Study Shows AI Boosts Personal Writing Creativity, Limits Group Diversity

A study authored by Anil Doshi from University College London and Oliver Hauser from the University of Exeter reveals the nuanced role of artificial intelligence (AI) in creative writing tasks, highlighting how AI can simultaneously boost individual creativity while curbing overall group originality. The research, detailed in Science Advances, delves into how AI assistance impacts short story writing. AI’s Role in Individual Creativity Participants were tasked with crafting short stories, each around eight sentences long, suitable for a general audience. They were split into three groups: a control group with no AI help, a group that could consult OpenAI’s GPT-4 for one story idea, and another that could employ up to five AI-generated suggestions. Findings indicated that those with lower initial creativity scores saw noticeable improvements in the quality, enjoyability, and novelty of their stories with AI assistance. Conversely, participants with higher creativity scores experienced minimal improvement or even slight declines in their story ratings when using AI. The researchers used OpenAI’s embeddings API to measure story similarities within each group. Stories generated with AI help displayed more similarity to one another, leading to a 9% to 10% drop in collective novelty. This homogenization suggests that AI tools, while beneficial on an individual level, might lead to less diverse creative output in group settings. Implications for Creative Industries These observations are particularly relevant for fields relying heavily on creative output, such as publishing and media. Speaking to TechCrunch, Hauser underscores the need for carefully assessing AI technologies rather than adopting them broadly based on assumed advantages. The study implies that while AI can support individual creativity, its widespread application might reduce the uniqueness of creative works. The research also points to AI’s potential as an educational aid, assisting individuals in enhancing their creative skills. Participants with initially low creativity scores produced stories that were rated higher in novelty, coherence, and engagement. Researchers suggest that AI tools could be tailored to offer more varied and less predictable suggestions to counteract the reduction in collective creativity. Acknowledging the study’s limitations, Hauser and Doshi call for future investigations into the use of generative AI in practical, real-world situations. They aim for their findings to inform the development and deployment of AI technologies, ensuring a sustained diversity of creative ideas across disciplines like writing, art, and music.