Adversarial proceduralism

Two systems are handed opposite sides of a question and argue it out in front of a judge who lacks what they have — the source passage, the evidence, the time, or the raw capability — and so cannot settle the question alone. Because each side gains by exposing the other's errors, the judge hears the strongest case against every claim before deciding, and this line of work measures whether judges actually end up more accurate: on passages the judge has not read, on contested COVID and climate claims before people with prior beliefs, and against a single unopposed consultant. One variant drops fixed sides altogether — the two systems are directed to locate the exact point they disagree on and hand the judge that crux rather than a finished argument.

A checker deliberately kept small and bounded is trained to interrogate a far more capable arguer it has no reason to trust, while a second arguer is rewarded for slipping wrong answers past it. Only answers the weak checker can follow and confirm survive the loop, so what comes out is reasoning a less capable reader — a small model, or a person working under a time limit — can verify. The distinguishing move is that the adversary sits inside the training loop and the checker is a lone gatekeeper facing one untrusted prover, not a judge picking between two contestants.

These are debate protocols written in advance so that a dishonest player cannot win merely by being able to compute more than an honest one. One design lets the honest side succeed using a manageable number of simulation steps even when its opponent is allowed exponentially many; another handles the trick of splitting a question into pieces so obscure that answering them honestly would be intractable, by pairing an arguer with an estimator and holding the honest player's cost comparable to its opponent's. The strategy is to fix the rules of the contest rather than to run it and see how the judge does.

The dispute here is over interests and values, where there is no fact to be right about, so each position is carried by its own agent and the procedure is built to stop the positions collapsing into one. Stakeholders are role-played by agents that bargain under a budget with a neutral mediator, and the outcome reports how the worst-off group fared rather than only the total; separately, giving each value perspective its own distinct underlying model keeps a deliberating council's answers spread out instead of converging on one. Two results sit behind the design: a study of what happens when a single model is made generator, debater and judge on moral dilemmas and contested political claims, and a proof that when majority preferences cycle no reward model can crown a winner at all.

Here the arguing is done by people and the model is the referee rather than a contestant: it facilitates, moderates and summarizes an online discussion, or bridges between the camps inside it. That role reversal is the distinction — in every other approach in this regime the model is one of the disputants, and here it is the procedure. The bet is that a mass argument can be kept legible and workable at a scale no human moderation team could reach.