Yes, we are open?! Responsible design of artificial intelligence with Open Science

September 2025 51 Katja Mayer, Jochen Knaus, Stefan Skupien et al. Yes, we are open?! Responsible design of artificial intelligence Conference results in the field of open science and AI Automatic Translation from German with DeepL - preliminary version 51 Yes, we are open?! Designing artificial intelligence responsibly AUTHORS Katja Mayer \\ University of Vienna \ [email protected] Jochen Knaus \\ Weizenbaum Institute \ [email protected] Theresa Züger \\ Humboldt Institute for Internet and Society Urs A. Fichtner \\ University Medical Centre Freiburg Katrin Glinka \\ Berlin University of Applied Sciences Jan Hase \\ Weizenbaum Institute Lambert Heller \\ TIB - Leibniz Information Centre for Science and Technology Lucic-Aiméc Kaficc \\ Hugging Face Sebastian Koth \\ Weizenbaum Institute Dominik Kowald \\ Know Centre Research GmbH, University of Graz Ilona Lipp \\ University of Leipzig Katharina Meyer \\ Digital Infrastructure Insights Fund Petra Ritter \\ Berlin Institute of Health at Charité – Universitätsmedizin Berlin Anne-Sophie Waag \\ Wikimedia Deutschland e. V., Stefan Skupien \\ Berlin University Alliance \ [email protected] With the collaboration of Lilli Iliev (Wikimedia Deutschland e. V.) and Charlotte Mysegades (Weizenbaum Institute) ABOUT THIS PAPER On 25 March 2025, the one-day conference "Yes, we are open!? Shaping artificial intelligence responsibly" took place in Berlin, organised by the Berlin University Alliance (BUA), the Weizenbaum Institute and Wikimedia. Over 100 participants from science, politics, administration, research infrastructures and research management discussed topics related to the tension between open science and AI in two panels and worked on them in a World Café. Subsequently, this discussion paper and a policy paper were developed by the organisers together with interested authors from among the participants of the event on the basis of photo protocols from the World Café. ABOUT THE WEIZENBAUM INSTITUTE The Weizenbaum Institute is a joint project funded by the Federal Ministry of Research, Technology and Space (BMFTR) and the State of Berlin. It conducts interdisciplinary basic research on the digital transformation of society and provides evidence-based and valueoriented options for action so that digitalisation can be shaped in a sustainable, selfdetermined and responsible manner. 51 Yes, we are open?! Designing artificial intelligence responsibly Weizenbaum Discussion Paper Yes, we are open?! Shaping artificial intelligence responsibly Conference results in the field of open science and AI \\ Summary The strategy paper "Yes, we are open! Designing artificial intelligence responsibly" is based on the results of the event of the same name held in March 2025 and shows how we can use the opportunities offered by AI courageously, openly and in the interests of the common good. Instead of viewing AI solely as "large language models," the paper emphasises the diversity of approaches – from rule-based methods to generative systems – and makes it clear that openness is the key to trust, transparency and participation. However, openness does not mean "everything for everyone", but rather a careful balance between accessibility, data protection and security. Only in this way can AI be democratically controlled, reproducible and oriented towards the common good. The vision developed is a digital culture in which AI technologies are developed and used in a transparent, sustainable and fair manner. Four areas of action have been identified to achieve this: knowledge spaces and networking for collaborative learning, competence building for critical and practical know-how, open digital infrastructures for sovereignty and sustainability, and innovative governance models that combine rules and freedom. With concrete measures – from exchange platforms and open science programmes to transparent benchmarks – the paper invites politics, science, civil society and business to work together to build a strong AI ecosystem. The goal is clear: AI should not lead to a concentration of power, but rather promote innovation, cooperation and participation. 51 Yes, we are open?! Designing artificial intelligence responsibly \\ Contents 1 Objective and background 5 2 Strategic fields of action 7 3 Measures 9 Imprint 13 51 Yes, we are open?! Designing artificial intelligence responsibly \ 5 1 Objective and reason This strategy paper was developed following the event "Yes, we are open! Shaping artificial intelligence responsibly" on 25 March 2025 (link), organised by the Berlin University Alliance (BUA), the Weizenbaum Institute and Wikimedia. The aim was and is to bring together representatives of the public sector from science, civil society, business, public administration and politics in order to better network existing expertise and infrastructures and jointly formulate strategies for responsible open research, development and application of artificial intelligence (AI) in the public interest. Even before the publication of large language models (LLMs), there has been a lively debate about the use, openness and orientation of AI and its regulation. Our discussions draw on many sources. This strategy paper summarises the discussions at the event in four key areas of action and proposes measures for the responsible design and implementation of open AI. Nomenclature: Artificial intelligence encompasses a broad spectrum of technical approaches: from ruleand symbol-based procedures to statistical methods and machine learning to generative systems. AI thus refers not only to large language models, but also to technologies in general that can represent knowledge, recognise patterns, support decisions and automate actions. In addition to data-driven AI, symbolic and rule-based methods have great potential: they can be developed in a transparent, verifiable and often resourceefficient manner and are suitable for critical fields of application such as administration, education and law. Their integration into hybrid approaches can help to strengthen the traceability and robustness of AI systems. Oficnhcit in AI means transparency and accessibility throughout the entire life cycle: open training data (as far as legally and ethically justifiable), open software and models, comprehensible documentation including resource consumption, interoperable interfaces, and transparent and verifiable infrastructures, standards and protocols. Only through openness in all phases – from data collection to model training to application – can democratic control and reproducibility be guaranteed. That is why the issue of governance is also central. The common good refers to the goal of aligning technological developments in such a way that they create collective benefits, promote social participation, uphold fundamental rights and ensure ecological sustainability. AI oriented towards the common good requires public funds and resources, should serve everyone and not increase the power of a few monopolies. The starting point: Artificial intelligence is increasingly changing not only our communication behaviour and the way we deal with knowledge in everyday life, but also the methods, processes and foundations of scientific research. 51 Yes, we are open?! Designing artificial intelligence responsibly \ 6 In many areas, comprehensive expertise and powerful open infrastructure for the development and application of open AI models already exist, but these have not yet been sufficiently visible or networked. In science, for example, research data infrastructures are available that curate high-quality data and make it available in the long term, or highperformance computing centres that enable complex modelling and data-intensive analyses. In addition, logic-based methods are also used, for example, to tap into previously unobvious knowledge connections from structured data with the help of rules. Beyond specialised knowledge and data spaces, there are also valuable resources and non-profit knowledge and data sources in more general social contexts (e.g. libraries or Wikipedia) that provide open, quality-assured and often participatory information. Reliable content alone does not make AI responsible – the entire development process is crucial. Open platforms (e.g. HuggingFace) show that transparent, verifiable open-source models can already form the basis for a collaboratively responsible AI ecosystem today. In order for its transformative potential to be effective in the public interest, clear political goals, better coordination between stakeholders and a common understanding of the opportunities, risks and limitations of these technologies are now needed. Vision and guiding principles: Our shared vision is a socially recognised digital culture in which AI technologies are designed to be transparent, comprehensible, sustainable and oriented towards the common good. The central guiding principles are based on open science: ӿ Openness, accessibility (meaningful access) and interoperability: Information, data and technologies are shared openly and made available in a sustainable manner so that they can be used meaningfully by as many people as possible over a long period of time. Particular attention is paid to data quality and management. ӿ Ethical responsibility and sustainability: AI systems must function without discrimination, be developed in a resource-efficient manner and be oriented towards the common good. ӿ Digital self-determination (independence and security): The ability to use and control digital technologies in a self-determined, secure, independent and data protection-compliant manner. As open as possible, as closed as necessary! The basis for this is a nuanced understanding of openness. Openness is not an end in itself and there is no one-size-fits-all solution: what is sensible and necessary in one specific context may be problematic or even counterproductive in another. While open licences, open interfaces and open data promote transparency and participation in many cases, other contexts – such as sensitive personal data in healthcare or critical security infrastructures such as power plants – require restrictions or controlled access. 51 Yes, we are open?! Designing artificial intelligence responsibly \ 7 Openness can both balance power relations and create new dependencies. Misunderstood or exploited openness – for example, when public data is transferred into proprietary AI systems without regulation – can even exacerbate existing power asymmetries. This is why a well-thought-out design is necessary: what kind of openness strengthens public knowledge commons, promotes cooperation and the common good? And what kind of openness opens the door to exploitation, monopolisation or discrimination? Responsible openness therefore means making context-specific decisions: What kind of access makes sense? Who is allowed to access which data or models, how, when and for what purpose? Is development open and transparent? And how can both scientific integrity and digital self-determination be preserved in the process? Openness should always go hand in hand with ethical reflection, legal diligence and clear goal orientation – in the spirit of a digital society oriented towards the common good. Key players for change: science (researchers and research institutions), libraries, archives and other memory institutions, civil society initiatives and organisations, educational institutions, public administration, infrastructure operators, funding and transfer organisations, but also companies in the digital economy. 2 Strategic fields of action The four areas of action identify the key levers for responsible AI design that is oriented towards the common good. They highlight specific starting points for cooperation, regulation and infrastructure – and serve as clear guidelines for making science, administration, justice, education, business and civil society actors fit for the future in the digital age . 2.1 Shared knowledge spaces and strategic networking Responsible AI design requires common terminology, coordinated standards and a shared understanding of key concepts such as AI, open science and digital commons. The success stories and value creation potential of these digital commons must be made more visible and recognised as a key resource. This knowledge base and the existing expertise from science and civil society should be systematically integrated into permanent, interdisciplinary networking structures and political decision-making processes. 51 Yes, we are open?! Designing artificial intelligence responsibly \ 8 2.2 Competence building and education In order to shape AI in all its diversity responsibly and effectively, both locally and internationally, targeted investments in education and training are necessary. This is not only a matter of technical know-how, but also of critical data and media literacy, ethical reflection and an interdisciplinary understanding of the associated processes. Learning processes should consciously incorporate past experiences – such as the reproducibility crisis in science, the appropriation and platformisation of the open access movement by profit-driven companies, or the concentration of power through social media. Historical analyses and the mapping of existing alternative infrastructures and practices can help to identify recurring problems and highlight sustainable solutions. 2.3 Sustainable development through open digital infrastructures The ecological, social and economic sustainability of AI technologies must be strengthened through binding standards, transparent impact assessments and the reusability of data and models. This requires stable, publicly accountable infrastructures based on open standards and codes that ensure interoperability across disciplines and institutions. Such open digital infrastructures (ODI) promote sustainability by enabling the shared use of resources, helping to reduce dependencies on proprietary solutions, and ensuring digital sovereignty and longterm accessibility and maintainability. This creates a solid foundation for technological independence and fair access to digital resources – complemented by targeted knowledge and skills exchange on the environmental sustainability of digital infrastructures. 2.4 Governance: combining responsibility and innovation Responsible AI design requires active further development of the legal framework with the aim of creating both legal certainty and scope for innovation. In doing so, it is important to openly discuss the potential and limitations of regulatory measures. In practice, there are uncertainties in various areas as to where the use of AI is actually sensible and legally compliant, while at the same time a multitude of applications are being promoted on the market under the banner of "AI" , with manufacturers generally portraying regulation as an obstacle to innovation. While companies often try to push through exceptions under the label of "AI", politicians and society must understand law and innovation as complementary forces. Particularly in the area of open science and open knowledge infrastructures, governance expertise already exists, for example in dealing with common goods, licensing models or participatory control. 51 Yes, we are open?! Designing artificial intelligence responsibly \ 9 governance mechanisms. These experiences should be systematically collated and transferred to the AI context in order to promote adaptive, context-specific and practical forms of governance that include clear responsibilities, transparency and the participation of a diverse range of stakeholders. 3 Measures Institutionalised cooperation between science, civil society, business and public administration is central to all measures. It enables needs to be identified at an early stage, knowledge from different areas to be pooled and innovations to be jointly designed in the interests of the common good. Shared knowledge spaces and strategic networking ӿ Exchange formats to highlight success stories relating to public welfare-oriented data infrastructures, digital commons and cross-sector knowledge building; Promotion of interdisciplinary cooperation and deeper understanding of concepts, standards and needs, e.g. in the development of networked knowledge systems (so-called knowledge graphs) or in the connection of generative and rule-based AI processes. ӿ Stronger links to standardisation organisations: Active involvement in the development of consistent open standards at national and international level, e.g. through audits of standards and standardisation processes. ӿ Promotion of real-world laboratories and dialogue formats between science, administration and civil society: Practical testing and joint reflection on open knowledge infrastructures in the application context. Such formats can help to make the practical potential of classic AI approaches – such as rule-based systems or linked data – visible and usable for public-interest-oriented digitalisation in the public sector . ӿ Mapping existing knowledge and data commons (e.g. open knowledge platforms, alternative infrastructure projects). Possible next step: ӿ Initiation of a community forum for open AI infrastructures by educational or research institutions, e.g. closer scientific cooperation with Public Interest AI Lab, start-ups, accompanying research, development of new evaluation metrics for the quality and openness of AI infrastructures; connection to existing exchange formats, such as AI-on-Demand of the European Commission.