Recently, I conducted an interview with the radio station rbb3 on the question of how music generated by artificial intelligence (AI) is protected. Are there actually copyrights here? Can rights be violated by this? Who can earn money with this? Right now, AI-generated music productions are uploaded to streaming platforms in high numbers. The music business needs to find a strategy for how to handle this and to make sure that fair remuneration is received by authors and musicians.
Copyrights and neighboring rights to music
When we look at music productions, we have to distinguish between two groups of rights:
Copyrights exist to the compositions, texts, and arrangements. Accordingly, the authors are the composers, lyricists, and arrangers. In practice, most rights of use are administered by collecting societies (such as GEMA).
In addition, there are neighboring rights to the music. These arise, for example, in the performances of artists and in the produced sound recordings. Most of these rights are in the hands of the record producers (labels) and are administered by them.
Input and Output in Artificial Intelligence
When it comes to AI, we need to look at two areas:
First, there’s the input. The AI generators scrape the internet for training purposes (machine learning). In doing so, they also use copyrighted works (films, music, texts, images, graphics, etc.). Here the question arises as to what extent this input is legally permissible.
Secondly, there’s the output. This is the result generated by the AI (a new image, new music, a new video film, etc.). Here the question is whether these AI-generated results can actually be copyrighted.
Input violations (training of AI generators)
If AI generators process copyrighted works for training purposes, this is generally permissible as data mining, § 44b German Copyright Act. Nevertheless, the rights holders have the option of applying a reservation of rights to the works in machine-readable form (opt-out). However, the question is to what extent this can then be technically implemented. In the European judicial area, this provision is based on an EU directive of 17 April 2019 (Art. 4). Currently, feeding the AI generators is permitted in the European legal area in the form of data mining. In the U.S., there are currently some court proceedings in which the legal situation is to be clarified.
Beware of the violation of personal rights by AI music
It’s more problematic when the AI imitates the voices of existing singers. For example, you can specify in the prompt that a song should be sung in the voice of Paul McCartney. In this case, the artists’ personal rights are violated and they can take action against this if you use the AI output.
Copyright protection of the output of the AI generators
According to German law, only natural persons can acquire copyrights. If the text, composition, and arrangement are developed entirely by the AI, no copyright exists. The situation is different if the AI user writes texts or text passages themselves and sets further concrete input commands (prompts) for text and music, revises the result again and again, makes artistic selection decisions, and thus influences the final result individually. Then the AI becomes more and more a mere tool and the AI user can acquire the copyright.
For the neighboring rights of the record producer, it is not necessary that a copyrightable work exists or that a human being has been involved here. Thus, the AI recording can give rise to neighboring rights of the producer. These neighboring rights would then lie exclusively with the providersof the AI application. However, the general terms and conditions of the AI services usually distinguish between commercial and non-commercial use. If you pay a fee, you might acquire all commercial rights. If you use the free version, you usually acquire only non-commercial rights.
Can I make money with AI music?
We have seen that AI users can become rights holders, e.g., by writing texts or text passages, making individual selection decisions when creating the music composition, and acquiring the neighboring rights to the music recording (through the general terms and conditions of the AI providers). Then such titles can also be commercially exploited, e.g., via DSP (Digital Service Providers, such as Spotify, Amazon, etc.). However, their terms of use usually stipulate that the use of AI in music must be labeled. AI providers also sometimes require the naming of their company. It remains to be seen whether the revenue models will be adapted to AI music in the future (e.g., whether lower royalties will be paid).
According to recent reports, the DSPs are currently flooded with millions of AI-generated music productions. Some DSPs allow that; others try to remove AI productions. The question arises of who the actual producers of this music are and whether they should be compensated like every other record producer of “real music”. Anyone can create an entire album including artwork in minutes. Neither musical education nor artistic talent is needed. A great part of the streaming revenues go into the pockets of persons who are not part of the traditional music industry, while many creatives – on whose works and performances the AI music is based – receive nothing.
Collecting societies (such as GEMA) will make demands on politicians to implement legal adjustments here, so that the creation of AI music will also be subject to collecting societies. Music labels may also push for AI providers to conclude licensing agreements. After all, the AI applications are trained by the protected copyright works and recordings protected by neighboring rights. Exciting developments are still to be expected here.
I would be glad to talk to you about your legal needs. Contact me
here.
