An interview with Lukas Jürgensmeier by Veronica Burbulea, University of Groningen
Educators have long faced the challenge of providing timely, high-quality feedback, especially in large-scale or resource-constrained environments. This difficulty becomes more pronounced in courses with technical or quantitative components, where detailed feedback is critical for student progress.
This challenge inspired Lukas Jürgensmeier (Goethe University Frankfurt) and his co-author Bernd Skiera (Goethe University Frankfurt) to delve into an educational setting where GenAI meets practice, as presented in their recent IJRM paper “Generative AI for scalable feedback to multimodal exercises”. The paper highlights the transformative potential of integrating GenAI into an app designed to offer feedback at scale.
I had the chance to speak with Lukas to learn more about their innovative study, the development of the app, and its implications for education’s future.
The feedback challenge in education
Feedback is central to effective learning. It helps students correct errors, refine their skills, and deepen their understanding of course material. However, providing personalized feedback becomes increasingly challenging in large classes, where the sheer number of students overwhelms the instructor's capacity to respond to individual needs.
This was precisely the issue faced in Bernd’s and Lukas’s marketing analytics course taught at Goethe University Frankfurt. With over 100 students learning to solve marketing problems by analyzing data in R (a common statistical and programming facility), feedback was critical for helping them navigate technical errors and complex assignments. But traditional methods couldn’t keep up with the demand.
This practical problem set the stage for the development of an app powered by GenAI. "We asked ourselves, how can we use this technology to provide feedback that is not only fast but also meaningful?" Lukas recalls. Their solution combined the adaptability of GenAI with a deep understanding of the specific challenges faced by educators and students.
Developing the feedback app
The app’s development began with a collaboration with Zavi AI, a startup that included a former student from the authors’ marketing analytics course. This insider perspective proved invaluable in tailoring the app to the needs of both educators and learners.
Using OpenAI’s tools, the team rapidly prototyped an app that could analyze student submissions, generate qualitative and quantitative feedback, and scale effortlessly to large cohorts. Within two months, the app was ready for testing in a real classroom setting.
The app’s functionality is intuitive yet powerful. Educators upload exercises and their solutions, while learners submit their work. The app then provides scores alongside detailed feedback, highlighting areas of success and opportunities for improvement. Importantly, it accommodates multimodal assignments, which require coding, statistics, and written economic reasoning, making it versatile across a range of subjects.
Privacy was another key consideration during development. The app does not relay personal data to the GenAI models’ APIs, ensuring that no personal or sensitive data is exposed during interactions with the AI models. This design aligns with best practices for ethical AI deployment in education.
The role of GenAI
At the heart of the app is GenAI, which powers its ability to analyze diverse inputs and generate feedback that is personalized to students’ submissions. By comparing the performance of seven different AI models, the researchers ensured that the app’s recommendations were both accurate and cost-effective.
Scalability was a key design principle. Whether serving a small class or hundreds of students, the app can seamlessly provide instant quantitative (i.e., points awarded for the exam) and qualitative feedback (i.e., textual or verbal feedback an educator would provide). The quantitative feedback was compared to points awarded by humans. The qualitative feedback was assessed in 3 dimensions- correctness, sufficiency and appropriateness. While the app’s performance is slightly more variable in cases of more complex exercises, the performance of the app using GPT-4 with correct solutions, remains highly accurate for both types of feedback.
Use of the app and things to know
The good news is that everyone can use the app. Students have responded positively to it, particularly appreciating the app’s immediacy. Unlike traditional feedback methods, which often involve delays, the app allows students to address mistakes almost instantly, accelerating the learning process. Nonetheless, Lukas highly recommends that educators communicate to their students that, while the feedback is quick and helpful in most cases, the app is not 100% accurate.
"The app is an excellent tool, but, like human graders, it is not error-free.”
- Lukas Jürgensmeier
While the feedback provided by the app is highly accurate for computational and technical tasks—such as programming exercises, data analysis, or numerical problem-solving—it may be less suited for open-ended or judgment-based assignments. For instance, the app could excel in courses like marketing analytics, computer science, or business management, where the exercises have clear solutions and defined criteria. In contrast, courses involving nuanced case analyses or strategy discussions may require human judgment to complement AI-generated feedback.
This is why keeping a human in the loop is crucial to ensure fairness and contextual sensitivity, especially for high-stakes evaluations, such as grading exams. The key lies in integrating AI thoughtfully, ensuring that it complements rather than replaces traditional teaching methods.
A model for the future of education
The study highlights broader implications for education. By automating repetitive feedback tasks, tools like this app free up instructors to focus on higher-order teaching activities, such as fostering critical thinking and guiding complex discussions. In this way, GenAI doesn’t replace educators but enhances their ability to support students.
While the app was designed for a marketing analytics course, its potential applications extend far beyond this domain. By design, the app is subject-, exercise type-, and language-agnostic and ready to use for any other exercise outside of marketing analytics.
Lukas also encourages educators to rethink how they design assignments. As AI becomes more prevalent, formats like essays may no longer be the best way to assess learning outcomes.
Reflecting on the project, Lukas emphasizes the value of collaboration, adaptability, and an openness to experimenting with new ideas. By combining cutting-edge technology with practical classroom insights, their work demonstrates how educators can harness GenAI to solve some of their most persistent challenges.
Read the paper
Interested in reading all the details about using GenAI to generate scalable feedback? Read the full paper here.
Want to cite the paper?
Jürgensmeier, L., & Skiera, B. (2024). Generative AI for scalable feedback to multimodal exercises. International Journal of Research in Marketing, 41(3), 468-488.
Meet Lukas
Recent Ph.D. graduate from Goethe University Frankfurt’s Marketing Department. Since September 2024, Graduate Programme Participant at the European Central Bank (Frankfurt, Germany)*.
What is the number one question you hope to answer in your career?
“While I do not necessarily have a number one question to answer, my aim is to always stay curious and open to new ideas. So far, this attitude has ensured that I have fun in the academic process—especially when trying to answer difficult questions with no obvious answer.”
What is the best advice you have ever received, and how has it influenced your career or life?
“One piece of advice I remember is from my former supervisor and co-author, Bernd Skiera. His advice is to ‘give many ideas a chance, and some will ultimately fly.’ This advice encapsulates that an individual idea—whether in research or in life—might have a low chance of success. If you put all your bets on one idea, sometimes you are lucky and you succeed. But it is also very likely that many ideas turn out to be subpar. That means you should not depend on a single idea or single project but create many opportunities to succeed. This advice has encouraged me to give our Generative AI feedback research idea a chance, which eventually turned into this IJRM publication”.
*Disclaimer: Lukas currently works at the European Central Bank. The article discussed was published before taking up this new role. The views expressed in his research and in this article are his personal ones and do not necessarily reflect those of his employer.
This article was written by
Veronica Burbulea
Ph.D. candidate at the University of Groningen, the Netherlands
Comments