“AI can make open science more transparent and, at the same time, easier to learn”

Matthias Söllner on his experiences with open science

You conduct research into trust and the acceptance of digital innovations. Where, in the debate on Open Science, has there been too little discussion of trust so far?

MS: First of all, we should distinguish between trust and acceptance. Trust can be a prerequisite for acceptance, but it is not the same thing. The perceived benefits are crucial for the acceptance of Open Science. Researchers are more likely to adopt open practices if they recognise concrete added value in them. That is why we need to focus more on what benefits individual researchers derive from Open Science and what incentives the scientific system provides. The processing and long-term provision of research data involves considerable effort, yet this has so far been given scant consideration in appointments and career decisions. In these contexts, publications and third-party funding are what count most. As long as open data means extra work without yielding scientific rewards, the demand for it will remain asymmetrical.

However, trust also relates to publication structures. Author fees can encourage business models in which high publication numbers are economically more attractive than rigorous quality assurance. This raises questions about the trustworthiness of publishers, journals and peer-review processes.

I have, for example, looked at the journal *Sustainability*, published by MDPI. Given the high number of articles published, the question arises as to the priority given to scientific selection versus the goal of high publication numbers. Similar mechanisms also exist at established publishers. If an article is rejected by one journal, the author is sometimes offered the option of submitting it to a sister journal, often in return for an open-access fee.

This is not inherently problematic. Journals can have different profiles and quality standards. However, it becomes problematic when the impression arises that virtually every article within a publisher’s portfolio is intended to find a place for publication. In the short term, all parties benefit. Researchers secure a publication, the publisher receives a fee, and more articles are freely accessible. In the long term, however, academic integrity may suffer. The more methodologically weak papers are published, the more difficult it becomes to distinguish robust findings from less robust ones. For outsiders, the quality of a study is often difficult to assess. Simply being published in an academic journal lends credibility to a finding. If quality assurance does not function reliably, this mechanism can undermine trust in academic publications and, ultimately, in the academic system itself.

Funding practices can also create conflicting incentives. At the University of Kassel, I observed that fees for articles in prestigious hybrid journals were sometimes not covered, whilst pure open-access journals tended to receive funding, regardless of their quality or publication practices. However, the decisive factor should not be solely whether a journal offers open access. More important are the publisher’s credibility, the quality of the peer-review process and the business model. Otherwise, public funding may end up supporting structures that formally comply with Open Access but do not sufficiently guarantee scientific quality.

I understand your point. For some time now, there has been a discussion under the banner of ‘scholarship-led publishing’ about returning greater responsibility for journals to the academic community. At the ZBW, a dedicated team is working on this issue. This highlights just how difficult this transition is. Familiar, albeit unpopular, procedures often seem easier to publishers than seeking out alternative funding models. What role do such habits play in the acceptance of research-led publication models?

MS: Habits are a classic barrier to change. Added to this is the fact that many researchers are probably unaware of what alternatives exist and what is fundamentally possible. However, there are already academic societies that publish major journals. These include, for example, the Academy of Management and, in the field of business informatics, the Association for Information Systems. MIS Quarterly, one of the leading journals in the field, is also published within the academic community. However, even such journals sometimes operate using hybrid publication models. This, in turn, leads to inconsistencies in funding. In such cases, my university does not always cover the open-access fees, even though the journal is academically recognised.

If you look at the academic world from a socio-technical perspective, what do researchers need in order to change their working practices?

MS: First and foremost, individual incentives are crucial. When it comes to appointments and performance-related bonuses, publications and secured external funding are the main factors. Whether researchers contribute to Open Science, on the other hand, has so far played hardly any role. It is therefore hardly surprising that many tailor their work to the criteria that are decisive for their careers. At the same time, the concrete benefits of Open Science need to be better communicated . Researchers weigh up whether to invest time in preparing and publishing a dataset or in another project that is more likely to lead to a publication or new third-party funding. That is why institutional recognition for open science activities is needed.

This also includes rewarding contributions in a transparent manner. If a dataset is frequently reused and leads to further publications, the person who made it available should receive academic recognition for this. One possibility would be to introduce appropriate metrics that are taken into account in evaluations. Fundamentally, we should measure research output less in terms of quantity. Rankings and evaluations are often based on the number of publications, citations and third-party funding. This encourages a publication culture in which quantity counts for more than quality. Anyone whose paper is rejected by a prestigious journal is then often incentivised to publish it elsewhere as quickly as possible, rather than further developing the project from the ground up.

The focus on quantity is, of course, also due to the fact that the number of academic publications is growing rapidly. In the field of economics, too, there is an increasing number of papers from countries such as China and India. This increases the workload involved in peer-review processes. It therefore seems more manageable for reviewers to examine quantitative indicators. How is this development changing academic quality assurance?

MS: Generative AI is likely to further intensify this trend. At major computer science conferences, the number of submissions is already rising sharply. At a leading conference on business informatics, too, the number of papers has apparently almost doubled compared with the previous year. It is hardly plausible that the number of researchers has also doubled over the same period. It is more likely that generative AI is accelerating the production of academic texts. However, more submissions do not automatically mean more high-quality research. It may simply mean more average research.

You also deal with trust in data and platforms. Would you agree with the argument that transparency is becoming increasingly important in the age of generative AI? In other words, the traceability of how a study was conducted, what data and methods were used, and on what code the results are based?

MS: Trust is the foundation of our actions. It plays a role not only in times of transformation, but fundamentally at all times. With generative AI, however, the question of how results were arrived at is becoming increasingly important. This is already evident in students’ coursework and dissertations. In future, it will probably no longer be sufficient to assess only the submitted result. The process by which it was produced will be just as important. How did the person go about it? Why did they make certain decisions? What intermediate steps can they document? And can they explain their own work and contextualise it within the discipline? We are therefore considering supplementing written work more frequently with discussions or oral examinations. This is not merely a matter of determining whether a piece of work was produced independently. It is also crucial that students have understood what they have done and can justify their approach. Some journals are already adopting this approach and supplementing articles with interviews with the authors, in which they explain their reasoning, their methodology and the key findings.

What distinguishes trust in a public research infrastructure from trust in a commercial digital system? Do different standards apply in each case?

MS: Fundamentally, similar mechanisms of trust apply. One difference, however, lies in the organisation responsible and the associated interests. Behind a commercial system stands a company pursuing economic objectives. In the case of a publicly funded research infrastructure, one can assume that the common good, integrity and the interests of users play a greater role. This can influence the assessment of individual dimensions of trust. Goodwill and integrity are perhaps more readily attributed to a public operator than to a company whose business model is based on profit-making. In my view, however, there is no fundamental difference in how trust is established.

This is also evident in the PriDI project, which focuses on an open European web index as an alternative to the proprietary infrastructures from the USA, China and Russia. It is often very difficult to convey the added value of an open infrastructure to users. There are good reasons for such an open infrastructure, such as transparency, independence and public oversight. However, these advantages do not automatically lead to acceptance. Once people have become accustomed to an existing system and consider it to be working reliably, making a change is difficult. In this respect, the mechanisms differ little from those of commercial systems.

What lessons can be learnt from the PriDI project for research infrastructures?

MS: Firstly, the project shows that the perceived diversity of digital offerings can be misleading. At the application level, we have numerous services and apps at our disposal. However, the deeper one looks into the underlying technical layers, the more the market is concentrated among a small number of players. This is where the PriDI project comes in. We are not developing the Open Web Index ourselves, but are investigating an infrastructure that is being established as part of a European initiative. The starting point is that Europe has not yet established its own foundation for search services. The existing infrastructures are primarily controlled by providers from the US, China and Russia, which apply different standards when it comes to data protection and regulation, for example. And here we must recognise that speed is simply a key factor in the digital world. Network effects mean that early providers can rapidly establish a dominant market position. This favours monopoly- or oligopoly-like structures.

There are already numerous platforms on the market today where researchers make their work publicly available. Anyone wishing to create further services or infrastructure must therefore consider from the outset how these can actually be used. This is because established platforms often have a considerable head start. They are well-known, integrated into workflows and consolidated by network effects. Even if they do not fully align with one’s own scientific policy or societal ideals , people have simply become accustomed to them and incorporated them into their routines. A switch therefore usually only succeeds if the new infrastructure offers clearly recognisable added value.

Do state- or publicly-funded initiatives have a realistic chance of establishing alternatives that can compete with commercial services in terms of speed and user-friendliness, whilst also offering recognisable added value? Is that even possible given the structures of the public sector?

MS: Under the current circumstances, I am sceptical. Nevertheless, it is necessary to develop such alternatives. They do not necessarily have to be operated by a public-sector institution. One possibility, for example, would be to entrust the operation to a European company whilst embedding public requirements and oversight mechanisms. In my view, the greater challenge lies in user behaviour. One can hardly expect researchers to switch to an unknown alternative solely for reasons of data protection or the common good, unless it delivers usable results quickly. A new infrastructure must therefore not only uphold different values, but also be at least competitive in terms of quality, speed and usability.

Suppose Europe wants to achieve greater data sovereignty and reduce its dependence on the major technology conglomerates. How should it go about this? Should a European company be commissioned to build an infrastructure, with its roll-out subsequently being specifically promoted and publicised? Or is a different approach needed?

MS: First of all, there is the strategic question of whether Europe actually wishes to build up a provider that is intended to be competitive with the existing technology giants in the long term. We are trying here to catch up with a massive lead that has built up over decades. One alternative might be to skip this stage of development and invest early on in an emerging technology of the future. If, however, the decision is made to develop a European alternative, I consider open source to be a sensible approach in principle. Openness and transparency could be a distinguishing feature compared to proprietary offerings. To achieve this, however, resources would need to be pooled. If every federal state or every university operates its own open-source solution, too many fragmented structures will emerge. These can hardly compete with providers such as Microsoft in terms of scalability, further development and user-friendliness.

One conceivable solution would therefore be a sort of ‘digital Airbus’ – that is, a European company that develops and operates open digital services. At the same time, European countries would have to commit to using and funding these services over the long term. Such a company would certainly need reliable contracts and investment for five to ten years in order to build up technical capacity and further develop its services. Only such a long-term perspective would provide the necessary planning certainty. In this way, a European provider could emerge that is initially supported by public demand and can later compete with established companies on the international market. That would, at the very least, be one possible strategy for gradually reducing Europe’s lag.

As a business IT specialist, you also deal with issues relating to data literacy. What skills do researchers need when working with open data?

MS: I am a business IT specialist, but I originally studied business administration. My research focuses, amongst other things, on digital learning. I am currently also working with researchers in the fields of mathematics and biology education on a project on data literacy, which also touches on issues relating to open data. Data literacy has always been important for researchers working empirically. This includes preparing and analysing data appropriately. However, the use of open and existing data requires additional skills, both before and after the actual analysis. The first step is determining how to find data that is relevant to a research question. Researchers need to know where to look and how to assess, as efficiently as possible, whether a dataset is suitable for their project. Those who collect data themselves can tailor its structure and content to their own research question. With secondary data, on the other hand, it is first necessary to examine the conditions under which it was collected, which variables it contains and what its interpretative value is.

Similar requirements arise when making one’s own data available. Researchers must document and publish datasets in such a way that others can find, understand and assess their suitability. Metadata plays a central role in this. However, I am not sure whether all these skills still fall under the term ‘data literacy’. One could also speak of ‘open science literacy’, which builds on data literacy but additionally encompasses skills for locating, evaluating, documenting and sharing research data.

Let’s call this set of tools ‘open science literacy’ for now. Do researchers have to acquire this knowledge themselves, or is a clearer division of labour emerging, for example through research data managers and data stewards? Are new tools and specialised roles already taking the strain off researchers, or has the workload so far mainly been an additional burden?

MS: We are in the midst of a process of professionalisation. Specialised roles are gradually emerging, and the division of labour is also improving. However, this depends heavily on a university’s resources. Institutions with additional funding are better placed to create posts for research data management or data stewards than smaller or less well-funded universities. At the University of Kassel, for example, there is a team dedicated to data and data management. However, this team is comparatively small. Under current conditions, Open Science therefore still means additional work for many researchers. At the same time, the processes and tools are gradually improving. This allows some tasks to be carried out more efficiently, and the additional workload may decrease. Nevertheless, the relevant skills are becoming more important for researchers. In my view, this is increasingly becoming one of the fundamentals of scientific work.

Is Open Science already an integral part of the training of early-career researchers?

MS: At least, I am not aware of any compulsory course in which Open Science is systematically taught. Whilst our degree programmes do include introductions to ‘ ’ scientific work, the focus there has so far been on other fundamental aspects. In my view, however, open scientific practice should be taught as a cross-disciplinary skill. The difficulty lies in the fact that the volume of subject-specific content is increasing rather than decreasing. It is therefore hardly feasible to include an additional course in the degree programmes for every new requirement. It makes more sense to integrate such skills into existing modules. We have tested this approach, for example, with regard to argumentation skills. During an introductory course in business informatics, students wrote several papers and received targeted support in the process. Our studies showed that their argumentation skills improved over the course of the semester, whilst at the same time they deepened their understanding of the subject matter.

A similar model could be envisaged for Open Science. In a course on academic work, students could not only learn the basics but also go through a research process themselves, from research and data use to documentation and publication in accordance with Open Science principles. In this way, the relevant skills could be taught without necessarily having to set up a separate course. This requires teaching formats that separate the transmission of knowledge from its application more clearly. Fundamentals can increasingly be taught digitally and with flexible timings. After all, the good old lecture dates back to a time when books were a scarce commodity. We are no longer in that situation. The time spent together in class can then be used for practical exercises, discussions and reflection. Why is open scientific practice important? What obstacles are there? And how can they be overcome? Formats such as the flipped classroom offer suitable approaches for this.

In your Komp_HI project, you are investigating how AI can support the development of skills. Can this approach also be applied to open science literacy?

MS: Yes, in principle it can. AI is, above all, a means of enabling students to learn a particular skill in the most adaptive and personalised way possible. In terms of content, the approach involves teaching cross-disciplinary skills alongside subject-specific content. Particularly in large lectures, it is virtually impossible for lecturers to provide individual feedback. However, for learning success, it is important that students receive personalised feedback on their respective progress. An AI system can take on part of this task, provided it has been developed for the specific use case and trained accordingly. This could also be applied to Open Science.

What opportunities do personalised learning programmes offer for academic work and Open Science? One possibility, for example, would be an AI-supported learning system that first assesses a student’s level of knowledge and then suggests suitable exercises. Is this a realistic model for the future? And are there already initiatives in this area?

MS: In principle, yes. Whether you need a dedicated lab for this is another question. Ideally, every learner would have their own private tutor who knows their level of knowledge and provides individual support. However, this is difficult to scale with human tutors. In the long term, AI could take on the role of a smart digital tutor here. To do so, the system would first need to identify which skills are already in place and where gaps still exist. It could then offer suitable learning modules, exercises and feedback. Research is already being carried out into such adaptive learning pathways , for example on the question of which learning module makes the most sense for a person to tackle next. However, as far as I am aware, there are as yet very few specific research studies on the topics of academic writing and open science.

What experience have you had so far with AI-supported feedback in the Komp-HI project? How do students respond to it, and how well does it work?

MS: We have not yet comprehensively investigated this form of personalised feedback within the project. However, we have developed an AI system for legal writing. Legal case analyses follow a prescribed style of legal report. In addition to the subject-matter content, students must therefore also learn a specific structure for argumentation and text organisation. Our system supports them in recognising and applying this structure through personalised feedback. Put simply, the content may be legal nonsense, but the text is at least structured according to the legal opinion style – so, in the worst-case scenario, it is well-structured nonsense. Our colleagues from the Faculty of Law remain responsible for providing feedback on the subject-matter accuracy of the case solution. Together, this results in a mix of AI-based and human feedback, which we also refer to as hybrid intelligence. So, students first write a solution. The system then highlights which elements of the legal opinion style it has identified in the text. This allows learners to see, for example, whether the weighting of the individual sections is correct or whether certain sentences cannot be assigned to an expected function. This gives them clues as to where they should revise their text.

It is important that the system does not immediately dictate how a passage should be correctly formulated. Students receive an analysis, but must draw their own conclusions from it and improve their text independently. The feedback thus supports reflection without replacing the learning process. This approach already works well for structural feedback. In several studies, we have been able to show that students learn the relevant skills more effectively. Compared with traditional tutorial groups, in which model cases are discussed, the system variants we tested were consistently more effective. Assessing the subject-specific content – that is, the semantics and the legal quality of an argument – is more difficult. We are currently conducting further studies on this. I cannot yet cite any reliable results in this regard.

You are a member of the newly appointed VHB working group on ‘Open Science’. What motivated you to get involved, and what do you hope to contribute there?

MS: For me, Open Science is an issue that will shape business studies in the coming years. At the same time, we can see that the current transformation is fraught with problems, such as business models in which publishers charge high publication fees. In principle, the results of publicly funded research should also be publicly accessible. This is not merely a question of transparency. Many societal challenges can only be tackled within larger research consortia, where data, methods and results are shared and documented in a traceable manner. I would therefore like to gain a better understanding of how Open Science can be effectively structured, what obstacles exist and what framework conditions researchers require. Being part of the working group gives me the opportunity to provide expert guidance on this development and to help shape it.

Thank you very much!

The interview was conducted on 29 May 2026 by Dr Doreen Siegfried.
This text was translated on 24 August 2026 using DeeplPro.

About Prof. Dr Matthias Söllner:

Prof. Dr Matthias Söllner is Head of the Department of Business Informatics and Systems Development at the University of Kassel and Director of the Scientific Centre for Information Technology Design. His research focuses on the design of information systems that promote trust and acceptance, digital teaching and learning innovations, and hybrid intelligence. He previously held positions at the Universities of Kassel and St. Gallen. He has received several awards for his academic work, including the 2020 Early Career Award from the Association for Information Systems.

Contact: https://www.uni-kassel.de/forschung/iteg/startseite/personen/direktorium/prof-dr-matthias-soellner.html

LinkedIn: https://www.linkedin.com/in/msoellner/

ORCID: https://orcid.org/0000-0002-1347-8252

Google Scholar: https://scholar.google.com/citations?user=9gD2bNwAAAAJ&hl=de

to Open Science Magazine