Ratcliff works with single-celled yeast. Sometimes those cells make copies of themselves that don’t separate but stay attached, forming lacy multicellular structures called snowflakes. In his initial tests, Ratcliff put the yeast cells under pressure by selecting those that fell fastest to the bottom of a test tube—a race that snowflakes tended to win—and discarding the rest. Over time, a strange thing happened: Instead of favoring genes that improved individuals, the yeast began to turn down the expression of some genes and turn up the expression of others, in ways that made it harder for cells to split off from the group. That allowed the snowflakes to grow larger and evolve into greater complexity.