Are the original creators of musical works used to train AI models entitled to a stake in the resulting derivative songs? Germany’s GEMA believes so, and it’s announced a new licensing model in pursuit of the objective. Photo Credit: Luca Bravo
Once-off payments are inadequate for authors whose works have been used to train generative AI models – at least according to Germany’s GEMA, which says it’s created the “first licensing model” tackling royalties racked up by derivative songs.
The Berlin-based collecting society and PRO reached out with an overview of the royalties framework, which it initially unveiled at the Reeperbahn Festival earlier in September. At the top level, it’s worth noting that this push for bolstered author protections has arrived amid the implementation of the EU’s sweeping AI Act.
Among many other things, the latter is expected to compel generative-model developers to disclose precise details about the media used to train their systems. That will presumably set the stage for the relevant recording and compositional rightsholders to seek payments for their IP’s (unauthorized) use.
But what about when AI factors prominently into works performed in public establishments? Just scratching the surface here, far-reaching questions remain with regard to measuring the percentage of each creation that’s attributable to AI.
That’s a departure from the comparatively straightforward existing process of identifying public plays (preferably with exact measurements as opposed to extrapolations) and then compensating the appropriate authors accordingly.
Perhaps more pressingly on the AI side, what about the public performance of derivative works that only exist thanks to generative models that were trained (with or without permission) on protected music?
Of course, there aren’t any direct answers at present – with even larger unknowns when it comes to developing a system to register the usages, particularly in light of the ongoing legal battles over where the AI-training copyright line will be drawn.
Nevertheless, GEMA says it’s “the first collecting society worldwide to develop a licensing approach aiming to balance technological progress and the protection of creative work.”
“Pure remuneration through a buyout, i.e. a one-off lump sum payment for training data,” the organization proceeded, “is not nearly sufficient to compensate authors in view of the revenues that can be generated. The model provides for fair remuneration at a high level while keeping in mind that the market and its technical developments can change dramatically and rapidly.”
Rather, “authors must be adequately involved in the subsequent generation of AI content based on their creative work,” the entity emphasized.
It’d be an understatement to say that fleshing out, implementing, and ensuring compliance with this system will prove a tall task. DMN requested additional details from GEMA, which offered an overview in its formal release, but didn’t receive a response in time for publishing.
In any event, the push raises interesting questions about yet another component of the AI explosion. Pre-cleared music for use in public establishments (and specifically those that are unconcerned with playing today’s top hits) is more widely available than ever, and many of the involved companies are harnessing AI.
But multiple AI developers say their models didn’t train on protected media at all, and in the bigger picture, AI tracks will undoubtedly make a mainstream commercial splash at some point.
