Academic output as input for AI

It is not disputed that artificial intelligence, and in particular large-language models, are trained, among others, on copyright-protected works. In the majority of cases, however, the authors of these works do not receive remuneration for the use of their works in Generative Artificial Intelligence (GenAI) training, even though the output these AI models are meant to produce are likely to compete with human creativity and may eventually even kill the demand for it. Intellectual property law in general and copyright law in particular serves as a tool to achieve a balance between prevalent societal interests. I argue that copyright law can and should play a role in ensuring that, firstly, GenAI models are trained with human copyright-protected works, and thereby provide the basis for GenAI models to generate high quality output that reflects diverse values and beliefs, and, secondly, that human creativity is stimulated by ensuring that the use of copyright-protected works for GenAI training is remunerated.

The author argues that universities’ role is to foster the wide dissemination of research by promoting the use of academic output as input for GenAI as part of their policy, albeit with respect to certain safeguards. First, universities and research institutions collectively should require publishers to pay researchers an appropriate and proportionate remuneration for the use of academic publications as input for GenAI training. And second, universities must emphasize the need for attribution of the works used in GenAI training, by supporting the use of GenAI models with strong citation capabilities.