Principles for regulating the copyright protection of AI-generated content

The possible new regulatory approach on the copyright protection of AI-generated content must follow the following principles:  

  • Neutrality: Instead of building sectoral approaches, the possible new regulations should strictly follow technology-, content- and business-model neutrality.
  • Harmonisation: The regulatory framework for the copyright protection of AI-generated content should be fully harmonised.
  • Uniformity: The European Union must provide clear and unambiguous definitions for key concepts of the possible new regulations (e.g., AI). The definitions should be the same across policy sectors (e.g. currently drafted Artificial Intelligence Act). The legislative environments for the different sectors must constitute a consistent whole.
  • Pro-innovation:  Any new regulatory environment should be built to encourage investment in R&D on AI-generated content, and the use of AI and AI-generated content by human creators. On the one hand, this requires good copyright protection of AI-generated content. And on the other hand, it requires a regulatory environment that encourages the use of human-generated content in the training of AI. 
  • Clarity and reliability: In order to build a predictable regulatory framework, the regulation itself has to be clear and understandable, not just the guidance documents. The current significant market uncertainty around the implementation and impact of the DSM Copyright Directiveis an excellent example of the harmful outcomes of poor regulation. 
  • Global approach: Digital markets are, by their nature, global rather than regional. Therefore, EGDF fully supports the efforts to minimise the fragmentation of global digital markets by pushing worldwide regulatory standards on the copyright of AI-generated content. Even when the European Union and third countries disagree on a joint approach, the different regulatory approaches must be as transparent as possible for European companies that operate globally. 

Many of the key copyright questions related to the use of AI in the games industry are connected with the different ways EU member states are going to implement the DSM Copyright directive). Therefore, the Commission should carefully map the current state of the European copyright framework during 2022 before going forward with preparations for further regulatory changes.  

The key areas of further clarification 

The key areas that would require further clarification are:

  • The threshold of originality in the context of copyright protection for AI-generated content: Currently, the threshold of originality needed for copyright protection is defined differently in different countries. In some countries, human involvement is required, and in some not. In different EU member states, different levels of skill, labour and judgement are needed to achieve copyright protection.
  • Implementing the text and data mining (TDM) exception to allow training the AI with copyright-protected content: All Member States should clarify that processing data for AI training should be encompassed within the TDM exception. Furthermore, it is still unclear how far commercial activities will fall under the scope of the TDM exception in different member states and under what conditions. 
  • Moral rights and AI
    • First of all, it is essential to keep in mind that copyright exceptions also apply to moral rights. Consequently, the national implementation of the text and data mining exception will also clarify the extent to which authors are allowed to use their moral rights to limit the use of their copyright-protected content for training an AI.
    • Secondly, the applicability of moral rights should be evaluated on a case-by-case basis. The moral right of integrity, for example, is more likely to apply when an AI is used to mimic or slightly modify an artistic work. It is less likely to apply when AI is used, for example, to make the work available in a higher resolution. 
  • A new sui generis economic right to protect AI-generated content: In the long run, to secure investment in R&D on AI-generated content that does not include any human involvement, the European Commission should explore the possibility of creating fully harmonised and transferable sui generis economic rights for AI-generated content.

Implementing the text and data mining (TDM) exception to allow training the AI with copyright-protected content

 All Member States should clarify that processing data for AI training should be encompassed within the TDM exception. Furthermore, it is still unclear how far commercial activities will fall under the scope of the TDM exception in different member states and under what conditions. 

Moral rights and AI

First of all, it is essential to keep in mind that copyright exceptions also apply to moral rights. Consequently, the national implementation of the text and data mining exception will also clarify the extent to which authors are allowed to use their moral rights to limit the use of their copyright-protected content for training an AI.

Secondly, the applicability of moral rights should be evaluated on a case-by-case basis. The moral right of integrity, for example, is more likely to apply when an AI is used to mimic or slightly modify an artistic work. It is less likely to apply when AI is used, for example, to make the work available in a higher resolution. 

AI mimicking an artist style There is no silver bullet in the fight against copycat games

Already now, in the pre-AI era, copycat games made by human beings are a common problem. Extending copyright protection to cover style is not the solution as it would cause significant uncertainty about when one is committing a copyright infringement while developing further an artistic style.  Consequently, this question should not be seen as an AI-specific problem. Instead, it should be approached as a general challenge that requires a multitude of different tools to be tackled (e.g. strong IPR protection and consumer protection tools)

AI-made or AI-assisted labels  

Information obligations towards end-consumers are more a consumer protection or AI regulation topic than a copyright issue. AI-based tools are already widely used in creative industries, and their use is only likely to increase in the near future. Therefore, an AI-made or AI-assisted label would be so common that they are not likely to bring added value for the end-user. The use of such labels should be limited to high-risk use of AI that poses significant and evident risks to health and safety or fundamental rights.   

The full position paper can be accessed here: http://www.egdf.eu/documentation/5-fair-digital-markets/5-a-digital-ready-copyright-framework/copyright-and-ai/ or as a pdf from here http://www.egdf.eu/wp-content/uploads/2021/05/202105-EGDF-approach-Copyright-and-AI-statement.pdf