It is difficult to distinguish AI-created poetry from poetry written by human poets. This is shown by a new study from the University of Pittsburgh, published in the journal Scientific Reports.
1,634 people were given ten poems - five written by artificial intelligence and five written by poets such as Lord Byron, Emily Dickinson, and TS Eliot. The participants largely guessed that the AI poems were written by a human, and that the "real" poems were not.
In another survey, 696 participants assessed the poems based on criteria such as quality, feeling, rhythm, and originality. The participants who did not receive information about the poems' origin generally gave the AI-written lyrics higher scores.
According to the researchers, the study shows that the participants perceived the AI-generated poems as more direct and accessible, and underestimated how human-like generative artificial intelligence can appear.
