If the stories written by AI are more in line with readers’ tastes

Texts generated by artificial intelligence are now virtually indistinguishable from those written by a human being. So far, nothing new. But there’s more: people appreciate AI-generated work more, and this is particularly true when they mistakenly believe that the texts were written by a real-life author. This is the finding of a study by Villanova University, authored by Sydney Sears and Deena Skolnick Weisberg, which was picked up almost immediately by Time and The Guardian.

A few figures to give us a better idea. Researchers asked 1,682 participants to read six short stories and rate both their quality and how engaging they found them. Each participant was told – correctly or otherwise – whether the text had been written by a person or by an AI system.

Stories generated by artificial intelligence are rated 6 per cent higher in quality and 8 per cent more engaging than those written by humans. Furthermore, the researchers note, they receive even higher ratings when participants believe they were written by a human author; in such cases, the rating increases by a further 3 per cent.

The study continues with a second experiment and offers further food for thought. A further group of around 900 people were presented with other texts, some generated by artificial intelligence, others written by humans. When asked who the author was, participants answered correctly in 39 per cent of cases in one test and 52 per cent in the other. “Previously reported familiarity with AI systems seemed to help in recognising the typical patterns of AI-generated writing,” says Weisberg. “This suggests that improving AI literacy could help people navigate the new world in which we live, one that is increasingly shaped by AI. Those who claim to have a literary background, on the other hand, proved to be less accurate in their assessments.”

The study also reveals another finding: AI-generated texts often better suit readers’ tastes and are easier to understand. “The models that power AI,” observes Cameron Jones, a psychologist at Stony Brook University who was not involved in the research, “learn to become a sort of average, inoffensive and accommodating individual, capable of appealing to just about everyone.”

Writing in The Guardian, Luke Kennard, a lecturer in Creative Writing at the University of Birmingham, goes a step further: “As humans, we tend to personify AI and imagine it as a sort of glowing robot with a benevolent, neutral face, when in reality it is merely predictive code based on a vast, stolen database of real-world texts, physically housed in vast, expensive and destructive data centres with catastrophic repercussions for the surrounding communities.” Kennard adds that artificial intelligence is not a ‘tool’ comparable to a spell-checker or a thesaurus, but rather ‘an existential threat’. Weisberg does not share this view, emphasising that artificial intelligence systems are already capable of generating short stories considered at least as good, if not better, than those written by humans, to the extent that we need to update our understanding of what AI is actually capable of.

In this exchange, however, it seems to us that the issue is not linked solely to the theme of ‘artificial writing’ – which certainly carries its own weight – but also to our notion of authenticity: the idea that a story is worth more because behind it lies a biography, hard work, who knows what insight or what ‘pen’. When put to the test using a scientific method, rather than relying solely on indignation, that idea reveals itself to be what it probably always has been: an aesthetic prejudice – reassuring and almost impossible to recognise when it is one’s own.