More than two years after it was introduced, and after a great deal of negotiation, the European Union’s AI Act is headed for a vote in the European Parliament.
Parliament’s Internal Market and Civil Liberties Committees on Tuesday (February 13) voted 71-8, with seven abstentions, to approve what is now likely to become the democratic world’s first comprehensive set of laws governing the development and use of AI technologies.
The AI Act now heads for a full vote in the European Parliament, which is scheduled for April. It will then have to gain endorsement from the European Council to become law.
Lawmakers are under pressure to have the law finalized ahead of EU elections in June, which polls show will likely significantly change the parliament’s makeup.
For intellectual property-dependent industries, such as music, the most salient parts of the law have to do with the rules on copyrighted materials used in the training of AI.
The AI Act is clear on transparency requirements: Developers of “general-purpose” AI models will have to provide a “detailed summary of the content used for training the general purpose model” in order to “facilitate parties with legitimate interests, including copyright holders, to exercise and enforce their rights under Union law.”
However, the law is less clear on what constitutes a “general-purpose” AI model: are AI models designed to create music, and trained primarily on music, “general purpose”?
“This regulation aims to protect fundamental rights, democracy, the rule of law and environmental sustainability from high-risk AI. At the same time, it aims to boost innovation and [establish] Europe as a leader in the AI field.”
European Parliament Press Release
The law also leaves some doubt as to whether or not AI developers will be required to license the copyrighted material they use to train their models.
The proposed law states that “any use of copyright protected content requires the authorization of the rightholder concerned unless relevant copyright exceptions and limitations apply.”
The issue, of course, is which “relevant copyright exceptions and limitations” might apply.
According to an analysis from multinational law firm Morgan Lewis, in the case of US-based AI companies that provide AI services in Europe, the proposed law “does not state clearly whether an exemption under US copyright law, such as a finding of fair use, would be sufficient… It is possible that only authorization of the rightholder as recognized under EU copyright law will count.”
A number of lawsuits against AI developers are working their way through US courts today, including one brought by Universal Music Group, Concord Music Group and ABKCO against Anthropic, developer of the Claude chatbot, which the music companies allege has produced copyrighted lyrics for users.
Many of the AI developers involved in these suits have argued that their unauthorized exploitation of copyrighted materials to train their AI models constitutes “fair use” under US law.
European Union copyright rules — which seek to guide the copyright laws that EU member states can adopt — don’t have a “fair use” doctrine; instead, there is a specific list of exceptions to copyright. Among those exceptions is one for research purposes; some AI companies have argued that their training of AI models is research.
OpenAI, the maker of the popular ChatGPT chatbot, launched as a non-profit research company, before switching to being a for-profit commercial enterprise.
Another exemption to copyright law in the EU involves “temporary acts of reproduction which are an essential part of a technological process.” Some of the arguments used by AI companies facing lawsuits could fit this exception; they say their use is essential, and that the use of copyrighted works is temporary.
AI companies have argued that their AI bots are not intended to be used to reproduce the copyrighted works the models were trained on. However, rights holders have alleged that these AI models do, indeed, reproduce copyrighted works at times.
The ambiguity in the EU’s proposed AI law may have to do with the fact the law is intended to achieve two things simultaneously: To protect EU citizens from the potential negative consequences of AI, while at the same time making Europe competitive in the AI field, in which the US is currently the global leader.
“This regulation aims to protect fundamental rights, democracy, the rule of law and environmental sustainability from high-risk AI,” declares a statement from the European Parliament, issued on Monday. “At the same time, it aims to boost innovation and [establish] Europe as a leader in the AI field.”
During negotiations on the proposed law, France, in particular, lobbied intensely to give AI developers as much leeway as possible, on the argument that placing burdens on them would cause European AI companies to be less competitive compared to their peers elsewhere.
Part of the solution to that appears to be that the AI Act seeks to apply EU copyright law on all AI services that are accessible in the EU, regardless of where they were trained.
According to the Morgan Lewis analysis, that provision could lead to conflicts between the EU and other jurisdictions.
“Such an approach disregards both the country of protection principle… and the internationally recognized territoriality principle for IP law, which states that the copyright laws regarding infringement of the place where the infringing activity occurred are applicable,” the analysis stated.
“The implications of the breach of these principles by the AI Act are enormous, and could have a major impact in favor of domestic providers of [general-purpose AI] models, e.g., in the public procurement of certain AI services. It could also lead to trade disputes with the EU and retaliation by other non-EU countries against EU businesses.”
In other words, as in the US, the shape of EU policy on AI and copyright could end up being settled in the courts.Music Business Worldwide
