A recent study shows that AI can help less creative people write more unique short stories, but at the same time it also stifles collective creativity.

The research was carried out by Anil Doshi and Oliver Hauser from University College London and the University of Exeter, and was published in the journal Science Advances. Despite focusing only on short stories, the study confirms many people's observations that AI may be useful but doesn't bring anything truly new to the creative sector.

“Our research is an early look at the big question of how large language models and generative AI more generally will affect human activities, including speech,” Hauser told TechCrunch. creative".

In the experiment, hundreds of participants were asked to write short stories (about 8 sentences) with a free theme. The first group wrote their own, the second group was given only one plot suggestion from GPT-4 (although they could repeat this idea for as many stories as they wanted), and the third group was allowed to use up to 5 hint from GPT-4.

Image source: Hauser, Joshi

The stories were then rated by the writers themselves and another group of readers who were not told what content was suggested by the AI. They judge based on the novelty, usefulness (publishability) and emotions the story brings.

The results showed that people with low creativity scores in terms of quality, attractiveness or novelty in group one still showed poor results.

However, when given ideas from GPT-4, their scores increased significantly.  And when they were allowed to refer to five ideas, their scores were even higher. This shows that AI can effectively support people who have difficulty finding ideas.

On the other hand, for people with high creativity scores, AI does not seem to bring significant benefits, and may even lead to poorer results. They can create their best work when not using reference AI.

Image source: Hauser, Joshi Analyzing the impact of AI

The researchers used OpenAI's API to evaluate the similarity of the stories. The results show that stories with AI become more similar and less diverse. This may be because less creative writers often rely on pre-existing ideas from AI, while creative writers build ideas from scratch themselves.

Although the difference was only about 9% to 10% compared to the group that did not use AI, this is enough to raise a warning: AI can reduce creative diversity. Researchers fear that if more and more people use AI to create, the result will be a "homogenization" of works.

This study serves as a wake-up call against claims about the limitless creativity of AI. The authors also acknowledge that their study is just the beginning, and that more in-depth research is needed.

“We hope this research will guide both AI technology and how we interact with it to ensure diversity in creative ideas, whether it is in literature, art or music.”