Study finds ChatGPT writes better literature
A new study published in the journal Judgment and Decision Making by the University of Cambridge found that short stories generated by ChatGPT were perceived as more qualitative and enjoyable than those written by humans. 1,682 adults rated the stories, and those created by AI received higher scores. However, participants preferred stories they were told were written by a human. A senior researcher called for making room for AI-generated books, while a film professor criticized the use of 'stolen' databases.
A new study published today (Wednesday) in the journal Judgment and Decision Making by the University of Cambridge found that short stories generated by ChatGPT were perceived as more qualitative and enjoyable than those written by humans. In the study, reported by The Guardian, 1,682 adults participated and were asked to read one of six short stories – three written by humans and three by ChatGPT. Each AI-generated story dealt with a topic similar to one of the human stories. After reading, participants were told how the story was written, but the information was not always correct. They rated the level of interest, engagement, and writing quality. Those who read an AI-generated story described it as more engaging and qualitative, but participants gave higher ratings to stories they were told were written by a human. It was also found that attitude toward AI had an impact: those with a more positive attitude gave higher ratings to AI stories. Dr. Dina Skolnik Weisberg, a senior researcher at Villanova University, said that AI systems already produce stories perceived as at least as good as human stories, and called for updating the view of AI capabilities. She added that there is room to make space for AI-generated books, but there is still room to appreciate human creativity. In contrast, Professor Luke Kennard from the University of Birmingham emphasized that AI is "code based on stolen databases of human writing," with catastrophic implications for communities. In additional experiments, 905 adults were asked to guess who wrote each story; in one, about 40% guessed correctly, and in another, 52%. The researchers wrote that there is no evidence that participants could distinguish between the stories, and that skill with AI tools helped guess correctly. Weisberg noted that AI writing has a style that can be learned to identify, and that the preference for human stories is related to a desire for authenticity, but in practice people preferred AI stories because they were easier to read.