A study published on Friday warns that if creative industries accept artificial intelligence to help compose stories, future books and movies may begin to feel the same.
The study, which involved hundreds of volunteers and was published in Science Advances, comes as concerns grow about the influence of widely available AI programs that transform simple text prompts into rather sophisticated music, art, and writing.
“Our goal was to study to what extent and how generative AI might help humans with creativity,” co-author Anil Doshi of the University College London told AFP.
Doshi and co-author Oliver Hauser of the University of Exeter recruited approximately 300 participants to act as “writers” in their experiment.
These were persons who did not write for a living, and their intrinsic creative capacity was evaluated using a typical psychological test that required them to supply ten significantly different words.
The scientists then randomly divided them into three groups and asked them to create an eight-sentence story about one of three topics: an adventure on the wide seas, an adventure in the jungle, or an adventure on another planet.
Participants were also randomly assigned to three groups that received varied levels of AI assistance.
The first group received no assistance, the second was given a three-sentence tale concept from ChatGPT, and the third may receive up to five AI-generated story ideas to get them started.
Individual benefit, collective loss
After finishing their stories, participants were asked to rate their own work’s inventiveness using criteria such as how novel it was, how pleasant it was, and how likely the idea was to be transformed into a published book.
An additional 600 external human reviewers evaluated the tale using the same criteria.
The authors discovered that, on average, AI improved an individual writer’s inventiveness by up to 10% and the story’s enjoyability by 22%, notably in terms of structure and narrative twists.
These impacts were especially noticeable for writers who were assessed to be the least innovative during the initial work, “so it has this kind of leveling the playing field effect,” Doshi explained.
However, on a collective level, they discovered that AI-assisted stories appeared considerably more similar to one another than those written without AI assistance, since writers “anchored” themselves too strongly to the proposed concepts.
Hauser said this created a “social dilemma.” On the one hand, “you make it easier for people to get in; lowering barriers is good.” But if the collective novelty of art decreases, “it could be harmful down the line.”
Doshi said the research also showed that, just like introducing pocket calculators to children too early could prevent them from learning how to do basic arithmetic, there was a danger that people could rely too much on AI tools before developing underlying skills in writing, music or more.
People need to start thinking about “where in my workflow can I insert this tool to get the most benefit, while still inserting my own voice into the project or outcome.”