Consider a public poll. Especially one where you subject the subjects to some subjective questionnaire (pun intended). How may you interpret the results of that poll? What conclusions you can draw from those results?
Generally, we see a trend in the results. Most of the participants agree on one thing. That’s consensus. It tells that those individuals have that fact bored into their minds, perhaps through regional history, family backgrounds, childhood story, actual event etc. Typically, it is fail safe to rely on this information.
But, there is a lower degree of disagreement too, almost always. Some people always seem to think that actually it was something different than what others are saying. Call them conspiracy theorists or whatever. The disagreement is always present.
Now consider, how much information is contained in that disagreement. It represents the perturbations present in the opinion pool. You can use those diversions to target minor classes of people, having quite radical ideas about what you are looking for. For example, consider a poll which asks for some kind of personal identifier and, on a scale of 1 through 10, how much they can strive for their country. Anyone with criminal or terrorist motivation would certainly go for people with low ratings.
Using similar procedure, you may want to go for the lower percentage splits when selecting candidates. Depending on your query, they may be more susceptible to bullying than others.
As another example, consider searching. While those results which have a high relevance certainly have value, but those which have low relevance may represent unorthodox concepts. An example of this technique was in ‘Digital Fortress’ where ‘non-confirmatory search’ was conducted to find data which appeared more random. It was in essence a search for random regions on hard disk and then pin pointed to those regions which were not found because they were more random than the found results.