Alternative creativity: why generative AI is creative, and why it is not like us

Abstract Whether generative artificial intelligence is creative is usually argued as a yes-or-no question, and both camps share an unexamined assumption, that there is one kind of creativity, the human kind, which the machine either has or lacks. This article defends a third position, alternative creativity. For any fixed medium the set of possible artefacts is finite and astronomically large, and almost none of it is valuable. Creativity, wherever it occurs, is therefore a compression of that space into a small manifold of meaningful candidates, followed by a value-guided search across the manifold. On the leading neuroscientific account, human brains work this way, and generative models demonstrably do. Neither creates from nothing, both recombine what they have internalised, both use constrained randomness, and both hold their skills in the strengths of their connections. In these respects the two are less different than is commonly supposed. Where they differ is in how the connections were shaped and in what each creator can reach at the moment of creation. A person's parameters are formed by a body acting in a physical and social world, a model's by an archive of what people have said and shown, and this difference in accessible context, together with a difference in how value is grounded, explains more of the divergence in what the two creators make than the difference of substrate does. In domains with checkable answers, machines already search beyond the human distribution under verifiable rewards. In the arts, where no verifier exists, they must borrow taste from people. From this account the article derives the homogenisation of machine output, the general intractability of tracing outputs to training inputs, the migration of authorship towards the specification of value, and the logic of human–machine collaboration, in which each creator supplies what the other constitutionally cannot.