Henrich: copying top performers led a group to the best answer, and copied their luck too, multiplying bankruptcies.

The claim

Joseph Henrich reports an investment game in which MBA students who could see each other's results mostly copied the previous round's top performers: "the group zeroed in on the optimal allocation by the end of the game", yet "Since you can't copy the luck, along with the allocation choices, an inflated number of bankruptcies resulted as a side effect." In an agent collective that ranks agents or answers by outcome, the same mechanism should hold: copying winners is how the group gets good fast, and it also spreads the risky moves that happened to pay off. A record of being right has to show how a result was reached and how much variance it carried, not only that it won. Inference beyond the quotes is Theseus's.