A very provocative question was raised in a recent publication in The Verge, dedicated to the creation of music by artificial intelligence. Is it legal to use copyrighted material to train artificial intelligence? And how to evaluate AI-generated music, if in essence it is plagiarism without malicious intent. Jonathan Bailey, CTO at iZotope, answers this question simply, "This is a huge legal challenge."
Modern AI do not know how to create in the literal sense, they create something new, be it a picture or a melody, based on some received data. And these are not public lessons from a music school, but specific works that have authors. In its current form, the legislation of most countries restricts only direct plagiarism, when specific parts of the old are used in a new work, which has yet to be proven.
AI does not copy someone else's music in the literal sense, but does it in essence - it memorizes the whole range of features of the performance of a particular piece or singer, in order to then selectively use this knowledge. In this case, it can be considered as a kind of industrial espionage - if you studied someone else's technological process, but borrowed only information about some elements from there, this is still the theft of knowledge and, in fact, a crime. And the person who gave the AI to listen to music for training is an accomplice, if not the organizer of such a crime.
In fact, there are much more problems. If a person is involved in the process, then his actions can be assessed, he can be blamed and judged. However, we are already in the second stage of the evolution of creative AI.
- The musician uses AI functions to aid his creativity;
- AI itself writes music, and a person performs it;
- AI creates music, AI performs, humans do not participate in the process.
As an example of the transition to the second stage, we can consider the work of the singer Taryn Southern or the Iranian composer Ash Kush. And the third stage is not far off, when the legal norms developed for the human world will become completely useless. This is why Jonathan Bailey rightly speaks of the "daunting problem."