AI-based Technologies: HowTo-4-Teaching

On 25–26 September 2025, the AI Society at the University of Padua organized the Fall School “Generative Artificial Intelligence – Problem Solving in Physics”, an intensive two-day educational initiative designed for students enrolled in the Bachelor’s Degree in Physics.

The Fall School represented the Society’s flagship training activity for physics students, combining theoretical lectures, hands-on workshops, guided exercises, and interactive discussions. The program introduced participants to the foundations of generative artificial intelligence and explored its practical use as a tool for scientific reasoning and problem solving.

Throughout the two days, students worked on topics including:

• The principles behind modern generative AI models;
• Effective prompting techniques for scientific applications;
• The use of AI systems to support physics problem solving;
• Comparative analysis of different large language models;
• Critical evaluation of AI-generated solutions;
• Methodological and epistemological limitations of generative AI.

A distinctive feature of the Fall School was its strong emphasis on experiential learning. Rather than simply introducing AI technologies, participants were encouraged to experiment directly with multiple AI models, compare their behavior, identify strengths and weaknesses, and reflect on when AI can effectively support scientific reasoning—and when human expertise remains essential.

Prior to the event, an open communication campaign introduced the initiative to the student community, presenting it as a pilot project aimed at developing advanced AI competencies that could later be extended to a broader audience within the University.

Participation was based on an open application process. The School was offered free of charge, required no prior experience with artificial intelligence, and was designed to be fully inclusive. Learning activities were adapted to participants with different backgrounds, and no personal software licenses or specialized hardware were required. At the end of the program, participants received a certificate of attendance recognizing the competencies acquired during the School.

The Fall School demonstrated how generative AI can become a valuable educational companion for physics students—not as a replacement for scientific thinking, but as a tool to enhance problem-solving skills, stimulate critical analysis, and promote a deeper understanding of physical concepts through active exploration.

SERIES OF TWO SEMINARS
This seminar will start from introducing the Technology Acceptance Model (TAM) to evaluate how generative AI (GenAI) is perceived and used by students and teachers. Through a series of physics problems, we will test and discuss the ability of GenAI systems for physics teaching and learning. The seminar also reviews institutional guidelines (comparing Imperial College and Padova) for ethical AI use. It concludes with survey data showing that while students use AI for coding and brainstorming, they are highly uncomfortable with AI being used for automated marking or replacing personalized feedback from lecturers.

SERIES OF TWO SEMINARS
This seminar provides a comprehensive overview of how Machine Learning (ML) and Natural Language Processing (NLP) are used to analyze both quantitative and qualitative data in physics education. After briefly tracing the history of text analysis from traditional “bag-of-words” classifiers to the current era of Large Language Models (LLMs), the seminar will focus on recent developments and uses in Physics Education Research. Specifically, it will present the use of LLMs to automate the coding of student lab notebooks, comparing the performance of different models against human inter-rater reliability.

Imperial College London has invested in a variety of projects to explore the use of generative artificial intelligence (GAI) and machine learning (ML) in higher education The main project which I will discuss aims to test the effectiveness of Large Language Models (LLMs) in assisting large-scale qualitative research and consequently whether workflows incorporating LLMs can provide useful and timely feedback to instructors on the impact of their teaching. The focus of the work is on student use of scientific argumentation in lab reports. We have constructed a codebook for scientific argumentation to identify elements of argumentation, the relationships between those elements, and their veracity. Through building a training dataset of over 250 lab reports from students in their first- and second-year lab courses, we aim to analyse the whole set of 1899 lab reports from two cohorts of students by utilizing fine-tuning of open-source LLMs to automate the coding process (Fussell et al., 2025). The specific purpose here is to test how changes to teaching between the two cohorts affected students’ scientific argumentation skills, and, therefore, help us to evaluate the impact of the changes.

In addition to this, I will provide an overview of the projects that are currently being undertaken across Imperial related to AI and education. These institutionally funded projects include: (1) the use of GAI in providing feedback to students through the in-house problem sheet web-interface Lambda Feedback; (2) the use of LLMs to analyse and provide in-the-moment feedback to students on the structural elements of their lab reports; (3) using ML to identify at-risk students through analysis of interaction data from Lambda feedback; and (4) collecting data on students’ views about how instructors use GAI in their teaching.