6 ideas
When solving hard problems, a randomly selected group of diverse problem-solvers can outperform a group of the highest-ability individual solvers. This holds because the high-ability group tends to share similar perspectives and gets stuck in the same local optima, while diverse solvers approach from different angles and cover each other's blind spots.
Cognitive diversity decomposes into four distinct tools: perspectives (ways of representing a problem), interpretations (ways of categorizing situations), heuristics (rules for generating solutions), and predictive models (ways of inferring cause and effect). Productive group diversity means variation across these cognitive tools, not variation in demographics or identity per se.
A perspective is a way of mapping a problem into an internal representation, which determines where the 'peaks' and 'valleys' of good and bad solutions lie. Two people with different perspectives see different landscapes, so a solution that looks like a dead-end local optimum to one person may be merely a foothill that another can climb past.
A crowd's predictive accuracy equals its average individual accuracy minus its diversity of predictions: collective error falls as members' predictions become more varied. This means a group can be collectively wiser than any of its members, and adding a diverse but individually mediocre predictor can improve the group more than adding an accurate but redundant one.
The benefit of adding one more person depends on whether their added diversity outweighs their lower ability. As a team grows, the best remaining candidates become increasingly redundant with existing members, so beyond a point an individually weaker but cognitively distinct addition contributes more than a stronger but similar one.
Cognitive diversity only delivers its gains on genuinely difficult, non-obvious problems where the best approach is unknown; on simple tasks with a clear correct method, diversity adds noise rather than insight. The value of assembling varied thinkers is therefore conditional on the problem's difficulty and the existence of multiple plausible approaches.