Moral Mirror — A Builder's Notebook
Can an AI make people less certain, not more, in domains without right answers? A builder's serialized notebook tracing every corner of the question and how each idea was stress-tested.
About this series
Anyone who has hesitated even once in Sandel's Justice lecture hall knows the feeling — the moment your own words collide with your gut is where real learning happens. This notebook starts from a vague hunch: could that mechanism be recreated one-on-one with AI?
It is a record of a thinking trajectory, not a results report. How an unnamed category was found, how a 10-persona roundtable stress-tested it, and how the manipulation, sycophancy, and mental-health risks Anthropic flagged first were met — every corner of the deliberation, published biweekly.
Written for those interested in AI education and literacy, evaluation in domains without ground truth, multi-agent architecture, and a pedagogy made possible only by AI. Twenty installments add up to one book — an in-progress book.
1 episodes
- 01