Big, linked, diverse groups ratchet their skills up

Each dot holds a skill level and an approach, shown as its colour. Each generation every learner looks for the most skilled model it can see: always the best in its own town, and the best of each other town with a chance set by interconnectedness. It copies that model imperfectly, landing below it on average by an amount set by individual ability, with a random spread (a Gumbel draw). This is Henrich's (2004) Tasmania model: because a few learners get lucky and overshoot, a pool where many learners copy the same expert ratchets skill up, and a small or cut-off pool slides down, since too few lucky copies arrive to offset the average loss. Skill stops at zero, the point where a town has lost the craft.

After learning, some dots meet a partner from their town, or from a linked town. When the two hold different approaches they recombine: the result sits above the average of their skills by a random jump, and counts as an invention when it beats both. Learners who copy a fresh inventor lose less in copying that generation, the feedback from inventions to individual ability. More approaches mean more pairs that differ, so diversity raises the rate of inventions; with one approach there is nothing to recombine. This follows the argument of Muthukrishna and Henrich (2016, Philosophical Transactions of the Royal Society B, "Innovation in the collective brain"): innovation depends on the size, links and diversity of the population more than on lone geniuses. The slide comes from a talk by Joseph Henrich. The numbers here are simulated to show the mechanism; the skill scale is arbitrary.