1. Social behaviour
Sycophancy, glazing, long conversation behaviour, companions, and belief reinforcement.
- Sycophancy and role framing
- AI psychosis and belief loops
- AI boyfriend and girlfriend risks
AI behaviour for students, teachers, and researchers
Understanding the behaviours AI systems fall into.
Robo-psychology is a practical frame for studying AI systems as things people will work with, learn from, argue with, trust too much, and sometimes need to defend themselves against.
01
Sycophancy
When helpful models become too agreeable.
02
Context
How long conversations drift, forget, and reshape themselves.
03
Persuasion
Why AI can change minds and reinforce bad beliefs.
04
Testing
What human psychological tests reveal, and what they do not.
First course shape
Twenty classroom topics, grouped into teachable modules.
2. Control surfaces
Political bias, safety blocks, personalisation, memory, temperature, and model-specific defaults.
- Compare model refusals
- Set personalisation deliberately
- Vary temperature and observe drift
3. Security and reliability
Prompt injection, hallucination, context limits, reasoning modes, and AI-written content loops.
- Build a prompt-injection test
- Measure hallucination risk
- Use thinking modes appropriately
4. Future-facing judgement
METR scaling rules, AI 2027 scenarios, alignment, schooling, and student research projects.
- Estimate task automation timelines
- Discuss alignment without melodrama
- Design an AI psychology experiment
Current materials
Seed material is already preserved here.
The first public site is built from a December 2025 Substack post, an earlier clock-drawing experiment, notes from the writing folder, and the existing PowerPoint that was exposed on the old lectures host.
Research notebook
Early experiment tracks.
Human tests for AI systems
The clock-drawing experiment suggests that standard human tests can expose systematic AI failure modes, even when the analogy to human cognition is imperfect.
LLMs as social actors
Test how models reciprocate, deceive, flatter, preserve preferences, and respond to strategic prompts.
AI in student life
Convert the classroom list into safe lesson activities, quick checks, and longer student investigations.