It seems that in principle a version of debate where only one agent makes statements and the other chooses which statements to expand could work, but it seems like it requires the judge to be very strict that the statement is 100% true. It seems hard to apply this kind of system to statements outside of formal mathematics.
Systems where both agents can make statements seem like they might be less vulnerable to judges accepting statements that aren’t 100% true. For one example, if both agents take turns being the arguer, then if both agents submit a path that is judged to be correct, you can stipulate that the agent with the shortest path wins (like imposing a simplicity prior).
It seems that in principle a version of debate where only one agent makes statements and the other chooses which statements to expand could work, but it seems like it requires the judge to be very strict that the statement is 100% true. It seems hard to apply this kind of system to statements outside of formal mathematics.
Systems where both agents can make statements seem like they might be less vulnerable to judges accepting statements that aren’t 100% true. For one example, if both agents take turns being the arguer, then if both agents submit a path that is judged to be correct, you can stipulate that the agent with the shortest path wins (like imposing a simplicity prior).