Assessing the Potential of AI for Scientific Writing Techniques

Due to the wide availability and rapid development of AI-based tools such as Microsoft Copilot, ChatGPT, and Google Gemini, lecturers and students have many questions, uncertainties, and challenges: Are such tools allowed at ETH? If so, what can they be used for, and what are their limitations? How can they be used, and how can this be taught and learned? In the Innovedum project “Assessing the Potential of AI for Scientific Writing: Recommendations and Findings” (3829) run by the Department of Environmental Sciences, together with the Zurich-Basel Plant Science Center (PSC) and ETH Library, we actively supported our lecturers and students at the department (D-USYS) in using AI-based tools in their teaching and learning, especially in scientific writing. We offered practical examples, curated teaching and learning materials, and created guidelines. Subsequently, we developed a wide range of support material (overview of use cases and tools, exercises with AI-based tools for literature search, exercises to use these tools for drafting and revising and for evaluating the outputs, learning sequences) for a total of 18 courses (BSc, MSc, PhD; course sizes of 10-150 students) at D-USYS in HS23 and FS24. Our support also included presentations, consultations, co-creation of course materials with lecturers, student surveys, networking/exchange events, and a Moodle platform for lecturers, which serves as a knowledge hub.

One of the main contributions of our project was the support of selected lecturers in adapting their courses to the new AI landscape (total of 18 courses – BSc, MSc, PhD; course sizes of 10-150 students). As the emphasis of our project was on scientific writing, we selected courses where this was a main component of the course.

HS23 751-0013-00L World Food System, BSc, B. Studer
HS23 751-1010-00L Introduction to Sci-entific Methods Part II, BSc, R. Kölliker
HS23 701-0007-00L Tackling Environ-mental Problems I, BSc, M. Mader
HS23 701-1302-00L Term Paper 2, MSc, L. Winkel, M. Müller
HS23 701-1211-01L Master’s Seminar: Atmosphere and Climate 2, MSc, A. Merrifield
FS24 751-2312-00L Agricultural Policy, BSc, R. Huber
FS24 701-0909-00L Seminar Environmental Systems, BSc, A. Carminati, L. Pellissier
FS24 751-0201-00L Scientific Methods Part I, BSc, R. Kölliker
FS24 751-5118-00L Global Change Biology, MSc, N. Buchmann, O. Diaz Yenez, K. Kohonen, M. Costa
FS24 701-1303-00L Term Paper 1: Writing, MSc L. Winkel, M. Müller
FS24 701-5001-00L Ethics and Scientific Integrity for Doctoral Students, PhD, M. Paschke
FS24 Scientific Writing II, PhD, R. Mihálka

Recommendations for the institutional level
1) The coordination of services (e.g. workshops, seminars, information material, individual consultation services for lecturers) should be performed by a central unit of ETH Zurich.
2) nstitutional support for applying genAI to teaching and learning at ETH Zurich will become more necessary to keep pace with current developments.
3) The training of basic skills for both students and lecturers in the application of genAI (AI literacy) must be coordinated and implemented at the curriculum level.
4) Essential and effective structures for knowledge sharing and expansion are communities of practice of lecturers who share their teaching practices, but also institutional experts who collect and take up these examples, analyze them and prepare them for sharing within the community. The coordination of these communities of practice should also be centralized.
5) The same applies to communities of practice that develop from students at all levels (Bachelor’s, Master’s, doctoral) and that should be encouraged or stimulated to present experiments and experiences with workflows for science (e.g. image generation and composition for scientific illustrations, programming/customization of AI models, but also the inclusion of AI in the scientific writing process or in data extraction). As experience grows, we expect that new AI-based tools and workflows that involve scientists and AI working together will become more prevalent in the academic environment.
6) The use of genAI is actively integrated into the learning process and the lecturer, together with the students, deals with its influence on the skills to be learned and acquired through a close supervision of the process. For example, when writing scientific papers, students need enough time and practice to learn the skills of summarizing texts, writing content, and analyzing information or data despite using genAI.
7) The responsible use of genAI should be taught repeatedly and at different levels, especial-ly in classes on scientific integrity. Here, compliance with data protection regulations or the question of what constitutes a secure genAI environment should be addressed as a cen-tral topic, as compliance with scientific integrity will remain a fundamental responsibility of students and scientists, even or especially in the age of genAI.

Recommendations for teaching scientific writing
In scientific writing, most tasks can at least be aided by generative AI tools, but to varying de-grees. Figure 1 provides an overview of typical scientific writing tasks and their suitability for AI assistance, developed on the basis of our teaching and first-hand experiences. It is appar-ent that even seemingly closely related tasks such as drafting results and conclusions can be supported by AI-based tools, with varying degrees of success. Generally, tasks that require a higher level of precision and depth (e.g., methods, results, and discussion sections) benefit less from generative AI support. Similarly, tasks that require consistency and attention to detail (e.g., methods, broad revision, revising the manuscript in response to the reviewers’ feed-back) can only be accelerated at the expense of quality.