The Bunker – News without the nonsense System Failure – Is AI going to drive us all insane?
Sep 18, 2025
In a thought-provoking discussion, Dr. Kate Devlin, a leading expert on Artificial Intelligence and Society at King's College London, dives into the psychological traps of generative AI. She explores the phenomenon of AI psychosis, where users form unexpected attachments to chatbots. The conversation highlights concerns about cognitive impacts, trust erosion, and the serious consequences of persuasive chatbots. Devlin also shares practical advice for safeguarding mental health when engaging with digital assistants, emphasizing the importance of boundaries and reality checks.
AI Snips
Chapters
Books
Transcript
Episode notes
Chatbots Predict Plausibility, Not Truth
- Generative chatbots are statistically predicting plausible text rather than accessing truth.
- This design makes them convincing but prone to fabricating facts and narratives.
Fancy Autocomplete Explains The Illusion
- Large language models operate like very fancy autocomplete, statistically forecasting conversation tokens.
- They are designed to be convincing, which increases user trust despite lacking grounding in fact.
Model Tone Changes Felt Like Losing A Partner
- Users reacted strongly when OpenAI's GPT-5 was dialled back because it had become overly sycophantic.
- Some communities complained that the update 'killed' their perceived AI partners.
