Can AI be creative or is it just remixing what it has seen?

It depends on a definition of creativity that humans haven't settled either — which is exactly why the argument never resolves. But the 'just remixing' framing is too dismissive, and 'genuinely creative like a person' is too generous. Something real is happening in between. The case that it's more than remixing: these models produce combinations that appear nowhere in their training data, and they generalise to genuinely novel problems. They can produce a working solution to a problem no one has posed before, in a style no one has written in. Interpolating in a learned high-dimensional space produces outputs that are new in any ordinary sense of the word. If a human produced them, we wouldn't hesitate to call it creative. The case that something is missing: - **No intention.** Human creativity typically involves wanting to express something, to a purpose. The model has no stake in the output. - **No lived grounding.** Human creative work draws on embodied experience — grief, hunger, a specific afternoon. The model has patterns of text describing these things. - **No taste in the strong sense.** It can generate a thousand variations; deciding which one matters is currently supplied by a human. Selection is arguably where most creativity lives. - **Bounded by its distribution.** It's very good at 'more of this, combined unusually'. It does not appear to make the discontinuous leaps that reframe a field entirely — though it's fair to note humans do that rarely too. A useful distinction from creativity research: *combinational* creativity (novel combinations of familiar ideas), *exploratory* creativity (finding new points within an existing conceptual space), and *transformational* creativity (changing the rules of the space itself). Current systems are strong at the first, decent at the second, and show little evidence of the third. Most human creative work is also the first two, which is the uncomfortable part of the comparison. Where this lands practically for anyone doing creative work: the model is a very fast generator of options and a poor judge of which matter. Human contribution shifts toward direction, curation, taste and knowing what's worth making — which was arguably always the scarce part. The people getting the most from these tools treat them as a tireless collaborator producing raw material, not as a replacement for having something to say. And the honest meta-point: our confidence that human creativity is categorically different rests partly on not understanding how our own works. That's not an argument that they're the same — just a reason for humility in both directions.