Private Law Theory

ABSTRACT
AI has sparked controversy. The backlash against AI runs the gamut, including the charges that AI-generated works are job-displacing, soulless, and based on the unethical and illegal use of billions of authors’ works without their permission. Some prominent AI researchers even fear AI will lead to the annihilation of humans. The charge of illegality stems from copyright infringement claims. More than twenty pending lawsuits in the United States will decide whether AI generators infringe the copyrights of artists based on their ‘ingestion’ of copyrighted materials to train the AI models, and their creation of allegedly substantially similar works.

Given the backlash against AI and general fears of AI among the public, we examined whether these negative views might foster a bias against AI generated works simply because they involve AI. Would the mere fact a work is described as AI-generated change a person’s view not only of its quality, but also whether it was infringing or fair use?

Through a behavioral experiment involving a simulated lawsuit involving allegedly infringing visual works, we examined this question. The results matched our hypotheses. First, consistent with past studies, subjects rated the artistic merit of a visual work lower if it was described as AI-generated, and higher if it was not, even though the underlying work was the exactly same. We call this lower rating of AI-generated works an AI merit penalty: people tend to view works described as AI-generated more negatively. Second, subjects were significantly more likely to find a defendant’s work infringed a plaintiff’s copyright if the defendant’s work was described as AI-generated. Likewise, subjects were significantly less likely to find such AI-generated works qualified as a fair use. We call this cognitive effect, identified here for the first time in a study, an AI litigation penalty: people are more likely to view a defendant’s visual work as infringing, and not fair use, if it is described as AI-generated. We offer several recommendations on how courts can mitigate the AI litigation penalty in copyright cases.

Lee, Edward and Moshirnia, Andrew, The AI Penalty: Is There a Bias Against AI-Generated Works? (February 1, 2024).