In this paper we explore whether Large Language Models (LLMs) can exhibit scientific creativity. To address this question, we first provide a thorough overview of the philosophical literature on the nature of creativity. We argue for a definition of creativity that distinguishes between product and producer. Using work by Boden and Gaut, we call products creative if they are novel, valuable and surprising, while producers are creative if they produce creative products with `flair' (which include…
Read moreIn this paper we explore whether Large Language Models (LLMs) can exhibit scientific creativity. To address this question, we first provide a thorough overview of the philosophical literature on the nature of creativity. We argue for a definition of creativity that distinguishes between product and producer. Using work by Boden and Gaut, we call products creative if they are novel, valuable and surprising, while producers are creative if they produce creative products with `flair' (which includes purpose, understanding, judgement, and evaluation). We then focus on LLMs, which are often argued to be mere `stochastic parrots', fundamentally incapable of creativity. We review the experimental literature on LLMs and find that, while exhibiting some parrot-like tendencies, they can produce creative scientific products and demonstrate a limited but real capacity for scientific creativity.
