Recent advances in fast development have raised complex questions about copyright because AI can generate new content based on old content. The questions may be even more challenging when personal biometric data is used. This is a legal question but also a question about technologies and also a business opportunity.
Young Paul McCartney or digital Eva Herzigová
We have seen many examples of how AI can generate new music based on someone’s style or voice. There are quite impressive examples in that video, not only recreating similar music but also making new music with the voice of a young Paul McCartney or John Lennon. These are illustrative examples of how AI is not just generating new content from old but also using personal data to produce it.
Another example is where brands can hire a supermodel’s digital twin, in this case, Eva Herzigová, for campaigns. It is basically a digitized version of her that can then be used to create new content. This creates opportunities for brands, but it also raises many questions.
Regulators and lawmakers have woken up to this situation, but the reality is that laws and regulations always follow technology development, not lead it. Emerging AI tools are opening new business opportunities and solutions like this, and copyright owners will need to manage not only their personal content but also versions modified by AI. The appearance, voice or other characteristics of stars like this involves their own biometric data.
New regulations around copyright
The EU and US are working to clarify which rules apply to generative AI content and what kind of new legislation is needed, and the EU has a specific new regulation draft for AI. The EU Parliament is considering adding obligations on providers of generative AI models, including requirements for transparency, disclosure of training data, and safeguards for AI-generated content moderation.
The transparency obligation in the proposed AI Act requires providers to document and publicly disclose the use of training data protected by copyright. Specifically, the questions are linked to the copyright implications of generative AI from the perspectives of input (training) and output (generated content).
In the EU, the Commercial TDM (The Exception for Text and Data Mining) exception in the Copyright in the Digital Single Market (CDSM) Directive provides an opt-out mechanism for rights holders. Some commentators see this approach as a way to increase the bargaining power of rights holders and facilitate licensing deals, while others raise concerns about market concentration and exploitation of creators. Transparency requirements regarding copyrighted works in datasets are being called for to address these concerns.
But the question about the concentration of the copyright market also raises other questions, specifically who are the original copyright owners.
Not only content but our biometrics, too
I have written previously that distribution networks try to dominate the content value chain. This might be clear and fair for created content, e.g. music and films that are produced by production companies or distributors. But now we are talking about a larger set of data rather than ‘just created’ content. Should a production company own the singer’s voice, their ‘style’ or a digital twin? This is not only about content copyright; this is about personal rights and digital rights.
So, one aspect is to find models that help to define copyright ownerships of works co-created with AI tools, but it is also to better categorize different ‘assets’ and their copyrights. For example, we can have a singer who produces songs and works with a record label. This record label may own the copyright of the songs, and the singer gets compensation based on their contract and role in creating the song.
But the same singer has a personal voice, appearance and certain characteristics. The singer can also give interviews, make comments and make public appearances. These can be used for the generative AI, too, to create characters, singing and new content. But the copyright cannot be based only on the models used for the singer’s songs. The same applies to sports stars; they also have their personal data.
Need for practical services to manage personal copyrights
These potential problems are not only linked to laws and regulations. In practice, it can be complex for the star or anyone managing personal copyrights. If you are a star, would you like to license your data to digital marketing with generative AI? How do you it? You can hire a law firm to consider such an opportunity and prepare an agreement, but this is very slow and expensive, especially if there are many small campaigns or services where it could be used. The digital age needs digital solutions for this.
Sooner or later, this will become relevant for each of us: how should we manage our own digital characters and data? I wrote in 2021 that avatar ownership becomes relevant to all of us. It is not only linked to metaverses, but it becomes relevant to all digital services, from healthcare digital twins to games and services that can be personalized with your data. Just think if someone creates a digital character that looks like you, speaks like you and behaves like you.
Individuals must have rights to their biometrics data. These are not new issues, but the last few weeks have made these considerations more relevant. Earlier, it was more of a theoretical discussion to consider options and optimal regulation. Now we have seen so many AI solutions based on existing content and biometric data that this is no longer theoretical; it’s happening now. We also need models and platforms where people can manage their own content and data. If there are no platforms and services to do it properly, the whole situation will be a mess. It would be like no models to manage film and music copyrights, but actually much worse when it is about your personal characteristics.