Study: Artificial intelligence improves individual creative writing, but reduces creative diversity overall

Researchers are exploring the existential implications of generating artificial intelligence, including whether advanced technology will actually make humans more creative or whether it will narrow our horizons.

The new study, published in Science Advances by two researchers from University College London and the University of Exeter, combined hundreds of short stories created only by humans with the creative help of ChatGPT’s generative artificial intelligence. Created short stories are compared. One group of writers only gets their own ideas, a second group can ask ChatGPT for a story idea, and a third group can use a set of five ChatGPT-produced prompts. The stories were then rated based on “novelty, usefulness (i.e. likelihood of publication) and emotional enjoyment,” according to the report TechCrunch.

“These results suggest that an increase in individual creativity may lead to a loss of collective novelty,” the study states. “This dynamic resembles a social dilemma: With generative AI, writers are individually better off. , but overall the scope of the novel’s content has narrowed.”

Prior to the writing session, participants’ creativity was “measured” through a commonly used word production task that established a standard of creativity for the respondent. Those who tested lower on the Creativity Agent test scored higher on personal writing when presented with AI-generated ideas. But for those who already had high creativity scores, AI ideas had little to no benefit to their story scores.

Furthermore, the AI-generated prompt-assisted story pool was seen as lacking diversity and exhibiting less unique writing characteristics, suggesting limitations to ChatGPT’s overall originality. Mashable’s Cecily Mauran reports that the new study’s literature findings add to concerns about AI self-consuming training loops, or the problem that AI models trained solely on AI output degrade the AI ​​model itself.

Study author Oliver Hauser said in a commentary TechCrunch: “Our study represents an early view on a very big question, namely how large language models and generative AI will impact human activity more broadly, including creativity… Importantly, AI is actually undergoing rigorous evaluation, rather than just widespread implementation, assuming it will produce positive results.