Writers, actors, musicians… More and more artists are taking legal action against companies developing generative artificial intelligence. The latter nevertheless believe that no additional regulations on copyright are necessary, and their arguments do not lack nerve.
Different arguments for a similar purpose
Thus, Meta argues that imposing a licensing regime for copyrighted content in AI training would be chaotic and not very beneficial, because the royalties would be minimal, given the insignificance of each work in a training dataset. Google, for its part, compares training models to reading a book. Therefore, the extraction of ideas and facts from copyrighted works should be considered as such and not a violation of the law.
Microsoft assures that obtaining licenses for each work used would greatly hinder the development of small AI companies, while Anthropic asserts that the current legislation is sufficient. As for Hugging Face, she explains that the use of a work allows the creation of a distinctive and productive model of Al. This is capable of creating a wide variety of different outputs that have nothing to do with the underlying copyrighted expression, analyzes the company, making it fair and non-harmful use.
Some countries are already legislating
Adobe goes in the same direction and specifies that this use is covered by fair use. StabilityAI, for its part, mentions countries where copyright legislation has been reformed to create free zones for training AI, like Singapore, Japan, South Korea, or again from the United Kingdom.
Companies had until October 18 to respond to the federal agency. For the moment, he has not communicated on a possible future decision. In Europe, future artificial intelligence legislation is expected to require companies to publish summaries of the copyrighted data used to train their models.
Source : The Verge
2