By Will Orwig and Daniel Schacter

Creativity in the Age of AI (generated by DALL-E)

What makes for a compelling story? Do computers have the capacity to be creative? These are evocative questions, especially with the recent explosion of interest in the capabilities of Artificial Intelligence (AI); however, there has been surprisingly little scientific research to address them. We tried to fill this gap by studying creative expression in the context of story writing. We conducted several studies comparing short stories written by humans and Large Language Models (LLMs). Specifically, we looked at GPT-3 and GPT-4, which are two of the most powerful LLMs that have fueled the recent fascination with the capabilities and limits of AI. In our experiments, both humans and the LLMs were prompted with three-word cues and instructed to write a five sentence creative story that includes and elaborates upon those cues. The stories were then rated by both humans and AI on a scale from 1 (very uncreative) to 5 (very creative).

We found that a broad range of ideas, or increased “semantic diversity,” as well as the presence of perceptual details, resulted in stories that were rated as more original and imaginative. This was true for stories written by humans and those generated by AI. Furthermore, AI ratings of creativity were largely consistent with those provided by humans. Stories written by human participants and LLMs did not differ in terms of their creative quality, suggesting that current AI systems can replicate aspects of human creativity. Such findings can potentially reshape our fundamental beliefs about creativity and raise the possibility that LLMs could function as collaborative partners in crafting engaging narratives. They also provide new insights into what makes a compelling story and hint at potential applications for these AI tools in creative industries. In short, our study suggests that the integration of human imagination with AI capabilities represents a new frontier in storytelling, challenging traditional views of the creative process.

Will Orwig is a PhD candidate in the Psychology Department at Harvard University.
Daniel Schacter is the William R. Kenan Jr. Professor of Psychology at Harvard University.

Learn more in the original research article:
The Language of Creativity: Evidence from Humans and Large Language Models.
Orwig, W., Edenbaum, E.R., Greene, J.D. and Schacter, D.L. J Creat Behav. 2024 Jan 11