In the comments on [my post mocking Granville Sewell’s dreadful article][sewell], one of the commenters asked me to write something about why evolution is frequently modeled as a search process: since there is no goal or objective in evolution, search seems like a strange approach. It’s a very good question, and so I’m going to do my best to give you a very good answer.
As I’ve said before, math is all about *abstraction*: taking something, stripping it to its bare-bones essentials, and seeing what it means: for example category theory, which I’ve been writing about for the last month, is about taking the basic idea of a function as a thing that maps things to other things, and seeing what that means when it’s stripped of everything but the most basic properties of that notion of mapping.
When we build mathematical models of real-world phenomena, we’re doing the same kind of thing: we’re looking at some particular aspect of the phenomena, and trying to see what it means when the complexities of reality are abstracted away to let us see what’s going on in a fundamental sense. When we build bridges, we do static-mechanical analysis of the bridge structures; the mast basic static mechanical analysis is remarkably silly in some ways, but it does manage to capture the basic properties of tension and force distribution through the structure. It lets us understand the most basic properties of the structure without having to worry about wind, about gravitational bowing of metal rods, vibration, etc. (Obviously, when a bridge gets built, all of those things need to get factored in; but you start with the basic analysis.)
So when a mathematician looks at modeling evolution, the way that we approach it is to ask: What are the basic fundamental things that define the process of evolution?
The answer, roughly, is two (or three, depending how you count them ) things:
1. Reproduction with mutation;
Evolution is a continual process in which those steps are repeated: you reproduce a pool of individual, either mutating while you copy, or copying then mutating. Then you select some subset of them as “winners” and keep them to copy for the next generation, and you discard the rest.
What’s the selection criteria? From the starting point of a mathematical model, *it doesn’t matter*. What matters is that it’s a cyclic process of reproduction with mutation followed by selection.
So we have a basic model. Now we look at it and ask: does this resemble any problem that I know how to solve? Let’s try drawing a picture of that: we’ll draw labelled dots for individuals; with arrows from a parent to a child. The children that were selected and reproduced, we’ll paint green; the children that were not selected we’ll mark red; and the current generation, which we haven’t selected yet, we’ll leave white.
So, does this look familiar to a mathematician?
Yes: it’s a mathematical structure called a graph. We know a lot about the mathematics of graphs. And evolution looks a lot like a process that’s building a graph. Each node in the graph is an individual; each edge in the graph is a reproduce with mutation step. The leaf nodes of the graph are either successful individuals (in which case the graph is going to continue to grow from that leaf), or it’s an unsuccessful one, in which case that branch is dead.
When you describe it that way – it sounds like you’ve described a classic graph search problem. To complete the model, you need to define the mutation and selection processes; the criteria by which successful children are selected and future generations are created are, of course, a crucial piece of the model. And to be complete, I’ll note that the mutation process is a *random* process, and so the mutation paths generated by running a simulation of this model will be different in each run; and further, selection is undirected: the selection process does *not* have a specific goal. (Note: this paragraph was edited from the original to clarify some points and to expand on the properties of mutation and selection.)
There are other possibilities for models of evolution; but the most obvious mathematical model for evolution is search: either a graph search or a terrain search, depending on exactly how you want to model the process of mutation.