A recent study shows that readers find it difficult to distinguish between poems created by human poets and those generated by artificial intelligence. Surprisingly, many participants rated AI-generated poems even higher than works by renowned poets. This research was conducted by Brian Porter, a postdoctoral researcher at the University of Pittsburgh, and his team, with results published this week in the journal Nature Scientific Reports.
Image Source Note: Image generated by AI, licensed through service provider Midjourney
The research team selected ten famous poets from English literature, including Geoffrey Chaucer, William Shakespeare, and Walt Whitman, encompassing nearly 700 years of literary works. To explore the differences between AI and human poetry, the researchers had OpenAI's ChatGPT 3.5 model generate five poems for each poet. Notably, these generated poems were not filtered by humans; the researchers directly selected the first five produced by the model.
The study was conducted in two parts. In the first part, 1,634 participants were randomly assigned to one of the poets and read ten poems in random order—five generated by AI and five created by humans. Participants had to determine whether the author of each poem was AI or human. The results showed that participants were more inclined to believe that the AI-generated poems were created by humans, while they considered the poems they thought were likely human creations to actually be human works.
The second part of the study involved nearly 700 participants who rated 14 characteristics of the poems, including quality, beauty, emotion, rhythm, and originality. Participants were randomly divided into three groups: one group was told the poems were from humans, another was informed they were AI-generated, and the last group received no hints. The results indicated that participants without any hints generally gave higher scores to AI-generated poems, while those who knew the poems were AI creations tended to give them lower ratings.
The researchers stated that these findings suggest that participants used some common but unreliable criteria when judging AI and human poetry. AI-generated poems, being relatively simple, may be easier for non-experts to understand, leading them to prefer AI's works while misinterpreting the complexity of human poetry as meaningless. The results of this study indicate that AI's capabilities in poetry creation have reached a level that is nearly indistinguishable from human works.
Key Points:
🌟 The study found that readers struggle to distinguish between poems by human poets and those created by AI, often favoring the latter.
📊 Participants often misidentified AI-generated poems as human creations when judging them.
📉 Ratings for AI poetry were generally higher than for human poetry, especially when the authorship was unknown.
