Logarithmically Right

Seth Roberts has a great new term:, stolen from a book he thinks has some weaknesses:

What’s interesting is that logarithmically right is a good way of describing how one’s beliefs should be transformed to be a fair approximation of the truth. When you think you are right, you probably are — but logarithmically. Much less than you think.

This should be the base point for most folks when holding ANY position that any smart person disagrees with.  Unless, of course, lots of smart folks disagree, and then you should decrease your likelihood of being right below even further. 

I had meant to write, earlier today, about a peculiar idea I'd had with respect to consensus, that just happens to fit nicely in with this line.  As far as I've been able to tell, any portion of science wherein you can't specify a simple experiment that will determine the answer...the consensus is at best logarithmically right.  The fact is status feedback leads to consensus opinions carrying much heavier weight than is appropriate. 

In Bayesian terms (the science of how you update your opinions), a consensus opinion with status to be gained from being in consensus (most such opinions) is equivalent to a whole bunch of prior trials leading to a 60% confidence.  1, or even 10 pieces of evidence should not convince you that the consensus view is wrong, but it also shouldn't take very many pieces of evidence either before a view should move into the point of view: this is suspicious, and I can't trust the consensus any longer.