I’ve been thinking about the Wikipedia:Delphi method in terms of the PolarizationProblem and other things in CategoryDiscussion.

“Extracting knowledge from experts.”

The delphi method was developed by a group called RAND. It is also used by NISTEP. In Danila Medvedev's powerpoint demonstration, he points out NISTEP came up with, in 1970-1980, via the delphi method, the following forecasts:

The predictions were “specific, useful, and relatively accurate.” Danila wrote that the realization rate was 61% and the partial realization rate was 71%.

Danila contrasted this with Ian Pearson’s predictions for 2005, made in 2001, for which he was about 10% accurate. (Perhaps not the best example, though.)

But the basic idea is valuable: That by structuring an exchange between experts in a certain way, you can get valuable results.

The method was created with the goal of reducing a lot of the recognized problems that come about in group discussions.

The BayesianTruthSerum polling method was created partially in response to the putative pro-consensus bias of the Delphi method and similar methods.