.png){kind=logo}
A groundbreaking study by the University of Pittsburgh reveals surprising results: AI-generated poetry has been rated higher than poems written by humans. The research, which included non-expert poetry readers, shows that poems created by AI, specifically using ChatGPT 3.5, were often preferred over those composed by famous poets like Shakespeare, Whitman, and Sylvia Plath.
The study involved showing participants poems written by 10 iconic English-language poets, along with AI-generated imitations of those poets’ styles. The results were striking. Participants were more likely to attribute AI-generated poems to human authors, and in many cases, they ranked the AI poems higher in terms of overall quality.
In comparison, poems written by human poets were judged as being human-authored only 75% as often as AI-generated works. This finding challenges previous research, which generally found that human poems were rated more highly.
The authors of the study suggest that AI-generated poems are appealing to non-expert readers due to their simplicity and accessibility. While human-written poems are often complex and require in-depth analysis, AI-generated poetry is praised for its clarity and ability to directly communicate moods, images, and themes.
According to the study authors, AI-generated poems lack the “complexity and opacity” of human poetry, which often “rewards in-depth study and analysis.” However, they argue that AI’s straightforward style is better suited to readers who may not have the time or inclination for the deep analysis human poetry demands.
Some of the responses to the study were gathered from an interview with the poet Joelle Taylor after examining its TS Eliot prize-winning work. She vehemently dismissed the poem as being an algorithm by arguing that there must be more to it than that and that its purpose is being able to cause a meaning, empathy, emotion and the like to stand out. Lastly, Taylor noticed that AI’s poetry samples are only from the poems of white, male, upper class poets meaning that poetry in its entire sense is restricted with AI poetry.
The conclusions of the work raise several questions to the author regarding AI in creative professions. Therefore, according to such researchers, with the intent behind AI works gradually becoming imperceptible from human-generated individual AI governance and openness may help the reader understand the actual author of what they are consuming.
Ultimately, the study opens a broader conversation about the nature of poetry itself: Isn’t it simply structure and meter, or is it the true essence of people, passion, and disorder, which can not be translated into an algorithm?
