The wisdom of crowds and rank ordering problems. Probabilistic knowledge retrieval and group communication for complex tasks
In 1907, Sir Francis Galton observed in Nature that given the estimates of a large group of people, the median response was better than that of any expert. He begrudgingly concluded that, “the trustworthiness of a democratic judgment is more credible than might have been expected.” Today we know that, more than just being credible, the “average” response (be it mean, median, or mode) of a large group of people is quite often the best estimate available (Surowiecki, 2004). This phenomenon, commonly known as the “Wisdom of Crowds,” relies on averaging responses so that individual error cancels out (Wallsten et al., 1997; Yaniv, 1997). But what happens when we can’t simply average our responses? When given more complex tasks, we must rely on other means. Are there ways that the idea behind the wisdom of crowds - using response independence to cancel out individual error and uncertainty - can be brought to bear on these problems? We will explore this question in a hierarchical Bayesian modeling framework, and introduce several computational-level rational models in attempt to describe the observed human subject behavior.