diff --git a/_posts/2026-08-25-ai-governance.md b/_posts/2026-08-25-ai-governance.md new file mode 100644 index 0000000..b8bc188 --- /dev/null +++ b/_posts/2026-08-25-ai-governance.md @@ -0,0 +1,255 @@ +--- +title: "AI for Research: Governance and Infrastructure — RSE Day 2026" +layout: post +author: "Carolin Odebrecht, Stephan Druskat, Johannes Schäffer" +menulang: en +--- + +- *DOI: [10.5281/zenodo.22076822](https://doi.org/10.5281/zenodo.22076822)* + +The disruptive impact of generative Artificial Intelligence (GenAI) +systems and methods concerns all research areas and disciplines, as well +as many roles within research, albeit in different ways. In general, +research conducted using AI tools introduces new responsibilities, +relies on new social and technical infrastructure, and requires new +skills and competencies. + + + +#### Responsibilities + +Research conducted using GenAI requires stakeholders to evaluate new — +and re-evaluate existing — requirements on research ethics and +integrity, legal requirements and IT security. Where research software +and data outputs become (critical) research infrastructure and related +services are being provided, researchers may become accountable as +service providers, in addition to their responsibilities as software +developers and data creators. Despite this, roles such as Research +Software Engineers are often overlooked in discussions around +responsibility profiles, as well as requirements for regulation and +(research) infrastructure, in the context of GenAI in research. This +means that the scholars who actually develop, adopt or fine-tune such +systems do not get to share their first-hand experience, provide +practical insights and influence the policies that will govern their +work in this area. The workshop "AI Research Governance. Responsibility, +regulation and guidelines in context of research developments", +conducted by Carolin Odebrecht at the first *Research Software Day +Berlin & Brandenburg* \[[7](#rseday)\], therefore +aimed to open discussions around governance of GenAI in research to +research software engineers, whose specific expertise makes them an +important part of the social infrastructure required for GenAI in +research. + +#### Social infrastructure + +Just like other computational work, the adoption of GenAI in and for +research needs to be cross-functional and inter-disciplinary, and +requires the close collaboration of individual research teams with many +stakeholders, such as other research groups, computing centers, +libraries and central RSE groups, all of which may be located within a +team's organizational unit or institution, or outside of it. These +collaborations need to be coordinated and often formalized. In contrast +to well- or better-established scenarios such as research data, research +software, and the collaboration on and sharing or provision of these +digital research outputs, there are still only few experiences and good +practices established for collaboration between stakeholders on +GenAI-related activities, policies, ethics and strategies. + +#### Technical infrastructure + +Similarly, the design, development and use of technical infrastructure, +such as hardware, AI-as-a-Service, and computational capacities for +pre-training, fine-tuning, instruction, inference, etc., requires new +models for collaboration and identification of responsibilities. These +need to be developed in academic organizations with the participation of +researchers. Some fundamental questions in this area include: *Which +technical infrastructure is needed for our research life cycles?* *Can +we build on existing internal or external infrastructure?* *How do we +design the FAIR and CAREful access and use of technical infrastructure?* +*How do we measure, and account for, the cost and ecological impact of +GenAI?* + +An AI governance therefore should cover the different social and +technical infrastructures. In general, AI governance is defined commonly +as a + +> \[...\] a system of rules, practices, processes, and technological +> tools that are employed to ensure an organization's use of AI +> technologies aligns with the organization's strategies, objectives, +> and values; fulfills legal requirements; and meets principles of +> ethical AI followed by the +> organization. \[[9, p. 604](#matti)\] + +Research organizations need to develop and implement a governance +framework that clarifies the responsibilities and accountability with +regard to the adoption and use of GenAI, and provides guidelines and +guardrails with respect to the technical and social infrastructure +required to support it. This framework must be operationalized through, +e.g., guidelines and policies targeting all relevant roles, job +descriptions and staff planning in research and administration, and IT +infrastructure strategies that safeguard independence from proprietary +solutions, avoid vendor lock-in, and support digital sovereignty in +research. Importantly, the development of governance frameworks should +be participatory, in that the process must include all relevant users, +producers, providers and other stakeholders, including RSEs. + +
+ +

Figure 1: Focal points for AI Governance in research contexts; purple: +research cycles and workflows; yellow: people; pink: organisational +infrastructure; green: organisation with goals and existing governance; +petrol: external regulation.

+
+ +AI governance systems should therefore: + +1. inform about relevant aspects in terms of and ethical, legal and + research integrity requirements, as well as IT security; + +2. establish references to stages and objects of research life cycles + (and/or teaching) across disciplines and roles; + +3. define the scope of accountability and responsibilities of + researchers, digital research technical professionals (dRTPs), other + staff, and the organization as such; + +4. define and implement corresponding social and technical (research) + infrastructures that supports researchers with regard to points + (1)–(3). + +# Raising awareness and fostering collaboration + +To raise awareness of the importance of AI governance for research, and +the lack of inclusion and participation of RSEs and other dRTPs in their +development and implementation, Carolin Odebrecht ran the workshop "AI +Research Governance. Responsibility, regulation and guidelines in +context of research developments" at the *Research Software Day Berlin & +Brandenburg 2026* on 3 June 2026. + +Participation from RSEs and researchers from biology, computer science, +physics, chemistry, digital humanities, earth sciences, economics, +sociology, life sciences, as well as dRTPs from museums and further +interdisciplinary backgrounds clearly showed the cross-disciplinary +relevance of the topic. For introduction, Carolin Odebrecht had prepared +an input on (AI) governance that served as the basis for our discussion +throughout the workshop. The session was enriched with interactive +elements using a feedback app (Particify \[[1](#particify)\]). + +As not everyone in the room had a definition of the term 'governance' to +hand, the introductory presentation began with a definition: Governance +is the framework by which an organization is controlled, and how +decisions are being made within the organization \[[2](#benz)\]. +Importantly for the context of the work presented in this blog post, +governance is also a tool to transfer external requirements — such as +laid out by law or funding requirements - into an organization's +internal context, making them actionable. + +Using Particify, we discussed our (the participants') backgrounds and +contexts — especially the roles we hold in our respective organizations +and whether we were aware of guidelines for AI use. As AI guidelines are +currently developed at many organisations, only half of the participants +stated that they are aware of an AI guideline. Surprisingly, a third was +not sure whether such guidelines exist at their organisation. This +points to a general discussion about responsibility profiles: whose +responsibility covers the dissemination and implementation of such +guidelines? In more depth, we collected our multiple roles via Particify +and discussed them ([Fig. 2](#fig2)). + +
+ +

Figure 2: Participants' roles in interdisciplinary research fields +show a wide range of responsibility areas and functions. The roles were +collected via Particify.

+
+ +Participants mentioned unclear responsibilities, i.e. the ambiguity +regarding who in an organization should take the lead in implementing AI +governance. Among others the following question came up: If everything +is already addressed in legal frameworks, what is the need for a +governance framework? Additionally, the lack of visible, immediate +benefits of implementing AI governance made it difficult to justify the +effort of defining a governance framework. Another point that was +mentioned was the fear of limiting oneself (or the organization), and +missing out on the opportunities that AI may provide. On the ethical +level, some argued that the moral integrity implied in guidelines for +good scientific practice alone - along with the assumption that +researchers inherently use AI responsibly - reduced the perceived +urgency for formal governance. Subsequently, we explored the potential +content for an AI governance document by identifying requirements for AI +use in our organizations, covering technical, personnel, and +administrative infrastructures ([Fig. 3](#fig3)). + +
+ +

Figure 3:Answers from participants identifying the topics relevant to +discuss in the context of AI guidelines.

+
+ +At the end of our workshop, we discussed multiple (partly hypothetical) +challenges and obstacles on the path to developing and adapting AI +governance standards. + +# Related Work + +Although AI for research is a broad topic, the specific focus on +governance and infrastructure in the context of roles and responsibility +areas is currently getting increasing attention, e.g., with the BUA +Workshop "AI Policies at Universities: Hands-On Workshop on Guidelines, +Implementation, and Open Science" on 2 June 2026 \[[3](#bua)\], the +de-RSE workshop on "AI-supported Research Software Engineering" in +September 2026 \[[4](#derse)\] and the Research Software Alliance's +workshop "Research Software Engineering in the Age of Generative AI: +Building a Community Vision" in Edinburgh (UK), March +2026 \[[10](#resa)\]. The evolution of the RSE role is also discussed in +the blog post "Research Software Engineers in the Age of GenAI: Same +Value, Changing Practice" \[[5](#blog)\]. +Discussing roles and responsibility areas is not new to RSE contexts, as +we also discussed this together with colleagues from King's College +already in June 2025 \[[8](#izd2m)\]. + +# Conclusion + +GenAI challenges many areas of responsibility and infrastructure - +especially for RSE and dRTP in their various roles dealing with genAI +directly in their research contexts. We argue that governance and +infrastructure are heavily interdependent. There is no living governance +without accessible infrastructure. Vice versa, establishing +infrastructure without governance risks the creation of responsibility +vacuums \[[6](#freeman)\]. Governance and infrastructure that are +developed with reference to each other, and specifically to serve the +inclusion of GenAI in research life cycles, enable research integrity, +sovereignty and therefore independent research. Governance and +infrastructure should be developed and adapt in a user-centred way to +address the requirements of administrative, social and technical +infrastructure mentioned above. However, governance and the development +of guidelines or policies is often centred around leadership roles which +in turn might risk a gap between researchers needs and leadership's +focal points. Interdisciplinary discussions with researchers fulfilling +their responsibility roles is needed to first raise awareness of topics, +second to exchange in depth knowledge and experience - literacies +exchange for both groups - and third to develop and adapt +collaboratively living guidelines that also take infrastructure into +account. + +# Acknowledgements + +The authors would like to thank Alexander Struck and Claudia Göbel for +organising the RSE Day 2026 in Berlin! This contribution was created in +the context of the [Interdisciplinary Centre for Digitality and Digital +Methods](https://izd2m.hu-berlin.de/) Campus Mitte, Humboldt-Universität +zu Berlin. SD's work was supported by the [Lower Saxony Digital Science +Support Space (DS³)](https://ds3-nds.de) project as part of +Hochschule.digital Niedersachsen, funded by zukunft.niedersachsen. + +# References + +- \[1\] Anon. [n. d.]. Particify. Particify GmbH. +- \[2\] Arthur Benz, Susanne Lütz, Uwe Schimank, and Georg Simonis (Eds.). 2007. Handbuch Governance. VS Verlag für Sozialwissenschaften. doi:[10.1007/978-3-531-90407-8](https://doi.org/10.1007/978-3-531-90407-8) +- \[3\] Berlin University Alliance. 2026. KI-Policies an Hochschulen: Praxisworkshop zu Leitlinien, Umsetzung und Open Science. +- \[4\] de-RSE – Gesellschaft für Forschungssoftware. 2026. De-RSE Collaboration + AI in RSE Workshop in Germany 2026. +- \[5\] Stephan Druskat, Michelle Barker, Ian Cosden, Cunliang Geng, Robert Haines, Daniel S. Katz, Joseph Shingleton, and Ben van Werkhoven. 2026. Research Software Engineers in the Age of GenAI: Same Value, Changing Practice. Technical Report. Zenodo. doi:[10.5281/zenodo.20320179](https://doi.org/10.5281/zenodo.20320179) +- \[6\] Jo Freeman. 1972. The Tyranny of Stuctureless. +- \[7\] Claudia Göbel and Alexander Struck. 2026. First Research Software Day Berlin & Brandenburg, 3 June 2026. +- \[8\] Henrik Schönemann. 2025. Retrospect: RSE Networking Event June 24th/25th. +- \[9\] Matti Mäntymäki, Matti Minkkinen, Teemu Birkstedt, and Mika Viljanen. 2022. Defining Organizational AI Governance. AI and Ethics 2, 4 (Nov. 2022), 603–609. doi:[10.1007/s43681-022-00143-x](https://doi.org/10.1007/s43681-022-00143-x) +- \[10\] Research Software Alliance. 2025. Research Software Engineering in the Age of Generative AI: Building a Community Vision. \ No newline at end of file diff --git a/assets/img/blog/2026/2026-08-25-ai-governance-fig1.png b/assets/img/blog/2026/2026-08-25-ai-governance-fig1.png new file mode 100644 index 0000000..f22c0fa Binary files /dev/null and b/assets/img/blog/2026/2026-08-25-ai-governance-fig1.png differ diff --git a/assets/img/blog/2026/2026-08-25-ai-governance-fig2.png b/assets/img/blog/2026/2026-08-25-ai-governance-fig2.png new file mode 100644 index 0000000..8eea2c7 Binary files /dev/null and b/assets/img/blog/2026/2026-08-25-ai-governance-fig2.png differ diff --git a/assets/img/blog/2026/2026-08-25-ai-governance-fig3.png b/assets/img/blog/2026/2026-08-25-ai-governance-fig3.png new file mode 100644 index 0000000..6be39b2 Binary files /dev/null and b/assets/img/blog/2026/2026-08-25-ai-governance-fig3.png differ