A few months ago Bjørn Østman and I were having a discussion about the degree to which selection-for must be thought of as an abstraction rather than a reality, and I presented the following (admittedly somewhat over-the-top) example:
Let's say in the year 5673 BC, one peacock was saved from certain death when his tail got caught in some branches, and that slowed him down enough that an impending avalanche that would otherwise have hit him now missed him -- whereas his brother with a slightly smaller tail continued unimpeded and was killed in the avalanche. Do we now have to include in our analysis a 10-10% contribution to the selection of the tail by "stops peacocks from walking into the path of avalanches"? If not, why not? (Without invoking agency, that is)
Don't get too distracted by my use of the peacock's tail -- that just happened to be the trait we were talking about at the time. Anyway, I want to flesh out my thoughts on this somewhat in light of the present discussion regarding neutralism.
First, a brief digression on two major schools of thought in probability theory. Let's say I roll a die three times, and I get the sequence 1-3-5. After I have finished the experiment I say, "Now what are the odds I would get that sequence?" A frequentist answer would be 216:1, because, assuming a fair die, if I perform this experiment enough times, I should get the sequence 1-3-5 approximately that often. An extreme Bayesian answer would be 100% -- as long as you were asking the question after performing the experiment. After all, we know that's what I rolled; it can no longer be anything else. Given the state of information we have now, the odds are 100%.
The Bayesian interpretation is not very useful there, but it's quite useful in regards to things like cancer screening. And nobody is so extreme of a Bayesian that they would stubbornly argue for the position in the previous paragraph. Anyway, you can get piles of information on this by Googling, and probably most people who have read this far are already more or less familiar with the concept, so I will not expound on it further.
If we wanted to take the really extremely stubbornly uber-Bayesian interpretation of natural selection, we would have to argue that every single extant trait is adaptive. After all, the a posteriori probability of evolution having taken the path that it has is 100%. Given the information we have now, every trait which exists in the present has perfect inclusive fitness, and every trait which is no longer extant is perfectly unfit.
I am making this rather stubborn and perhaps overly philosophical point in order to drive home the idea that any useful description of selection must inherently be an abstraction. It's not just that we need to use an abstraction in order to analyze it in practice; even in principle, one cannot say anything meaningful about natural selection without taking a frequentist approach, abstracting out the actual environment in which a given organism lives, reproduces and dies, in favor of an idealized, "typical" environment analogous to an idealized fair die roll. Without taking this important step, the most meaningful thing we can say about evolution is, "What happened happened."
(As a brief aside, I think the primary mistake made by Fodor and Piattelli-Palmarini was their failure to take that next step from the stubborn philosophically pure position to recognizing that the whole enterprise can be handily salvaged by just drawing a few abstractions. In the spirit of full disclosure, I must admit I haven't read their book, but I've read critiques and defenses of it, and unless there's some amazing revelatory concept in the book, this seems to be the rather obvious problem with their entire argument.)
I think this already somewhat undermines the dichotomy of the selectionist/neutralist debate, since both are exposed as abstractions which are (again, by necessity) pretty far removed from the most literal reality of evolution. Because each organism (or each allele if you want to take the gene-centric view) exists in its own actual environment rather than being repeatedly tried in some ideal environment, to even pose the question of whether a particular trait is adaptive or not by necessity admits a sliver of teleology-esque reasoning into your model. By no means am I objecting to this! But it is worth observing.
Of course we can still ask which model is more useful, even if it turns out that the answer is "it varies" -- so there is meaningful discussion to be had. As a lay person, I'm not going to attempt to answer that. But what I will do is point out one problem with the intensely1 neutralist view espoused by folks like Larry Moran.
Consider the example of the rhinoceros' horn discussed in the comments at Sandwalk. Let's say for sake of argument that, in comparison to each other, the Indian rhino's one horn has no selective advantage over the African rhino's two, and vice-versa. Let's go one step further and assume that sexual selection doesn't even play a role, and that if you could magically give some fraction of Indian rhinos an extra horn, or take away one from some African rhinos, it wouldn't affect the afflicted rhinos' inclusive fitness one iota. Does that necessarily mean that "neutral" is the best way to describe the trait(s)? I'm not so sure.
This is a pretty unrealistic hypothetical, but run with me here for a moment: Let's say the common ancestor of the Indian and African rhinos had a sort of bumpy ridge where the nose is. Two sub-populations become separated, and (here's where the unrealistic part comes in) a single transposition event in one sub-population causes the ridge to grow into a single horn, and a single transposition event in the other causes the ridge to grow into two horns. In each sub-population, only one of those two possible mutations is available, and in both cases the transposition event is so unlikely that we would not expect it to occur again. Assume further that horns -- whether one or two -- have a tremendous selective advantage over the bumpy ridge.
How can one argue with a straight face that either trait is "neutral"? Both mutations are tremendously adaptive over and above the available alternatives in the gene pool. If by the time the two sub-populations are reunited they can no longer produce viable offspring, then we now have two species, each with a mutually exclusive trait that is no better than the other's, but both of which are clearly adaptive. Genetic drift doesn't even come into it, except in the trivial sense that all novel alleles have to arise via random mutation.
I am again being somewhat stubborn and over-the-top, I think, but the point I am making is that in the long term and in a large enough population, every single trait that goes to fixation is adaptive, in the sense that it is superior to the immediately available alternatives present in the gene pool. Now the manifestations of those immediately available alternatives are influenced by effects such as genetic drift, of course. And it's probably fair to argue that it is not uncommon to have multiple available alleles in a population which are (within the limits of our abstraction) no more adaptive than each other, and in those cases the neutralist account would indeed be the most sensible. What I am saying, though, is that just because two potential traits of an organism have no particular selective advantage over one another if both had been readily available in the same population at the same time, that doesn't at all invalidate a selectionist account.
This is all probably hopelessly redundant, covered territory in evolutionary biology, and as a layman I should probably just not blather on about it. But oh well, those are my thoughts on it.
1I struggled with the wording here... I initially had "extreme" rather than "intensely", but I felt that was unnecessarily -- and unintentionally! -- pejorative. Given our human penchant for promiscuous application of teleology, I think the "extreme" (not in a pejorative sense!) neutralist position is a very necessary part of the tapestry of evolutionary inquiry. Without neutralists like Moran, the adaptationist model could easily get completely out of control. Even great minds like Dawkins, who has been entertainingly referred to as an "uber-adaptationist", have at times seemed a bit too eager to prematurely embrace an adaptationist account. All deserved respect to Larry Moran here, by all means!