Network Structures of Collective Intelligence: The Contingent Benefits of Group Discussion


Speaker


Abstract

Research on the “wisdom of crowds” has found that the average belief in a group can be remarkably accurate even when individual group members are wildly inaccurate. This phenomenon has been observed for domains ranging from financial forecasting to medical diagnoses, and a common theoretical claim is that group beliefs are most accurate when they are collected from individuals who are socially and statistically independent. However, the requirement for independence poses a challenge in many social and organizational settings where interaction and communication are an intrinsic part of decision-making.  In contrast, I show that social information processing can produce beliefs that are even more accurate than the collected beliefs of independent individuals—under the right conditions.  This talk will present formal models and behavioral laboratory experiments to identify when, and why, group interaction can help (or hurt) numeric belief accuracy. The main focus of this talk compare mediated information exchange (i.e., the “Delphi method”) with unstructured discussion, showing how network theory can resolve longstanding contradictions in previous research.