Thursday 9 October 2014

Natural selection is not an algorithm

The modern use of the term "natural selection" has some issues, from my perspective. The problems I see are:

  • Natural selection crudely combines the production and elimination aspects of Darwinism;
  • Natural selection is defined as being "non-random" change;
  • Natural selection is widely and frequently characterized as being an "algorithm";
To summarize the first objection, I think it is better to teach evolution by separating its productive and eliminative aspects. I've gone into this many times before, in particular:

This split seems like a simple matter of good epistemic hygene to me.

My second objection is about philosophical difficulties with the concept of randomness. I've gone into that in detail in:

My third objection is the topic of this article. Starting with Lewontin (1970), the term "natural selection" has been frequently characterised as being an "algorithm". Of course, there is a kind of abstract algorithm associated with Darwinism - involving the iterative application of both production and elimination. The issue described here is concerned with using the term "natural selection" to characterize this algorithm.

The term "natural selection" suggests that nature is choosing something. We can imagine nature choosing which organisms live, which die, which organisms are selected as mates, and which organisms are selected as victims. However all these usages omit the production of variation.

The Darwinian algorithm includes the production of variation. This is mutation. I think that mutation should be a different category from selection. It is possible - at a stretch - to characterize mutation as a form of choice: nature "chooses" which mutation to use. However, the biggest problem with combining them is that this conceptually muddles together two very different things.

IMO, even those who swallow the first two items I am objecting about should reject using the term "natural selection" to describe the Darwinian algorithm. It includes something else besides selection - namely the production of variation. Variation and selection should be conceptually separated out. If you are going to use "natural selection" to describe an algorithm involving selection and mutation, you've polluted the term for selection alone.

No comments:

Post a Comment