Copyright protection in the age of generative AI (COPY.AI)

COPY.AI investigates how creators can be recognised and compensated when their work is used to train generative AI, while supporting the responsible development of new models.

Creative space. Woman sat at white desk working on a tablet in front of her laptop. Surrounded by sheets of calligraphy and magazines. — Photo: Antoni Shkraba / Pexels.

A growing challenge for creators and AI developers

Generative AI systems such as ChatGPT are trained on vast amounts of online content, including copyrighted books, articles, photographs and illustrations. However, creators often receive little information about whether or how their work has been used, and are rarely credited or compensated.

Consider an illustrator who shares their portfolio online. The images may be included in a dataset used to train a generative AI model without the illustrator’s knowledge. Once the model has been trained, it can be hard to establish whether those particular illustrations were used, how they influenced its outputs and if and how the illustrator should be credited or compensated.

As a result, many creators are restricting access to their work or choosing to opt out of AI training. This may help them regain control, but it can also lead to unintended consequences. If their work becomes less accessible, creators find it harder to reach audiences, be discovered by potential clients and earn money from their work. AI developers, in turn, may struggle to obtain reliable, diverse and representative training data. This could affect the quality of the models and their outputs.

What will COPY.AI investigate?

How can copyright and creators’ rights be protected while maintaining access to the data needed to develop generative AI responsibly?

COPY.AI brings together researchers in natural language processing, computer vision, machine learning, intellectual property law, technology and economics. With a particular focus on Nordic and Baltic cultural heritage, the project will explore approaches that balance creators’ rights with responsible AI development.

The researchers will investigate:

How copyright law should apply when protected material is used to train AI models without the creators’ permission.
How we can determine whether specific copyrighted works were used to train a model, and whether outputs can be traced so that creators’ contributions can be identified
When AI-generated content may infringe copyright.
How generative AI systems can be designed to reduce the risk of infringement.
How creators might be compensated, and which approaches are legally, economically and techincally feasible.

The way we address these challenges will influence the digital culture we all share. When human-created content is valued and used on fair terms, it can provide AI systems with a stronger foundation for continued development and reduce the risk of the internet becoming dominated by repetitive, low-quality content.