»We want to facilitate the use of AI«

Generative AI tools are conquering many areas of society at present. Yet, while the use of machine learning techniques in extensive artificial neural networks promises to bring about scientific advances, it also presents challenges for university researchers and teaching staff. In this interview, Prof. Dr Christoph Steinbeck outlines the position the University of Jena has adopted on this issue and explains how its researchers are contributing to the development of AI approaches through the German National Research Data Infrastructure (Nationale Forschungsdateninfrastruktur – NFDI).

Interview by Ute Schönfelder

The University of Jena has been pursuing a digitalization strategy for some time now. What role does AI play in this?

A comprehensive digital transformation is underway at our University. We are approaching this process through three sub-strategies focusing on digitalization in studies and teaching, in research and library, and in administration and infrastructure.

In all three areas, this includes engaging with AI applications. This ranges from developments in research data management and infrastructure to the use of AI tools in teaching and studying to AI applications in administrative settings, where they can help our staff to respond to queries and provide information swiftly and efficiently.

How is AI changing the work of scientists?

AI is undoubtedly one of the »disruptive« technologies changing our society—not just the world of science. Of course, the term »AI« is itself in a state of flux. AI has been a topic in the context of research for several decades and is now in widespread use, such as through machine learning applications. Even in the 1990s, researchers were able to use algorithms to train small neural networks to solve very specific problems. For some years now, we have been able to use extremely large artificial neural networks—including large language models (LINK) like ChatGPT, which so many people are excited about.

That said, AI also poses challenges for us in research, as machine learning tools are only as good as the data used to train them. Training data of sufficient quantity is not yet available for all areas of research. However, realizing the potential of big data analysis involves training algorithms with huge quantities of data. If this isn’t possible, the trained tools often deliver results that only appear reasonable. We primarily see major scientific breakthroughs with AI in fields where large volumes of data are available.

The German National Research Data Infrastructure (NFDI) aims to build precisely these data pools. To what extent are researchers from the University of Jena involved in this project?

The NFDI is funded and structured by the German Research Foundation (Deutsche Forschungsgemeinschaft – DFG). Its purpose is to systematically connect and open up all databases from the fields of academia and research for the entire German science system. It brings together and networks data that has only been available to date on a decentralized, project-specific or temporary basis, thereby making it accessible on an ongoing basis and for all manner of research topics.

There are now 26 different NFDI consortia nationwide, each focusing on a different field of research. Researchers from the University of Jena are involved in a number of these consortia, including in the fields of biodiversity, geosciences, microbiology and history, along with my field, chemistry. Together with my colleague Oliver Koepler from the Leibniz Information Centre for Science and Technology in Hanover, I am leading the NFDI4Chem consortium.

Do you use AI in your own research?

Yes. Even in the 1990s, my team and I used machine learning to predict nuclear resonance spectra from chemical structures. These methods make it possible to determine the structures of previously unknown substances. However, we only had a limited amount of data available to train the algorithms, which remains the case to this day. Extensive pools of data simply do not exist in this area. If nothing else, that’s what we’re hoping the NFDI will achieve.

We are currently really enjoying our work in which we are using deep learning to automatically translate chemical structural formulae from specialist publications into machine-readable code. In this way, we’re uncovering »old knowledge« in past publications and making it available to the world of science in open databases (From game boards to chemical AI tools - LINK).

Nowadays, AI has a role to play not only in research but also in teaching. In 2023, the »AI in Teaching« task force was established at our University. Who are its members?

It was an initiative led by the Academy for Teaching Development (Akademie für Lehrentwicklung – ALe) when it became clear that we would need to formulate recommendations for handling generative AI tools, which are increasingly used by teaching staff and by students. The task force comprises members of the ALe as well as representatives of faculties, students, the Vice-President for Learning and Teaching, the Service Centre for Higher Education Didactics (LehreLernen), the Michael Stifel Center, the Multimedia Centre, the Legal Office and the Student Affairs Division. As Vice-President for Digitalization, I chair the task force.

What topics doeas the task force deal with?

Very generally, we consider how AI tools such as ChatGPT and Dall-E can be used in teaching and learning, and how we can anchor knowledge of these applications in our curricula. In addition, we primarily focus on very specific issues. Our first topic, for example, was modifying the declaration of independent work our students routinely submit together with their theses. In this declaration, students confirm that they have completed their thesis independently and only drawn on permitted tools and resources. Given the emergence of various AI tools, we had to reflect on how to proceed.

What general regulation does the University's management consider necessary in relation to AI applications in teaching?

As the University’s management, we resolved to regulate the use of AI as little as possible and adopt a positive, enabling position. Rather than creating restrictive policies, our aim is to support lecturers and students in the application of these tools.

This also means that teaching staff can decide for themselves the extent to which they wish to allow their students to use AI tools in writing their theses. In the »AI in Teaching« task force, we have developed an interactive form that enables lecturers to create a suitable framework of rules and adapt the student declaration accordingly. The consequence is, of course, that we continue to punish cheating and attempts to cheat.