When writers get ideas for novels from AI, it improves their creativity, but it also creates big problems
As the author of the Akutagawa Prize-winning work revealed, '
Generative AI enhances individual creativity but reduces the collective diversity of novel content | Science Advances
https://www.science.org/doi/10.1126/sciadv.adn5290
AI prompts can boost writers' creativity but result in similar stories, study finds | Artificial intelligence (AI) | The Guardian
https://www.theguardian.com/technology/article/2024/jul/12/ai-prompts-can-boost-writers-creativity-but-result-in-similar-stories-study-finds
Experiment finds AI boosts creativity individually — but lowers it collectively | TechCrunch
https://techcrunch.com/2024/07/12/experiment-finds-ai-boosts-creativity-individually-but-lowers-it-collectively/
In a study by Anil Doshi of University College London and Oliver Hauser of the University of Exeter, an experiment was conducted to see if generative AI could create creative short stories as a first step to understanding the relationship between generative AI and human creativity. The researchers believe that generative AI has two roles in influencing creative writing: providing a starting point as a springboard for ideas and providing new ideas to overcome writer's block. Therefore, the researchers explored the potential of generative AI by having a generative AI provide ideas to writers and evaluating whether the short stories created from those ideas were highly creative.
In the experiment, 293 participants were divided into three teams of about 100 people each, and all were asked to write a short story of about eight sentences for teenagers to those in their early twenties. Group A could write the story on their own, Group B could receive one idea from GPT-4 that would inspire them to write the story, and Group C could receive up to five ideas from GPT-4. The completed stories were evaluated by another group of 600 people on items such as 'story novelty,' 'publishability,' and 'emotional characteristics.' At this time, the evaluators read the stories without knowing in advance whether they contained parts written by AI, and then guessed which stories were written using generative AI after reading them.
Participants were free to choose whether or not to adopt the ideas provided by the generative AI, but overall, 82% of Group B accepted the ideas provided by the AI. In addition, of Group C, which could receive up to five ideas, 93% adopted at least one idea, and about 25% requested all five ideas.
As a result, it was found that Groups B and C, which included ideas from generative AI, had improved evaluations of the novelty and publishability of stories. Group B, which received one idea, had an average evaluation score of 5.4% higher in novelty of stories and 3.7% higher in publishability than Group A, which did not use the help of GPT-4. Furthermore, Group C, which had five ideas to choose from, had an average evaluation score of 8.1% higher in novelty and 9% higher in publishability than Group A.
Next, on items assessing the raters' affective characteristics of the story based on categories of common reader interest, such as how well the story is written, whether it is enjoyable, interesting, boring, and how many twists and turns the story has, both Groups B and C rated it higher than Group A. However, only on the question 'Is this story interesting?' did Group C, which received many ideas, score lower than Group B, which received a single idea, and both of those scores were lower than the work of Group A. However, the difference was small and 'not statistically significant,' the paper states.
Regarding the results of the study, Doshi said, 'In terms of the creativity provided by generative AI, while the effect size was not particularly large and scores did not increase, we did find a difference that could be considered statistically significant. However, we do not believe that the stories of inherently creative people are being 'enhanced' by the AI's ideas. Rather, people who scored highly on the creativity index did not receive any additional recognition from generative AI, suggesting that they may not benefit at all, or may be negatively affected. The ones who benefit most from generative AI are writers who are naturally less imaginative.'
Furthermore, the researchers point out that one of the caveats of using generative AI for creative ideas is the problem of 'the homogenization of works.' In addition to the participants' evaluations, the researchers used Open AI's API to analyze how similar each story was to other works in the same category. As a result, the group that used generative AI had a high similarity to stories in the same category created by other participants, resulting in low overall diversity.
Hauser acknowledged that 'our research is still in its infancy, and there are many factors that have not been included,' but added, 'Ideally, our research will help guide both the technology and how we engage with it to ensure continued diversity in creative ideas, whether in writing, art, or music. Evaluating our use of AI will be essential to ensure we reap the benefits of this potentially transformative technology without falling prey to its potential drawbacks.'
Related Posts:
