Since artificial intelligences began to gain popularity thanks to the emergence of ChatGPT, one of the main issues has been the authorship of the created works. Whether it’s texts, images, videos, or audio, all AIs are trained using millions of files that often have copyright protections.
Obviously, many authors have systematically denounced these practices because many companies, like OpenAI, are quite opaque regarding the source of their training data.
This week, the latest major case of alleged copyright violation is being played out by the record label Universal Music Group (UMG), which has sued Anthropic, a company that uses artificial intelligence to generate song lyrics. According to UMG, Anthropic would have violated the copyrights of several artists from their catalog, such as Katy Perry, Taylor Swift, or Drake, by creating lyrics similar to theirs and offering them on their online platform.
UMG claims that Anthropic has used the GPT-3 language model to train its algorithm with the lyrics of its artists and then generated new lyrics based on them. The company argues that this constitutes an act of unauthorized copying and unfair competition since Anthropic charges a monthly subscription to its users to access its lyrics.
Anthropic, on the other hand, defends itself by stating that its lyrics are original and creative and do not infringe any copyright. According to the company, its algorithm does not copy lyrics from other artists but uses them as “inspiration” to create new word combinations. Anthropic also argues that its lyrics are transformative and parodic works protected by the right to freedom of speech.
This case raises several legal and ethical questions about the use of artificial intelligence in the musical field. Who is the author of the lyrics generated by Anthropic? What criteria should be used to determine if there is a copyright infringement in a work created by AI?