If you let your record company license your recording for AI algorithmic music a la Orwell’s “versificator”, it’s like Silicon Valley making you train your replacement.
By Chris Castle
After the money, one of the most important parts of a recording artist negotiation is the “marketing restrictions”. These are restrictions on what the record company or music publisher can do with your work–what type of licenses they can, or more frequently cannot, grant to third parties, for example. Essentially, whatever is not prohibited is permitted.
Marketing restrictions also have a temporal element–during or after the term, recouped or not recouped. There are some restrictions that are acknowledged to be verboten and are usually easy and unrestricted concessions. An example of these would be licensing for certain types of commercials such as tobacco, firearms, grooming or hygiene products and alcohol.
Stewart Dredge has an excellent article this week in the Guardian which brings to mind Laura Kobylecky‘s post on MusicTechPolicy drawing comparisons between Spotify’s “fake artist” problem and “The Next Rembrandt” with echoes of the fictional “versificator” operated by Big Brother’s “Music Department” in 1984. According to Stewart, there are dozens of AI music startups getting funded that all essentially do the same thing. Using a library of recordings (sometimes called a “corpus”), the algorithms “create” new recordings based on the songs and recordings in the corpus. Google is, of course, a leader in the space (not that different from how they used Google Books to train their translation algorithm, a process called “corpus machine translation”–the librarians will be next).
Those recordings can then be sold or licensed at a very low price which, as Laura and others have noted, can be used to drive down the royalties payable to all other artists on digital music services.
This is, of course, not dissimilar to Silicon Valley companies hiring lower paid foreign workers and ordering the employees who they are to replace participate in training their replacements. The difference is, of course, that those recordings have to come from somewhere.
It’s time to start adding to the list of marketing restrictions that the song or recording cannot be licensed for AI purposes of any kind.