IPN-Kolloquium zu KI-gestützten Tutorensystemen: Wie verändert Generative AI das Lernen mit digitalen Systemen?

Am 28. Juli ist Prof. Dr. Zachary A. Pardos von der University of California, Berkeley zu Gast am IPN. In seinem Vortrag „AI-Enhanced Intelligent Tutoring Systems: Bridging Authoring, Feedback, and Student Pathways“ gibt der Bildungsforscher und Informatiker spannende Einblicke in aktuelle Entwicklungen rund um KI-gestützte Lernsysteme und ihre Potenziale für die Hochschullehre.

Prof. Pardos lehrt an der School of Information sowie am Graduate School of Education der UC Berkeley und forscht im Bereich Learning Analytics, Adaptive Learning und KI in Bildungssettings. Zentrales Thema seines Vortrags ist die Frage, wie generative KI (GenAI) eingesetzt werden kann, um zentrale Elemente intelligenter Tutorensysteme (ITS) wie Aufgabenstellung, Feedback oder Kompetenzzuordnung automatisiert zu generieren – und welche Chancen und Herausforderungen sich daraus für Forschung, Lehre und die Gestaltung individueller Lernpfade ergeben.

Internationale Zusammenarbeit mit dem IPN


Der Vortrag steht auch vor dem Hintergrund einer neuen, transatlantischen Zusammenarbeit: Die IPN-Doktorandin Marlene Steinbach war im Frühjahr zu Forschungszwecken bei Prof. Pardos in Berkeley zu Gast. Was als Austausch im Rahmen einer Konferenz begann, entwickelte sich dort zu einer gemeinsamen Studienidee mit ersten Daten – die Grundlage für weitere Kooperationsprojekte.

Das Kolloquium richtet sich an alle Interessierten innerhalb und außerhalb des IPN.

Das Wichtigste in Kürze:

  • Termin: 28. Juli 2025, 13:00–14:30 Uhr
  • Ort: IPN-Hörsaal, Olshausenstraße 62 oder per Zoom

Wenn Sie online teilnehmen möchten, wenden Sie sich per E-Mail an steinbach@leibniz-ipn.de, um den Zugangslink zu bekommen.

Mehr erfahren: Im Kurzinterview sprechen wir mit Prof. Pardos über seine aktuelle Forschung zu GenAI und Tutorensystemen, über die Chancen von Open-Source-Ansätzen im Bildungsbereich und über seine Zusammenarbeit mit dem IPN. (Interview auf Englisch)

IPN: You are working on intelligent tutoring systems supported by generative AI. What makes this approach particularly exciting for education today?

Prof. Pardos: It's a theoretically grounded use of GenAI that we aren't seeing so much of in chatbots deployments and other popular approaches in education. It's taking a new technology (i.e., GenAI) and seeing how it can solve an existing problem in education (i.e., the scalability of ITS). For many who have been around in education, that is more exciting than promises of new technology revolutions. 

IPN: Why did you choose to develop OATutor as an open-source platform, and how does that approach benefit research and collaboration in the education field?

Prof. Pardos: There are more engineers, learning scientists, and education psychologists interested in technology now than ever before. If everyone wanting to work on adaptive tutoring has to start from scratch, re-producing all the tutor components, then we are going to have a lot of wasted funding and productivity. Open-sourcing the system means getting to the novel extensions, re-designs, and experimentation faster. It also enables greater inclusivity in tutor research, allowing research to be conducted by PIs and teams that don't have the resources to develop a tutor from the ground up, but do have enough funds to do recruitment and hire a student. For educators, it means getting an effective mastery-based tool, refined by the empirical findings of the field, with no licensing fee that can inhibit adoption. Because it is open license and relatively simple to maintain (i.e., there's no backend database), adopters can run the system on their own - thus eliminating dependency on tech providers to continue use of the system.

IPN: You recently started to collaborate with Marlene Steinbach during her visit at Berkeley. How did that collaboration come about, and what role did your collaboration with Marlene Steinbach play in advancing your current research or inspiring new directions in your work?

Prof. Pardos: The collaboration with Marlene has been great. There's a popularly espoused concern about the harms GenAI can have on learners because of hallucinations. If GenAI is 80% accurate in a domain, is that enough for it to be used in tutoring a student? A new publication from our collaboration, recently presented at Learning @ Scale in Palermo, suggests that hallucinations from GenAI don't affect learners in obvious ways and that erroneous feedback can be more helpful than fully correct feedback, though more time consuming. Why this may be the case is the subject of future work.

IPN: Looking to the future: What current developments in AI and education do you find most promising or important?

Prof. Pardos: I'm most interested in conversations around theory building with AI and personalization and studying its role and effects on agency, motivation, and engagement. These are of course constructs from education psychology and I think that collaborations between those on the computational and psychological sides of education are most promising when it comes to advancing AI in education.  Colleagues Jennifer Meyer and Olga Viberg recently facilitated discussions around a subset of these topics that brought scholars from different disciplines together. More of this is needed.