The Mirror, the Whetstone and the Black Hole: How Should We Think with Large Language Models?
We do much of our thinking in language – making sense of what we perceive, regulating emotions, reflecting on ourselves, and working through problems. That self-directed conversation once stayed silently inside the mind. Now, large language models (LLMs) are joining the stream of thought as fluent, on-demand thinking partners.
Whether conversing with an LLM sharpens or erodes human reasoning depends on how people engage with it. The same model can act in three very different ways:
- A mirror, reflecting a user’s existing view back with added fluency and conviction.
- A whetstone, offering counterarguments and prompts that challenge a line of reasoning in ways a solitary thinker might miss.
- A black hole, absorbing the reasoning so that the user keeps a polished output while giving up the cognitive work that builds understanding.
About the workshop
Planned for spring 2027 at Lancaster University (exact date to be confirmed), this one-day interdisciplinary workshop will bring together around 30 researchers across cognitive science, psychology, educational research, management science, linguistics, philosophy, and human–computer interaction. It is generously funded by DSAIL
Organised by Dr Bo Yao (Department of Psychology), Dr Yanyao Deng (Department of Educational Research), and Dr Shiyan Zhang (Department of Management Science), the day will focus on three core questions:
- Concepts: How should we model human–LLM dialogue – as an externalisation of inner speech, a form of distributed cognition, or a substitute for thought – and how does it reshape decision making in individuals and teams?
- Measurement: How can interaction logs, eye-tracking, think-aloud protocols, and brain recordings (EEG) track, turn by turn, whether a model is mirroring the user, sharpening their reasoning, or taking over the thinking?
- Practice: How can we design educational and workplace practices that encourage active, challenging reasoning with AI while protecting privacy, equity, and diversity of thought?
Programme and participation
The day will open with a debate on how conversing with LLMs shapes human thinking, followed by three focused sessions on concepts, measurement, and practice, and a closing synthesis session to plan a shared conceptual framework and pilot studies.
Dates, speaker confirmations, and registration details for spring 2027 will be posted here once finalised. If you are a researcher or postgraduate student interested in taking part, please contact Bo Yao.