Evolution for Everyone

Group selection was decisively rejected on theoretical grounds, according to the patriotic history of individual selection theory. Richard Dawkins declared in 1982 that group selection had “soaked up more theoretical ingenuity than its biological interest warrants” and compared further inquiry to the futile search for a perpetual motion machine. Richard Alexander stated in 1987 that “a great deal of convincing theory suggests that any such view [the beneficence that evolves by group selection] will eventually be judged false” (see T&R V for more).

Blustery statements like these have a bad smell about them. In truth, the models that led to the rejection of group selection were little more than sketches on the backs of napkins. Models supporting the plausibility of group selection began to appear as early as the 1970’s, but that did not stop the patriotic propaganda machine from rolling onward.

Before I support my claim, a word about theoretical modeling is in order. Some models attempt to approximate reality in as much detail as possible, but most evolutionary models are more like caricatures. Just as a good artist can capture the likeness of a person with a few lines, a good modeler attempts to capture the essence of a problem such as group selection with the fewest possible assumptions. This kind of model tells us what can happen, given the assumptions of the model. It says nothing about what might happen, given other assumptions. Many models are required to support the sweeping claim that “between-group selection is invariably weak, compared to within-group selection.”

Against this background, let’s take a look at the most influential model that led to the rejection of group selection, which is affectionately known as the haystack model. The year was 1964. Two years earlier, V.C. Wynne-Edwards had published his book Animal Dispersion in Relation to Social Behavior, which made sweeping claims about group selection (see T&R III). One year earlier, a British graduate student named William D. Hamilton published a note in the journal American Naturalist outlining how altruism can evolve among individuals that share the same genes. John Maynard Smith, one of the top evolutionists in the UK, was familiar with Hamilton’s work and thought that it provided an alternative to group selection. He therefore wrote a critique of Wynne-Edwards in a letter to the prestigious journal Nature titled “Group Selection and Kin Selection”. The letter was only three pages long and included a brief model, which was truly little more than a sketch on the back of a napkin.

Like Dr. Seuss, Maynard Smith asked the reader to imagine a cartoon world in which a species of mouse lives entirely in haystacks. Each haystack is colonized by a single fertilized female, whose progeny reproduce for a number of generations. At some point, all of the mice from all of the haystacks disperse, mate on a population-wide basis, and the cycle is repeated for a new set of haystacks. There are two genes in the population, one coding for aggressive behavior and the other coding for docile behavior. Aggressive mice outcompete docile mice within each haystack, but they also decrease the growth rate of the group. Specifically, every haystack is colonized by four genes; two from the female and two from the male who fertilized her. Maynard Smith assumed that if one or more of these genes is the aggressive gene, then the aggressive gene completely replaces the docile gene by the time dispersal from the haystack occurs. When a haystack is colonized by four docile genes, however, more mice are produced by the time dispersal occurs than from the other haystacks. Imagine those docile mice streaming away from their haystacks singing “Kumbaya”, while the aggressive mice limp forth from their haystacks nursing their wounds!

Let me praise the haystack model before presenting and critiquing the results. The essence of the group selection controversy — what I called “the original problem” in T&R II — is whether a trait that is “for the good of the group” can evolve in the total population, despite being selectively disadvantageous within groups. The haystack model elegantly captures the essence of the original problem. In this sense, it is a good cartoon. Docility is for the good of the group. Aggressiveness is selectively advantageous within each group. Which level of selection prevails?

Maynard Smith concluded on the basis of his model that aggressiveness almost always wins. Group selection is an evolutionary force, but it just can’t prevail against the opposing force of selection within groups. In contrast, Maynard Smith felt that Hamilton’s theory provided a more plausible explanation for the evolution of altruism. It was Maynard Smith, not Hamilton, who coined the term “kin selection” to contrast it with “group selection”.

The haystack model had an enormous impact on the rejection of group selection and the conceptualization of kin selection as an alternative to group selection. Nevertheless, at least one of its many assumptions is fatally biased. Nobody noticed at the time. Can you figure it out in hindsight?

To be continued.

Comments

  1. #1 Mike
    October 30, 2009

    Dear Dr. Wilson,

    I hope you will present to us the population-biological mathematical models of group-selection, because to be honest, that’s what I still have some problems with.

    I’m guessing that the success of groups is measured in population growth, or perhaps in the percentage it has of the total population.

    Now, we would need to know what constitutes a group of altruists – is it a group where practically everyone is an altruist, or one with more than 50% altruists, or is there some other threshold? And wouldn’t that depend on the group-fitness value of a specific altruistic disposition?

    Furthermore, if egoists out-compete altruists within groups, wouldn’t that mean that an altruist group will over time become a group with too little altruists to out-compete groups of egoists?

    And furthermore, if the altruist group grows more rapidly than the egoist group, but the altruists are getting proportionately less in the altruistic group, how will the altruistic traits (we needn’t assume they’re genetically specified) get fixed in the population?

    I guess if you start with a relatively large group consisting solely or almost solely of altruists, and no free-riders, this group could grow faster than a competing group of egoists, and thus the entire population will get more and more altruists… but only until free-riding becomes a problem.

    So, I hope you will post the mathematics in the future.

    Best,
    -Mike

    I really think a mathematical model would help me understand these issues better.

  2. #2 piker
    October 30, 2009

    “There are two genes in the population, one coding for aggressive behavior and the other coding for docile behavior.”

    This involves a fatally based assumption, although not the one you will later tell us about, or apparently have noticed yourself. Because again you have presented us with a system where possession of one alleged gene requires the absence of it’s opposite number.
    And again the same old mutual exclusivity paradox that arises when you tie the effectiveness of the trait to the individual that carries it, rather than tie the effectiveness of the individual to the choice of that trait in a particular group dynamic.

  3. #3 Kevin S.
    October 30, 2009

    My stab at the “fatally biased flaw”: The model seems to tacitly assume that every mouse, male or female, always mates with the most aggressive member of the opposite sex without regard to other characteristics. In the real world, this isn’t necessarily true. The most obvious example is that a female may reject a psychotically aggressive male (one with four aggressive genes working overtime) for fear that he will harm her or her children.

  4. #4 stripey_cat
    October 30, 2009

    The most obvious flaws, to me, are that there is no competition between groups in such a way that would allow one group to entirely obliterate another: aggressive, low-population groups still survive because they’re in independent haystacks and there’s no way for the docile, large-population groups to squeeze them out; and that at the mating-time, they all become one large group and intra-group competition is the mechanism for choosing mates (when you’d expect aggression to win).

  5. #5 daedalus2u
    October 30, 2009

    My guess at the flaw is the assumption that the aggressive mice become homozygous in the aggressive gene. If the aggressive gene is dominant, then only homozygous passive mice are removed from the haystack. Heterozygous aggressive/passive mice remain.

    When the founder mice are AP and PP, the first generation is either AP or PP. AA mice can only occur if there are both male and female AP mice. Only ¼ of the first generation is AP. Even less of the next generation is AP unless PP mice don’t reproduce.

    For the entire haystack to become AA, there needs to be high replication of mice and then deletion of PP and AP mice. If there is high replication in haystacks with A mice, then there should be the same replication (but without deletion of P mice) in haystacks with P mice or the A gene is doing more than just aggression.

    The number of mice produced per haystack of PP mice should be many times larger than a haystack containing A mice.

  6. #6 InfuriatedSciTeacher
    October 30, 2009

    Not going to deal with the mutual exclusivitiy paradox, as others have, but wouldn’t the relative success of docile mice as compared to aggressive not only depend on the gene frequency in the population but also the reproductive strategies of that species? The main flaw here is that in order to streamline the model Wynne-Edwards removed so many real-world parameters that the results are meaningless if you want to extrapolate the data to a natural population. Climate change models run into the same issue, because they leave out forcings or feedback loops if they’re inconvenient to program or not as well understood as we’d like. It’s wonderful that we can predict that there will be more or better fit docile mice emerging from each haystack, but how does that play into the reproductive success, and, given the mutually exclusive assumption (playing by the rules of the model, however poorly constructed), which allele is dominant? This model doesn’t raise a situation where each haystack is necessarily in competition with the others if the gene pool is redistributed after x generations. Wouldn’t gene flow have to be reduced if we’re to see a strong selection effect?

  7. #7 Shane Horan
    October 30, 2009

    The model seems to tacitly assume that every mouse, male or female, always mates with the most aggressive member of the opposite sex without regard to other characteristics.
    The model (I think) assumes no sexual preferences: mating is random. Females are simply more likely to mate with whichever variant is more common.
    It’s already well-established that if the ‘generation time’ of the haystacks (time before dispersal) is comparable to the individual mice’s generation times, then group selection may be significant and the altruists may come to dominate. Of course, if very many generations of mice occur before a dispersal the altruists become extinct in every haystack, and disapppear altogether.

  8. #8 Leo Martins
    October 30, 2009

    Richard Dawkins (…) compared further inquiry to the futile search for a perpetual motion machine.

    If you are referring to the passage in The Extended Phenotype mentioned in T&R V then no, he didn’t make this comparison. Quoting his passage, from your post:

    Something about the fact that this has been proved to be impossible is seen as an irresistible challenge by a certain type of intellectual dilettante. Perpetual motion machines have a similar fascination for some amateur inventors. The case of group selection is hardly analogous: it has never been proved to be impossible, and never could be.

    He simply mentioned perpetual motion machines as an example of irresistible challenges, immediately clarifying that group selection cannot be compared to these machines – since it cannot be proven to be wrong, for a start.

    Or am I missing something, since I haven’t read this book?

  9. #9 InfuriatedSciTeacher
    October 30, 2009

    Leo> I’ve not read that one in a while, but Dawkins appears to be comparing the intellectual draw of the two concepts, not the feasibility. The quote David gives above corroborates this idea: according to Dawkins, group selection is possible but rare and/or a weak force for evolution that pulls in a disproportionate amount of research time. If he’s made the analogy to perpetual motion machines in the other context elsewhere, I’ve not seen it…

  10. #10 abb3w
    October 31, 2009

    DSW: Can you figure it out in hindsight?

    In addition to the homozygous/heterozygous issue alluded to by daedalus2u, there appears to be the implicit presumption of only a binary response of “aggressive/nonagressive”; thus, leaving out mixed dominance, or other gradations. EG: varying numbers of copies of the gene for testosterone production.

    It also assumes the aggression isn’t moderated by detecting degree of relation (which would explicitly favor kin-selection) or more general INGROUP-type moderator mechanisms.

  11. #11 Bob O'H
    October 31, 2009

    There are two genes in the population, one coding for aggressive behavior and the other coding for docile behavior. Aggressive mice outcompete docile mice within each haystack, but they also decrease the growth rate of the group.

    ??????

    I assume you’ve mixed up “gene” and allele”: I’m pretty sure the haystack model was a single gene model with two alleles.

  12. #12 piker
    October 31, 2009

    Single gene with two alleles?
    In which case there would be no mutual exclusivity of traits, but then the model wouldn’t work as intended.

  13. #13 piker
    October 31, 2009

    For more on the subject of mutual exclusivity and its application to coevolution (which is more likely to be going on here), see the paper A Game Theoretic Memory Mechanism for Coevolution, included in the publication, Genetic and Evolutionary Computation – GECCO,
    by Erick Cantú-Paz, James A. Foster

    Also available online at:
    http://books.google.com/books?id=_IlWhIKahmsC&pg=PA287&lpg=PA287&dq=mutual+exclusivity+of+traits&source=bl&ots=6GTi9bjxJw&sig=2qgjPj8kf5VuSufLTjCaxdKggsc&hl=en&ei=DpzsSo2NJYKOswOeltR4&sa=X&oi=book_result&ct=result&resnum=8&ved=0CCkQ6AEwBw#v=onepage&q=mutual%20exclusivity%20of%20traits&f=false.

  14. #14 David Sloan Wilson
    November 1, 2009

    The model assumes a single locus with two alleles. With respect to Dawkins’ perpetual motion machine comparison, everyone acknowledges that group selection is possible in principle; its rejection was based on the claim that it is invariably trumped by individual selection in practice. Dawkins was saying that this claim had become so well established, theoretically, that to search further would be as futile as searching for a perpetual motion machine.

  15. #15 piker
    November 1, 2009

    Single locus with two alleles, yet somehow the alleles are separated out by breeding yet not for their strategic effectiveness? The assumption still seems to be that the selections are blindly choosing the strategies instead of letting the individuals have that option. This model is even less logical as a thought experiment than the two-gene model (or should I say your two gene models), which at least have some logical consistency as far as selecting for exclusive strategies would be concerned.
    And if Dawkins is saying that group selection is possible in principle but not in practice, then what the hell does that mean except that the machine won’t work as proposed? Does he perhaps intuit that its parts won’t evolve without some sort of intelligent choice function? Or that so far no life forms have evolved that resemble the ones whose family names have been appropriated for the purposes of the experiment?
    Although I’ll concede that he’d just as likely to be wrong intuitively as you would, considering the meme fiasco.

  16. #16 Guy
    November 3, 2009

    Well, one thing the haystack model ignores is the tendency for groups of altruists or cooperators to evolve policing mechanisms. This would make the fitness of the aggressive individuals a function of their frequency in the group because the altruists could effectively punish the rare aggressor.

New comments have been disabled.