Aumann's agreement theorem proves that two rational agents with common knowledge of each other's beliefs cannot agree to disagree — if they share the same priors and each knows the other's posterior, they must converge. In practice, persistent disagreement signals that at least one party has different priors, different evidence, or is reasoning incorrectly. The Rationalist approach to disagreement: take the other person's belief as evidence (their brain processed information you have not seen), update toward them proportional to your assessment of their reliability, and investigate the crux — the specific factual or inferential disagreement that drives the difference. Productive disagreement requires identifying cruxes rather than repeating arguments.
In your next substantive disagreement, try to identify the crux: what is the specific factual claim or inference where you and the other person diverge? State it explicitly and check whether resolving that point would change both your minds. Practice taking the other person's confidence as evidence — if a domain expert disagrees with you, how much should you update?
From steelmanning, you know how to engage with the strongest version of an opposing argument. From intellectual humility, you know that calibrated confidence means being as certain as the evidence warrants, no more and no less. Disagreement and rational updating brings these skills to bear on one of the most common epistemic situations: someone you respect disagrees with you. What should you do?
The theoretical foundation is Aumann's agreement theorem, which proves that two rational agents with common priors and common knowledge of each other's posteriors cannot agree to disagree -- they must converge on the same probability estimate. The conditions are strict: shared priors, and each party knowing what the other believes (and knowing that the other knows, and so on). These conditions are rarely fully met in practice. But the theorem's practical lesson survives the idealization: persistent disagreement between reasonable people is a signal that something is going on -- different priors, different evidence, or reasoning errors on one or both sides. Treating disagreement as merely a difference of opinion misses the information it contains.
The Rationalist approach treats another person's belief as evidence. Their brain has processed information you have not seen, made inferences you have not made, and arrived at a conclusion that encodes all of that processing. When a well-calibrated domain expert tells you she estimates the probability of H at 30% and you had it at 70%, her estimate is a data point -- not as authoritative as a controlled experiment, but not as dismissible as mere opinion either. The rational response is to update toward her estimate by an amount proportional to your assessment of her reliability and relevant expertise. You do not blindly average (that ignores differences in information quality), you do not fully defer (that ignores your own information), and you certainly do not refuse to update (that treats your prior as immune to evidence).
The most productive move in any substantive disagreement is to identify the crux -- the specific factual claim, inference, or value judgment that actually drives the divergence. Most disagreements cycle endlessly because both parties restate their conclusions rather than locating the premise where they part ways. A crux-finding approach asks: "What would have to be true for you to update toward my view?" and "What would have to be true for me to update toward yours?" This locates the actual point of divergence, where evidence could potentially resolve the dispute. Once you have identified the crux, you can investigate it directly rather than rehearsing downstream disagreements that neither party will concede without the upstream issue being settled. This turns unproductive argument into collaborative inquiry -- which is what rational disagreement is supposed to be.
Topics in reflective domains aren't scored by quiz answers. Read, reflect, and mark when you've thought it through.
No topics depend on this one yet.