If anyone’s got their finger on the pulse of the latest developments in music copyright and AI it’s John Phelan, the Director General of ICMP, the global trade body that represents the music publishing industry worldwide.

Here, Phelan discusses the organisation’s latest initiative, RightsAndAI.com, and details the steps required for AI and music markets to “symbiotically thrive in future.”

What is RightsandAI.com and why is it needed?

This site is an open, online resource, freely available for music companies and organisations worldwide to take steps to protect their businesses and their creative partners against illegal scraping of the world’s music by AI companies and big tech companies with AI divisions.

The need is clear. Our industry is witnessing many AI companies trying to misuse copyright exceptions in laws around the world such as Fair Use and Text and Data Mining (TDM) exceptions to essentially build a business off the back of the world’s music without securing prior authorisation or licenses from rightholders. Such practices are legally, commercially, ethically and wholly unacceptable to us and our members.

Let’s be clear: Artificial Intelligence doesn’t make decisions to respect or neglect copyright. The management of companies who deploy AI make those decisions. And companies who try to neglect copyright are liable under law and will be under incoming AI laws. Consequently, ICMP is taking a number of steps around the world to protect the industry’s rights. This is one such step.

There’s a temptation to anthropomorphise AI. We need to avoid that and demystify it. Cold eyed, commercial decisions are being made as to whether to respect copyright or not. That reality requires risk management of course, but it also should reassure everyone in the business that where there are problems, there are practical pathways to effective solutions. I often think of the old adage: ‘AI is whatever doesn’t work yet, because once it works it’s just software.’ There’s a hefty kernel of truth in that.

What is the goal of the platform?

When people read and hear of ‘AI guardrails’, well this is a real and practical one.

RightsAndAI.com provides an open resource which operates on two levels: Legally it enables a music company to make an effective ‘rights reservation’ on behalf of its songwriters, composers and artists. Technically a .CSV file is available to upload the company’s business ID and data in ‘machine readable’ form, thereby ensuring online content crawlers and bots can easily and automatically recognise the company information. These two combined will help put a stop to unauthorized access of digital music around the world and scraping.

The site removes any ambiguity that AI companies’ use of the world’s music without a license is possible, and it provides a convenient central resource for AI companies to comply with the rules of the road.

“The site removes any ambiguity that AI companies’ use of the world’s music without a license is possible, and it provide a convenient central resource for AI companies to comply with the rules of the road.”

The site is available in 7, and soon 8, languages and it applies as much to an AI start up as it does to Google and OpenAI.

After all, prevention beats cure every day of the week and twice on Sundays.

There is so much noise currently around the threat AI poses to the music and wider creative industries. Can you summarize the key points?

Sure. The vital issues are practical ones, don’t let anyone tell you they are abstract.

The combination of tech, market and governmental developments in AI epitomise exponential risk and opportunity. Get AI right, growth will flow. Get the laws and rights enforcement wrong, the risks are profound.

In general, there are 4 main AI business activity types: content access, dataset or Large Language Model training, b2b administration, and Generative AI. Each comes with its own needs – such as:

• Ensuring accurate metadata for LLM training.
• Watermarking GenAI.
• Ensuring transparency of rightholders’ datasets used across the value chain.
• Avoiding machine generated works flooding the licensed digital market and diluting the royalty pool.
• Ensuring all rights are respected in AI usage – mechanical, graphic, performance, synch.
• Maintaining full flexibility of options as to how to license.
• Protecting the existing and future growth of the UGC and short form SVOD markets which we’ve fought so hard in recent years to underpin with legislative wins such as Article 17 – after all a full 65% of all internet traffic today is driven by SVOD content, up more than 10% in 3 short years and with few signs of any imminent reverse.

But as with all problems sets, sorting the source is the real key.

What I mean by that is that we need to end any unlicensed access of music for any of the above activities. Some of the biggest companies in the world are essentially scraping music from DSPs in order to train and or make GenAI, then rolling that out across their services. One of the big issues with that source problem, aside from it being systematic infringement, is it causes a metadata blind spot. They are just scraping raw audio. Even willing business partners cannot act properly on such bad material. Such scraping is essentially glorified stream ripping. The EU AI Act will help put a stop to such malpractices.

“Some of the biggest companies in the world are essentially scraping music from DSPs in order to train and or make GenAI, then rolling that out across their services.”

Always recall that although there are many tools for transparency here, the best form of transparency is prior authorisation by the AI company. Get these things right, then AI and music markets can symbiotically thrive in future.

What progress are you seeing, legislative and otherwise, across the industry?

It’s as significant as it is rapid. Our team are working on these issues daily in Europe, the UK, Asia, Australia, North America and just about everywhere else. Just last week we were in Mexico speaking with Senators and government about how to optimise AI laws there. All governments are exchanging ideas with each other at unprecedented levels of cooperation and speed.

It has been a huge challenge for our industry. In 2023 far too many governments, Prime Ministers, Presidents, and institutions were either in thrall to AI companies, rolling out the red carpet for them, panicking about what to do, or a mixture of all three. None of this was good for the music industry. We even saw in Germany and France there was huge and unprecedented pressure to give AI companies a free pass, right up to the business end of the EU AI negotiations. I’m glad to report this has not been allowed to occur.

The EU AI deal now contains a robust and extensive set of positive obligations – for example AI companies will be obliged to comply with copyright law and all that entails, will have to retain detailed records of what they’re training their datasets on, must be transparent as to what is involved in GenAI content. The transparency obligation will apply irrespective of where in the world the training has taken place – a remarkable tool, and we will have hefty sanction tools to enforce our rights against misuse. ICMP will also be working directly with the new AI Office on behalf of our members – that’s a responsibility which started last month and will last for decades.

“The EU AI deal now contains a robust and extensive set of positive obligations – for example AI companies will be obliged to comply with copyright law and all that entails, will have to retain detailed records of what they’re training their datasets on, must be transparent as to what is involved in GenAI content.”

Following these recent outcomes, it’s no coincidence – and it’s also highly welcome – that India’s Economy Minister has officially followed suit and demanded prior authorization by AI companies from rightholders, China has introduced detailed and helpful AI obligations and the G7 is shoring up its own AI principles. We remain in discussions with many other governments around the world.

We need to make these standards a springboard, not a ceiling. Only then can creators and companies have full confidence in AI. But we have to plan to get the bases right.

How do we go about developing the licensing frameworks required for generative AI? Is this something the ICMP is currently working on?

This is a key question of course, one which goes to the heart of how the AI and music industries will thrive in the near and longer term. Several core criteria are critical:

• First, you need to identify which AI business activity we’re talking about: content access, content training, b2b administration, or GenerativeAI.
• Then you need to apply all the pertinent rights.
• Then you need to factor in the relevance of which jurisdiction, or jurisdictions.
• Next you need to ensure optionality for the rightholder – there should be no compulsion, no statutory obligation, nor any limitation as to how or who with the music publisher wants to license and administer the activities.

So, yes we need to be fast on all this, but we need to be fulsome, futureproof and efficient too. These are some of the areas we’re focused on in order to ensure a) exclusive rights are exactly that b) authorisation happens before any usage c) growth potential is maximised.

How much potential for industry growth do you think there is in licensing music for AI training?

I think this depends which part of the industry we mean. There’s clearly significant scope for AI licensing as an assistive tool to the creative process, that takes many forms – audio, visual, lyrical. There’s also a vein of value flying under the radar of more high-profile GenAI issues, such as how companies can license datasets or leverage AI to optimise their business admin. And there’s no doubt there’s a place for legal, licensed and therefore valuable GenAI – that’s not theory, it’s already happening, even with household name artists in their live shows.

But, judging by what we see, it’s not a platitude to say music fans pretty much everywhere want human created music. For example, data on visit recurrence to ‘soundalike’ AI music shows that people are curious of the novelty of the tech, but often unconvinced of the superficiality of some of it – certainly of machine generated works. That makes sense to me and I find it reassuring. Who wants to listen to an audible algorithm? Let them do the administration, artists do the creation, publishers drive the value, and fans love and engage with music.

” Technologies come and go, but this latest one must not be allowed to harm creators’ rights nor jeopardise our industry’s economic value.”

AI companies too often deliberately deploy jargon. Sometimes it is intended to try to deflect legal liabilities. Often it is intended to entice VC investment. We hear talk of “hallucination”, “learning”, “ingestion”. At the end of the day, whether it’s upstream or downstream AI, the water source of value is always in songs. Technologies come and go, but this latest one must not be allowed to harm creators’ rights nor jeopardise our industry’s economic value. We’ll get there.

How can rightsholders get involved in RightsandAI.com?

Simple. Get over to RightsAndAI.com. Check it out in English, Chinese, Spanish, French, Arabic, German, Japanese or Italian. Then a choice of two actions are available.

Rightsholders of any kind – music publishing companies, CMOs/PROs, labels, DSPs – can reserve their rights.

Non-rightholders or organisations can endorse the work towards a legal, ethical and fair future of AI and music.

Everyone should spread the word. Talk to industry partners and share in your socials. There isn’t a music company or creator anywhere unaffected by these issues.

Found this article useful? Why not check out:

• Anticipating the Future of Music and Media: What Lies Ahead in 2024?
• Why Music Search is So Complex, and How AI is Helping Us Navigate Surging Content Volumes
• AI of the Storm: Artificial Intelligence and the Music Industry in 2023