Made, Not Meaning: Why AI Generation Is Impressive But Not Creative

Introduction: The Creativity Question

When OpenAI's GPT-4 produces a sonnet or Midjourney renders a haunting landscape, the results can be genuinely difficult to distinguish from human work. This surface-level indistinguishability has prompted a wave of claims that artificial intelligence is not merely simulating creativity but actually practicing it. The argument deserves serious engagement, because the question is not trivial: if creativity is just the production of novel combinations, then large language models and diffusion networks are creative by definition. But if creativity requires something more — intention, experience, the pressure of meaning — then AI-generated work, however technically impressive, is something categorically different. This essay argues that AI generation is fundamentally derivative rather than genuinely creative, not because its outputs lack sophistication, but because the process that produces them is untethered from the conditions that make creativity meaningful. Where human creativity emerges from constraint, risk, and the negotiation of a self with the world, AI generation optimizes patterns within a distribution. The difference is not one of degree but of kind.

What Each Process Requires of Its Creator

The most useful place to begin the comparison is with what each process actually requires of its "creator." Human creativity, across every tradition of aesthetic theory, has been understood as inseparable from experience. The poet does not merely arrange words; she brings to those arrangements a history of losses, desires, and embodied encounters that give the choices their particular charge. As philosophers of creativity have noted, the act of creation involves what Margaret Boden calls "exploratory," "combinational," and "transformational" creativity — but crucially, Boden insists that even the most rule-bending transformational creativity in humans operates within a conceptual space that the creator has genuinely inhabited (Boden 4). A jazz musician who subverts harmonic convention does so by first internalizing that convention through years of listening and practice, and then finding it inadequate to what she wants to express. The subversion is meaningful because it costs something. AI systems, by contrast, require nothing from their operators except a prompt. The "training" of a large model is not analogous to a musician's apprenticeship, because the model has no stake in what it learns. It accumulates statistical regularities without experiencing any of the content those regularities encode.

Process, Resistance, and the Meaning of Making

This matters because creativity is not only about output but about process, and process shapes the meaning of what is made. Consider the difference in compositional process between a human novelist and a language model asked to continue a story. The novelist, as E.M. Forster famously described, often finds that characters "take over" — that the internal logic of a character's psychology resists the plot the author intended (Forster 85). This resistance, this back-pressure from the material, is constitutive of serious literary fiction. It indicates that the author has built a world with genuine internal consistency, one that pushes back against arbitrary authorial choices. Large language models have no such experience of resistance. They do not find their outputs surprising or troubling; they cannot discover, mid-generation, that a character would never say the thing the model is producing. What looks like narrative coherence in AI fiction is statistical plausibility, not the hard-won consistency of an imagined world. The model produces what is likely, not what is true to an internally maintained vision.

Defenders of AI creativity often respond that the distinction between "statistical plausibility" and "aesthetic judgment" is murkier than it appears. After all, human aesthetic judgments are also shaped by exposure to prior work; human writers also draw on patterns absorbed from reading. The difference, this argument runs, is quantitative rather than qualitative — humans are just slower, smaller language models. Margaret Boden's framework is sometimes recruited for this position, since she defines creativity in terms of novelty and value within a conceptual space, without requiring that the creator be conscious. But this reading of Boden flattens her argument. Her concept of transformational creativity requires not just producing a novel output but restructuring the conceptual space itself — a move that requires the agent to recognize inadequacy in the current space and to have reasons, grounded in experience, for reconceiving it (Boden 78). A language model does not recognize its conceptual space as inadequate; it has no concept of inadequacy relative to an intent, because it has no intent. It has objectives, set by its trainers, but objectives and intentions are not the same thing. An intention implies a subject who can be satisfied or disappointed; an objective is an optimization target.

Conclusion: Impressiveness Is Not Creativity

What this comparison ultimately reveals is that the question of AI creativity is not primarily a technical question but a philosophical one — specifically, a question about what we think creativity is for. If creativity is purely instrumental, a mechanism for producing novel and valuable artifacts, then AI may well qualify as creative, or will qualify soon. But if creativity is expressive — if it is a way of being in the world, of making meaning under conditions of uncertainty and finitude — then AI generation is not creative in any meaningful sense, however technically accomplished it becomes. The danger of conflating the two is not abstract. When we treat AI outputs as equivalent to human creative work, we implicitly devalue the conditions that make human creativity possible and important: vulnerability, experience, the slow accumulation of a life. AI generation is impressive, and it will become more impressive. But impressiveness is not creativity. The sonnet that moves you because a human being reached for language in a dark moment and found it is doing something categorically different from the sonnet that moves you because a model assigned high probability to those particular tokens. Both sonnets may make you cry. Only one of them earned the right to try.