Transparency is a big topic in AI music, and not just in terms of rightsholders demanding AI-music companies reveal what they trained their models on.
There has been a lot of talk about transparency on the listener side: making sure that music-streaming users know when they’re hearing an AI-generated track. Now a group of music-industry bodies have proposed a system for exactly that.
The Wall Street Journal broke the news this afternoon, with the initiative led by US labels body the RIAA and global labels body the IFPI. A2IM, The Grammys, SAG-Aftra and the Human Artistry Campaign are also backing the system, according to the report.
Per the WSJ, the system will involve two tags displayed alongside tracks on streaming services: one for wholly AI-generated tracks – or where “the lead vocals or key instrumental performances” are AI – and a second for “AI-assisted” tracks where AI has been used only for certain elements of the otherwise-human creative process.
The report includes the graphics for the two labels: an upper-case ‘AI’ in white text inside a black box for the first category, and a lower-case ‘ai’ in black text inside a white box for the second.
This will have to be explained very clearly to music listeners so they understand which is which. We imagine artists might worry that if they disclose AI assistance on tracks, some listeners may see the label – the second one – and mistake it for meaning the track was fully AI-generated.
Update: the official announcement has now been sent out, which includes more details and quotes from the industry bodies.
“Fans want to know whether and how generative AI has been used in the music to which they listen. Given how important human artistry and authenticity is to music lovers all over the world, these labels will provide an immediately understandable and easily scalable approach to transparency,” said IFPI boss Vikki Oakley and RIAA chief Mitch Glazier in a joint statement.
The announcement includes two more industry bodies who are backing the new labels: global independent labels body WIN and European indies organisation Impala.
It also includes the full definitions for the two categories of labelled track.
Those getting the upper-case AI treatment are tracks where “generative AI was used to generate the entirety or the primary portion of the creative elements of the recording: this would include, for example: lead vocal performance generated by AI; key instrumental performance generated by AI; entirely prompt-generated AI music”.
Meanwhile, those getting the lower-case ‘ai’ label are tracks where “the recording was created substantially by humans and expresses human creativity; however, generative AI was used for some expressive elements. Humans performed the lead vocal and primary instruments“.
The obvious question now is what the big streaming services make of this. Several have already launched (or at least started work on) their own systems for labelling AI-generated or assisted music.
Spotify revealed its plans last September, and then rolled out its nuanced transparency-credits system in beta this April.
It revolves around artists disclosing how AI has been used for specific contributions including vocals, lyrics and production. According to Spotify’s updated blog post on the launch, “tens of thousands” of these credits are already being submitted on a daily basis.
Apple Music revealed its own ‘transparency tags’ system in March, meanwhile. Labels and distributors will be required to disclose whether AI has been used in artwork, tracks, compositions and music videos.
Tidal and Deezer, meanwhile, are both using their own technology to identify and tag wholly AI-generated music on their services, with Tidal recently stressing that “we expect – and will begin to enforce – that content distributors identify AI-generated content before it reaches our platform”.
In theory, the new labelling system proposed by the industry bodies can slot into the DSPs’ strategies, if they are willing to adapt their own systems and collectively adopt the proposed badges.
It is unclear whether any have already agreed to do this. If so, that’s not mentioned in the Wall Street Journal’s story – if they had, we’d expect it to have been reported. In the coming hours, we expect to have more clarity on their responses though.
The report this afternoon does include a quote from their lobbying body DiMA’s CEO Graham Davies which is best described as a cautious welcome for the news.
However, it is made very clear that the new system needs to be backed up by (in the WSJ’s paraphrasing) “more detailed and accurate AI metadata” delivered with tracks.
Update: we’ve now been sent the full statement from Davies:
“DIMA has long advocated for the creators, owners, and distributors of music to provide accurate and timely metadata on all music released and distributed to streaming services,” he said.
“We are following today’s announcement closely and look forward to receiving more detailed and accurate AI metadata, which will strengthen our ability to give fans the transparency they deserve.”
“That information flows best when it travels the entire path from creator to fan, and our members rely on industry partners to make that possible,” he added.
“Our members look forward to continuing to work with the labels, producers, artists and distributors, as well as other industry stakeholders and standards bodies such as DDEX, to build a robust supply chain in which consumers can trust.”
(As far as we’re aware, DDEX is not yet involved with the system proposed by the industry organisations.)
