•  
  •  
 

Abstract

Generative AI can rapidly output vast amounts of expressive content, some of which has great value for society. This new computational process also raises a deep question of fairness: Will the original creators of the content used to train these systems share in the value they create? The question becomes particularly urgent as the potential effects of generative AI on markets for creative works become clearer. Artists with a distinctive style may find it nearly impossible to sell their new work when very low-cost substitutes can be generated automatically. News publishers whose content can now be paraphrased by generative AI systems without violating copyright laws may lose significant advertising revenue from readers who no longer need to click through to publishers’ websites. Millions of workers may be wholly or partially displaced by generative AI trained on their works.

Numerous scholars have begun to address this issue. Some have focused on challenging generative AI providers’ claims that their ingestion of copyrighted works for training models and outputting new works is fair use. Others have conceded or bracketed the fair use question and proposed levies or compulsory licenses to compensate for these uses. We take a distinct approach, proposing a new right for copyright holders with respect to AI training using their work. This protection is appropriate given massive AI systems’ ability to process vast amounts of information far faster and less expensively than humans can. An exclusive right to license AI training, called a “learnright” for short, would enable copyright holders to claim some share in the revenues arising out of automated systems that learn from covered material.

This essay examines the rationale and potential mechanisms for implementing such laws. It explains the high degree of legal uncertainty surrounding the many current lawsuits against generative AI providers, and it proposes learnrights to complement the existing exclusive rights guaranteed to copyright holders. Given the many sources from which AI can “learn,” market mechanisms would likely permit a fair and reasonable degree of revenue sharing pursuant to copyright holders’ assertion of their learnrights. Compensation for learnrights would also redress some striking imbalances apparent in current copyright policy that favor mechanical processing of texts over human engagement with them.

Download

Share

COinS