UMG has had a long history of embracing new technologies, and you can read what our leaders have had to say about the topic in a variety of articles, such as Sir Lucian Grainge’s interview with The New Yorker or Michael Nash’s keynote interview at Music Ally Connect.

If you’re an AI startup looking to work with UMG, you might wonder how we evaluate and prioritize potential partners or licensees. This article aims to answer those questions and provide some guidance.

High-level, we look for 4 things:

• • AIs that have been built and trained in an ethical and intellectual property (IP) safe manner;

• • Platforms built with artist copyright owner and other rightsholder consent in mind;

• • Platforms that avoid user and artist harms;

• • Uses of AI that empower people and address a concrete artist or label need.

 

Responsible Development: AIs & Training Materials

UMG’s business is built on copyright, which exists to reward and encourage creativity and innovation by protecting the creators’ works and giving them control over how those works are used.

To understand more about copyright, head over to the Learning section to Episode 3 of the 5 Things You Should Know series, focused on Rights & Licensing.

As a matter of principle, this means that we seek to protect the works created by our artists, but that we also respect the rights of all other artists or creators.

When it comes to AI, this principle means that we look carefully at how AI models were built, what materials were used to train them, and whether those materials were properly licensed. This applies to Generative AI of all kinds, whether music, text, images, video, or even code.  Because we do not want AIs to train on our content without authorization, we cannot use AIs trained on other people’s content without authorization.  Therefore, when a company comes in with an AI product or service, we want to know what AIs are being used and how they were trained, from the foundation model up through fine-tuning.

We will also want to review the terms and conditions to make sure that no unintended rights are being granted in the output created.

Many of the popular AI systems including Stable Diffusion, MidJourney, Llama 2, and ChatGPT are known to have trained on unlicensed copyrighted content.  As a result, UMG avoids these tools, as well as 3^rd party tools built on top of them.

While the UMG AI team does extensive vetting in-house to determine if AI companies meet the goals above, a straightforward alternative to direct evaluation by UMG is the FairlyTrained.org certification – which you can read about here.

 

Value Artists and their Art: Consent, Credit, Compensation, Control

Everything we do at UMG has the interest of artists at its core. When it comes to AI, this translates to 4 principles that cover fairly valuing artists and their art. This applies across the use of copyrighted works, as well as well as artists’ names, images and likenesses (including voices).  It also applies whether the use is for training, fine-tuning, or as some other form of input.

These are:

• • Consent: Consent must be obtained to include our and our artist’s content in any AI model or tool. Consent means that we and other relevant rightsholders have opted-in, rather than being told we must opt-out.

• • Credit: artists/rightsholders must be given credit and their contribution labelled appropriately on outputs that feature them or their work.

• • Compensation: we and our artists must be fairly rewarded for the value brought to a service or feature.

• • Control: if rightsholders authorize use of their content, they will have requirements about how it can be used. For example, if an artist does not like an output generated with their likeness (e.g. because it is derogatory or harmful), they should have the ability to take it down.

We expect our AI partners to respect the principles above.

 

Respect for Users: Harm Avoidance

Many start-ups are working on AI-powered tools intended to use artist likenesses for fun and compelling purposes.  Done responsibly, such products and services can be great for an artist as well as enjoyable and memorable for fans.  To avoid consumer confusion and other harms, these scenarios will require that AI-generated output is clearly labeled.

It should be no surprise that UMG will avoid AI products that have the risk of creating consumer harm or deception, or harm to people whose name or likeness is included in AI-generated output. This includes but is not limited to: misrepresenting the source or nature of goods or services (“passing-off”); tricking a user into believing they are interacting with a real person when in fact they are interacting with an AI; emotional manipulation or social engineering that can lead to a negative financial impact, or loss of privacy, agency, or dignity – or creation of misinformation or disinformation.

 

Copyright Implications and Appropriate Uses of Generative AI

UMG believes ethically trained AI can be a valuable assistive tool to human creativity – like many technologies that preceded it.  Rather than replacing human creators, we seek tools that empower them to do their best work more efficiently, allowing them to make works of art we can distribute to fans on their behalf.  Our artists’ creations depend on copyright protection in order to generate revenue.  Copyright law from around the world, including the US and EU, only grants copyright protection to human created works and we will not use generative AI to create the primary assets we distribute, including a song’s underlying composition and lyrics, music, music videos, cover art, and merch designs.

Generative tools can still be helpful for ideation, story boarding, planning, internal pitches, minor asset clean-up, certain types of marketing assets, and for many other purposes.

Conclusion

UMG believes the responsible use of AI will be beneficial to both artists and fans alike.  The pointers above intended to help start-ups understand some of the issues to consider when building or partnering with generative AI products and services, but UMG’s support for human creators is much broader.  UMG is actively working with AI companies, governments, and several industry bodies and organizations in support of human creativity and responsible AI use.  One such organization is the Human Artistry Campaign, and you can read more about their broader set of principles here. UMG has also recently announced a strategic relationship with The Roland Corporation, to empower human artistry. The collaboration kicked off with the publication of “Principles for Music Creation with AI”, for the responsible use of AI within the music creative ecosystem. You can read more here.

 

***

If you’re looking for more general information on what we look for in a startup – or advice on how to put your best foot forward if you’re looking to engage with UMG, head over to the Learning section to check out the 5 Things You Should Know series.

If you have any other questions, head over to the Talk to Us section and drop the team a line.