Can AI Explain Science Without Understanding It?

When AI is used in science, are we learning from a knowledgeable assistant or misinterpreting the capabilities of a system that is simply very good at producing convincing language?

This interactive workshop explores how large language models (LLMs), like ChatGPT, Gemini, and Claude, work, and what it means to use tools that can communicate scientific information fluently but whose relationship to meaning, knowledge, and understanding remains contested. 

Through hands-on activities and group discussions, we will investigate what these tools are good at, where their limits come from, and what responsible use looks like in research and science communication.

Participants will explore:

• How LLMs generate their answers and why they can sound authoritative even when they’re not
• How AI is changing how scientific knowledge gets produced, communicated, and validated; and what this means for trust in scientific outputs
• Why fluency and accuracy come apart, and how to spot the gap in practice
• How to interrogate an AI output the way you’d interrogate a source: what questions to ask, and what failure modes to look out for