Roland and Universal Music Group partner on “AI for Music”

The world’s biggest music company has now teamed up with one of the world’s leading instrument and music creation tech makers to weigh in on the growing AI boom. Mostly what we get is a set of general principles – but talking about who made music for the data sets is central, and it could be an indication of musician/instrument/music industry partnerships to come.

Humans first?

There’s a lot of vague corporatespeak to wade through in this announcement – I’ve at least included the full principles below and their site. But we can cut to the chase. Basically, Roland and UMG say they’re both interested in AI; it sounds like machine learning-based technologies may well figure into future Roland products for musicians. But they are also making an unequivocal statement that data training sets need transparency, need to involve creators and musicians in consultation, and ultimately need to credit (and pay) artists.

I’m not going to open the can of worms around generative music – that deserves its other article – but suffice to say this is timely. For what it’s worth, while I found references to Suno.ai protecting copyright of material produced with its tool, and charging for commercial use, they seem to have no transparency about their data sets. (Maybe someone else can find more – see the FAQ and v3 announcement, as there’s nothing there.)

Let’s step back for a moment because while it’s still relatively novel in the music instruments business to see this kind of statement, there’s some larger context. Roland and UMG specifically talk about “aligning,” at least in part, with the so-called Human Artistry Campaign. That includes an insanely wide international group of co-signing organizations. You’ll find pretty much any music advocacy and performing rights group you can think of – from ASCAP and BMI to the Society of Finnish Composers, Black Music Action Coalition, Turkish Association of Independent Labels, Bulgarian Association of Music Producers, and the Indian Music Industry.

But in a sign of just how many industries are potentially impacted by AI automation and data set abuse, you’ll also find the NFL and NHL (yeah, the football and hockey ones), Communications Workers of America, stage directors and choreographers, writers, actors, and STOMP. (You can also sign that petition and sign on with AI for Music, if you want.)

The AI For Music statement is not anti-AI – it even references the potential for utility for musicians. But it’s particular about transparency, credit, consent, and compensation. For instance:

We believe that human-created works must be respected and protected, and they uniquely merit copyright protections. The use of copyrighted works – and use of the music artists’ name/image/likeness/voice by others – must be authorized in advance of the use and music artists should be given recognition for their contribution.

See the full statement with supporters (so far launching with UMG and Roland, but others are expected) and a place to sign up:

https://aiformusic.info/

So, what does this mean? Does it mean Roland’s sample libraries will now involve generative AI trained on UMG artists? Or thinking really speculatively, could we see Roland devices that have a “don’t train on me” switch in menu settings to digitally watermark your next SP-404 set or keyboard jam automatically? (Hey, things are getting weird fast, so who knows? I’d love an anti-AI stage piano. Roland didn’t comment on this case specifically – that one’s free for y’all if you want it.)

They do say that they’re working on joint research. But that research is not only about how to build new creative AI tech – it’s also now, with an apparent sense of urgency, about protecting artists. From the press release:

Additionally, the companies have prioritized initiatives, including establishing a joint research and development hub, a collaborative research project focused on developing methods for confirming the origin and ownership of music, and the integration of Roland products and services in select Universal-owned music production facilities around the world.

That’s probably the most important bit – I’d keep an eye on what the research projects around creation and ownership/identification look like. (Maybe that “don’t AI me” switch isn’t so crazy, after all.)

You can read the full press release, which includes the usual press release-style quotes from Masahiro Minowa, Chief Innovation Officer at Roland, Naoshi Fujikura, President & CEO, Universal Music Japan, and Michael Nash, Chief Digital Officer, EVP at Universal Music Group.

Roland Corporation and Universal Music Group Form Strategic Relationship to Empower Human Artistry

It’s nice to see Masahiro Minowa’s name, as he has worked extensively with everything from bleeding-edge drum and piano products at Roland to working at RPG, the group that gave us AIRA, the TR-8/8S, Boutique instruments, and a lot of our other favorite stuff in recent Roland history. And he’s been very musician-focused and customer-oriented, so think the opposite of a lot of the Valley AI types.

I spoke to long-time Roland veteran Paul McCabe, the company’s Global Vice President of R&D and Strategic Partnerships, as apparently he was a leading force behind making this deal a reality. (Interestingly, I see on Paul’s LinkedIn that a number of industry folks have signed on – I think that suggests these values resonate, and that people trust Paul on a personal level as well as Roland.)

I asked Paul directly about what Roland is planning in AI – which makes sense, as this is the company that has pioneered a lot of digital and modeling tech (I mean, a huge chunk of the basis of everything I write about here.) “Roland is currently researching and prototyping many responsible applications of artificial intelligence and machine learning to support human-centric music creation,” Paul tells CDM.

So, what about this generative stuff?

Paul: At Roland, we are interested in creating new solutions to help people create music. We support the innovation that generative AI provides if it is used to enhance the creative process without infringing on the rights of others and is used responsibly. We believe the potential process of creating music and expressing yourself through music will become much more accessible for everyone with the advancement and evolution of generative AI tools.

And what are they actually building, to whatever extent they want to say?

Paul: Roland is researching and prototyping multiple responsible applications of artificial intelligence and machine learning to support human-centric music creation. As part of our research exploration, we receive input from music creators representing diverse perspectives to ensure that anything we develop is well understood and seen as truly complementary.

There’s at least one reason to keep an eye on the relationship between instrument makers and the music recording industry: this interaction is at the roots of copyright law. At a time when the phonograph was almost unusably primitive for music, reproducing pianos – pianolas – were able to mechanically “perform” music without the aid of musicians. That’s a delight, but it left the musician out of the process.

One of my favorite places in Amsterdam is the incredible unsung gem that is the Pianola Museum:

They also have terrific historical resources on those instruments:

The Reproducing Piano – An Overview

So, what have these old-timey pianos got to do with copyright, let alone futuristic Artificial Intelligence? Well, a lot – once you trace the history of copyright law. Here’s a nice timeline for all of us who aren’t intellectual property lawyers (I definitely am a musicologist, not a laywer):

COPYRIGHT THEORY AND HISTORY [University of Oregon]

International copyright law is closely tied to US copyright law, which in turn is based on early 18th century British law. And securing rights for “Authors and Inventors” in the USA is as old as the competition – it’s a power vested in Congress right in Article I, which may come up again soon as AI discussions evolve. We’re still talking essentially written documents, which may ultimately be linked to machine learning data training sets, but don’t evolve any technology. Music sheet music gets its say in 1831, thanks to Congress.

Technology makes its first appearance entirely thanks to the invention of instruments like those pianolas. Ever wondered why you hear about “mechanical” royalties? The term is derived literally from these machines – no joke. That goes back to 1908; film makes an entry in 1912 but incredibly sound recording doesn’t even get a mention in the law until 1971. The US landed on the moon before Congress really dealt with the phonograph.

Here’s the mechanical part, also from that UOregon article:

Compulsory Mechanical License. In 1908, the United States Supreme Court ruled in White-Smith Music Publishing Company v. Apollo Company that pianola music rolls (cylindrical rolls with holes punched in them that served as the “software” for player pianos in the early 20th century) and other reproductions that are part of a mechanical music playback process are not eligible for copyright protection as copies of printed music because they are not intelligible as music notation. In response to this ruling, Congress included in the 1909 act one of the foundations of modern music copyright law, the compulsory mechanical license. 

So, historically speaking, I can’t think of a more fitting marriage symbolically. UMG is the Dutch-American descendant of the American branch of Decca Records; Roland has shaped drum machines, digital sequencers, pedals and effects, and MIDI as we know them. And what we’re really talking about with AI is not so much new machine intelligence as it is new generative, autonomous technology and how the law adapts to technology in new forms. The worries about people having their music ripped off in data sets without credit, consent, or compensation isn’t just metaphorically similar to the worries of performers and composers being replaced with machines. It’s got the same issues around human involvement with machine music at its core.

Now, whether this sort of blurry corporate manifesto will make a difference, or anything substantial will come out of R&D (and a bit more Roland gear in UMG-owned studios), that’s another question. But are we at another juncture of machine and human music making? Totally.

Though to be honest, those pianolas are more fun that these text prompt things. Better sound quality, too. So maybe if the AI takes all our business, we’ll just go back to building robots instead. Moritz Simon Geist should be in great shape.

That’ll be especially important if it takes Congress 100 years to update the law.

Here’s one more great read on the (current/immediate future) state of the law, by Sunny Duan last year:

Copyright in the Music Industry — a new age of scrutiny [Medium]