Generative AI

Dates: 23., 24. & 30. September 2026
Registration Link
Location: ETH Center
Lecturers: Réka Mihálka, Jeanine Reutemann, Paulina Zybinska, Alexis Shakas, Robin Staab, Melanie Paschke and others
Target group: Priority will be given to PhD students enrolled in PhD Programs of Life Science Zurich Graduate School and DUW students at UNIBAS. 

Course Abstract 2026

Generative AI tools such as large language models (LLMs) are rapidly transforming how research is conducted, communicated, and evaluated. While many researchers are already familiar with tools like ChatGPT, Claude, Copilot, effective and responsible use in a scientific context requires more than basic prompting skills. This course is designed for PhD students in the natural sciences who want to move beyond casual use and develop a structured, critical, and practical approach to working with generative AI.
Over three days, participants will explore how GenAI can support key research workflows, including scientific writing, literature synthesis, data analysis, coding, and visual communication. The course combines conceptual input with hands-on workshops, enabling participants to experiment with tools and apply them directly to their own academic workflows.
A central focus is placed on understanding the limitations of these technologies. Participants will learn to identify common failure modes such as hallucinations, bias, and overconfidence, and develop strategies to critically assess and validate AI-generated outputs. Rather than treating prompting as a set of isolated techniques, the course emphasises iterative interaction, problem-solving, and the development of reliable workflows.
The program also addresses broader implications of GenAI in academia, including authorship, transparency, privacy, and ethical use. Through applied exercises and real-world scenarios, participants will reflect on when and how AI can be used responsibly in research.
Between sessions, participants are encouraged to apply GenAI tools to their own work, documenting both successful applications and limitations. The course concludes with a reflection phase, where participants share their experiences and derive practical insights for integrating GenAI into their research practice.
By the end of the course, participants will have developed a clearer understanding of how to collaborate effectively with generative AI, recognising both its potential and its boundaries, and will be equipped with practical approaches that can be directly applied in their PhD projects.

Detailed program will follow during summer

Generative AI is rapidly becoming part of everyday academic work. But how are PhD researchers actually using it, and where do they still need support?

To help shape the upcoming course Explore the Responsible Use of Generative AI in Academic Work, the Zurich-Basel Plant Science Center (PSC) surveyed PhD researchers enrolled in the Life Science Zurich Graduate School programs, including the PhD Program in Plant Sciences. In total, 96 PhD researchers responded to the survey, representing 16 PhD programs. The largest groups came from Neuroscience (27%), Plant Sciences (15%) and Evolutionary Biology (14%), with respondents from a broad range of other life science programs.

The results suggest that generative AI is no longer an occasional experiment. Almost 69% of respondents use LLM-based generative AI at least daily, including 44% who use it several times per day. A further 20% use it several times per week.

AI is already part of the research workflow

PhD researchers use AI across a broad range of academic activities. Common applications include brainstorming and understanding new concepts, literature searches and summarising scientific papers, scientific writing and editing, as well as coding and debugging. Some respondents are also experimenting with AI for data visualization, project planning and experimental design.

Coding stands out in particular: 71% identified coding as one of the three AI applications currently providing the greatest value in their work, followed by debugging code (43%) and language and grammar editing (35%).

Researchers are also investing in these tools themselves. Around half of respondents reported using personal paid subscriptions. ChatGPT and Claude were among the most widely used tools, while university-provided services were used considerably less frequently. Notably, 36% of respondents did not know that their university might provide access to AI tools or related services.

Yet the survey also reveals a group of researchers moving well beyond conventional chatbot use. Around 29% have already used AI agents, 14% have built a custom GPT or AI assistant, 16% have worked with AI APIs, and 8% have fine-tuned an LLM. At the same time, 39% had done none of these things. This contrast points to an emerging training gap: while some PhD researchers are already developing sophisticated AI-assisted workflows, these approaches remain unfamiliar to many of their peers.

AI adoption outpaces awareness of guidelines

Supervisors appear largely open to this development: 57% of respondents described their supervisor as generally supportive of AI use. At the same time, 87% reported that their research group has no defined rules for using AI. While universities have introduced guidelines and recommendations for the use of generative AI, awareness of them remains uneven: only 39% of respondents said they were familiar with their institution's guidelines, while 28% knew they existed but had not read them, and 33% were either unsure or unaware of them.

Importantly, guidance from the wider research community is not simply about restricting AI use, but about establishing conditions for its responsible use. The European Commission's guidelines encourage the responsible adoption of generative AI in research, emphasizing transparency, accountability and responsibility^[1]. EU guidance also permits the use of generative AI in preparing research grant proposals, provided researchers remain responsible for the content, verify AI-generated information and disclose its use. Similarly, publishers such as Nature Portfolio^[2] and Elsevier^[3] allow AI tools to support aspects of manuscript preparation, while emphasizing transparency, verification, human oversight and accountability.

This makes awareness and practical understanding of the rules increasingly important. Data privacy was the most frequently reported limitation to effective AI use (51%), followed by verification of AI outputs (40%) and uncertainty about institutional policies (37%). Respondents also expressed concerns about hallucinations, copyright, scientific integrity, bias, overreliance on AI, reproducibility and the potential loss of critical thinking.

At the same time, the survey suggests that PhD researchers approach AI-generated information with considerable caution. Almost 89% said they always or usually independently verify AI-generated information before incorporating it into their academic or scientific work. Yet confidence in recognising incorrect AI-generated information was more mixed. Learning how to document AI use, protect sensitive information, verify outputs and remain accountable for the resulting work is therefore becoming an important part of research practice.

From survey insights to hands-on training

The survey findings have helped shape the upcoming course Explore the Responsible Use of Generative AI in Academic Work, taking place on 23, 24 and 30 September 2026. The course combines areas where PhD researchers already see considerable value in generative AI with those where our survey identified a need for further training.

Participants will explore how GenAI can support scientific writing and literature research, coding and problem-solving, scientific communication and data visualisation. The course will also look beyond conventional chatbot use, introducing AI agents and research workflows that combine different tools and approaches. Responsible use is integrated throughout the course. Questions around data privacy, hallucinations and verification, copyright, scientific integrity, bias, reproducibility and overreliance on AI will be discussed alongside practical applications, helping participants consider not only what AI can do, but when and under which conditions its use is appropriate in research.

Importantly, the learning will continue between the course days. Participants will be encouraged to test GenAI approaches in their own research, documenting both successful applications and cases where the tools fall short. In the final session, they will bring these experiences back to the group to compare approaches, discuss limitations and derive practical lessons for their future work.

The aim is not to promote AI use for its own sake, but to equip PhD researchers with the knowledge, practical experience and critical perspective needed to decide when GenAI adds value to their work and how to use it responsibly while maintaining scientific integrity.

References

1. European Commission ↩
2. Nature Portfolio ↩
3. Elsevier ↩

I initially signed up for the Generative AI course mostly because I found the topic interesting. Since then, I have completed my PhD at ETH and started working as a Senior Tech Consultant, where my daily work revolves heavily around building and deploying LLM and RAG systems. To my surprise, many of the topics discussed in the course, such as hallucinations and sycophancy, have become practical challenges that we deal with every day. It has been fascinating to see how directly these concepts translate into real-world applications. Looking back, the course material has held up remarkably well in practice.

— Former ETH PhD student, now Senior Tech Consultant