AI + Creativity
Today I was thinking about whether or not Ai can be authentically creative. To get to the bottom of this topic, I asked ChatGPT to generate a short essay based on literature arguing that AI can be authentically creative. I shared that essay with Claude and asked for an evidence-based response. I iteratred back and forth for a few turns, and then asked each LLM to generate "the truth" based on the existing arguments.
ChatGPT said:
The most defensible conclusion is that AI is genuinely creative in a functional, product-based sense, but not creative in the full human sense that includes consciousness, lived experience, intrinsic motivation, and authentic self-expression. That is why the two sides seem to clash while both capture part of the truth. Under the standard research framework of novelty, surprise, and value, AI clearly qualifies often enough that denying its creativity altogether is hard to sustain. But critics are also right that this does not show AI has a self, intentions, or human-style authenticity. So the truth is not “AI is absolutely creative” or “AI is constitutionally incapable of creativity.” It is that AI displays real but limited creativity, and the disagreement is mostly about whether we are using a broad evaluative definition of creativity or a narrower, human-centered one.
And Claude said:
The rebuttal is largely correct. The anti-AI essay I wrote argues its case well, but it wins primarily by definitional maneuvering — Runco's updated framework is a serious contribution, but it's also a response to the inconvenient fact that the older, widely-used definition couldn't exclude AI. The honest reading of the literature is closer to the original essay you provided: under the operational criteria that creativity researchers have actually used for decades, AI demonstrably meets the standard, and the empirical record — including the 2025 meta-analysis across 28 studies — supports that conclusion. The deeper philosophical objections about consciousness, intentionality, and embodied selfhood are genuinely important and unresolved, but they describe a contested narrowing of the concept, not a settled scientific verdict. The most defensible position is that AI is authentically creative by the dominant empirical standard, while the question of whether it is creative in the richer phenomenological sense that humans mean when they talk about their own creative lives remains genuinely open — and may be unanswerable with current tools.
While largely coming to the same conclusion, Claude's seem to be a bit more interesting to me largely because it "agreed" with ChatGPT claiming that ChatGPT's rebuttal of is largely corredt.
Ultimately, I agree with the idea that under existing creativey frameworks that can be studied and tested, AI demonstrates an equivalent creative ability. However, there is still much debate about whether or not those frameworks are even an accurate way to think about creativity.
Here are some of the essays:
The literature strongly supports the claim that AI can be creative, at least under the most widely used research definitions of creativity. A foundational definition from Margaret Boden treats creativity as producing ideas or artifacts that are novel, surprising, and valuable, and she argued decades ago that AI systems could achieve this through combinational, exploratory, and transformational processes. Recent philosophy and survey work still uses that framework because it fits how contemporary generative systems actually work, from recombining concepts in new ways to exploring vast conceptual spaces and generating outputs that human evaluators judge as original and worthwhile. More recent surveys of AI creativity likewise conclude that the central question is no longer whether machines can generate novel and valuable outputs, but how to evaluate different forms of machine creativity across artistic, linguistic, scientific, and problem-solving domains.
The empirical evidence is especially striking. In a 2024 Scientific Reports study, GPT-4 outperformed human participants on several standardized divergent thinking tasks, with researchers concluding that current AI language models showed higher creative potential than the human comparison group on those measures. A 2025 systematic literature review and meta-analysis synthesizing 28 studies and more than 8,000 participants found no significant overall difference between GenAI and humans in creative performance, which is itself powerful evidence that AI now performs at a human-comparable level on many creative tasks. At minimum, the literature no longer supports the older assumption that creativity is a uniquely human performance capacity. It supports the stronger claim that AI can already produce outputs that satisfy operational creativity criteria used in psychology and computational creativity research.
The case becomes even stronger when we move from isolated tests to real creative practice. Studies of writing and design show that AI assistance often increases the judged creativity, quality, and usefulness of outputs, especially for less experienced creators, while recent work in design education and co-creative workflows finds meaningful gains in idea generation and refinement. The main limitation identified in this literature is not that AI fails to be creative, but that widespread use can make outputs more similar at the collective level. In other words, the best objection in the current literature is a diversity objection, not an incapacity objection. Even critics who worry about intention, authenticity, or homogenization generally acknowledge that AI systems can produce artifacts that audiences experience as creative and that evaluators score as creative.
So the strongest conclusion from the literature is this: yes, AI absolutely has the ability to be creative, if creativity is judged by the standard scholarly criteria of novelty, surprise, and value. What remains contested is not the existence of creative capacity, but its source, meaning, and cultural status. Some scholars reserve “true” creativity for agents with consciousness, intention, or lived experience, but that is a philosophical narrowing of the term, not the dominant empirical standard used to measure creative performance. On the evidence now available, AI is not merely imitating creativity from the outside. It is generating original, valuable, and sometimes superior outputs across multiple domains, which is enough to conclude that creativity is a real capacity of contemporary AI systems, even if human creativity still differs in its phenomenology and social meaning. -- ChatGPT 5.4
And Claude:
The Unbridgeable Gap: Why AI Is Constitutionally Incapable of Authentic Creativity
The most powerful argument against AI creativity begins not with technology but with philosophy, and specifically with what the word "authentic" requires. Mark Runco, publishing in the Creativity Research Journal in 2023, argued that recent developments in AI compel a fundamental revision of the standard definition of creativity, adding authenticity and intentionality to the criteria of originality and effectiveness — and that even when AI output is original and effective, it lacks these two essential properties. Taylor & Francis Online This is not a peripheral objection. Runco goes on to argue that authenticity is impossible for AI because authenticity requires a self that expresses ideas and feelings without manipulating them for the sake of others, and there is simply no self in an AI system to do the expressing. ResearchGate Philosopher Berys Gaut reaches the same conclusion from the theory of agency: Gaut and Runco both contend that the challenge in acknowledging AI creativity on the same level as human creativity stems from AI's lack of intentionality, agency, and authenticity, with agency understood as the capacity to initiate events through genuine intention. Wiley Online Library What makes this philosophical convergence decisive is what happened to the existing theoretical framework when tested against AI output. A 2025 paper in AI & Society (Springer) concludes that until 2023, the standard theory of creativity — centered on novelty and usefulness — did not offer a definition sufficiently robust to exclude AI-generated outputs from the domain of human creativity, and that the theory required explicit reformulation precisely because AI had exposed its incompleteness. Springer In other words, the scholarly field itself acknowledged that its working definition was too shallow, not that AI had achieved creativity. The revised, philosophically adequate definition — one that incorporates consciousness-dependent properties like intentionality and authentic self-expression — excludes AI on structural grounds. What AI produces is better described, as Runco writes, as "pseudo-creativity": outputs that carry the surface profile of creative work while lacking the generative inner life that makes human creativity meaningful.
The deeper reason AI cannot satisfy those revised criteria lies in what John Searle identified decades ago and what contemporary cognitive science has extended: AI systems operate purely at the level of syntax, with no access to semantics, and therefore no access to the understanding that authentic creativity requires. Searle's Chinese Room argument, first published in 1980, maintains that a digital computer executing a program cannot produce real understanding, because computation is defined purely formally or syntactically, whereas minds have actual mental or semantic contents — and we cannot get from syntactic operations to semantic content simply by adding more of them. Stanford Encyclopedia of Philosophy A 2025 study in AI & Society formalizes this through the Minimal Cognitive Grid (MCG), establishing that generative AI systems such as large language models are non-cognitive — and that a structured argument based on that framework definitively excludes the possibility of intentional states in such systems. Springer This is the point where the empirical results cited by AI-creativity enthusiasts become deeply misleading. The same AI & Society analysis notes that GPT-4's consistent scoring in the 99th percentile across most metrics on standardized creativity tests strongly suggests prior exposure to those test materials during training, raising critical questions about the adequacy of current creativity assessments for evaluating AI. Springer The test instruments being used were designed to measure cognitive processes in human minds; applied to a system that has likely ingested those instruments as training data, they measure recall and pattern-matching, not the divergent thinking they were built to assess. As the literature on the stochastic parrot hypothesis emphasizes, an LLM is a system for stitching together sequences of linguistic forms observed in its training data according to probabilistic information about how they combine — without any reference to meaning. ResearchGate Creativity without meaning is a contradiction in terms. A system that produces statistically likely continuations of human-generated text is not being creative; it is being interpolative, and the difference matters absolutely when what is at issue is whether there is any authentic generative agency behind the output.
The third pillar of the case against AI creativity concerns the role of embodied, lived experience as the non-replicable substrate of genuine creative work. A 2025 paper in the Journal of Cultural Cognitive Science argues that the limitations of large language models in replicating human creativity are inseparable from the importance of embodied cognition, emotional depth, and the deeply personal and relational dimensions of creativity — drawing on thinkers from Rollo May to Judith Butler and Kimberlé Crenshaw to underscore that creativity emerges from the particular situatedness of a human life. Springer Human creative breakthroughs — the ones that actually transform artistic or intellectual domains — emerge from the tension between an individual's specific biography, desires, grief, obsessions, and the available cultural material. An AI system has none of those. Runco argues that intrinsic motivation, mindfulness, and choice are enormously important components of creativity, and that AI does not choose to produce novelty — like a porpoise conditioned to perform tricks, it may produce something that observers attribute with creativity, but what is actually present is at most mere novelty, and even that novelty is not intentional. ResearchGate This point has teeth precisely because it targets the process, not just the product. Theories of creative process — from incubation and insight to the Default Mode Network literature in neuroscience — describe phenomena that are grounded in subjective experience, emotional valence, and the sense of a problem felt from the inside. Reviews of current evidence confirm that AI still falls short in transformational creativity — the radical, paradigm-shifting insights that redefine domains — precisely because of its lack of curiosity, anomaly detection, emotional grounding, and embodied experience. Exploratio Journal These are not incidental features of human creativity that could in principle be swapped out. They are constitutive of what it means to face a creative problem as a being with something at stake — and no large language model has anything at stake.
Finally, the structural character of what AI actually produces, when examined at scale and over time, reveals not a creative system but one that is systematically opposed to what creativity most fundamentally means: departure from the norm. Research published in The Conversation in early 2026, summarizing a new empirical study, shows that generative AI systems tend toward homogenization when used autonomously and repeatedly — producing content that is compressed and generic, converging toward the familiar, the describable, and the conventional, even without any retraining on AI-generated data. The Conversation Research published in Science by Doshi and colleagues (2024) confirms that while access to AI can increase individual creativity scores, it reduces the collective diversity of outputs — stories written with AI assistance are more similar to each other than those written without any assistance. arXiv This matters because authentic creativity, at its highest expression, is precisely the act of making something that could only have come from this mind, at this moment, with this history. As the homogenization literature makes plain, large language models are by design built to predict the most statistically likely next word or idea — and that predictive strength becomes a creative limitation, compressing creative possibility into a narrower band of outputs rather than multiplying the idea space. Hupside A system optimized for likelihood is constitutionally oriented away from genuine novelty. The creative act, in the richest sense studied by psychologists, philosophers, and artists, involves courage — Rollo May's term — the willingness to venture into territory that has no probabilistic precedent. That is precisely what a system trained to minimize surprise cannot do. What the literature, taken together, establishes is that AI has produced a sophisticated and useful simulacrum of creativity. It has not produced creativity itself, and the gap between the two is not a matter of scale or compute: it is a matter of the presence or absence of a mind.
A note on the opposing case: The above represents the strongest arguments defenders of this position would offer. There is a serious and unresolved counter-literature. Empirical studies do show AI performing at human-comparable levels on standardized divergent-thinking tasks, and some philosophers argue that the revised definitions excluding AI amount to moving the goalposts rather than identifying a genuine boundary. Whether intentionality and authenticity are truly absent in AI systems — or simply different in form — remains one of the genuinely open questions at the intersection of philosophy of mind and cognitive science.