New research on AI-enhanced creativity highlights a noteworthy social dilemma

Generative artificial intelligence (AI) is transforming various areas of our lives, from customer service to creative endeavors. A recent study in Science Advances explores how these AI systems, particularly large language models, influence human creativity in writing. The research indicates that AI can enhance the perceived creativity and quality of short stories, especially for writers who may not be inherently creative, though it also raises concerns about the risk of standardizing creative outputs.

Creativity, a fundamental aspect of human expression and innovation, is increasingly challenged by the rise of generative AI technologies, which question the uniqueness of human-generated content. In their study, Anil Doshi, an assistant professor at UCL School of Management, and Oliver Hauser, a professor and deputy director at the Institute for Data Science and Artificial Intelligence at the University of Exeter, sought to understand how generative AI affects people’s ability to create imaginative written content, with a focus on short fiction.

“We were both intrigued by the potential of generative AI,” the researchers shared with PsyPost. “We saw an opportunity to explore a common interest. Our focus on creativity stems from the fact that generative AI is a novel and potentially transformative technology, and we wanted to examine how it impacts a fundamental human trait—our capacity for creativity and the generation of new ideas and outputs.”

The researchers enlisted 500 participants from the Prolific platform, an online pool for research subjects, ensuring a dependable sample by selecting only those with high approval ratings and located in the United Kingdom. After adjusting for dropouts and exclusions, 293 participants completed the study.

Participants were randomly assigned one of three writing prompts: an adventure at sea, an adventure in the jungle, or an adventure on another planet. They were asked to craft an eight-sentence story aimed at a teenage and young adult audience. The participants were then split into three groups based on the level of AI assistance they received:

• Human-only group: This group wrote their stories without any AI support.
• Human with one AI idea group: Participants in this group could request a single, three-sentence story idea from OpenAI’s GPT-4 model.
• Human with five AI ideas group: Participants here had the option to request up to five story ideas from the same AI model.

Upon completing their stories, participants rated their own work on various stylistic attributes, such as creativity and enjoyment. Subsequently, a separate group of 600 evaluators from the same platform assessed the stories’ creativity, quality, and originality without knowing whether AI assistance had been used.

The researchers discovered that stories created with AI-generated ideas received higher ratings for creativity, quality, and enjoyability compared to those created without AI assistance. This effect was especially pronounced among participants with lower inherent creativity. For these less creative writers, access to multiple AI ideas led to significant improvements in the novelty of their stories, elevating their work to a level similar to that of more naturally creative participants.

“We found that AI-generated ideas enhance story creativity,” Doshi and Hauser told PsyPost. “What surprised us was that nearly all of the boost in creativity was seen among the least creative writers in our sample. Furthermore, receiving multiple AI ideas made their stories as creative as those from the most creative participants in our group. We observed a clear 'leveling the playing field' effect on story creativity due to AI assistance.”

However, a drawback of using AI-generated ideas was the increased similarity among the stories. The researchers noted that stories from the AI-assisted groups were more similar to each other and to the AI-generated ideas. This raises concerns about the potential for homogenization in creative outputs if AI tools become more prevalent. While AI can enhance individual creativity, it might also reduce collective diversity and novelty in creative works.

The researchers discovered that stories created with AI-generated ideas received higher ratings for creativity, quality, and enjoyability compared to those created without AI assistance. This effect was especially pronounced among participants with lower inherent creativity. For these less creative writers, access to multiple AI ideas led to significant improvements in the novelty of their stories, elevating their work to a level similar to that of more naturally creative participants.

“We found that AI-generated ideas enhance story creativity,” Doshi and Hauser told PsyPost. “What surprised us was that nearly all of the boost in creativity was seen among the least creative writers in our sample. Furthermore, receiving multiple AI ideas made their stories as creative as those from the most creative participants in our group. We observed a clear 'leveling the playing field' effect on story creativity due to AI assistance.”

However, a drawback of using AI-generated ideas was the increased similarity among the stories. The researchers noted that stories from the AI-assisted groups were more similar to each other and to the AI-generated ideas. This raises concerns about the potential for homogenization in creative outputs if AI tools become more prevalent. While AI can enhance individual creativity, it might also reduce collective diversity and novelty in creative works.

Another intriguing discovery was the gap between participants’ self-assessments and the external evaluations of their stories. Participants who used AI assistance did not rate their own stories as more creative or enjoyable compared to those who did not use AI. However, external evaluators consistently rated the AI-assisted stories higher. This indicates that individuals might not fully perceive the enhancements that AI provides to their creative work.

“Generative AI tools, like ChatGPT, enhance the average creativity of a writer’s story,” Doshi and Hauser noted. “However, stories with AI-generated ideas tended to be more similar to one another compared to those without AI assistance. This has significant implications—both positive and negative—for individuals and society.”

The researchers also highlighted a social dilemma: “While individual stories may be rated as more creative with AI help, widespread use of AI could reduce the overall novelty of ideas. This decrease in collective creativity might not be ideal from a societal perspective.”

The study underscores both the potential advantages and drawbacks of AI-assisted creativity, but it also comes with some limitations.

“Our study focused on a specific application of AI to better control the experiment,” the researchers clarified. “We standardized the prompt and did not allow participants to interact with the AI. This approach was designed to prevent scenarios where more skilled writers might generate better prompts to obtain superior ideas from the AI, which could skew our findings on the causal effect of AI ideas on creativity. Future research could explore how various prompts and interactions influence the creative process.”

“We are developing a research agenda to explore how generative AI might be used across different economic activities,” Doshi and Hauser continued. “For instance, we’re considering a project to investigate how AI aids in the creation of new ideas in various contexts, such as corporate strategy development. We’re also examining how different types of people might react to generative AI. Ultimately, our goal is to provide valuable insights for organizational and societal leaders when formulating their AI policies and strategies.”