IT for Change contributed to the U.S. Copyright Office's call for comments in relation to AI and copyrights. Our input spoke to the significant issues and challenges around consumption and utilization of knowledge – both public and copyrighted – in relation to the developments in the field of AI, and generative AI in particular. Our submission highlighted, above all else, that AI-produced works must be classified as uncopyrightable. The summary of our recommendations is below:
a. The permissibility of the use of copyrighted material for training AI models must be evaluated against a new ethical standard of ‘fair learning’. In the case of collective knowledge, particularly traditional and indigenous knowledge, the risk of cultural appropriation and enclosure must also be a significant factor in the assessment of ‘fair learning’.
b. Regulation must provide for the following minimum obligations:
- Mandate labeling of AI-generated work as distinct from human creations, disclosure of training data, specification of sources used for the same, and a clear, legible outline of the methodology of training.
- Build standard contract requirements that provide for standardized definitions (for example, ‘prompt’ in relation to the original idea entered into a Gen AI system to derive a response) and multidisciplinary collaboration. Standardization could be a stepping stone towards addressing bargaining power inequities by offering organizations a series of viable options for contractual rights, duties and risks.
c. The promotion and maintenance of open datasets should not create a ‘free for all’ mandate for data sharing. Protections outside of copyrights could be granted to ensure that open datasets do not allow for-profit entities to enclose public knowledge for private gain, thereby depleting the intellectual and knowledge commons.