Boden's three kinds of creativity
Much of the clearest thinking here comes from the cognitive scientist Margaret Boden, who died in 2025 after decades studying machine imagination. She argued that creativity is not one thing but three. Combinational creativity puts familiar ideas together in unfamiliar ways. Exploratory creativity works within an accepted style or rule set and finds new possibilities inside it. Transformational creativity changes the rules themselves, opening a space of ideas that was literally unthinkable before.
This framework is useful because it lets us be specific instead of arguing about 'real' creativity in the abstract. Today's AI is genuinely strong at the combinational and exploratory kinds. The transformational kind, where a system rewrites its own conceptual space, is where machines remain weakest and where human breakthroughs still stand out.
The evidence: move 37 and generative models
The most famous example is move 37. In 2016, during a match against the champion Lee Sedol, DeepMind's AlphaGo played a stone that professional commentators first thought was a mistake. It turned out to be brilliant, a move no top human would have chosen, and it helped win the game. Whatever you call it, it was novel, valuable, and surprising, the three properties most working definitions of creativity require.
Generative models now write, paint, compose, and design at scale, and some outputs are striking. By the novel-plus-valuable-plus-surprising test, a lot of this clears the bar. Boden herself held that the interesting question is not whether machines can appear creative, which they plainly can, but how they do it and whether we should credit the machine or the people whose work trained it.
The originality objection
The strongest pushback is that these systems recombine rather than originate. A generative model learns statistical patterns from enormous amounts of human work and then produces new arrangements of what it absorbed. Critics say that is sophisticated remixing, not invention, and note that human creators also constantly recombine, so the line is blurry. The debate has sharp practical stakes: lawsuits and copyright fights turn on whether training on copyrighted work, and generating outputs from it, counts as transformation or copying.
There is also the question of intention. Human creativity usually involves wanting to say something, caring whether it works, and understanding why it matters. Current AI has none of that; it optimizes an objective without any stake in the result. So the fair answer is that AI is creative in the measurable, output sense, and not obviously creative in the deeper sense of an agent expressing meaning it grasps. Which one you mean decides the argument.