Open science and the use of artificial intelligence (AI) in research are linked in many ways. Two aspects will be mentioned here.

First, open data form an important basis for AI. The latest AI successes, and the considerable advances in this field, are due in no small part to the enormous growth in digital data production – undoubtedly not only but also in the research context. The development of AI models and applications can benefit from open science practices insofar as these practices result in high quality, openly accessible, and easily and legally reusable data – provided common standards are maintained.

The second aspect concerns the openness of the AI systems, models, and instruments themselves. These can be distinguished not least by the extent to which the underlying training data, methods, algorithms, etc are made transparent. The current discussions about “trustworthy” or “responsible” AI are very much focused on this question of transparency. As the transparency and verifiability of the research process are at the same time key principles of research integrity, the use of AI in research presupposes that it is also possible to understand the functioning of the models and instruments – at least to the extent that serious distortions can be avoided (see DFG, 2023). Openness plays a decisive role in this respect.

The use of generative AI can undoubtedly facilitate the creation of research publications, provided it takes place in a regular and responsible way. However, we are also seeing an increase in AI-generated fake publications, which then place a burden on the publishing system. Many publishers are expanding their publication- checking departments and are also using AI tools for this purpose.

The DFG supports open access to findings resulting from research projects, for example through guidelines for handling research data or activities for the open licensing of research publications. In the case of publications, it is important that their reusability is ensured but that the exploitation rights remain with the authors. In this context, the efforts of some academic publishers to secure AI rights of use in publications – even in open access publications – and to contractually agree restrictive limitations or AI usage bans that in some cases remain below the statutory permissions could prove problematic. For researchers, this would mean that, according to these contractual provisions, they would not be permitted to conduct any AI training etc. with their own publications. This contributes to the already existing asymmetric distribution of power in publishing and will manifest itself even more strongly – particularly and precisely because of the innovative and future-shaping AI applications and developments.

A recently published call for proposals by the DFG’s Committee on Scientific Libraries and Information Systems (AWBI) is aimed directly at the development of data corpora for training AI. Thus, although the promotion of excellent, knowledge-driven research remains its primary task, the DFG is also strengthening indirectly and directly the foundations for future AI developments. AI-generated results are reliable only if they are also based on comprehensive research data and publications.

• Deutsche Forschungsgemeinschaft (DFG, German Research Foundation). (2023). Statement by the Executive Committee of the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) on the influence of generative models of text and image creation on science and the humanities and the DFG’s funding activities. https://www.dfg.de/resource/blob/289676/230921-statement-executive-committee-ki-ai.pdf

Bilic-Merdes, M., Brandt, S., & Lentze, M. (2025). Perspektiven der DFG auf KI und Open Access, Teil II. Beziehungen zwischen Open Science und Künstlicher Intelligenz. open-access.network. doi.org/10.64395/njk7k-gpm12.