It seems to me that I can accurately determine which of two people is smarter by just listening to them talk if at least one person is less smart than I am, but that this is very difficult or impossible if both people are much smarter than me. When both people are smarter than me, I fall back on crude heuristics for inferring intelligence. I look for which person seems more confident, answers more quickly, and corrects the other person more often. This, of course, is a very flawed method because I can be fooled into thinking that people who project unjustified confidence are smarter than timid but brilliant people.
In the intermediate case, when I am only slightly dumber than at least one party, the problem is reduced. I am better able to detect over-confidence, often because I can understand what’s going on when the timid smart person catches the over-confident person making mistakes (even if I couldn’t have caught them myself).
(To an extent, this may all be true when you replace “smarter” with “more skilled in domain X”.)
This suggests that candidate voting systems (whether for governments or otherwise) should have more “levels”. If we all want to elect the best person, where “bestness” is hard to identify by most of us mediocre participants, we would do better by identifying which of our neighbors are smarter than us, and then electing them to make decisions for us (possibly continuing into a hierarchy of committees). This is an argument for having federal senators chosen by state legislatures.
Of course, there are many problems with additional levels, e.g. it is difficult to align incentives across just two levels. Each additional level exacerbates the problem.
Is there any way to get the smartness-discrimination advantage of many levels with reduced incentive issues?