Study Finds People Prefer AI-Written Stories, Especially When Attributed to Humans

New research indicates that people tend to prefer stories written by artificial intelligence, particularly when they are led to believe a human authored them. This preference may stem from the way AI models are trained to optimize their output for widespread appeal.

According to the study, large language models used in generating AI content are trained to produce responses that resonate with a broad audience. This is achieved through a process where models generate text snippets, and human testers provide feedback, indicating whether they like or dislike the content. This iterative feedback loop effectively trains the AI to align with user preferences.

"We basically get the model to generate little bits of text, and then we get people to read them, and they give a thumbs up if they like what the model's saying and they give a thumbs down if they don't like what the model's saying," explained Jones, a researcher involved in the study. He added that it is unsurprising that people prefer results when the models are optimized against their preferences.

While this optimization can lead to oversimplified output, it results in writing with higher overall appeal. In contrast, human writers may produce exceptionally unique content that caters to more specific tastes. "Real humans are weird and idiosyncratic, and they're very different from one another," Jones stated. "But models can kind of learn to be this kind of milquetoast everyman who appeals to everybody."

The findings align with the increasing exposure to artificial intelligence and the prevalence of algorithmic reward systems that favor broadly approachable content on platforms like TikTok. However, researcher Weisberg suggests that the current digital landscape is not the sole cause. "Technologies amplify existing tendencies, rather than creating them," she said.