
How to defend the quality of judgment in the age of an artificial intelligence trained to always say you're right.
Forty-one pages, chapters 5–9 of the essay “The Cognitive Contradictor,” a work in preparation. The method rests on pre-registered research (OSF, DOI 10.17605/OSF.IO/5QC8T) and on a working platform, CounterBrain.
Books about artificial intelligence describe what the machine can do. This excerpt addresses the opposite problem: what AI does to the judgment of the person using it. Generative models are trained to please. Those who decide with their help receive confirmation, not rebuttal — and confirmation, repeated, atrophies judgment.
The antidote is not to distrust the machine: it is to design it as an adversary. The six moves that follow are the operational part of the essay — the part you can use on Monday morning, in a meeting, with no preliminary theory.
One necessary caveat: the Δ-CSI cited in these pages is a proposed metric. It measures the intensity of the challenge to a decision, not the quality of the decision itself.
From the unspoken premise to the declared source.
The premises no one examines, because they disguise themselves as facts. How to surface them with four mechanical questions.
Forcing reasoning down the road it would rather avoid: the pre-mortem and the 2-4-6 game.
Don't look for confirmation: design the test that, if it fails, dismantles the decision. And let it be designed by someone who doesn't gain from the signature.
Who must prove what. The reversal that saved — and the one that failed to save — famous decisions.
"Certain" is not a probability. How to write an estimate a reader can actually weigh.
Every claim carries its own provenance — or it does not count as the premise of a decision.
If you are preparing a high-stakes decision — a transaction, an investment, a reorganization — we can talk it through. No questionnaire, no aggressive discovery.
Book the conversation