Copyright and technology are intertwined. This started with the printing press, then photography, the phonograph, moving pictures, broadcasting, until the internet and software. Now, we have artificial intelligence (AI), used by almost everyone for any purpose. The intersection of AI and copyright focuses on two main issues: whether AI-generated outputs can be protected by copyright, and whether using copyrighted works to train AI models constitutes infringement. The law currently resolves the first issue: purely AI-generated works cannot be protected by copyright as the law requires human authorship. The second issue — whether training AI models on copyrighted works constitutes infringement — remains actively contested. Unless there is some sort of solution, the use of copyrighted works by tech companies without permission constitutes infringement. A solution would require setting out what is protected and what is public domain.
Using AI requires training a model by feeding it copyright protected works. The threshold for copyright protection in the UAE is relatively low. The work must be sufficiently original, that is it cannot be derived from previous works of others and it must reflect the author’s personality. Law grants copyright owners a bundle of exclusive rights, including the rights of reproduction, distribution, and preparation of derivative works.
Most recently, it was announced in the UAE that royalty fees will be charged for using music in commercial establishments, for example, cafés, restaurants, hotels, radio stations, TV channels, malls and airlines (public performance of music). Collective societies — Music Nation and the Emirates Music Rights Association — will manage the collection and distribution process. This is the start of collective administration in the UAE. The same could apply to generative AI, especially as the UAE competes to become a global AI powerhouse. When using generative AI, the issue that arises is the use of copyrighted materials as inputs for training AI models, ultimately creating new outputs such as literary, dramatic, artistic, and musical works. For example, music can be generated using an AI music generator trained on protected recordings. The process requires making copies of the copyrighted works for training purposes. Training requires an initial copy during data collection, then a series of copies during model training, and finally a copy permanently incorporated into the model’s parameters. This type of use requires authorization from copyright holders unless there are exceptions in the copyright law, otherwise there is an infringement of the rights of authors.
As a matter of principle, authors do not enjoy exclusive rights over their works. Countries, including the UAE, balance the rights of authors and public interest. For instance, as an exception, “private” copying is permitted under the UAE Federal Decree-Law on Copyright and Neighboring Rights, subject to certain conditions relating to the subject matter, the identity of the parties, and the scope of coverage. As a setback for AI commercial companies, only natural persons can benefit from this exception. No company could ever benefit from the private copying exception. Other exceptions in the law do not fit the use of inputs in AI. These exceptions permit copying small parts of a work in a written, audio-recorded or audio-visual-recorded form for educational, cultural, religious or vocational training purposes. In other countries, such as the US, the central battleground in AI training litigation is the fair use defence. However, fair use determination differs from case to case, depending on the nature of the copyrighted works, how the work is used in the training process, and the purpose and functionality of the AI system being trained. AI systems can reproduce substantial portions of their works verbatim.
Permitting the free use of copyrighted works to train AI models will deprive authors of a source of revenue in the form of licensing fees. Billions of dollars are at stake as generative AI models are trained on copyrighted works to create outputs that may compete directly with, or as a substitute for, the original works in the market. However, AI developers could argue that AI outputs are not substantially similar to any specific training work and therefore do not substitute the originals.
The existing copyright framework in the UAE provides compulsory and collective licensing models for specific industries such as music, but this does not extend to AI training data. As the UAE starts its journey with the collective system for music, it could improvise a similar system for using copyrighted materials for AI training. Rather than negotiating with each and every copyright holder, an insurmountable task by itself, and be subject to lawsuits, AI companies in the UAE could resort to a collective mechanism whereby there would be payment of fair remuneration to copyright holders through collective societies. AI companies could negotiate with a representative society rather than numerous copyright holders. The system could incorporate an opt-out option whereby a copyright holder could decide not to join the process and exercise his rights individually if he desires to do so.
Collective societies could act as a blanket clearinghouse for handling data licensing for AI training thus ensuring that local copyright holders in the UAE are paid by tech companies. In addition, the designated collective society could have an enforcement authority whereby it could bring an action against an AI tech company for unpaid royalties. This would be an effective tool to address the asymmetrical power relationship between individual copyright holders and tech companies. In the absence of this system, the individual author behind the user generated content may think it is not worthwhile pursuing legal action for the unauthorised use of his work. The fragmented nature of copyright ownership, with individual authors holding AI training rights rather than publishers, makes individual licensing transactions costly, and a collective mechanism is the only practical way to ensure broad remuneration.
The success of Music Nation and the Emirates Music Rights Association shows that there is an opportunity to create collective licensing bodies for new technologies, and the AI context presents the same scale and fragmentation problems that justified those frameworks. Of course, the value of any individual work as AI training data is difficult to quantify and a collective remuneration system would face challenges in fairly distributing royalties among the hundreds, if not thousands, of copyright holders whose works may have been used. As such, distribution can be made based on market share data. The legal landscape at the intersection of AI and copyright is evolving rapidly. It is only a matter of time before we can expect a similar case using collective systems in AI inputs.
The views and opinions expressed in this article are those of the author and do not reflect the official policy or position of any agency of the government. The author made this article in his own personal capacity.
Bashar Malkawi is Legal Counsel and Head of Research and Publications at the Government of Dubai Legal Affairs Department.
Email your news and story ideas to: [email protected]
