Wednesday, August 1, 2007

Dawkins #6: One Goal to Rule Them All

[ This post is number 6 of a series. They will be of interest mainly to those who've read Richard Dawkins' book, The Selfish Gene. I recommend they be read in order. ]

Richard Dawkins ascribes two different goals to genes: "survival" and "increase". In previous posts, I pointed out that these produce different models, which in turn predict different results. I expanded the "goal" metaphor out into literal terms, and showed that if this "thought problem" is worked out with genes taken on a "snapshot" basis, their goal is clearly seen as survival.

In the last two posts I showed that "increase goals" will not work without a well-defined condition or "target". Pure "increase goals" had several problems, the worst of which showed up when I expanded the "goal" metaphor into literal terms. Pure increase goals require us to believe genes have an intent, not just metaphorically, but in literal fact. Unlike limited increase goals, where we can justify our statements that genes have those goals by pointing to the actual effect of the genes in the material world, pure increase goals require us to make statements about what genes "would" do if they "could". That takes us from metaphor into fantasy.

It remains to look at "increase goals" with well-defined conditions. Testable (if not necessarily realistic) examples might be "Increase the germline gene count if foraging produces more than the target level of N calories per average day, otherwise regulate the germline gene count to maintain or restore that level" or "Maintain a germline gene count which keeps the species at 9% of the epipelagic Antarctic biomass."

In one sense, these are not goals of "increase" as stated by Dawkins, but "maintenance goals". But I will give any goal based on germline gene frequency, honorary status as an "increase goal". More important than whether they track Dawkins language (Extended Phenotype, p. 84) closely is whether they make a satisfactory alternative to the survival goal. That's not a question that should be answered with a semantic quibble.

The basic problem with the limited versions of "increase goals" is best shown by their confusion in the face of outlaw genes. Assume that genes do have gene-frequency goals which, at least at points, involve maintaining a certain population level. What's to stop a new mutation that cheats? At first glance, it would seem that if the genome of a species has settled on a rule that says "stop increasing gene count at level X", any gene that ignores it will have a selection advantage over its law-abiding fellow genes.

I won't dwell on the practical side of this problem: how the law-abiding genes in a genome can "police" these outlaws. (Dawkins deals with "outlaws" in The Extended Phenotype, ch. 8.) I'm more interested in the definitional problem which the practical issues highlight.

"Population outlaws" are outlaws because they don't share the limited increase goal which is supposed to be the general goal of natural selection. But if a limited increase goal is going to be the general goal of natural selection, it needs to describe the behavior of law-abiding genes and outlaw genes alike.

The general goal of natural selection has to be a law not in the sense of the law against train robbery, but in the sense of the laws of physics. Train robbers can decide to ignore the criminal code, but they've got to shoot straight or suffer the consequences, the same as any lawman. The general goal of natural selection must impose itself on outlaw genes the way the laws of physics impose themselves on gunmen.

Ideally, in fact, the general goal of natural selection will tell us why outlaw genes do what they do. A survival goal would seem to describe "population outlaws" well: All genes try to increase their probability of survival. Population outlaws are genes that attempt to survive at the expense of the other genes in a genome by cheating on their population discipline.

The existence of "population outlaws" shows that a limited "increase goal" cannot stand in even as the law governing the genome of a single species. Limited increase goals are even worse when applied to the genes outside of a genome. Even if there were no such thing as "population outlaws", a limited increase goal would describe the behavior of, at most, a narrow range of species.

A pure "increase goal" is general enough to cover all life, but we were forced to rule it out because it was incoherent. We have to rule out impure, or limited increase goals precisely because they are limited and specific.

With the elimination of both limited and unlimited gene-frequency goals, my argument that survival is the general goal of natural selection is complete. "Increase" goals and gene-frequency targets are tactics employed by certain genes at certain times in the battle for survival. Gene-frequency targets themselves evolve and go extinct. The goal of survival will not go extinct before life itself does.

Saturday, July 28, 2007

Dawkins #5: "As Much As Possible" is Vacuous

[ This post is number 5 of a series. They will be of interest mainly to those who've read Richard Dawkins' book, The Selfish Gene. I recommend they be read in order. ]

Richard Dawkins ascribes two different goals to genes: "survival" and "increase". In previous posts, I pointed out that these produce different models, which in turn predict different results. I expanded the "goal" metaphor out into literal terms, and showed that if this "thought problem" is worked out with genes taken on a "snapshot" basis, their goal is clearly seen as survival.

In the last post I showed that pure "increase goals" (goals of increase without condition or limit) have several serious problems. By far the worst is that when the pure "increase goal" metaphor is expanded into literal terms, it's seen to require genes to have states of mind not just metaphorically, but literally. This makes pure "increase goals" incoherent, and that's a show-stopper.

So far I have not discussed "increase goals" with some kind of condition or limit. These can be coherent, and a number of them can be seriously meaningful. At first, a goal of "increase as much as possible" may seem to be an obvious choice among them.

But "increase as much as possible" is deceptive. It probably works, but that's because it's not really an "increase goal". "Do X as much as possible" means "Do X subject to condition Y." "Increase as much as possible" means "Increase, subject to condition Y." The only other goal Dawkins mentions is survival and "as much as possible consistent with survival" is the natural reading of the meaning hidden inside "as much as possible."

If we translate this goal into terms easier to model mathematically, we get "First maximize the probability of survival, and having done that, maximize the number of germline genes." Survival is a demanding goal. Since we are talking about our goal as the candidate for the general goal of natural selection, our gene will be battling for survival against genes selected over eons for their skill at surviving, including their skill in surviving at our gene's expense whenever it's in the slightest bit useful. Unlike increase, which must happen in units at minimum one amino acid in size, survival is a probability. Arbitrarily tiny advantages built up over time can make a difference.

The issue may be clearest if we look at the goal of "first maximize survival, then decrease the count of germlines genes as much as possible". This goal is coherent, and we don't have to worry about a gene with this goal decreasing itself to extinction. The first priority for the gene's efforts is survival. No effort will be available for any second priority.

If the gene's goal is "first maximize survival and then do Z as much as possible", Z can in fact be nonsensical. For example, Z might be "attend Rolling Stones concerts" and the fact that genes don't know how to buy tickets will make no difference -- they'll never have the chance to worry about it. The "do Z as much as possible" clause is completely vacuous.

Perhaps the reader never had any taste for slippery language like "as much as possible", and has been wondering why I'm going on about it. If so, I apologize. Goals like "increase the germline gene count until you've used 9.7601% of the Antarctic biomass, then sustain that biomass level" are both coherent and substantive. Unlike the incoherent and vacuous versions of the "increase goal" that I've been discussing, which can be dismissed a priori, specific gene-frequency goals are likely to reward empirical investigation. Even so, I think they are inferior to survival as a choice for the general goal of natural selection, and in my next post, I'll show why.

Wednesday, July 25, 2007

Dawkins #4: Pure Increase is Incoherent

[ This post is number 4 of a series. They will be of interest mainly to those who've read Richard Dawkins' book, The Selfish Gene. I recommend they be read in order. ]

Richard Dawkins ascribes two different goals to genes: "survival" and "increase". In previous posts, I pointed out that these produce different models, which in turn predict different results. I expanded the "goal" metaphor out into literal terms, and showed that if this "thought problem" is worked out with genes taken on a "snapshot" basis, their goal is clearly seen as survival.

But I have not shown what happens if we run through the "thought problem" using methods of sampling that take into account the frequency of genes. And while I showed that "survival" works as a goal, I did not show that "increase", that is, increasing the number of germline genes, does not work even better as a goal.

If we look at multiple "snapshots" of genes from a frequency point of view, then the genes might be seen as having a goal of appearing in as many snapshots as possible. In this case we might be able to work through the "thought problem" in a way that justifies us in saying they have a goal of increasing their frequency in the sampling.

Devising an sampling method that's right for calculating gene frequency in the "thought problem" is not easy. There are genes deep in the ocean and deep inside the earth of which we know very little. And what of genes that may be traveling on meteorites, or elsewhere in the solar system? The sampling method needs to make sense as something that genes would seek to optimize. The sampling won't make sense as a "goal" for genes if it is biased in favor of genetic material that's easy for humans to find, or even in favor of genetic material humans know about.

Since we are dealing with a "thought problem", the sampling method can be one difficult or even impossible to carry out in practice, but that still leaves issues of definition. And if we can't measure what we say genes are optimizing, we can't test our assertion that they are selfish.

Picking a sampling method is not an issue for the "survival" goal. All the survival goal asserts, is that any gene having survived many generations of intense competition to end up in a snapshot can be said to have the "goal" of ending up in the snapshot. The logic of the survival goal is on a snapshot-by-snapshot, gene-by-gene basis, and sampling issues do not come up. Only if we are using frequency as determined by a sampling, which we have to if we want to define an "increase goal", does it matter whether the sampling is arbitrary, or biased.

The sampling issue may be solvable, at least in principle, so I will assume that it has been solved and move on. Increase goals have other, more serious, problems.

The pure form of the "increase goal" is incoherent. Nothing that comes in finite pieces can increase indefinitely. Every increase in the number of germline cells must be of a minimum size, certainly no smaller than the smallest possible amino acid. Long before the entire mass of the solar system is used up by germline genes in a single locus and for a single phenotype, increase must come to a halt.

At first glance, this may look like an unfair quibble. Take the statement that "Charles Ponzi's goal was to increase the amount of money in his Security Exchange Company." Ponzi's scheme itself may not have been sound or coherent, but it is certainly coherent (and probably true) to say that unlimited increase was Ponzi's goal. It is not incoherent to state that someone has an incoherent goal.

But there's a crucial difference between Charles Ponzi and a gene. A gene's goal must be coherent in terms of our "thought problem". That is, the "goal" ascribed to a gene must still make sense when the metaphor is translated into literal terms. Genes do not have minds in which to form schemes, and if translation of a goal into literal terms requires genes to have a mind of their own, then saying that genes have that goal is incoherent.

We are entitled to decide that Charles Ponzi had certain plans even if he never fully carried them out, in fact even if it would have been impossible for him to carry them out. We can do this because we believe Charles Ponzi was, in literal fact, a conscious being with a mind in which such a plan could exist. Based on Ponzi's actions and statements we can ascribe unrealized and even impossible goals to him.

We are not entitled to ascribe any "goal" to a gene if it means that we are stating that the goal was only in the gene's "mind". The gene has to actually carry out its goal, or we have no basis to say it was the gene's goal. We can't say that genes have the goal of unlimited increase, because we could never see them actually carry this goal out. Humans can be said to have impossible or unrealized goals, but genes cannot. Genes don't think and we can't go beyond metaphor and use the language of "goal" in a way that requires genes, literally and as a matter of actual fact, to have minds of their own.

I did not originally intend to use Charles Ponzi for my example. I'd wanted to use the statement "Richard Dawkins' goal is to increase the number of atheists" instead. But I couldn't, and the reason I could not is relevant. It's very difficult to come up with an example of a "pure increase goal" that doesn't involve insanity or delusion, because in the real world most such goals hit limits.

Dawkins, once he had converted every last man, woman, and child to atheism, would stop. (At least I think he would.) I assume, for example, that he would not go on to try to create a population explosion in order to further add to the count of atheists. Ponzi was the kind of wild optimist who does not think about potential obstacles. Deluded people like Ponzi can more easily be spoken of as having pure increase goals. A real gene with anything close to a pure increase strategy, in practice, would be likely to drive itself to extinction by overexploiting the resources it needs to survive, a fate not unlike that of Ponzi's scheme.

To be coherent, an "increase goal" must be impure, that is to say, limited in some way. In the next post, I'll comment on some "increase goals" which are limited and coherent, and explain why they also don't work as alternatives to the survival goal in explaining natural selection in general.

Saturday, July 21, 2007

Dawkins #3: Do Genes really Care?

[ This post is number 3 of a series. They will be of interest mainly to those who've read Richard Dawkins' book, The Selfish Gene. I recommend they be read in order. ]

In previous posts I identified two distinct goals Dawkins ascribes to genes: "survival" and "increase". I pointed out that these produce different models, models that will often predict different results.

I've been speaking of genes which "strive" to achieve "goals", and which "care" about outcomes. In this I've followed Dawkins, who finds it a helpful metaphor. We're dealing with foundations, however, so we need to be careful. Neither Dawkins or I imagine that genes are really conscious, much less that they actually have goals. It's time to slow the pace and speak more literally.

Let's look at a snapshot with a single gene in it. In the rare case our snapshot is of a recent mutation, we toss it in the trash and select another. That guarantees that our gene is the result of a long process of natural selection. Our gene, typically as part of a complexly organized team of genes, has passed a long series of stringent tests.

For example, if it's a gene in a cuckoo, a bird that lays its eggs in the nests of other birds and whose nestlings must trick the other birds into rearing them, it "is descended from a long line of ancestral cuckoo nestlings, every single one of who must have succeeded in manipulating its foster-parent. Any cuckoo nestling that lost its hold, even momentarily, over its host would have died as a result." (Dawkins, The Selfish Gene, 30th Anniversary Edition, 2006, p. 250). None of the genes in those nestlings who failed momentarily and died would, of course, have survived to be in our snapshot.

The gene's success is not because it, or anything or anyone else, planned for it. All the gene did was produce the amino acids which its chemistry forced it to produce. Fortunately for the gene, these were animo acids which helped produce successful cuckoos, generation after generation.

The gene in our snapshot originated long ago in a mutation, a random, purposeless accident. There were many other mutations, random like the one which created our gene, and almost all of them were disastrous. Those other mutated genes usually killed any cuckoo unlucky enough to inherit them, and disappeared along with the birds they killed. Because it is in our snapshot, our gene was one of the very rare ones which did not result in failure. It somehow survived alongside other, already successful germlines. Our gene may have added useful variation to the genes already in the cuckoo gene pool and found a comfortable niche among them. It may have survived by pushing other genes to extinction. Because it is in our snapshot, we know that whatever it took to survive, our gene did it.

At no point did our gene, the cuckoo, any other gene or anything or anybody else think this out, form any plans or set any goals. The gene in our snapshot was the result of a long series of accidents, each one by itself completely without purpose or meaning.

But over the millenia, this random series of separately meaningless events became something our brains see as having a purpose. This cuckoo gene produced amino acids which resulted in an intricate mix of behaviors and traits which resulted in the gene being in our snapshot. As for any of its comrade-genes that during that long period produced even slightly less successful amino acids, they are nowhere to be seen.

So it feels natural to us to speak of the gene as having undertaken the project of surviving so that it could appear in our snapshot. The length of the process which brought it into our snapshot makes it natural for us to speak of the gene "striving", of it "caring" about whether it wound up in our snapshot or not, and of it having "survival" as its "goal". The complexity of that process makes it natural for us to speak of the gene "planning" that success. The idea of genes which "care", and are "selfish" or "altruistic" comes naturally to our minds and lets the non-linear parts of our brain assist in putting together a picture of natural selection. But we need to remember this anthropomorphic language is a metaphor, and we must always be prepared to translate the metaphor into the unconscious, random processes which are the actuality of natural selection.

The reader who's been following me carefully will have noticed that "snapshot" was a crucial choice of words in the above. If we think of our selection of a gene for our thought problem as a "snapshot", the "goal" that falls out of following Dawkins logic through the "thought problem" is "survival", instead of an increase of the number of germ-line cells.

"Snapshotting" is the simplest way to picture sampling genes for our thought problems, and the above description shows it's a natural way of looking at the problem. But while simplicity and naturalness can and should give the feeling that we are on the right track, they are not proofs. Might not other ways of sampling genes for our thought problem result in the "goal" appearing to be an increase in the germ-line population?

If we view our sampling not as a "snapshot", but as some kind of "random sample", concepts of gene frequency enter into the picture. As natural and simple as the snapshot/survival viewpoint is, might not the viewpoint based on "random samples" be more revealing? And if so, wouldn't that support seeing the gene's goal as "increase"? In my next post, I'll examine that issue.

Wednesday, July 18, 2007

Dawkins #2: Unfair to Elephants?

[ This post is number 2 of a series. They will be of interest mainly to those who've read Richard Dawkins' book, The Selfish Gene. I recommend they be read in order. ]

From Richard Dawkins' Extended Phenotype, pp. 254-255 [references removed]:
The many-celled body is a machine for the production of single-celled propagules. Large bodies, like elephants, are best seen as heavy plant and machinery, a temporary resource drain, invested in so as to improve later propagule production. In a sense the germ-line would 'like' to reduce capital investment in heavy machinery, reduce the number of cell divisions in the growth part of the cycle, so as to reduce the interval between recurrence of the reproduction part of the cycle. But this recurrence interval has an optimal length which is different for different ways of life. Genes that caused elephants to reproduce when too young and small would propagate themselves less efficiently than alleles tending to produce an optimal recurrence interval. The optimal recurrence interval for genes that happen to find themselves in elephant gene-pools is much longer than the optimal recurrence interval for genes in mouse gene-pools. In the elephant case, more capital investment is required to be laid down before returns on investment are sought. A protozoan largely dispenses with the growth phase of the cycle altogether, and its cell divisions are all 'reproductive' cell divisions.

The delights of reading Dawkins include frequent passages where he submits his own views to unsparing examination. Certainly it seems here that the proboscis touches on a problem of pachydermic proportions, one that if allowed to get out of control could trample his theory into the dirt.

In my previous post I noted that while Richard Dawkins usually states that genes seek to "survive", sometimes he describes their goal as being to "increase" their numbers. I also noted that in applying his theory he reverses this preference -- in his examples he usually has genes striving to increase the number of germline genes. I pointed out that "survival" and "increase" are not the same thing. They produce different mathematical models and, in real life, they will often predict different results. Where the models predict different results, it cannot be the case that genes are optimizing both for "increase" in numbers and "survival". Of the two models, at least one must be wrong.

Now why, if the goal is to increase germline gene count, does the elephant propagule produce an animal weighing several tons and almost entirely made up of gene-copies which are reproductive dead ends? Why does it keep that animal alive for decades? Why do the few DNA copies that reproduce, do so at such a leisurely pace? The gestation period of an elephant is over a year.

If we regard the "goal" of the gene as increasing its numbers, the elephant is a evolutionary failure of comic proportion, and it takes no very intelligent designer to see that a "fix" for the elephant just isn't in the cards. Yet elephants are clearly the product of a long process of natural selection. If it is the goal of germline genes to increase their population, nobody told the elephant or his ancestors.

But what if Dawkins' elephant genes are in fact even more selfish than he claims? What if an elephant germline gene just doesn't give a hoot whether there are any other copies of itself, so long as it survives? From this point of view, elephants make sense. An elephant is the way for elephant germline DNA to maximize its chances of being around a hundred, a thousand or a million years from now. It doesn't care if it's present in one copy or a billion, so long as it survives.

From a "survival" point of view, the elephant biomass expended in dead-end cells is no loss for the elephant's germline DNA. It costs an individual gene-copy in the elephant germline nothing to take over as much dead-end biomass as it finds useful for increasing its chances of survival. Neither does the germline gene-copy care how long reproduction takes. It seeks only to maximize the chances that it will still be around when they rake in all the chips.

If we make Dawkins' genes more selfish, so that every individual gene-copy is out for itself and nobody else, not even it's exact "brother" copies, elephants are no longer a paradox. Elephants no longer look like losers in comparison with Antartic krill. The krill germline gene uses fast, easy, high-risk reproduction to increase its survival chances, and the germline cells don't pause to take on a lot of extra biomass. Elephant genes are risk-averse. They seek out a slower, surer path, and they find a lot of non-germline biomass can be useful in the process. If we take the goal as "survival" and not "increase", and the "unit of selfishness" as the individual germline gene-copy, elephants are no problem to explain.

I've been following Dawkins in speaking of the "goals" of genes, what they "care" about, and what their "plans" are. This is a metaphor. Neither Dawkins or I imagine it as anything except a useful fiction to allow our intuitions to assist our brains in understanding natural selection. But metaphors and intuitions can mislead. We need to take this metaphor apart and look at the reality in more detailed and material terms. I'll do this in my next post.

Dawkins #1: Are His Genes Selfish Enough?

[ This post in number 1 of a series. They will be of interest mainly to those who've read Richard Dawkins' book, The Selfish Gene. I recommend they be read in order. ]

In The Selfish Gene, his fascinating bestseller, Richard Dawkins advances the theory that natural selection is not the result of competition among species. Instead, he says, genes compete with each other, and the creation, extinction and evolution of species is an indirect result of the competition of these "selfish" genes.

I am running an extreme risk of presumption (I'm not a biologist), but I'm going to propose a "friendly amendment" to Dawkins argument, one that more closely defines the "goal" of the "selfish" gene. It's my hope that it increases the explanatory power of Dawkins' brand of neo-Darwinism, solves a few mysteries and adds precision to his argument. First, I want to very briefly sketch that part of Dawkins' argument most relevant to mine.

Again, Dawkins identifies the gene, not the species, as the "unit of selection". He's not splitting hairs, but distinguishing real differences. The "selfish gene" explains a lot of behavior that from the species-selection point of view is mysterious. For example, as I write this I am looking at two nests put up as part of an effort to restore bluebirds to their original habitat here in Berkshire County, Massachusetts. Bluebirds compete for these nests with tree swallows, who are small enough to fit nicely into the nest, and large enough to win most fights with bluebirds.

So far, it sounds like the old "survival of the fittest species" story and bad news for bluebirds. That's why two bluebird nests were put close together. A tree swallow family may get the first nest, but let another tree swallow family show up and the first tree swallow family will chase off the newcomers. Tree swallows don't want competition for the airborne insects they feed on. Bluebirds feed on ground-dwelling insects, so tree swallows don't compete with them for food and don't mind if they nest nearby. It is easy to interpret the tree swallow's behavior as promoting the survival of its own genes, and very difficult to see in it any concern for the survival of the larger gene pool of its species.

The difficulty that I found in Dawkins is this: he ascribes two different goals to his "selfish genes". Most of the time they are said to be maximizing "survival" (The Extended Phenotype, p. 233) but sometimes they are described as striving to "increase" (Phenotype, p. 84). When he turns from theoretical statements to examples, Dawkins seems to reverse this preference. The genes in his examples usually strive to increase the number of "germline" genes. (Germline genes are those actually in reproductive lineages.)

Just as with species vs. gene, survival vs. increase is a distinction with a difference. To make an analogy with investment, some investments are best for capital preservation (survival), while others promise higher returns (increase) in exchange for higher risk. Under normal economic conditions, no investment is optimal for both goals. A mathematical model optimizing for the probability of survival of a gene will not be the same as one seeking to increase gene count. Which of the two models genes obey should be testable, but in advance of that I believe we can make a very strong guess.

Closely related to the "increase" vs. "survival" question, is the question of the exact nature of "the unit of selfishness". It's the "gene", but "the gene" can either be the collection of all genes with the same phenotype at the same place in the chromosome (which I'll call the gene-type), or one individual member of that collection (which I'll call the gene-copy). While the English language is ambiguous, Dawkins is not. His examples and his more detailed explications clearly show that he thinks the gene-type is what is "selfish", and that gene-copies subordinate their individual interests to the gene-type.

It's possible Dawkins sees the interests of the individual gene-copy and its gene-type as identical in practice. They are not. The individual germline gene-copy has no interest in increasing the germline population count of its gene-type -- it's only interested in its own survival. If survival is really the goal, Dawkins is not allowing individual gene-copies to be selfish enough. The gene-copy has no reason to "care" about identical copies of itself elsewhere, unless it can use them as a means to the end of its own survival.

Of course, DNA and genes aren't conscious and don't "care", "plan" or have "goals", except within the framework of a metaphor I follow Dawkins in finding useful. (In a later post, I'll "lift the hood" on this metaphor.) In the next posts, I will show that in the struggle between these two goals, "survival" and "increase", to describe natural selection according to Dawkins, one emerges as clearly the fittest.