AI Plausibility Risks in Cultural Content: Libreriamo Study

A research conducted by Libreriamo, established by Saro Trovato, examined nearly 1,500 interactions with generative AI systems. The findings indicated that only 38% of the responses were accurate and verifiable, while 44% were plausible yet unverifiable, and 19% contained inaccuracies. The analysis encompassed literature, poetry, books, author quotes, philosophical maxims, artworks, and Italian grammar. Within literature, 35% of responses were correct; for books, the accuracy was 25%, with 59% deemed plausible. Author quotes and philosophical maxims had correct rates of 15% and 13%, respectively, with false attributions at 23% and 22%. Grammar accuracy ranged from 85-90%. Trovato cautioned against AI's potential to distort literature and suggested a ten-point guideline for its appropriate use in cultural and educational contexts.