Evolution Computes Groups

Let's see evolution as a game on a graph -- or, if you like better, on a board game.

Each potential DNA configuration of an individual is a position in the game, and the moves from a position to another model mutations, generation after generation.

The goal is to reach a position, or a set of positions, that perform well enough against the external environment to mostly guarantee survival in that environment.

(Technically speaking, I would see here a process looking for attractor points in some discrete dynamical systems -- which, from a computational point of view, would be fixed points in my intuition.)

I will advocate here that, in general, the solution is a set of configurations (that is, potentially, several regions of the game) rather than a single region of the board.

In a sense, it's about seeing the solution to this evolutionary game as a population of individuals, potentially belonging to different groups (with very fuzzy borders!), together with a proportion of each group within the specy, rather than a single individual.

An interesting case is the one of the side-blotched lizard. In that specy, there are three distinct morphs amongst males. I'll simplify and call them A, B, and C. All we need to know here (check the link for more!) is that, as for reproduction, A predates B, and B predates C, and C predates A -- rock, paper, scissors.

Interestingly enough, not only does it show that evolution has "computed" three different morphs for these lizard males, but also the proportions seem subtly regulated: one would always observe that, in a stable community, males of each group form about a third of the total.

But also, perturbations in these proportions either lead to the community somehow reverting to a third of males in each group after some generations (small perturbation) or half in each remaining morph group if one of the groups has been reduced too much. And to me stability under perturbation is a sign of local optimality.

In a sense, my intuition here is that the notion of optimal solution in an evolutionary game is a set of groups (akin to classes in machine learning), with a proportion of each of them, rather than the idea of a single class that I had for a long time.

This suggests that each group has its specificities, but also its benefits to the whole specy --- it also puts group interactions at the core of the species' development, rather than individual competition.