Doctoral Student Experimenting Lab (Block Course): Explore the responsible use of AI in generating scientific texts, images, audio and code (1 ECTS)

The project Doctoral Student Experimenting Lab: Explore the Responsible Use of AI in Generating Scientific Texts, Images, Audio and Code was developed to address a critical need: empowering early-career researchers with both the technical skills and ethical grounding to use generative AI based on large-language-models (LLMs) – GenAI responsibly in academic contexts.

Our motivation was threefold. First, PhD students lacked safe, guided environments to experiment with GenAI in research-relevant settings. While many were already using such tools informally, they had few structured opportunities to test them, raise questions, and receive feedback. Second, we observed growing risks around uncritical adoption. As GenAI become more accessible, their academic use often outpaces understanding of their limitations and implications. Without guidance, researchers may unintentionally compromise academic integrity, mishandle sensitive data, or misattribute authorship. Third, no structured, interdisciplinary training existed to cover the broad scope of GenAI across research workflows. Formal education is lagging behind technological change. We saw this as a chance to pioneer a model course that integrates hands-on exploration with ethical foresight.

Unlike existing workshops at ETH that focus on individual tools or tasks, our 4-day (1 ECTS) block course offered a holistic experience. Delivered twice, it combined technical training with embedded ethical reflection, structured around four pillars: (1) LLM customization and prompting; (2) AI-supported workflows (e.g., coding, literature research, image and data analysis); (3) generation of visuals, audio, and video; and (4) ethical data handling and privacy protection.

What set our course apart was its design: students didn’t just learn tools—they actively experimented, applied them to their research, and discussed feasibility and risks across disciplines. This problem-based, collaborative format enabled students to become critical, informed users of GenAI.

Ethics was not a separate unit but embedded throughout. Through real-world case studies, exercises, and debate, students explored dilemmas involving authorship, bias, copyright, and scientific integrity. We aimed not just to teach how to use GenAI, but when and why—preparing researchers to make responsible, context-aware decisions and uphold the role of the human-in-the-loop.

As experts in research integrity and AI, we view this course as a scalable model. Its resources—tutorials, guidelines, and Moodle platform—are openly shared to benefit the ETH community and beyond. Ultimately, our goal is to shape a generation of researchers equipped not just to use GenAI, but to critically shape its role in science.

The “Doctoral Student Experimenting Lab” training was implemented as a 4-day block course (1 ECTS), delivered twice (in November 2024 and May 2025). Each iteration admitted an interdisciplinary group of doctoral students—81 in total—representing 10 ETH departments. Courses took place across flexible teaching spaces (ETH main building, Student Project House, Media & Methods Lab), creating an environment conducive to structured input, experimentation, and transdisciplinary exchange.
The course followed a problem-based, student-centered model, combining short theoretical inputs, expert-led demonstrations, hands-on collaborative work, and moderated group discussions. Students applied new knowledge directly to their own research problems, data, or communication goals. Instruction was delivered by 17 experts from fields including machine learning, scientific writing, AI ethics, privacy, and research integrity.
A key feature was the involvement of computer scientists as co-instructors, who offered foundational insights into how LLMs work from a coding and systems perspective. Their tailored coding exercises and live demonstrations enabled students to probe how AI generates outputs, what causes hallucinations or biased results, and how technical design choices affect outcomes. These sessions deepened technical understanding and supported more informed, responsible tool use.
The course emphasized workflows over tool-specificity. While students explored tools such as ChatGPT, NotebookLM, Copilot, Claude, FloraAI, LeonardoAI, and DALL·E, the focus remained on transferable skills: how to construct structured prompts, align tools with research goals, and critically assess results. This ensured that students’ knowledge remains relevant despite rapid technological change.
Ethical reflection was embedded across the curriculum—not isolated in a standalone unit. Every technical module was grounded in case-based moral reasoning and scientific integrity and institutional frameworks, covering bias, misinformation, privacy, authorship, and copyright.
Based on planned student evaluations, the second iteration further enhanced interactive, hands-on learning. While the original format already emphasized experimentation and group work, we streamlined theoretical content into live exercises, panel discussions, Q&A sessions, and collaborative problem-solving. Each day concluded with dedicated time for debriefing, reflection, and student feedback—ensuring the course remained adaptive and learner-driven.
Assessment was formative. Students presented projects, discussed applied use cases, and reflected on ethical dilemmas. All materials—including tutorials, coding exercises, and case sets—were shared via Moodle.

The project confirmed that doctoral students are not only eager to engage with GenAI, but also represent a critical cohort for driving institutional adoption of this rapidly evolving technology.
The blended teaching format proved successful in fostering an active learning environment where students could explore GenAI, reflect collaboratively, engage across disciplines, and receive feedback. Crucially, the course embedded ethical reflection directly into technical and practical training. Students developed a deep understanding of how LLMs function—through coding exercises and system-level walkthroughs—which was in fact essential for recognizing risks (hallucinations, bias, and privacy violations). By the end of the course, students had acquired not only advanced digital competencies but also the ability to make informed, value-based decisions when using GenAI in their own research practices.
The diversity of the student cohort was a clear asset. Participants from 10 ETH departments brought complementary skills and perspectives. Computationally advanced students supported peers in coding, while others contributed ethical insights. Many shared field-specific use cases that proved surprisingly transferable across disciplines, enriching the course beyond its planned content. This diversity however, also introduced challenges. About 10–20% of students struggled with codding exercises, while others found them too basic. Interests also varied highlighting the difficulty of designing a one-size-fits-all course. Offering varied examples and encouraging flexible participation helped, but content calibration remains a challenge for future versions.
Based on student feedback, the second iteration included key refinements: more time for hands-on tasks, student-led discussion and moderated Q&A for course ongoing refinement. Demand also increased—23 students stayed on the waiting list—signaling raising demand for structured, ethical GenAI training at ETHZ.
An interesting insight was the rapid outward flow of knowledge. Students began sharing course materials within their labs, prompting further interest. This affirmed the role of doctoral students as knowledge multipliers within ETHZ’s academic ecosystem. However, the rapid evolution of GenAI poses sustainability challenges. Even six months between iterations required major updates. Maintaining the course requires broad expertise and stable support. We are actively seeking long-term funding to ensure the course remains sustainable, high-quality, and help broader GenAI adoption ETHZ efforts.
Students appreciated the holistic exposure to GenAI but called for continued learning. While informal exchange is already taking place, the need for a central, moderated platform—for sharing use cases, coordinating updates, and streamlining knowledge across ETH—remains unresolved. Addressing this would strengthen institutional readiness and long-term capacity in GenAI education.