I originally titled this post "The Less Wrong wiki is wrong about group selection", because it seemed wildly overconfident about its assertion that group selection is nonsense. The wiki entry on "group selection" currently reads:

People who are unfamiliar with evolutionary theory sometimes propose that a feature of the organism is there for the good of the group - for example, that human religion is an adaptation to make human groups more cohesive, since religious groups outfight nonreligious groups.

Postulating group selection is guaranteed to make professional evolutionary biologists roll up their eyes and sigh.

However, it appears that the real problem is not that the wiki is overconfident (that's a problem, but it's only a symptom of the next problem) but that the traditional dogma on the viability of group selection is wrong, or at least overconfident. I make this assertion after stumbling across a paper by Martin Nowak, Corina Tarnita, and E. O. Wilson titled "The evolution of eusociality", which appeared in Nature in August of this year. I found a PDF of this paper through Google scholar, click here. A blog entry discussing the paper can be found here (bias alert: it is written by a postdoc working in Martin Nowak's Evolutionary Dynamics program at Harvard).

Here's some quotes (bolding is mine):

It has further turned out that selection forces exist in groups that diminish the advantage of close collateral kinship. They include the favouring of raised genetic variability by colony-level selection in the ants Pogonomyrmex occidentalis and Acromyrmex echinatior—due, at least in the latter, to disease resistance. The contribution of genetic diversity to disease resistance at the colony level has moreover been established definitively in honeybees. Countervailing forces also include variability in predisposition to worker sub-castes in Pogonomyrmex badius, which may sharpen division of labour and improve colony fitness—although that hypothesis is yet to be tested. Further, an increase in stability of nest temperature with genetic diversity has been found within nests of honeybees and Formica ants. Other selection forces working against the binding role of close pedigree kinship are the disruptive impact of nepotism within colonies, and the overall negative effects associated with inbreeding. Most of these countervailing forces act through group selection or, for eusocial insects in particular, through between-colony selection.

Yet, considering its position for four decades as the dominant paradigm in the theoretical study of eusociality, the production of inclusive fitness theory must be considered meagre. During the same period, in contrast, empirical research on eusocial organisms has flourished, revealing the rich details of caste, communication, colony life cycles, and other phenomena at both the individual- and colony-selection levels. In some cases social behaviour has been causally linked through all the levels of biological organization from molecule to ecosystem. Almost none of this progress has been stimulated or advanced by inclusive fitness theory, which has evolved into an abstract enterprise largely on its own


The question arises: if we have a theory that works for all cases (standard natural selection theory) and a theory that works only for a small subset of cases (inclusive fitness theory), and if for this subset the two theories lead to identical conditions, then why not stay with the general theory? The question is pressing, because inclusive fitness theory is provably correct only for a small (non-generic) subset of evolutionary models, but the intuition it provides is mistakenly embraced as generally correct.

Check out the paper for more details. Also look at the Supplementary Information if you have access to it. They perform an evolutionary game theoretic analysis, which I am still reading.

Apparently this theory is not that new. In this 2007 paper by David Sloan Wilson and E. O. Wilson, they argue (I'm just pasting the abstract):

The current foundation of sociobiology is based upon the rejection of group selection in the 1960s and the acceptance thereafter of alternative theories to explain the evolution of cooperative and altruistic behaviors. These events need to be reconsidered in the light of subsequent research. Group selection has become both theoretically plausible and empirically well supported. Moreover, the so-called alternative theories include the logic of multilevel selection within their own frameworks. We review the history and conceptual basis of sociobiology to show why a new consensus regarding group selection is needed and how multilevel selection theory can provide a more solid foundation for sociobiology in the future.

From the other camp, this seems to be a fairly highly-cited paper from 2008. They concluded:

(a) the arguments about group selection are only continued by a limited number of theoreticians, on the basis of simplified models that can be difficult to apply to real organisms (see Error 3); (b) theoretical models which make testable predictions tend to be made with kin selection theory (Tables 1 and 2); (c) empirical biologists interested in social evolution measure the kin selection coefficient of relatedness rather than the corresponding group selection parameters (Queller & Goodnight, 1989). It is best to think of group selection as a potentially useful, albeit informal, way of conceptualizing some issues, rather than a general evolutionary approach in its own right.

I know (as of yet) very little biology, so I leave the conclusion for readers to discuss. Does anyone have detailed knowledge of the issues here?

New to LessWrong?

New Comment
15 comments, sorted by Click to highlight new comments since: Today at 11:43 PM

...Huh. That page is by Eliezer Yudkowsky, and apparently it was written back in June 2009. It seems that writing something up on the wiki is a very poor way to get it discussed.

The argument given seems to make sense, but with two noteworthy caveats. The first is that, while group selection doesn't work, kin selection does, and in an organism that mostly interacts with close relatives, altruism only towards relatives may not be meaningfully distinct from altruism towards all others of its species. The second issue is that group selection can interact with the founder effect, which is where a very small group enters a new environment and founds a colony that turns into a large population. In that case, altruistic genes in the original founders affect the probability that the colony succeeds; so if an environment is particularly hard to colonize, and many attempts are made, it may be that only a group with some altruistic genes could succeed.

I am no biologist so I may be confused at some point here, but isn't the point of this discussion article that research has demonstrated group selection in colonies of insects?

Is there some strong sense in which queen bees/ants constitute "founders" or make entire colonies "kin"? Or is there still conflict between the paper nhamann cites and the wiki?

Keep in mind that eusocial species are different in a way that strongly affects who is kin: members of the same colony share 75% genetic material because of how their reproduction works (compare with 50% for parent-child or between siblings -- or to 100% for cells within the same organism).

So as a non-biologist, what I'm hearing from this and nhamann's reply is:

Maybe/probably there is an effect in genetics that would cause the bees in the original study cited to be more closely related to one another in ways that would encourage specific group selection in ways that wouldn't extend smoothly to other species. However, this effect is losing ground in the scientific community so it is worth reexamining the case.

It's more complicated than that. It actually turns out that several mathematical analyses since the 80s have shown inclusive fitness theory and group selection theory to be mathematically equivalent*. In fact, both sides seem to readily acknowledge this fact, so the two groups' appear to be arguing over semantics (where does selection "really" occur?)

This argument over semantics is best illustrated by the following quote at the end of the supplementary information for the paper in the OP:

Finally, we propose that kin selection among social insects is an apparent phenomenon which arises only when you put the worker into the center of evolutionary analysis. Kin selectionists have argued that a worker who behaves altruistically by raising the offspring of another individual, requires an explanation other than natural selection, and this other explanation is kin selection. We argue, however, that there exists a more convenient coordinate system.

* There is, of course, ongoing debate about this, and whether one theory is more general than the other.

Okay so now what I'm getting is more like:

Some animals (esp. social insects) behave "altruistically," i.e. via raising others' offspring. Some people consider the explanation "an altruistic set of animals produces more offspring which inherit their altruism" sufficient, and some people require saying "animals that behave altruistically toward their kin produce more offspring etc." and calling it "inclusive fitness." People who think the first way point out that people who think the second way are ignoring phenomena like domestication (i.e. behaving altruistically toward other species) and being too specific.

I'm throwing the example of domestication in there because it came to mind for me as a way in which the first statement seems more general than the other.

Mmm, I'm probably going to have to try to write a post explaining what I think I understand about the debate, because there's a lot of other points.

  • Researchers in the behavioral economics literature have been conducting experiments with human subjects playing public goods games. Time and time again, it has been shown that cooperation can be sustained in groups when subjects are allowed the ability to punish defectors and reward cooperators. They argue, at the very least, that inclusive fitness theory cannot explain such behavior between strangers.

  • Nowak and others seem to be proposing an expansion of the biological theory of genetic evolution to a mathematical theory of evolution. He and others propose that evolution can be genetic or cultural or any other kind of evolutionary process you can dream up, as long as it fits the mathematical framework they've developed (which is based on evolutionary game theory).

The whole controversy seems silly, however. Group selection effects have been observed among insects in laboratory conditions. Inclusive fitness theorists argue that these conditions are rare, so they can be safely ignored for the genetic theory of evolution. The multilevel selection theorists argue that evolution is a mathematical process which can explain a variety of phenomena beyond just genetic evolution. Both sides are strictly correct, they're simply talking past each other.

Okay, so, again parsing this in layman's terms:

-Some researchers say "let's use group selection for studying evolution."

-Some other researchers say "when we talk about genetic evolution we don't need group selection."

-The first researchers say "but it's useful in other contexts, why ignore it?"

-The second group says "because we research genetic evolution."

Is that about right?

I think that's about right, but this is what I understand from at most a weekend of skimming journal articles, so I might be off here.

What you said is only true for organisms which happen to engage in haplodiploidy. From the paper by Nowak et al in the OP:

By the 1990s, however, the haplodiploid hypothesis began to fail. The termites had never fitted this model of explanation. Then more eusocial species were discovered that use diplodiploid rather than haplodiploid sex determination. They included a species of platypodid ambrosia beetles, several independent lines of Synalpheus sponge-dwelling shrimp (Fig. 2) and bathyergid mole rats. The association between haplodiploidy and eusociality fell below statistical significance. As a result the haplodiploid hypothesis was in time abandoned by researchers on social insects

Interestingly, group selection has been observed in microorganisms:

Experiments with actual groups illustrate the point. Pseudomonas fluorescens bacteria quickly suck all the dissolved oxygen out of a liquid habitat, leaving a thin habitable layer near the surface. But some bacteria spontaneously develop a beneficial mutation. These group-saving individuals secrete a polymer that enables bunches of individuals to form floating mats. As a mat, all the bacteria survive, even though most of them expend no metabolic energy producing the polymer. But if the freeloaders get greedy and reproduce too many of their kind, the mat sinks and everybody dies, altruists and freeloaders alike. Among these bacteria, then, groups that maintain enough altruists to float outcompete groups with fewer altruists than that minimum number. The former groups survive, grow and split up into daughter groups. Thus, altruistic individuals can prosper, despite the disadvantage of expending precious resources to produce the polymer.

There are four letters in a later issue of Nature commenting on Nowak et al. (All of this is behind paywalls, sorry.) All of the contributors seem to agree (whether wholeheartedly admitting it or not) that inclusive fitness theory is, in effect, a theorem derivable from basic individual selection theory, not a separate principle. Nowak et al. further argue (mathematically) that the preconditions for inclusive fitness theory are rarely satisfied in practice, and that direct arguments from individual selection are a better tool for understanding the relevant phenomena.

Demathematised, one of the points Nowak et al. make is that individual fitness theory takes account only of the benefit an organism's costly actions provides to its relatives, but not of any reciprocal benefits. It therefore underestimates the amount of cooperation one might expect to see.

Eliezer's original article on group selection said some things that I think ought to be included in the wiki article: While the preconditions for group selection appear unlikely to hold in the wild (a similarity with inclusive fitness theory), they can be produced in the laboratory. The result of imposing selection for limited colony size of certain insects was not that the insects limited their own numbers, but that they limited each others' within their colony, by eating their eggs and larvae.

Novak co-authored a paper called "Transforming the dilemma". I'd say it's relevant:

Whenever evolution ‘constructs’ a new level of organization, cooperation is involved. The very origin of life, the emergence of the first cell, the rise of multi-cellular organisms, and the advent of human language are all based on cooperation. A higher level of organization emerges, whenever the competing units on the lower level begin to cooperate. Therefore, we propose that cooperation is a third fundamental principle of evolutionary dynamics besides mutation and selection. Evolution occurs in populations of reproducing individuals. Inaccurate reproduction can lead to mutation. Mutation can lead to selection. Selection can lead to cooperation.

The analysis that follows actually uses game theory:

In the Prisoner’s Dilemma, defectors dominate cooperators unless a mechanism for the evolution of cooperation is at work. We will discuss five mechanisms for the evolution of cooperation: direct reciprocity, indirect reciprocity, kin selection, group selection, and network reciprocity (or graph selection).

ETA: I hope it's alright that I use the comments section as a workspace; I'm trying to dig through the literature to see what the fuss is all about.

It seems like this is more introductory than the paper I posted about above. It features an interesting theoretical argument:

Consider a population that is subdivided into groups. The fitness of individuals is determined by the payoff from an evolutionary game. Interactions occur between members of the same group. We model stochastic evolutionary dynamics. In any one time step, a single individual from the entire population is chosen for reproduction with a probability proportional to its fitness. The offspring is added to the same group. If the group reaches a critical size, n, it will divide into two groups with probability q. The members of the group are randomly distributed over the two daughter groups, see Fig. 1. With probability 1− q, the group does not divide, but a random individual of the group is eliminated. Therefore, n resembles the maximum number of individuals in a single group. The total number of groups is constant and given by m; whenever a group divides, another group is eliminated. These assumptions ensure that the total population size is constrained between a lower bound, m, and an upper bound, mn.

Our simple model has some interesting features. The entire evolutionary dynamics are driven by individual fitness. Only individuals are assigned payoff values. Only individuals reproduce. Groups can stay together or split (divide) when reaching a certain size. Groups that contain fitter individuals reach the critical size faster and, therefore, split more often. This concept leads to selection among groups, although only individuals reproduce. The higher-level selection emerges from lower-level reproduction. Remarkably, the two levels of selection can oppose each other.

This presentation is a good source for the inclusive fitness side of the debate. It's by Stuart West, evolutionary biologist at Oxford.

In the talk, West, a proponent of inclusive fitness, says:

[There's] also the general point of [...] Occams Razor: If we can explain everything more generally without it, what use is it?

This debate is making me boggle. Both sides seem to claim that their theory is more general than the other, and that therefore by Occams Razor we should eliminate the other theory.