Following the adoption of the GPAI Principles of Co-Generated Data, approved by the ELI Council in December 2024, a high-level webinar was held to elaborate on the Principles intended to address new copyright or data protection rights in co-generated in- and output (available here). The event brought together leading experts to examine the implications and legal challenges of co-generation, data sharing, and governance in digital ecosystems.
Pascal Pichonnaz (ELI President; Professor, University of Fribourg) opened the webinar by highlighting the significance of AI co-generation in modern digital ecosystems and the need for robust legal frameworks to guide its governance. He acknowledged the collaborative efforts of ELI, GPAI, and CEIMIA in developing principles, which are intended to be of a preliminary nature for such a fast-moving field, that support ethical and legally sound AI applications.
Christiane Wendehorst (ELI Scientific Director; GPAI Project Co-Lead; Professor, University of Vienna) provided an overview of the GPAI ‘Cogen’ project, which has been running for two years and drew influence from the ALI-ELI Principles on the Data Economy. The current project builds on this foundation, exploring whether these principles can be extended beyond data to cover content or even AI itself. Additionally, Wendehorst raised the question of whether these principles, initially developed for individual rights, could be adapted to encompass collective rights. She emphasized that this exploration requires addressing both data law and intellectual property law.
Kyoko Yoshinaga (GPAI Project Co-Lead; Project Associate Professor, Keio University; Non-Resident Fellow, Georgetown Law Center) discussed the development of the principles to guide legislators in drafting legal provisions for generative AI, specifically in relation to copyright and data protection rights concerning co-generated inputs and outputs. Yoshinaga explained that these legal considerations must address the distinct roles of copyright, which incentivizes creativity and innovation, and data protection, which prioritizes individual autonomy and harm prevention. Drawing on legal frameworks from the US, EU, and Japan, and engaging with global experts, the project aims to create a coherent legal framework that integrates these two areas of law while accounting for the unique challenges of generative AI.
Alain Strowel (ELI Co-Reporter; Professor, UCLouvain; Partner, Pierstone Brussels) and Sebastian Schwamberger (ELI Co-Reporter; Professor, University of Rostock) provided a comprehensive explanation of the principles. Strowel mentioned the main report that preceded the Principles, which focused on how different jurisdictions, such as the EU, US and Japan deal with the issues of co-generation under the frameworks of copyright and data protection law. Strowel covered principles such as the use of publicly available data, the right to an economic share in co-generated content, and remedies for the unlawful use of co-generated input. Schwamberger, on the other hand, compared the regulatory frameworks of the EU, US, and Japan regarding the principles of the right to information and transparency in data processing, the right to authentic attribution for AI-generated output, and the transition from co-generated input to co-generated output. Schwamberger also discussed the economic aspects of these processes, highlighting the need for balancing rights across different jurisdictions.
Michele Woods (Director of the Copyright Law Division, WIPO) highlighted WIPO’s focus on fostering dialogue among stakeholders to create a human-centric and transparent AI system. Woods provided insights into WIPO’s ongoing work on policy responses and international frameworks aimed at reconciling the roles of human authors and AI-generated contributions. She discussed WIPO’s efforts to reconcile AI’s role in creative works without undermining human authorship and emphasized the need for legal clarity around ownership, liability, and moral rights. Woods also addressed the importance of data control in developing countries, fair remuneration for creators, and the challenges of integrating AI into existing copyright frameworks.
Christopher S Yoo (Professor, University of Pennsylvania; American Law Institute (ALI)) provided further insights from a US perspective. He discussed the US's lack of an omnibus privacy law and how publicly available data is largely unregulated, with exceptions like the Computer Fraud and Abuse Act. Yoo also addressed recent legal developments, such as Thomson Reuters v Ross Intelligence, which impact the use of data for AI training. Having compared US and EU approaches, Yoo noted that the US allows data use unless explicitly prohibited, while the EU requires affirmative consent. He raised concerns about the challenges of data rectification and the cost of retraining algorithms, emphasising the need for clear policies balancing innovation with privacy and intellectual property rights.
Joe Massey (Senior Researcher, Open Data Institute (ODI)) discussed the complexities of co-generation through six case studies from the ODI under the GPAI Co-Gen Project, including data labeling, crowdsourced data collection, and generative AI. He highlighted challenges in exercising rights, especially in industries like data labeling, where poor labour conditions are common. Massey emphasised the need for more transparency in AI data practices and the difficulty of implementing opt-out mechanisms for generative AI training. He suggested exploring alternative solutions like bespoke data licenses and new governance models to better support co-generators and ensure fair redistribution of financial benefits.
The webinar concluded with a lively discussion, where participants discussed various aspects of AI regulation, the role of human oversight in AI-generated content, and the evolving legal landscape for digital ecosystems. The session underscored the critical role of international cooperation in shaping effective AI governance policies.
The recording of the webinar is available below.
* This webinar aimed to disseminate the GPAI 2024 report, which was planned prior to the integration of the Global Partnership on Artificial Intelligence (GPAI) and the Organisation for Economic Co-operation and Development (OECD) mid-2024. Consequently, the report and this webinar were not subject to approval by all 44 GPAI and OECD members and should not be considered to reflect their positions.
