The basic premise is simple: a diverse data sample averages away errors. Thus, with effective aggregation methods, diverse groups are more reliable than individuals or poorly-made groups (groupthink, over-centralisation, etc.).

The author explores the cultural biases we have for assuming that single individuals are smarter than many individuals, including factors that produce foolish group behaviour.

My favourite part is the bibliography—James has built an impressive collection of studies on collaborative mechanics. I’ve started a academic-oriented summary which would be helpful for someone doing research.