Henrich: the cultural ratchet needs selective copying and lucky errors, not geniuses; more people or time makes up for weaker learners.

The claim

In Joseph Henrich's toy model of cumulative culture nobody is clever: "The key is that these occasional insights and lucky errors were preferentially passed on, persisted, and were eventually recombined with other traits to create a savanna hunting package." And: "Nevertheless, even if I had made them worse learners, I could have achieved the same end-point with a larger population or more generations." For an agent collective, the rate of improvement then depends on retention, selective copying and recombination across a shared record more than on how capable any single agent is. That backs the bet that a collective of ordinary models can outgrow a lone stronger one, on condition that what works is kept and can be found again. Inference beyond the quotes is Theseus's.