Scientific writing with AI support at HSG: "The focus is on critical reflection"

The HSG Writing Lab is a point of contact that supports students and faculty with academic writing. In 2024, the HSG Writing Lab provided around 700 consultations, a fifth of which revolved around the use of artificial intelligence (AI). “We are seeing a steady increase in demand from students for workshops and advice on AI,” says Carina Gröner, who holds a doctorate in German studies and is the administrative director of the lab. That is why the lab is offering several workshops on the topic of AI and scientific writing.

In addition, Gröner, together with HSG Lecturer and AI researcher Reto Gubelmann, has launched a bachelor's course entitled (Scientific) writing with AI support – interdisciplinary approaches to large language models. “The focus here is on practical writing exercises, trying out various AI tools and critically examining them,” says Gröner.

Critically evaluating what the AI produces

“Most students use AI in their everyday lives, for example for summaries. However, the proportion of those who also use tools like ChatGPT when writing academic papers is likely to be lower,” says Gubelmann. At the University of Zurich, his research includes the logical and argumentative abilities of Large Language Models (LLMs). LLMs are language models that generate texts using probability calculations.

Examples of LLMs include the models behind OpenAI's ChatGPT, Google's Gemini, Anthropic's Claude and, more recently, DeepSeek from DeepSeek AI. In the first step of the course, students learn what LLMs are and how they work. “It is crucial to understand that LLMs do not interpret words as right or wrong, but apply them according to probability laws. They can't really 'think' creatively,” says Gubelmann.

According to Gröner, an important learning objective of the course is for students to be able to evaluate the quality of texts generated by AI. “To do that, they need to know the requirements for scientific texts, which we also teach them in the course,” says Gröner.

And, adds Gubelmann, students need to recognise that they should only use AI in areas where they can independently verify the results. “There are countless examples of cases in which AI produces hallucinations. Critically examining the accuracy and correctness of AI content is therefore a key skill for the degree course, (academic) programme; studying, studies; higher education and working life,” adds the philosopher and AI researcher.

In addition to ChatGPT, students in the course look at various other AI tools – including, for example, those that specialise in literature research. To this end, staff from the HSG library will give a guest lecture in the course.

HSG promotes reflective use of AI

The course fits into developments around AI at the university: for example, HSG wants to continuously strengthen the teaching of AI competence in its course offerings. In 2024, the University Management formulated a vision according to which university members are encouraged use the transformative power of AI. The vision emphasises the importance of a culture that embraces technical innovations in a future-oriented way and reflects responsibly on their use.

Among other things, HSG has sent notification to all faculty that they can use the Brian learning app. This app was developed by HSG graduates in collaboration with the HSG's Teaching Innovation Lab and uses AI to create course content. At the same time, faculty can take advantage of ongoing training in AI application through HSG offerings.

HSG requires that students declare the use of AI tools in the list of aids when using them in their academic work. If AI content is used in the argument, this content must be cited literally or in essence according to the guidelines of common citation standards, such as APA, and the AI tool must be cited as a source.

AI can be a sparring partner

According to Gubelmann, the role of AI in the production of scientific texts is currently that of a tool to support authors.  “AI can help with literature research, it can be a sparring partner in the search for ideas and also help with writer's block,” says Gubelmann. However, LLMs would struggle to develop an argument over several dozen pages. “Scientific texts should be original and provide a new report, article, piece or insight in a subject area. This is where AI reaches its limits.”

In the course, students try out various text formats to see what support AI provides and what AI-generated texts are like. As an exam assignment, students excerpt and review a scientific article and compare their own excerpt with the AI summary. “Over time, scientific writing triggers a thought process in which authors connect information and gain new insights. AI can support students and researchers in this process, but it cannot replace it,” says Gröner.