EU copyright rules complicate AI development, researchers say

European Union copyright rules are less flexible than privacy regulations when it comes to using material to develop artificial intelligence, according to a study co-authored by legal scholar Aleksei Kelli. ERR reported that the researchers, Nele Nisu and Kelli, found that GDPR rules are relatively flexible for using AI in research, but copyright protections are not.

Kelli, a professor of intellectual property law at the University of Tartu, said the two legal frameworks serve different purposes: the GDPR protects people’s privacy, including their health data, while copyright law protects creative work.

That protection raises broader questions about what belongs to an author, Kelli said. He argued that people are shaped by what they have experienced, using the example of a fictional Mowgli raised by wolves who would be unlikely to discuss the philosopher Friedrich Nietzsche.

Kelli also said cultural ideas about creators have changed over time. In antiquity, people attributed creative work to the spirit of genius, while the Enlightenment brought the idea that a person is an author, he said. That view is reflected in the legal protection of copyright.

He compared the use of creative works to train AI with a recent literary controversy. In a newly published biography of Juhan Viiding, titled “Juhani evangeelium,” Mihkel Mutt used the poet’s farewell letters without permission from his heirs, according to the article.

The letters involve both personal rights and copyright, Kelli said. Copyright does not protect words individually, but it can protect an author’s selection of words. By contrast, an idea such as the one behind Mati Karmin’s sculpture “Kissing Students” is not protected, while the sculpture itself is.

Kelli also pointed to Italian Prime Minister Giorgia Meloni’s decision to trademark her voice to prevent its use in deepfakes. The trademark database includes a recording of her saying in Italian, “I am Giorgia Meloni.”

Kelli said the tension between copyright and the GDPR in the AI era was not deliberate, but argued that the EU’s approach had not been optimal. He described the challenge as a trade-off: people do not want to give up privacy, but tighter regulation could allow AI models in China and the United States to pull further ahead. The issue should be considered in terms of competition, he said.

He added that AI development also matters for the Estonian language, since no one else will digitise a small language on Estonia’s behalf.