Machine Bards

AI is giving poetry a high-tech twist.

A new study found that readers can’t reliably tell the difference between poems written by human literary legends such as William Shakespeare and those generated by AI.

Even more surprising? Many readers prefer the chatbot’s work, according to Smithsonian Magazine.

A research team gathered more than 1,600 participants and asked them to identify whether a series of 10 poems – five by humans and five written by AI imitating a human poet’s style – were real or machine-generated. The results showed participants guessed correctly only 46 percent of the time, slightly worse than chance.

“It is quite a weird phenomenon,” Edouard Machery, a philosopher at the University of Pittsburgh and co-author of the study, told the Washington Post.

The AI behind the verses, OpenAI’s ChatGPT-3.5, demonstrated a knack for emulating the styles of great poets such as Shakespeare, Walt Whitman, and T.S. Eliot. Researchers believe this skill stems from the chatbot’s training, which likely included large datasets of the poets’ original works.

But in a second experiment, the team asked 696 participants to rate poems on qualities, such as rhythm, originality, and emotional impact. They divided the volunteers into three groups – one group was told the poems were written by humans, another was told they were reading AI-generated poems, and a third received no information about authorship.

Unsurprisingly, those who believed they were reading human-written poetry gave higher ratings, regardless of the actual author. However, participants who didn’t know the poems’ origins consistently rated the chatbot’s creations higher than the human-authored ones.

The scientists suggested the chatbot’s straightforward style makes its poetry more accessible.

That’s because AI-generated poems lack the complexity of human-authored verse, which means they are better at “unambiguously communicating an image, a mood, an emotion, or a theme to non-expert readers of poetry,” the study explains.

However, this simplicity comes at a cost. Human poets often embrace quirks and complexities that challenge readers to think deeply. T.S. Eliot’s “The Boston Evening Transcript,” a satire on a bygone newspaper, was the most frequently misidentified as AI because readers found it confusing.

While the study has strengthened fears that AI will replace human artists and writers, Dorothea Lasky – the only living poet in the study – welcomed readers’ preferences.

“Poetry will always be necessary,” she told the Post. “If these people in the study read AI poems and liked that poem better than a human-generated poem, then that, to me, is beautiful. I feel there is room for all poets – even robot poets.”