Generative AI poses a risk to European culture

We risk losing intangible European cultural heritage if the generative AI of the future is not trained on European data. Generative AI produces texts and images that are statistically probable based on the data the model was trained on. This means that minority viewpoints are less likely to be reflected in AI-generated content, which could lead to a tyranny of the majority. This is a problem for democracy in general, because of the potential loss of diverse voices. It is a problem for Europe in particular because European content is underrepresented in popular generative AI models like ChatGPT. Policy recommendations for Europe Support the development of national and European AI models. Work to increase European content in international and commercial AI models while supporting creators’ rights to their content: Develop legal frameworks for collective rights so copyrighted material can be included in training data while creators are compensated and their intellectual property rights respected. Make European data that is not copyrighted available as open access datasets that can be used to train AI models Prioritise making data from small and minority languages accessible in collaboration with language communities. Increase the general public’s understanding of how generative AI works so we are better equipped to use it productively without succumbing to its biases and flaws.