Swiss universities: AI changes the assessment of theses

Context

In brief

• Zurich and ZHAW are focusing on oral exams, presentations and thesis defenses
• The thesis must also be assessed based on the process, not only the result
• SUPSI asks for a more detailed description of AI use

Key facts

• What → rethinking the assessment of written work
• Where → University of Zurich and ZHAW
• Who → Stefano Peduzzi, lecturer at SUPSI
• Objective → verifying the skills actually acquired
• New → providing more detailed information about interaction with AI
• Detectors → ZHAW and DFA/ASP have chosen not to use them

The University of Zurich and ZHAW are placing greater emphasis on oral exams, presentations and thesis defenses. The choice stems from a simple change to describe: creating summaries, organizing notes and writing a thesis now takes less time when students use artificial intelligence. If the final result can be produced with increasing assistance from technology, assessment cannot focus solely on the submitted document.

The objective is to verify more directly whether the skills have actually been acquired. AI is described as important: anyone looking at annunci di lavoro must also consider the process.

Why oral exams are not a complete answer

Stefano Peduzzi, a lecturer in the Department of Education and Learning / Teacher Education University of SUPSI and head of the Teaching Diploma for Upper Secondary Schools, recognizes the advantages of oral assessments, but limits their scope.

«The oral exam can test some essential skills, such as those related to presentation and language».

The thesis, however, requires more: searching for original sources, selecting and evaluating them, and distinguishing relevant information from irrelevant information. For this reason, according to Peduzzi, the oral exam does not cover all the skills required to write a thesis. The limitation is also organizational, because oral exams are difficult to manage in courses with hundreds of students.

The focus thus shifts from the product to the process. It becomes necessary to understand what the student has done independently, how they have used AI and whether, during the work, they have developed the necessary skills. The source does not provide a numerical threshold for measuring this relationship: 2% and 80% alone do not explain the share of work done by the student or AI if it is not clarified what they refer to.

The approach chosen by SUPSI

ZHAW has decided not to use software to recognize AI-generated texts, considering it insufficiently reliable. A similar choice was made at DFA/ASP of SUPSI, where the difficulty of determining precisely the extent to which the technology has been used is emphasized.

The answer is not prohibition. The department focuses on critical and informed use, with continuous discussion between the student and the supervisor. In the case of the thesis, the limited number of students supervised by each lecturer makes more frequent interaction possible than in a regular classroom.

Operational details

From finished text to verifiable work

The change described by Swiss universities modifies the meaning of the thesis. Previously, the submitted document could appear to be the main evidence of the work carried out; now the result must be read together with the decisions, sources and explanations that produced it. The text is not always enough to show how much the student reflected, selected and verified.

Two levels of the same assessment

The comparison does not lead to a single rule. An oral exam can assess presentation and language, but it does not cover the research of sources, their selection and the assessment of their relevance. A written exam may remain necessary, especially when the number of students makes it impractical to question everyone. The answer indicated by the source is therefore a balance between different forms of verification: presentations, discussions, oral exams and written work.

This balance also changes the relationship between teacher and student. If the supervisor has to understand how a paper was constructed, recognizable steps and dialogue throughout the thesis are needed, not merely an assessment at the time of submission. The practical consequence is greater attention to motivations: why a source was chosen, why information was considered relevant, which part of the text was developed independently and what assistance AI provided.

Why the detector does not solve the problem

The choice by ZHAW and DFA/ASP not to use detection software is linked to the difficulty of precisely establishing the degree of AI use and to the low reliability attributed to these tools. Monitoring therefore does not shift to an automated percentage.

For the university, the point is not merely to prove that a text was written without assistance. It is to understand whether the person still knows how to find, select, evaluate and apply information. For the student, this means that the use of AI must be explainable, not simply denied or hidden.

The link with the world of work extends this shift beyond the thesis. The source states that knowing how to use AI is becoming important, but the degree does not replace specialized knowledge. A credible educational path must therefore show both things: familiarity with the tool and the ability to reason without delegating everything to technology. Anyone considering the next step can complement this reflection with aziende che assumono.

Useful tools to protect your net income

To reduce FX leakage, compare CHF-EUR exchange options and banks for cross-border workers.

Key points

How to document the thesis process

For a student, the new approach translates into a more traceable workflow. The source already describes the direction chosen by the DFA/ASP: compared to last year, the interaction with AI during the drafting process must be indicated in more detail. From this, a concrete procedure can be derived, built upon the steps that the department asks to be made visible.

Five steps to make visible

1. Clarify before processing. The first step is to discuss with the supervisor in which phases AI can be used and to what extent. The definition must be transparent and shared. It is not enough to know that the tool is available: its use must be linked to the individual activities of the thesis.

2. Record interactions. During the drafting process, it is advisable to note how AI was used, without reducing everything to the final submission. The more detailed indication requested by the DFA/ASP serves precisely to distinguish what the person did from what was entrusted to technology. The recording should help understand what was actually learned and what, instead, was delegated.

3. Separate skills. The search for original sources, selection, evaluation, and the distinction between pertinent and non-pertinent information remain steps that one must know how to explain. The same applies to decisions made during the work. If AI accelerated an activity, the result must still be understood and applicable independently.

4. Prepare for the discussion. A presentation or a meeting with the supervisor requires illustrating the arguments, justifying choices, and critically analyzing the results. It is precisely these steps that allow for the evaluation of the process, not just the form of the submitted text. The oral exam may probe presentation and language, but the discussion must fit into the broader framework of the thesis.

5. Maintain ongoing dialogue. The relationship with the supervisor does not end with the final revision. The source indicates frequent interaction throughout the process, made possible by the limited number of students supervised by each lecturer. This dialogue allows for the correction of AI use as the work progresses and for the development of a critical and conscious application.

The logic also applies to those preparing for the professional world: knowing how to use AI matters, but it does not replace critical reflection. To transform this awareness into a concrete step toward employment, consult the salary calculator.

Source: rsi.ch