Public Opinions About Copyright for AI-Generated Art: The Role of Egocentricity, Competition, and Experience

TL;DR

The paper investigates how laypeople perceive copyright in AI-generated art, using an incentivized juried online exhibition to test judgments of creativity, effort, and skills, and to identify authorship and rights-holders. It finds that people attribute authorship and rights mainly to users and data contributors, and observe egocentric biases that are amplified by monetary rewards. The results suggest a move toward distributed ownership and licensing approaches that compensate data contributors, challenging model-centric copyright norms and informing future GenAI regulation and system design. The work has practical implications for policy, platform terms of service, and the design of AI art systems to ensure alignment with public expectations and fair compensation for training data creators.

Abstract

Breakthroughs in generative AI (GenAI) have fueled debates concerning the artistic and legal status of AI-generated creations. We investigate laypeople's perceptions ($N$$=$$432$) of AI-generated art through the lens of copyright law. We study lay judgments of GenAI images concerning several copyright-related factors and capture people's opinions of who should be the authors and rights-holders of AI-generated images. To do so, we held an incentivized AI art competition in which some participants used a GenAI model to create art while others evaluated these images. We find that participants believe creativity and effort, but not skills, are needed to create AI-generated art. Participants were most likely to attribute authorship and copyright to the AI model's users and to the artists whose creations were used for training. We find evidence of egocentric effects: participants favored their own art with respect to quality, creativity, and effort -- particularly when these assessments determined real monetary awards.

Figures (6)

• Figure 1: High-level overview of our experiment.
• Figure 2: Example screenshots of the study's user interface.
• Figure 3: Example images generated by creators for our AI art exhibition. The images in the top row were among the best evaluated, while those in the bottom row were judged poorly.
• Figure 4: A) Perceived creativity, effort, and skills involved in generating images using GenAI for our AI art exhibition. B) Perceived authorship of the user, the AI model, the company that developed the AI model, and data contributors (i.e., those whose creations were used to train the AI model). Gray bars present the mean value across all conditions, while circles and triangles represent mean values in each treatment condition according to the legend. Error bars correspond to 95% confidence intervals.
• Figure 5: Percentage of participants who chose the user, the AI model, the company that developed the AI model, data contributors (i.e., those whose creations were used to train the AI model), and anyone as right-holders. We present results separately for the rights to A) display commercially, B) to display non-commercially, C) to make copies commercially, and D) to make copies non-commercially. Gray bars present the mean percentage across all conditions, while circles and triangles represent mean percentages in each treatment condition according to the legend. Error bars correspond to 95% confidence intervals.