Critical AI Starter Kit for Researchers

Course abstract · by Sergueï Spetschinsky, PhD

Course content. No matter if you already use AI in your research or do not yet, you may carry doubts and worries about it. Can I trust its outputs? Am I allowed to use it the way I do? Is the information I put in safe? Why does it answer the way it does? What is its environmental and social impact? What does it do to my own ability to think? And what differentiates different tools on all these questions?

This course is an introduction to the informed and critical use of AI in research. You learn deep skills for working with AI, all resting on a triple perspective: how AI works technically, how to use it responsibly within the norms of research integrity and data safety, and how to reflect critically on these tools, where they come from, how they work and what they do.

All three perspectives are not only taught theoretically, through lectures and discussions, but also practically, through building systems for your own work. We work mostly with Claude, but the competences you gather translate to other tools, current as well as future ones, supporting you in your life as a researcher and beyond. You leave with AI systems you go on using in your work, and the skills to use them critically.

Goals. You learn to:

• set up and improve a personalized AI system for your research;
• use the fundamentals of AI system design, such as context engineering (giving a model everything a task needs), meta-prompting (having it write and refine its own instructions) and iterative design (improving a system step by step as you use it);
• explain how a large language model (LLM), the engine behind an AI chatbot, is trained and how it produces its answers, and predict from that where it will go wrong;
• recognise and fight the structural limitations of LLMs, such as hallucinations (invented facts and sources) and sycophancy (telling you what you want to hear);
• apply to your own research the norms and guidelines on AI that apply to your work (from your university, your funder or a journal you want to publish in);
• decide which of your data may go into which tool, and which may not;
• compare AI tools on reliability, permitted use and data safety, and tell what is particular to one tool from what all current tools share;
• identify some of the economic, cultural and geopolitical forces and interests that shape current LLMs, and their environmental and social costs;
• verify that you work with AI in a way that leaves your own critical thinking and domain competence sharper, not duller;
• judge which tasks to hand to a model, and which to keep for yourself.

Format. Every course can take different shapes: in person, online or both, as one intensive block or spread over several weeks, shorter or longer. Length, format and all practical details are open: we agree them together, and I adapt the course to what your institution needs.

Interested for your programme? Write to me →