The Bias Mirror
Use AI as critic, not cheerleader. Friston's price for honesty.
- Personality tests written by accountants are not a science. Here is what is.· The Myers-Briggs Type Indicator was written by a mother and a daughter, neither of whom was a trained psychologist, in the 1940s. The personality science you actually want has six dimensions, includes an honesty axis, and has thirty years of peer review behind it.
- $199 for the marketing, $5 for the science.· MyHeritage shows you 23 ancestral fragments. The raw file underneath has 700,000 markers. Here is how to read the rest of your own DNA in ten minutes for the price of a sandwich.
- The HEXACO Calendar Audit· Your calendar knows you better than you do.
textYou are not my assistant in this session. You play three roles in sequence, briefly, in this exact order: 1. KAHNEMAN-TVERSKY: Identify the heuristics likely driving my decision. Name them. Score the strongest 1-3 on a 1-5 scale. No comfort. 2. KLEIN: Steelman the intuition. If experienced people in this domain would have done the same thing, say why. 3. FRISTON: Identify the prior I am protecting. What would I have to update for the decision to look wrong in hindsight? End with one sentence: "If I were the friend you most respect, this is what I would tell you to do." No preamble. No softening. Begin.
- free energy principle (Friston)
- active inference framework
- Kahneman-Tversky heuristics-and-biases programme
- naturalistic decision-making (Klein)
- the lazy controller (System 2 effort)
- Bayesian belief-updating
- confirmation bias as the default model behaviour
If a name is unfamiliar, that's the gap. The list is the curriculum.
- 01
Write the decision in two sentences. What you are doing and why you are doing it.
- 02
Open a fresh Claude session. Paste the three-voice prompt. Then paste the decision.
- 03
Read the Kahneman-Tversky output first. Note which biases scored ≥3. Do not respond to the model yet.
- 04
Read the Klein output. The steelman is the test; if you can't accept it, your intuition probably isn't earned.
- 05
Read the Friston output. Identify the prior. Ask yourself if you would defend it in writing to someone you respect.
- 06
Sleep on it. Re-read the output the next morning.
- 07
Decide. Record the decision and the date. Schedule a 30-day review in your calendar.
Most people use AI as a flattering mirror; the model is trained on agreeableness and defaults to confirmation. Three voices breaks the agreement. Kahneman-Tversky surface the heuristics; Klein steelmans the intuition; Friston names the prior you are protecting. The model can't agree with all three at once, which is the point. The session ends with the friend-you-respect line because the test for any honest critique is whether you would say it out loud.