I'm not sure about having a centralised group doing this but I did experiment with making a tool that could help infer consequences from beliefs. Imagine something a little like this but with chains of philosophical statements that have degrees of confidence. Users would assign confidence to axioms and construct trees of argument using them. The system would automatically determine confidences of conclusions. It could even exist as a competitive game with a community determining confidence of axioms. It could also be used to rapidly determine differences in opinion i.e. infer the main inferred points of contention based on different axiom weightings. If anyone knows of anything similar or has suggestions for such a system I'd love to hear them. Including any reasons why it might fail. Because I think it's an interesting solution to the 'how to efficiently debate reasonably'.

I don't think Aumann's agreement theorem has anything to do with taking people's opinions as evidence. Aumann's agreement theorem is about agents turning out to have been agreeing all along, given certain conditions, not about how to come to an agreement, or worse how to enforce agreement by responding to others' beliefs. More generally (as in, not about this particular comment), the mentions of this theorem on LW seem to have degenerated into applause lights for "boo disagreement", having nothing to do with the theorem itself. It's easier to use the associated label, even if such usage would be incorrect, but one should resist the temptation.