Contribution by Ana Lazarova, KR21’s national coordinator for Bulgaria
On 14 June, I had the privilege of attending the conference on “Text and Data Mining, Artificial Intelligence, and Copyright” in Stockholm. The event was co-funded by the KR21 programme and organised by our own Eric Luth, National Coordinator for Sweden, in collaboration with Wikimedia Sweden, the Swedish Library Association, and Institute of Intellectual Property and Market Law at Stockholm University (IFIM). The conference delved into the complexities at the intersection of AI and research, offering insightful sessions on pressing topics. These included the application of Text and Data Mining (TDM) in legal scholarship, the strengths and weaknesses of implementation of EU TDM exceptions, as well as the current practical challenges linked to uptake of TDM by the researchers served by libraries. Additionally, it explored the dual role of public and academic libraries: as both beneficiaries of these exceptions and custodians of collections and databases crucial for AI training.
Setting the stage, the conference was opened by Dean Jane Reichel of Stockholm University, who provided insights into legal challenges and practical issues related to text and data mining. She highlighted AI’s impact on societal information dynamics and stressed the need for cautious navigation amid uncertainties to build digital trust. The organisers of the event – Wikimedia, the Swedish Library Association and IFIM – were represented by Eric Luth, Lisa Gemmel and Professor Frantzeska Papadopoulou respectively. They emphasised the importance of open access in AI development, highlighting evolving intellectual property issues that sparkled discussions among legal scholars, librarians, and researchers. The globally streamed event aimed to promote understanding and collaborative strategies in navigating these technological landscapes within ethical and legal frameworks.
Practical Opportunities and Challenges of Text and Data Mining
The first session of the conference focused on the “Practical Opportunities and Challenges of Text and Data Mining”. Felix Steffek, a professor of Law at the University of Cambridge and Senior Member of Newnham College, talked on the topic of “Creating and Publishing Large Legal Datasets for Data Mining and Artificial Intelligence”. He introduced the Cambridge Law Corpus, a collection of 320,000 UK court decisions spanning 450 years, crucial for AI applications predicting legal outcomes and extracting information. Prof. Steffek highlighted challenges like data availability, copyright issues, and data protection, advocating for open access to legal data, including in the form of withholding copyright protection from official texts. His team’s ethical considerations included restrictions on commercial use and anonymization concerns. Prof. Steffek called for governmental action to facilitate data sharing and underscored AI’s role in improving legal access and transparency globally.
Professor Johan Lindholm’s presentation focused on integrating AI and text mining into legal scholarship. He argued that AI could enhance empirical legal research by offering new insights into complex legal issues. Prof. Lindholm emphasised that despite traditional scepticism, AI and text mining were crucial for legal scholars to study law as a dynamic social phenomenon accurately. He discussed practical applications such as judgement analysis, predicting citations, and detecting legal overrulings. Prof. Lindholm also called for improved legal data accessibility, cleaner data curation, and collaborative platforms to advance AI’s role in legal research and education. He stressed the need for trained scholars to navigate these evolving methodologies effectively.
In the Q&A session, the speakers discussed the challenges of data quality in legal datasets for AI and TDM. They highlighted issues like errors and inconsistencies in court data across EU databases, emphasising the need for collaboration over immediate standardisation. They acknowledged concrete efforts for data sharing but expressed scepticism about achieving uniform standards soon. Concerns also extended to cross-border legal predictions and copyright complexities, indicating ongoing research and startup initiatives to address these issues globally. They concluded with a call for interdisciplinary collaboration to improve legal data integrity and accessibility.
Legal Opportunities and Challenges
The next session – and the one yours truly was most excited to attend – was on “Legal Opportunities and Challenges” and featured Branka Marušić, Stockholm Fellow at Oxford University and Christ Church, Kacper Szkalej, a Researcher in Intellectual Property Law at the Institute for Information Law (IViR) at the Amsterdam Law School, and Communia’s Legal Director Teresa Nobre.
Dr. Branka Marušić presented on “TDM Exception – Methodology of Implementation in the EU Member States: Creating Cohesion or Diversion?” exploring how the two TDM exceptions in the Directive on Copyright in the Digital Single Market (DCDSM) were implemented across Europe. Dr. Marušić discussed the assessment of the convergence or divergence in the results of the recent transposition. Her emphasis was on the methodology of implementation, exploring literal transposition versus flexible approaches. Addressing complexities in copyright law, she clarified that according to the way in which the assessments of transpositions is carried out by the European Commission, the two existing exceptions could be seen as full harmonisation in scope. i.e. maximum standards concerning the types of usage provided for in arts. 3 and 4 of the DCDSM. She critiqued the flexible definitions of the Directive and noted the uneven national implementations, as well as gold-plating provisions such as collective licensing. Conclusively, Dr. Marušić advocated against a flexible approach, foreseeing redundancy or legal conflicts without clearer guidelines or judicial interpretation.
A Moment of Reflection
One highlight of Dr. Marušić’ talk was the issue of rightsholders circumventing TDM exceptions by licensing contracts. Indeed, a study on the contractual override of core library missions, commissioned by KR21 and performed in 2023 (the final report is pending publication) on 100 licensing contracts between scientific publishers and other data vendors, on the one hand, and public libraries and research institutions, on the other, found that more than a half of the 100 studied contracts, which were concluded after the entry into force of the DCDSM in 2019, sought to prohibit towards text and data mining. Of these contracts, 31 prohibited TDM explicitly or implicitly (through the prohibitions of the use of robots, spiders, crawlers or other automated downloading programs; or of the continuous and/or automatical search or index of licensed materials/databases etc.). Furthermore, 9 out of 100 contracts provided for limited opportunities for TDM compared to the scope of the current research TDM exception, while 6 out of 100 contracts had unclear provisions regarding mining. Finally, 41 of the 100 contracts had no mention whatsoever of the use for TDM purposes, of which 11 contained an express general contractual override provision, explicitly prohibiting uses that were not expressly authorised by the contract.
While I respectfully disagree with the bleak forecast that contractual conditions imposed by rightsholders might render TDM exceptions redundant, they can surely account for a great deal of legal uncertainty amongst beneficiaries, potentially resulting in a chilling effect on the utilisation of these exceptions. This effect might be exacerbated by the inherent ambiguity of one of the major conditions for the application of the TDM exceptions – that of “lawful access”. Which brings us to the next topic, discussed at the conference…
The second speaker of the session, Dr. Kacper Szkalej, shared insights on “The Paradox of Lawful Text and Data Mining: Experiences from the Research Sector and Future Directions” addressing key challenges and advancements in TDM research practices focused around the ambiguous concept of “lawful access” as precondition to applicability of the exceptions. He highlighted challenges such as fragmented legal interpretations and the influence of right holders on access conditions and advocated for clearer legislative frameworks to balance interests between users and creators in the evolving landscape of AI and research. Dr. Szkalej talked about his involvement as part of the IViR team in the drafting of the recent study, commissioned by the Directorate-General for Research and Innovation of the European Commission on “Improving access to and reuse of research results, publications and data for scientific purposes – Study to evaluate the effects of the EU copyright framework on research and the effects of potential interventions and to identify and present relevant provisions for research in EU data and digital legislation, with a focus on rights and obligations” (Kacper has authored part 1.3. of the study, entitled “Overview of plausible policy options and areas in need of improvement”). He concluded his presentation with an announcement of his freshly published article on the nexus between TDM exceptions and copyleft licenses. The news presented a particularly exciting development for this National Coordinator, who has in the past taken part in exploring the topic within the Creative Commons network.
Finally, Teresa Nobre’s presentation entitled “Teaching the Machine: Policy Developments in Europe” focused on policy frameworks shaping AI and TDM across the continent. Nobre stressed the need for clearer and standardised opt-out mechanisms and data transparency in AI training. She advocated for the expansion of copyright exceptions for research purposes beyond TDM, addressing access barriers and liability concerns for researchers. Nobre also urged harmonisation across EU Member States to facilitate cross-border research collaboration and ensure fair use of digital resources, promoting public interest and innovation.
AI and Media and Information Literacy
The third session, “AI and Media and Information Literacy”, Jutta Haider, a professor at the Swedish School of Library and Information Science (SSLIS), University of Borås, and affiliate professor at Lund University (Sweden), delved into the role of AI in media and information literacy, examining its impact on public understanding and access to information. The conference concluded with reflections on the topic of “Libraries, AI and copyright” by Ms. Karin Grönvall, National Librarian of Sweden, and Mr. Wilhelm Widmark, Head Librarian at Stockholm University Library. Ms Grönvall emphasised the role of libraries in AI and copyright discussions. Highlighting the fact that their vast collections are essential for AI research, she stressed the role of the laws on legal deposit in connection to text and data mining. Despite challenges like copyright restrictions, she sees libraries as pivotal in balancing access and IP rights. Mr. Widmark, reflected on the crucial role of libraries in supporting researchers amidst copyright challenges. He advocated for libraries to enhance scholarly communication and assist with publishing. Widmark emphasised the need for collaboration across university administrations to provide comprehensive support for research data management and open science initiatives, stressing the importance of negotiating fair access to digital resources amidst evolving legal frameworks.
Closing Remarks
In summary, the Stockholm conference provided a comprehensive exploration of AI, TDM, and copyright, fostering interdisciplinary dialogue and paved the way for collaborative strategies in legal and technological advancements. In particular, it highlighted the need to look further at how to enable research into, and powered by, TDM in the context of wider efforts to realise the potential and manage issues linked to AI. Crucially, it advanced shared understanding of the issues among the wide range of stakeholders present, and should help facilitate a more research friendly agenda in upcoming legislative discussions.
8 July 2024
