given framework. Redefining the framework itself—the conceptual breakthrough behind the most transformative innovations—remains a human prerogative.
| Creative dimension | AI performance | Human performance | Advantage |
|---|---|---|---|
| Idea volume: fluency | Very high | Limited by time | AI |
| Response originality | 99th percentile: TTCT | Varies by individual | AI on standardized tests |
| Evaluating originality | Cannot distinguish ordinary from original | Natural differentiation | Human |
| Fixation bias | ~80% conventional ideas | ~80% conventional ideas | Comparable |
| Incremental discovery | Effective | Effective | Comparable |
| Conceptual breakthrough | Absent | Present: rare but possible | Human |
| Intention and strategic vision | Absent | Decisive | Human |
AI as an Amplifier: Use Cases That Work
Faster Exploration of the Design Space
The best-documented creative AI use case is rapid exploration of variations. Mattel uses generative AI in Hot Wheels product development: design teams now generate four times as many visual concepts, accelerating ideation without replacing designers' judgment about what deserves production.
The mechanism is simple: AI compresses the exploration, or divergence, phase, which traditionally takes weeks of brainstorming and mockups. A designer can ask Midjourney for ten variations in minutes, evaluate them, combine relevant elements and refine with the team. The freed-up time is reinvested in convergence, where the real creative value lies.
Personalization at Scale
Stitch Fix illustrates another high-value use. The company uses DALL-E to visualize clothing products based on customer preferences for color, fabric and style. Human stylists can quickly identify similar items in stock and suggest combinations the catalog alone did not reveal.
AI does not create the style; it makes a space of possibilities visible for people to explore with their expertise. This collaboration model also appears in architecture—ARCHITEChTURES generates building plans meeting geographic and budget constraints for architects to refine—music—composition models suggest harmonic structures for composers to rework—and marketing—AI generates advertising message variations for creative teams to evaluate and select.
Co-Creation and the Measured Augmentation Effect
Harvard Business School's 2024 study, “The Crowdless Future? Generative AI and Creative Problem Solving,” quantifies augmentation. Compared with traditional creative crowdsourcing, AI-assisted solutions:
- Reduce ideation costs by 99% and time by 99.8%
- Score higher in strategic viability, environmental value and financial value
- Are judged to have higher overall quality when all criteria are considered together
Notably, purely human solutions retain an advantage in perceived novelty. AI optimizes; humans surprise. This complementarity underpins a clear-eyed innovation strategy.
Practical guide — Five Questions to Assess Whether AI Can Amplify Your Creative Process
- Does your creative process include broad exploration such as brainstorming, mockups or variations? → AI can accelerate this phase.
- Does value lie mainly in selecting and refining ideas rather than generating them? → AI is a good candidate.
- Does your team have the domain expertise to evaluate AI proposals' relevance? → This is essential.
- Is the intended innovation incremental—improving something existing—or disruptive—a new paradigm? → AI excels at incremental innovation, not breakthroughs.
- Do you have data or a reference corpus AI can use? → The richer the corpus, the more relevant AI becomes.
When AI Fails: The Blind Spots of Statistical Creativity
Insidious Homogenization
A documented but rarely discussed risk is that widespread generative AI use in creative processes tends to homogenize outputs. Because users of the same model share its statistical biases, work converges toward an average style. The phenomenon is already visible in graphic design: Midjourney images share a recognizable aesthetic—golden light, hyperrealistic detail and centered compositions—that becomes an unintended stylistic signature.
The strategic risk for businesses is real: trying to differentiate a brand with the same tools as competitors, trained on the same data, mechanically produces less differentiation. In this scenario, AI becomes a source of sameness rather than competitive advantage.
The Absence of Productive Friction
Human creativity often emerges from constraints, errors and friction. A musician playing the wrong chord sometimes discovers an unexpected progression. A designer facing a technical limitation invents an elegant solution. A developer working within a tight budget designs a cleaner architecture.
Generative AI removes this friction. It produces satisfactory outputs by default, optimized for average statistical expectations. Yet true innovation emerges precisely where the expected result is inadequate, where dissatisfaction drives exploration of unconventional directions. Easy generation can paradoxically stifle the creative impulse born of difficulty.
The Problem of Anchoring and Lazy Thinking
Cognitive psychology studies show that the first suggestions received strongly anchor later decisions. When a product manager starts by asking ChatGPT for ten innovative features for a delivery app, those suggestions immediately frame exploration. Ideas missing from the list—perhaps the most disruptive ones—are statistically less likely to emerge later.
Time pressure and productivity's appeal amplify this cognitive anchoring. Why spend three hours thinking when AI provides an answer in three seconds? Time saved becomes depth lost, and the organization's creativity gradually weakens without anyone noticing.
The Impossibility of Intention
Perhaps the most fundamental point is the least technical. Human creativity is intrinsically intentional: it pursues a goal, expresses a vision, responds to personal frustration or translates an intuition shaped by experience. Picasso did not paint the most likely token; he painted what he had decided to show the world.
AI has no intention, frustration or vision. It optimizes a mathematical function. This absence of intention is not a flaw to fix in the next version; it is a structural feature of statistical systems. Understanding this limit means understanding why AI will always be a tool for creativity, never its author.
The Amplification Thesis: A Framework for Businesses
MIT's EPOCH Model: What AI Does Not Replace
In 2025, MIT Sloan formalized the EPOCH framework to identify human capabilities AI does not replicate:
- Empathy: emotional understanding of user needs
- Presence: grounding in a physical and social context
- Opinion/Judgment: the ability to choose between ambiguous options
- Creativity: creativity as conceptual breakthrough
- Hope: imagining a desirable future that motivates action
This framework clearly distinguishes automation, transferring a task to a machine, from augmentation, where AI strengthens human productivity. For creative tasks, MIT consistently recommends augmentation: AI supplies raw material, while people provide judgment and direction.
Structuring an AI-Augmented Creative Process
An effective innovation process integrates AI at specific stages without delegating the entire chain:
Phase 1 — Strategic framing: 100% human Define the problem, identify constraints and formulate the creative ambition. AI has nothing to contribute here: it does not know why you are innovating.
Phase 2 — Divergent exploration: AI + human Generate many possibilities: visual concepts, text variations and functional scenarios. AI excels at this phase. People define prompts, guide directions and identify unexpected avenues.
Phase 3 — Evaluation and selection: 100% human Sort proposals, identify what is truly original—which AI cannot do—and evaluate feasibility and strategic relevance. This is the most critical and most underestimated phase.
Phase 4 — Refinement and prototyping: AI + human Develop selected concepts, produce mockups and iterate quickly. AI accelerates production; people guide the creative direction.
Phase 5 — Field validation: 100% human Test with users, gather feedback and adjust. EPOCH's empathy and presence are irreplaceable here.
Metrics That Matter
Measuring AI's effect on creativity requires more than counting generated ideas. Mature organizations track more nuanced indicators:
| Metric | What it measures | Pitfall to avoid |
|---|---|---|
| Volume of generated ideas | Raw productivity | Confusing quantity with quality |
| Idea-to-prototype conversion rate | Relevance of exploration | Prototyping everything without filtering |
| Time from ideation to market | Cycle acceleration | Sacrificing field validation |
| Perceived differentiation: customer research | Real originality | Comparing against competitors' AI outputs |
| Creative team satisfaction | Adoption and ownership | Ignoring resistance to change |
Practical Examples: Three Business Scenarios Facing the AI–Creativity Dilemma
Scenario 1 — The Industrial SME Rethinking Its Products
A 200-person company manufactures professional kitchen equipment. Its four-engineer design office spends three months a year developing new products. The traditional process includes trend research, competitor benchmarking, brainstorming, CAD mockups and customer validation.
With AI: engineers use image-generation tools to explore dozens of shapes and configurations in a few hours. They supply technical constraints—dimensions, materials and standards—and receive variations they would not have considered alone. Design time falls by 40%, but more importantly, the range of explored options increases fivefold.
What AI does not do: it does not understand the ergonomics of a chef during service or the maintenance constraints of a kitchen serving 200 covers. An engineer with ten years of field experience knows which attractive on-screen ideas cannot work in practice. Their judgment remains the decisive filter.
Scenario 2 — The SaaS Startup Seeking Its Positioning
A founder is developing a project management platform for construction. They need a differentiating angle in a saturated market containing Asana, Monday, Notion and dozens of vertical competitors.
With AI: the founder analyzes thousands of user reviews of competing platforms, identifies recurring frustrations and generates feature concepts addressing them. AI maps the problem space faster than conventional market research.
What AI does not do: it does not tell the founder which problem deserves priority. Product vision—backing one angle rather than another, accepting that you cannot do everything, choosing a precise segment—is strategic judgment AI cannot exercise. Startups delegating this decision to AI produce average products that tick every box without surprising anyone.
Scenario 3 — The Banking Group Automating Customer Communication
A retail bank's marketing department produces 500 content items a month: emails, notifications, landing pages and social media posts. It deploys generative AI to reduce production time.
With AI: writers use AI for first drafts. Each item is generated in 30 seconds instead of 45 minutes. The team shifts from production to curation and strategic personalization. Volume triples and production time falls by 70%.
The identified risk: after six months, engagement metrics drop 15%. Diagnosis: AI-generated content is technically correct but lacks the distinctive touch that differentiated the brand. Stylistic homogenization has eroded editorial identity. The solution is to restore human rewriting for strategic content and reserve full automation for transactional communications with little creative significance.
Building a Clear-Eyed AI Creativity Strategy
Three Mistakes to Avoid
Mistake 1 — Believing AI creates and delegating innovation to it. AI recombines, quickly, effectively and at scale. But recombination is not innovation. Innovation begins when someone decides that a particular recombination deserves to exist in the real world and commits to making it happen.
Mistake 2 — Rejecting AI because it is not truly creative. This misses the point. Nobody faults a microscope for not being a biologist. AI is an instrument of creativity: it amplifies exploration and accelerates production. Rejecting it leaves you at an operational disadvantage against competitors using it intelligently.
Mistake 3 — Measuring AI creativity by output volume. The 68% of companies reporting time savings from AI in creative content development (Ipsos/Google, 2024) say nothing about content quality. The real success indicator is an organization's ability to produce differentiated results, not simply more of them.
A Decision Framework for Leaders
Before integrating AI into a creative process, ask three foundational questions:
1. What type of creativity is involved?
- Combinatorial creativity: new combinations of existing elements → AI is a powerful ally
- Exploratory creativity: exploring a defined possibility space → AI is a good partner
- Transformational creativity: redefining the rules → AI cannot help directly
2. Where does value lie in your innovation chain? If value centers on ideation—finding ideas—AI can be transformative. If it centers on selection and execution—choosing and realizing the right idea—AI remains an auxiliary tool.
3. Is your organization mature enough to use AI creatively? AI amplifies what exists. A strong creative team augmented by AI produces exceptional results. A weak team augmented by AI produces more mediocrity, faster. The tool does not compensate for missing expertise.
The Future: Creative Agents, Not Autonomous Creators
The global market for AI in art and creativity was valued at USD 4.8 billion in 2024, growing at 18.3% annually (Grand View Research). Generative AI investment increased sixfold between 2023 and 2024 to USD 13.8 billion. The direction is clear: creative AI will become ubiquitous.
But the trajectory is not toward autonomous AI creators. According to Gartner's 2025 Hype Cycle, the AI innovations closest to mainstream adoption are AI agents and multimodal AI, systems designed to collaborate with people rather than replace them. The future of AI-assisted creativity is collaborative: agents that propose, iterate and produce, guided by humans who decide, judge and direct.
FAQ
Can generative AI actually produce original ideas? Yes, in a statistical sense: GPT-4 reaches the 99th percentile for originality on the Torrance test. But this originality is combinatorial: AI recombines existing patterns. It cannot evaluate whether an idea is truly novel or merely unusual. Breakthrough originality remains a human capability.
Which creative professions are most affected by AI? Roles with substantial repetitive production—adapting designs, writing marketing content and producing graphic variations—are most affected. Roles whose value lies in artistic vision, aesthetic judgment or empathetic understanding of users remain largely protected.
Should my company invest in AI for creative processes? If your processes include broad exploration—ideation, mockups and variations—AI can accelerate that phase by 40–70%. Investment is justified provided your team has the domain expertise to evaluate and filter proposals. Without that human filter, AI produces volume without value.
Will AI replace designers and creative professionals? The data does not support this hypothesis. MIT's 2025 EPOCH framework identifies transformational creativity as distinctly human. AI changes creative roles from producers to curators and creative directors without eliminating them. Companies replacing creative teams entirely with AI generally see differentiation decline in the medium term.
Is There a Risk in Relying Too Heavily on AI to Innovate? The main risk is homogenization. Users of the same model share statistical biases, so outputs converge toward an average style. Cognitive anchoring from initial AI suggestions can also limit exploration of unconventional paths. The solution is to use AI as a starting point rather than a destination.
What Budget Should You Allow for Integrating AI into a Creative Process? Generative tools such as Midjourney, ChatGPT and Claude cost €20–€200 per user per month. The real investment is team training and process adaptation: allow 2–5 training days per employee and 3–6 months to stabilize an AI-augmented workflow.
AI Coder Squad: Where Technical Creativity Meets AI Acceleration
The most successful custom applications emerge from a strong business vision paired with technical execution that knows how to use AI tools without surrendering to them. That is exactly AI Coder Squad's position: senior developers who use AI to accelerate production while providing the technical and strategic judgment machines do not supply.
AI Coder Squad designs custom applications and AI agents for businesses that want to move quickly without sacrificing quality—with senior developers and an AI-powered approach.
→ Start your project and discover how AI Coder Squad can accelerate your next delivery.