By Ana Andrijevic*
In May 2024, the Organisation for Economic Co-operation and Development (OECD) updated its Principles on Artificial Intelligence (AI),[1] including the principle of transparency,[2] which has contributed to shaping policy[3] and regulatory debates on AI and generative AI (i.e. deep learning models that can create new content, such as text, computer code, and images, in response to a user’s short, written description - a “prompt”).[4] From a copyright perspective, the principle of transparency has become increasingly relevant in several respects: the transparency of the databases (or datasets) and their content used to train AI models, the transparency in AI models, and the transparency regarding the use of AI tools in the creative process.
This contribution focuses on the transparency of the databases and their content used to train AI models, which are increasingly being kept under lock and key and made inaccessible for perusal by AI companies to maintain their competitive edge. Since these databases contain a wide range of protected literary and artistic works (e.g. literary works,[5] photographic works,[6] paintings and drawings,[7] musical works,[8] and more), the interests of AI companies collide with those of authors and rights holders, whose works are used without authorization or compensation. In this contribution, we explore this discrepancy of interests from the principle of transparency angle.
Lack of transparency of databases and their content used to train AI models