Statement EVA about copyright and AI

Within European legislative bodies, a lot of hard work is being done on regulations that are going to apply to artificial intelligence: the so-called “AI Act”. Last week in the European Parliament another step was taken towards an agreement on this, which will also specifically address image-generating AI. In the second half of this year, additional efforts will be made within the EU on the definitive text. Pictoright is closely involved in this process through the European umbrella organization EVA.

Last week, EVA released the following statement on copyright and AI:

• We welcome the “Proposal for a Regulation laying down harmonised rules on Artificial intelligence” (AI Act), as there is an urgent need to identify risks and regulate uses related to AI.

• EVA considers it very important that European digital regulations protect the “Acquis Communautaire” in terms of copyright.

• EVA recalls the importance of maintaining a high level of copyright protection to protect European creativity.

• EVA warns the European Commission of the extremely serious consequences for visual artists of having a general exception allowing the free use, without permission or compensation to visual artists, of data, text, and image mining.

• EVA considers it inadmissible and contrary to the Charter of Fundamental Rights of the European Union that European digital legislation would allow, directly or indirectly, the recognition of AI-generated works as works protected by copyright and the extension of this human right to machines or companies owning such machines.

• EVA urges the European Commission to provide guidance on standardised opting- out from Article 4 DSM uses. Visual artists cannot reserve their rights by machine readable tools because such tools are, in general, not available, or not standardised and therefore are excluded from monetisation based on licensing.

• Articles 12, 15, 17, 18, 19 and 20 of the DSM directive should be the inspiration for European digital standards as they ensure the necessary safeguards and encourage collective management.

The impact of AI on visual artists

AI could become a tool for Artists in the processing and finalisation of their works.  CMOs (Collective Management Organisations) could also benefit from AI to facilitate the management of authors’ data and for authorship identification. However, at the same time, AI hurts human artists by, for instance, increasing the demand for low-cost AI- generated art and therefore causing unfair competition. Also, by using authentic works of art without the permission of the authors who do not get any remuneration, whereas AI companies make huge fortunes, raises questions on the legality, ethics and liability of AI applied to visual arts.

Visual authors are vulnerable, in a weak bargaining position and often self-financed. This vulnerability leaves them with a need of a high-level copyright protection which is more recently promoted by the EU Directive 2019/790 of 17 April 2019 on the Digital Single Market (DSM) and the Digital Market Act (DMA). While the DSM Directive has provided new copyrights and remunerations for authors[1],  it takes time until the new rights can show positive effects and develop their potential to improve the author’s situation. The efficiency of the European copyright framework must be doubted due to unfair practices, such as buy-outs, as has been reported by the Council Presidency that collected information from member states  [2], showing that authors are still not receiving appropriate remuneration.

More must be done to protect these vulnerable authors. While exploiting the enormous potential of AI, the rights of visual artists must be protected through effective regulation that includes compulsory remuneration and the possibility of opting out from AI uses.

Limited exceptions for text and data mining

The EU Directive 2019/790 (DSM Directive) provides for two exceptions allowing the use of protected data to train AI: for scientific non-commercial research purposes and cultural heritage institutions, as well as for data analytics, for any commercial use as long as the rightsholders agree (Art. 3 and 4). Authors may opt-out from commercial use only by machine readable means, but this solution does not work for the visual sector since no common standards are available and the EU does not provide any support or guidance on such standardization. The text and data mining exception is particularly harmful to authors, who are not aware of the use made of their works due to a lack of transparency from AI companies, who do not usually provide the necessary tools for artists to verify these uses. A transparency obligation should not be limited to providing a summary of the protected works used for training AI. Also, fragmented use of works in data mining affects moral rights of authors. Exceptions must not be further extended as that would lead to a new value gap.

EVA expresses its deep concern about the pressures that the technological industries developing Artificial Intelligence are putting on governments to obtain a “general exception” for text and data mining (TDM) that would allow any kind of use, including a commercial exploitation of copyrighted works of visual art. Copyright holders, who have invested their financial resources and effort in the realisation of their works, do not receive any compensation. Moreover, it must be kept in mind that since AI systems most likely cannot “unlearn” protected works that they have been trained with, the “opt-out”-option may not erase the damage already done to artists. This should be addressed, for instance in the form of remuneration as compensation, but also possibly by an opt-in system where AI companies must obtain the authors’ prior authorisation to exploit their works, and hence returning to the principle that any use of a copyrighted work requires the prior authorisation of its author.

The current AI legal ecosystem in the United Kingdom with regards to exceptions is concerning. In an attempt to make the UK the “world’s Artificial Intelligence Hub” there is a trend to obtain,  on the one hand, a general exception to copyright permitting the unlimited use of text and data mining of copyrighted works and databases, for commercial use and, on the other hand, to amend the current law on computer-generated works (CGW) so that computer-generated works, without a human author of the work, can be protected by copyright, i.e. to recognise human rights, such as intellectual property rights, in a machine. We kindly ask the European Commission to oppose these movements.”

AI-generated art would not exist without real artworks

AI systems generate new art by processing huge existing data sets, which are often protected by copyright. The images generated can be, more or less, similar to the initial images[3] or copy the style of other artists in new images[4], or even be generated starting from natural language [5], just to mention a few applications. Protected images are used in multiple ways: the scraping, copying and download from the internet, the use for training purpose and the output from the AI are several acts that require prior authorisation in copyright. Moreover, the output from the generative AI process regularly violates moral rights of authors because of the use of parts of protected works, as well as of the distortion, mutilation and/or modification of the protected works. The training and the generation of output are two processes with each qualifying for damages and interest.

Deepfakes and the need for transparency

AI can also be used to produce the so-called “deep fakes”. Deep fakes are AI-generated media created by manipulating existing copyrighted images, photographs, illustrations, and other visual works. In most cases, the works appear to be original, misleading people to believe they are authentic and undermining the trust placed in their real value, which is crucial for the visual arts sector. Generative AI must be transparent about deep-fakes and their true origin, to prevent the public being misled, which could have devastating effects. As was reported in the media, the European Parliament would presently be considering an amendment to the draft AI Act which does not require full transparency for certain cases where it would be obvious that AI images are fake, such as in parodies. EVA, however, supports full transparency to prevent damages to authors and society caused by deep fakes.

Copyright protection as a fundamental right

EVA welcomes the report of the Special Committee on Artificial Intelligence in the Digital Age of the European Parliament (AIDA) which emphasizes that the digital transformation must be shaped in full respect of fundamental rights and in such a way that digital technologies serve humanity. EVA supports the proposals of the Special Committee on Legislation, in particular the Better Regulation Programme, and we consider its proposal urging the Commission to develop ex ante impact assessments, in particular on the relationship of the AI with the copyright of visual creators, to be indispensable. However, the Committee failed to address the effects of AI on copyright protection.

It must be remembered that the right to protection of intellectual property is a fundamental right as it is defended in the Charter of Fundamental Rights of the European Union, the Universal Declaration of Human Rights, and the Universal Berne Convention.

EVA supports the parallel legislative process at the Council of Europe and welcomes provisions of safeguards for visual artist for rights and remuneration, which ensure that adequate oversight mechanisms as well as transparency and auditability requirements tailored to the specific risk are available and guarantee protection of Fundamental Rights.”

The upcoming Artificial Intelligence Act is an excellent opportunity to ensure a balanced regulation that respects the moral and economic rights of authors.