Google wants a copyright lawsuit filed by a group of independent artists over its Lyria music AI model to be thrown out of court. Not based on any tedious ‘fair use’ arguments, but because the YouTube terms and conditions clearly state that Google can do pretty much whatever it wants with the work of any musicians stupid enough to have uploaded their music to the video site.
In a new court filing, Google says that the artists involved in this lawsuit “each granted YouTube and Google - which provides the service - a broad licence” to use all the content they uploaded to their YouTube channels in an assortment of ways.
It doesn’t quite say “anyone who does this is a schmuck and we can trample on them as much as we like”, but it’s fairly clear what the lawyers were probably thinking when they wrote the submission.
That licence, continues the filing, “authorised the conduct alleged in the complaint”. The complaint being that Google used music that was uploaded by artists so that their fans could stream their content on YouTube to train Google’s Lyria AI model, which will then compete with the same artists by generating lots of new music.
Later in their court submission, Google’s lawyers clarify which term they are referring to. It’s the one where creators grant to YouTube - and its “affiliates” - a “worldwide, non-exclusive, royalty-free, sub-licensable and transferable licence” to use uploaded content, including to “reproduce, distribute and prepare derivative works”. And ‘affiliates’ means any other division of Google owner Alphabet Inc, they add.
Because the artists agreed to this term, Google - whose motto, remember, used to be “don't be evil” - have a licence to use their music to train Lyria and therefore didn’t need any other permission or to pay the artists any money for the AI training. And, as a result, the artists’ copyright infringement lawsuit should be dismissed.
It’s an interesting legal strategy for a company that previously said it would “embrace AI responsibly together with our music partners” and “partner with the music industry to empower creativity in a way that enhances our joint pursuit of responsible innovation”.
Which, it turns out, was code for “we’ll trick you into granting us wide-ranging and valuable rights in terms of service we know grassroots artists will never read”.
Of course, when Google and YouTube talk about “our music partners”, they really mean the record labels and music distributors who have bespoke licensing deals covering the use of music across the YouTube platform, not independent artists who are directly uploading videos onto the site.
Presumably the bespoke licences negotiated by labels and distributors do not grant YouTube and the rest of Google such wide-ranging rights to use their music to train AI models without asking.
Many independent artists will have got their music into YouTube’s music app and Content ID rights management system through a distributor with a bespoke licensing deal, but will have also directly uploaded music videos to the YouTube platform. Which is where the generic terms kick in.
If Google is right that it can use any music directly uploaded to a YouTube channel to train Lyria under the video platform’s existing terms of service, it doesn’t need to negotiate anything bespoke or make any payments to independent artists in order to develop its music-generating AI.
And it doesn’t need to employ the more common defence used by tech companies sued for making use of existing content for AI training without permission, which is that under US law AI training is ‘fair use’, which means no creator or rightsholder permission is required in the first place.
Though, even if Google is right here legally speaking, it’s a really bad look in reputation terms for YouTube, a platform that has gone to such great lengths over the years to schmooze and placate creators, including musicians.
The YouTube employees tasked with managing relationships with artists would presumably prefer it if their own lawyers didn’t go to court and tell those same music creators that they are idiots who willingly gave away all their rights by clicking a submit button without properly scrutinising the small print.
Some bits of YouTube have also tried a bit harder to live up to that commitment to “embrace AI responsibly together with our music partners”, treading more cautiously than many AI start-ups when it comes developing AI tools without involving the music industry.
Although Google DeepMind - the division that developed Lyria - has generally been less cautious and is certainly very vague about how it’s training its generative AI models and what content is being used.
Google’s lawyers also use that vagueness against the musicians in their response to this litigation, noting that the artists’ lawsuit is “based on the unsupported hypothesis that Google trained on their specific works”, and that “the only basis for this allegation is that their works were available on YouTube during the period Google developed Lyria”.
Of course, the reason why the musicians can only hypothesise that their music has been used to train Lyria is that Google DeepMind, like most AI companies, offers more or less zero transparency when it comes to what content it has used to build its training datasets.
Because, it turns out, “embracing AI responsibly together with our music partners” means not telling artists if their music has been used in AI training and then using that very secrecy as another reason to shut down any copyright claims.
We await to see how this all plays out in court - and whether YouTube’s reputation takes any sort of hit in the court of public opinion.
