Monday, July 5, 2010

Fodor, Godfrey-Smith and Natural Selection

Fodor has been arguing that the theory of natural selection is flawed, and many, including Godfrey-Smith, have claimed he was wrong. I'll respond to part of that response, which is representative I believe of many responses to Fodor.

In Peter Godfrey-Smith's example, trait T is that of having a camoflaged appearance, T* is an irrelevant property that is biologically linked to T ('linked' means that, due to practical biological constraints, the property passes together - perhaps they are adjoining on the same gene, etc.), and U is the trait of having an easily seen appearance, while U* is an irrelevant property biologically linked to U. Then, Godfrey-Smith's argument goes, we can examine the life of an individual, and see that it dies as a consequence of being seen - hence U was negatively selected for, and T positively selected for. Correspondingly, we can look at the population as a whole and see that many deaths occurred as a consequence of having trait U, and hence U was negatively selected for, while T was selected for, and T* and U* were just along for the ride. If this is correct, then Fodor's argument is flawed.

Instead, I will argue that this example is flawed. Consider the following case: There are two further traits that are biologically linked to both T and T*: T1 and T2. T1 is the property of making unnecessary noise - which makes you more likely to be heard by a predator. T2 is the property of being quick. Assume that it is the case that the noise T1 makes is detrimental in the same amount as T2 is beneficial: T1 makes you likely to be heard, but T2 makes you faster, so you have a better chance of escaping. So, when looking at just properties T1, T2, they 'cancel each other out' as far as natural selection is concerned. So, this apparently does not change the picture - T was still selected for, T* is along for the ride, and T1 and T2 are 'noise', not affecting things on the bottom line.

However, we can look at this in a different way. Imagine that the beneficial effect of T is the opposite of the negative effect of T1. So we can summarize the combination of T and T1 as having no net change to how likely you are to be detected by a predator: You are less likely to be seen due to T, but more likely to be heard due to T1, and these effects balance each other (remember we are assuming that the amounts are in the right magnitudes - which means the amount of predators, whether they use vision or hearing mostly, how effective they are, etc. etc.). That leaves T2, which is beneficial - being faster helps avoid predators when they do detect you - so the cluster of traits T, T*, T1, T2 will be selected for. So T's frequency will rise, as in the original example. However, in this interpretation, T does not seem to be selected for.

Arguably the latter interpretation makes more sense - T and T1 are two aspects of 'being detected', while earlier we considered T1 and T2 together only because we introduced them at a later time than T and T*. But the order of introduction makes no difference of course. Likewise, if traits T and T1 appeared together historically at exactly the same time, that does not matter to the present. From the present's perspective, it is the same as if T1 and T2 appeared together: In both cases, the change is a net zero, as two new traits appear together and they cancel each other out. So, despite there being various possible interpretations here, it seems we should either conclude that T is not being selected for, as per the second interpretation (which has perhaps a slightly more appealing intuition), or that there is something wrong with both interpretations.

Indeed, there is something wrong here. The problem is that statements such as 'T was selected for' implicitly assume that T is 'independent' of other traits, in the sense that natural selection can 'pick it out' by itself. Indeed, if T were not biologically linked to any other trait, then the frequency of T should rise over time, statistically speaking. However, if traits are linked as we have assumed above, then that is no longer true. All natural selection can 'work on', in the example above, is the combination of T, T*, T1, T2. When we had just T and T*, it seemed easy to say that T was being selected for. But when T and T2 are both beneficial, as we have seen finding the proper interpretation of what is selected for is less clear. From the perspective of natural selection, the combination of T, T*, T1 and T2 is beneficial, so it's frequency will tend to rise. That is the only safe claim we can make here, that does not depend on interpretation.

Still, perhaps we can say more, with more data? Imagine that the traits did *not* perfectly balance each other out, as we assumed before. That seems the more likely case in general. In that case, we might argue that one of the traits is indeed being selected for. However, if it turns out that it is T2 being selected for, and not T, then we still have a problem with the original argument - which arrived at the false conclusion that T was being selected for. Note that the issue is not that we simply didn't know enough about T itself. The problem is that, if we are not even aware of the existence of traits T1 and T2 and their link to T, then no amount of knowledge about the usefulness of T will avail us. T, considered by itself, seems useful. But only by knowing about all the relevant genes can we find out that T in fact is not selected for.

Let's make that more explicit. T is useful in itself - as we can see from counterfactuals. However, it is linked to trait T2 which effectively cancels it out, and also to trait T1 which is also useful, in fact slightly more useful than T. In other words, a counterfactuals argument on T itself leads to a faulty conclusion. It simply does not follow that, if counterfactuals show a trait to be useful, then it is selected for. Being actually selected for depends on something more.

So, there are two possibilities here:

  • There is a fact of the matter about which genes are selected for. But, to say anything with certainty about that, we do not just need a lot of data about the usefulness of the genes we are aware of, but all other genes that are biologically linked to it. Otherwise, we may be wrong, and we cannot at present even estimate the chance of our being wrong (perhaps once we know a lot more about biology, we can answer that).
  • There is no fact of the matter about which genes are selected for, if there are linked properties that have usefulness levels that are too close together. There may also be more complicated forms of linking than that we have considered here, that make things even worse.