Humans versus AI: whether and why we prefer human-created compared to AI-created artwork.
Level 2 - randomized trial
Individual randomized experimental trial manipulating artwork labeling across participant evaluations.
PubMed 37401999 · doi:10.1186/s41235-023-00499-6
What was done
Researchers presented AI-generated paintings to participants and manipulated the purported creator by randomly assigning either a "Human-created" or "AI-created" label. In Study 1, participants rated the paintings across four dimensions: Liking, Beauty, Profundity, and Worth. Study 2 sought to replicate Study 1 while incorporating additional appraisal criteria (Emotion, Story, Meaningful, Effort, and Time to create) and testing the moderating role of personal attitudes toward AI.
What was found
Study 1 showed higher positive ratings across all four criteria (Liking, Beauty, Profundity, and Worth) for paintings labeled as human-created compared to those labeled as AI-created. Study 2 replicated these main findings. In Study 2, perceived narrativity (Story) and perceived effort (Effort) significantly moderated label effects for sensory-level judgments (Liking and Beauty), whereas positive baseline attitudes toward AI moderated label effects for communicative judgments (Profundity and Worth). The abstract reports no numerical values, exact effect sizes, or test statistics.
Why it matters
These findings suggest that people exhibit a negative appraisal bias against AI-generated art, demonstrating that perceived human effort, narrative intent, and engagement fundamentally shape artistic appreciation.
Limits
The abstract does not state the participant sample sizes, demographic characteristics, or specific statistical magnitudes. The stimuli were restricted to AI-generated paintings, so findings may not generalize across other artistic media (e.g., music, literature) or fully account for naturalistic viewing contexts.