The Perils of Too Much Consensus?
This article discusses a type of cognitive bias or failure-mode concerning the “Paradox of Unanimity” that arises because of the incredibly small likelihood of every member of a large group agreeing on a thing. Bayesian analysis shows that for many phenomena, as the group of people agreeing increases beyond a certain point the chance of them being correct decreases until it is little better than a random guess. This occurs whenever there is any bias concerning the phenomena under consideration, even a small bias or preference for one feature or another.
Large unanimous agreement still holds as a good, powerful tool in situations in which there is zero or near-zero bias for any features.
I’m still trying to wrap my head around this. The article discusses fascinating examples including police line-ups, forensics tests, Volkswagen emissions, electoral outcomes, noise and lack thereof in scientific experiments, and decisions by committee. One interpretation is to recast “deniers” of any generally broadly supported complex model in a thoroughly positive light, not as “ignorant” but rather as those whose very disagreement or denial actually support the validity of the complex model that has achieved consensus; climate change is an obvious one that comes to mind.
I particularly like the cryptographic example and wonder if any computer scientists or cryptography experts can comment as to its validity. Apparently, even with cryptographic security dealing with probabilities of 2^-128 being typically held as acceptable, random bit flips due to cosmic rays can occur with a probability of 10^-13, which dominates over the 2^-128 security and so implying the cryptographic algorithms may be far less secure than they appear.
Article here: http://phys.org/news/2016-01-evidence-bad.html