The Boundaries of Artificial Creativity
I had a conversation the other day with a colleague of mine during our lunch break. We were discussing the difference in creativity between humans and LLMs (or other language-based AI systems). We reached a unanimous conclusion:
Anything an LLM generates is just a reshuffling, a re-creation of what humans have already made. That means humans will always be more creative, as this limit does not apply to them.
LLMs, or other language-based forms of AI, are trained on content that has been produced before by humans. After training, an LLM works by figuring out what word fits best as the next word in the given context. And because those words come from the training dataset, anything an LLM creates must, by definition, be some combination of words that humans have written before.
Humans are different. They can invent something completely new, something that has not been made before by anyone at any time. That makes the ideas humans come up with more creative, or at least different from what LLMs do.
The British philosopher Margaret Boden described creativity in three forms: combinational (making new connections between familiar ideas), exploratory (pushing the boundaries within an existing framework), and transformational (changing the very rules of the framework). In that sense, what LLMs mostly do is combinational: they shuffle and remix the knowledge they were trained on.
Humans, however, are capable of transformational creativity—changing the space of possibilities itself. That is why human ideas feel fundamentally different: they can break the rules rather than only play within them.
AI won’t replace us as long as this holds. We’ll always need a human in the loop.