AI creativity debunked: Generative systems lack genuine cognition

The author’s comparative analysis underscores that even when AI replicates the procedural logic of human creativity, it does not replicate the underlying cognition that gives creativity purpose. This analytical separation allows the concept of “artificial creativity” to stand on its own terms, distinct from the “human creative act” that depends on awareness, experience, and emotion.

A new study published in AI & Society challenges one of the most controversial questions in artificial intelligence research: can machines be truly creative if they lack cognition? The investigation scrutinizes the foundations of “artificial creativity” to determine whether generative systems like GPT-4 or DALL·E exhibit real creative capacity or simply imitate it.

Titled “Artificial Creativity: Can There Be Creativity Without Cognition?”, the study takes a philosophical and cognitive approach to defining the limits of machine creativity. It uses Antonio Lieto’s Minimal Cognitive Grid (MCG) to test whether current AI models possess cognitive properties essential for authentic creative acts. The conclusion is both clear and provocative: today’s generative systems are not cognitive, and while they can produce creative-looking results, they do so without awareness, intention, or understanding.

AI’s creative output without the mind behind it

The study dissects the notion that creativity requires a mind capable of thought, reflection, and purpose. Traditional human creativity, the author argues, arises from cognitive mechanisms such as memory, attention, association, and intentionality. These mental faculties allow individuals to connect ideas, perceive meaning, and assess value, processes that cannot be replicated by statistical models that only predict likely outputs based on previous data.

Using the Minimal Cognitive Grid, the author evaluates generative AI systems against criteria such as representational richness, learning, autonomy, and adaptability. His analysis finds that while AI demonstrates functional competence, creating poetry, visual art, or code, it does not meet the structural and representational requirements of cognition. It manipulates symbols without understanding them.

In practical terms, AI’s output may pass as creative in form but lacks the mental causation that defines genuine creativity. The author frames this distinction as the difference between a mechanistic simulation of creativity and creativity as a cognitive act. The latter depends on intention, the conscious drive to create something new and meaningful, which current AI systems simply do not possess.

Despite this limitation, the study acknowledges that AI can perform strongly on standardized creativity assessments. For example, large language models like GPT-4 score well on tests such as the Torrance Tests of Creative Thinking (TTCT) and the Alternative Uses Task (AUT). However, The author cautions that such tests measure output originality, not cognitive process. When a model recombines learned information in novel ways, it may produce surprising results, but these are statistical artifacts, not intentional acts of creation.

Redefining creativity through a computational lens

The study redefines creativity in non-cognitive terms, a minimal, positive definition of artificial creativity. Rather than dismissing AI’s generative power as imitation, The author proposes a framework that allows for a new class of creativity based purely on mechanism and output.

He defines artificial creativity as a non-cognitive, non-intentional, and non-authentic generative mechanism - a process capable of producing creative artifacts without consciousness or self-reflection. This reframing enables researchers to evaluate AI’s creative value without conflating it with human cognition.

To support this distinction, the study maps the creative workflow of AI systems to the classical Wallas–Jaoui model of creativity, which includes preparation, incubation, illumination, and verification. AI models mimic these stages: training corresponds to preparation, hidden-layer processing parallels incubation, generative output aligns with illumination, and feedback mechanisms act as verification. Yet this resemblance remains functional, not mental.

The author’s comparative analysis underscores that even when AI replicates the procedural logic of human creativity, it does not replicate the underlying cognition that gives creativity purpose. This analytical separation allows the concept of “artificial creativity” to stand on its own terms, distinct from the “human creative act” that depends on awareness, experience, and emotion.

Beyond simulation: The ethics and future of creative AI

In its broader implications, the paper raises ethical and epistemological questions about the nature of authorship and originality in a machine-mediated age. As AI-generated content increasingly floods art, music, and literature, societies must decide whether to value creativity as an outcome or a process.

The author warns that if creativity is judged solely by output, AI could eclipse human artists in recognition and production. However, if creativity is redefined as a cognitive and intentional process, machines can only ever simulate it. This distinction carries consequences for education, copyright law, and artistic recognition. It also challenges the anthropocentric tendency to evaluate machines by human standards.

The paper calls for a more transparent taxonomy of creativity, one that acknowledges machine-generated art as a distinct category rather than a counterfeit of human creativity. By doing so, AI research can avoid philosophical confusion while maintaining scientific rigor in measuring creative performance.

The author also connects his argument to emerging subfields like Explainable Computational Creativity (XCC), which aim to build AI systems that can describe their own creative reasoning. For such systems to be truly co-creative, he argues, they must incorporate memory, reasoning, argumentation, and metacognition, features absent in current architectures but essential to bridging the gap between generation and cognition.

This points toward a hybrid future where human and machine creativity may coexist, with AI acting as an amplifier rather than a substitute for cognitive imagination. In such a model, AI would handle combinatorial and generative tasks while humans retain evaluative, ethical, and emotional oversight.