David Queller on the evolution of eusociality

Liao, Rong and Queller just weighed in on the Nowak/Tarnita/Wilson 2010 model of the evolution of eusociality.

That paper was mostly smacked down by critics because of its delusional depiction of kin selection, but few bothered to criticize its model of the evolution of eusociality. Now, David Queller's gone through the model, and the results are not too pretty.

They had a modeling strategy that should work and should be fine, but they weren’t careful enough when they made claims about their models’ novel results

...David Queller is quoted as saying.

From the abstract of the article:

The claim of these authors was bolstered by a new model of the evolution of eusociality with novel conclusions that appeared to overturn some major results from inclusive fitness. Here we report an expanded examination of this kind of model for the evolution of eusociality and show that all three of its apparently novel conclusions are essentially false. Contrary to their claims, genetic relatedness is important and causal, workers are agents that can evolve to be in conflict with the queen, and eusociality is not so difficult to evolve. The misleading conclusions all resulted not from incorrect math but from overgeneralizing from narrow assumptions or parameter values. For example, all of their models implicitly assumed high relatedness, but modifying the model to allow lower relatedness shows that relatedness is essential and causal in the evolution of eusociality. Their modeling strategy, properly applied, actually confirms major insights of inclusive fitness studies of kin selection. This broad agreement of different models shows that social evolution theory, rather than being in turmoil, is supported by multiple theoretical approaches. It also suggests that extensive prior work using inclusive fitness, from microbial interactions to human evolution, should be considered robust unless shown otherwise.

Queller's paper certainly makes entertaining reading. However I can't help thinking that it takes the Nowak/Tarnita/Wilson paper too seriously. The most obvious response to that paper is, I think, ridicule. Jon Wilkins hit roughly the right note with Important Harvard Scientists Attack Kin Selection.

News coverage:

• Relatedness, Conflict, and the Evolution of Eusociality by Xiaoyun Liao, Stephen Rong and David C. Queller
• Did a scientific battle about altruism just end? by Diana Lutz (source)
• New paper shows that Nowak et al. were wrong: kin selection remains a valuable concept in evolutionary biology by Jerry Coyne
• Much Ado About Inclusive Fitness by Razib Khan

---

Update 2015-05-09:

Nowak and Allen reply in Inclusive Fitness Theorizing Invokes Phenomena That Are Not Relevant for the Evolution of Eusociality

Liao, Rong and Queller reply: Some Agreement on Kin Selection and Eusociality?

Nowak's papers on the topic are here.

Herbert Gintis vs inclusive fitness theory

Herbert Gintis has an elaborate attack on inclusive fitness theory on his web site. It's titled: Inclusive Fitness and the Sociobiology of the Genome. It says, among other things:

The general point is that if there is a conflict among loci concerning fitness maximization, and if the frequency of alleles at one locus affect the fitness costs and payoffs at other loci, then it is logically impossible that the allele at each locus maximize its inclusive fitness. Rather, the proper setting is evolutionary game theory [...]

I think this is a case of expecting too much from the concept of maximization. The idea of maximization does not imply that all parties reach the maximum they are seeking. It doesn't even imply that they increase the value they are trying to maximize. A hill-climber might be climbing a hill on a mountain that is sinking into the sea. Their elevation may decrease until it reaches zero and they die. This is quite consistent with the idea of maximization.

Gintis apparently claims that frequency-dependence invalidates the idea of maximization of inclusive fitness:

If the genome’s success is based on a pattern of cooperation, promotion, and suppression across loci, which will occur, for instance, if the production of a protein, RNA sequence, or social behavior requires the collaborative activity of many genes (Noble 2011), or if there are frequency dependent social interactions among individuals in a social species (Maynard Smith 1982), then neither genes nor individuals can be characterized as maximizing inclusive fitness. The conditions under which a population genetics model of gene flow implies fitness maximization at the gene or individual level has been carefully explored (Grafen 1999, 2002, 2006; Metz et al. 2008; Gardner and Wild 2011; Gardner West and Wild 2011). With frequency independence, they affirm the maximization hypothesis. With frequency dependence, the hypothesis it is in general false, and no careful researcher has ever claimed otherwise.

Frequency dependence typically means that genes are spending time in environments that differ from the average of the environments they evolved in. It's a case of adaptive lag - where organisms are adapted not to their current environment, but to a weighted average of ancestral environments. This is a classic case where organisms sometimes behave sub-optimally in ways that don't further their own interests.

In such cases, organisms don't always act to maximize their inclusive fitness - but rather act to maximize their inclusive fitness under the hypothesis that they are in an environment that is like the one that their ancestors evolved in. Of course, that hypothesis might be mistaken.

In fact, sufficiently sophisticated creatures might be able to evolve a superior strategy - where they sample the frequency involved and change their behaviour adaptively, based on the observed frequency. So, what frequency dependence often boils down to is that organisms can sometimes fail to maximise their inclusive fitness - because they have cognitive limitations and make mistakes.

In this case, I feel that Gintis is expecting too much from inclusive fitness theory. It doesn't claim that all creatures maximize inclusive fitness perfectly. Creatures have all kinds of imperfections and limitations that prevent them from acting as perfect maximizers.

Kin selection and its critics

Here's an interesting new paper: Kin Selection and Its Critics - Jonathan Birch and Samir Okasha.

Samir Okasha wrote a rather flawed book on group selection some years ago. However, this article shows that he is doing a good job of keeping up with developments in the field - and it isn't so easy to find significant mistakes in this large recent article.

I think one problem is that it takes the work of Nowak and Wilson a bit too seriously.

In one place the authors argue against equivalence, saying:

In one respect, the kin selection approach is arguably more general than the multilevel approach, because the latter requires that individuals be nested into nonoverlapping groups, as in figure 4; this is necessary for the decomposition technique in box 2 to apply (Hamilton 1975, Okasha 2006, Frank 2013). Groups of this sort exist in some taxa (e.g., the colonies of many social insect species). But in other cases, individuals engage in social interactions with their conspecifics, but there are no well-defined, discrete groups. The kin selection approach can handle such cases easily; indicative of this is that in deriving equation 4 above (box 1), we did not make use of the fact that the individuals were nested into nonoverlapping groups. Therfore, the claim that kin and multilevel selection are formally equivalent requires at least this qualification.

This doesn't seem like too much of a stumbling block to me. The modern "group selection" approaches depend critically on defining a "group" to include any collection of organisms - no matter how fleeting or ephemeral. You have to buy into this conception of a "group" for the approach to be worth considering in the first place.

The authors say:

The widespread preference for kin selection may be partly due to multilevel selection's association with the flawed good-of-the-group tradition of the 1950s and 1960s and the associated superorganism concept, of which many biologists remain suspicious. It is undeniable that the careless appeal to group-level advantage as a way of explaining a trait's evolution led to serious errors in the past, so biologists’ wariness of this mode of explanation is understandable.

That's about the size of it. However, this paints group selection's problems as being in the past. I think that this is inaccurate. A fairly cursory look at the evolutionary social sciences shows that misapplication of group selection is still widespread.

The essay closes with a plea for "causal aptness": use kin selection when you have relatives, use group selection when you have interacting groups. This proposal sounds reasonable - but I think it would do little to stem the existing misuse of group selection. The problem is that people see differential group reproduction, reach for group selection, and produce just-so stories about how group traits are the product of differential group extinction or reproduction. This is a systematically bad methodology that use of group selection directly encourages. Using "causal aptness" would probably boost usage of group selection. That seems as though it is likely to cause a range of negative outcomes associated with the misuse of group selection - and so I regard the proposal as suspect.

"Causal aptness" is one proposal. A big health warning relating to the misuses of group selection is another. I think that, if you adopt the first proposal, you should also adopt the second one.

West and Gardner on kin vs group selection

West and Gardner have not been shy about pointing out the virtues of kin selection over group selection. Here they are with a summary in 2013:

The most frequently used methods are neighbour-modulated and inclusive fitness. In particular, modern neighbour-modulated fitness methods allow the modeller to go from the underlying biology to an expression or fitness, in a way that facilitates the development of relatively general models [8,27,28,35,37]. In contrast, the group selection approach is used relatively little for modelling specific traits partly because as soon as one moves away from the simplest, most abstract models, and wants to add in real world biology, it often becomes analytically intractable — for example, when populations are structured into different classes of individual, according to sex, age, caste or ploidy [38–40].

I know these folk also have plenty of other objections to group selection. Here, forced to pick one, they went for analytic tractablility. It seems like an odd choice to me. As a veteran computer modeller, analytic tractablility comes relatively low on my list of virtues. I would probably list group selection's association with junk science as my number one complaint.

Identity by descent: a confusing concept

There's a lot of discussion of kin selection using the term "identical by descent" and "identical by state". The idea of "identical by descent" is that genes are shared as a result of direct descent from a common ancestor - without recombination or mutation. "Identical by state" just means that the DNA sequence is shared. It is said that 50% of their genes with their daughters "IBD".

I think the "identical by descent" terminology is confusing and not useful. In biology, if genes are identical, they are practically always identical through being copied from one (or more) shared ancestors. Mothers share more than 50% of their genes with their daughters - due to genes that have reached fixation, inbreeding and so on. However they still share these gene sequences due to descent from shared ancestors. As to a gene mutating into another form and then mutating back again. If you do the sums, for a gene of any reasonable size this rapidly becomes ridiculously unlikely. There's too much scope for neutral mutations elsewhere. In practice, when genes are identical, the odds are enormously in favour of this being due to shared ancestry. The idea that recombination with an identical gene makes genes not "identical by descent" is an awful one. They are still "identical by descent" - just descent from various different ancestors.

You can't say that mothers share 50% of their genes with their daughters "IBD". It is confusing and mistaken. If you want to use the 50% figure, you have to find another reason for doing so.

Analysis of "group selection and inclusive fitness are not equivalent"

I briefly analyzed the paper:

Group selection and inclusive fitness are not equivalent; the Price equation vs. models and statistics by Matthijs van Veelen, Julian Garcia, Maurice W. Sabelis, Martijn Egas.

This is one of the papers mentioned on my equivalence naysayers page.

The paper claims that group selection and kin selection are not equivalent. It argues that inclusive fitness requires fitnesses to be "additive". Additive fitness is a common assumption when deriving Hamilton's rule - and is indeed associated with inclusive fitness. However, inclusive fitness is a simplified model of kin selection. Kin selection enthusiasts are not too impressed with such critiques - the limitations of inclusive fitness are well known. The paper uses "kin selection" and "inclusive fitness" as though these concepts are interchangeable. I think this is not all that useful an approach.

Kin selection doesn't depend on fitness being additive. That idea is associated with inclusive fitness and Hamilton's rule. These are concepts associated with simplified models of kin selection.

Hamilton advocated using "inclusive fitness" instead of "kin selection". Hamilton (1975) "Innate social aptitudes of man" says:

The usefulness of the ‘inclusive fitness’ approach to social behaviour (i.e. an approach using criteria like (b K-k) > 0) is more general than the ‘group selection’, ‘kin selection’, or ‘reciprocal altruism’ approaches.

However, I think the pendulum has swung away from "inclusive fitness" and back towards "kin selection" as the term of choice. That's what Gardner and West use tend to use, for example. I'm with them.

Obvious types of kin selection

If you ask a student of kin selection about the most obvious aspects of the human phenotype that have been influenced by kin selection and are coded for in human DNA, I think you would generally get back a list looking something like this:

• Genitals;
• Breasts;
• Placenta;
• Umbilical cord;
• Female body fat;
• Maternal love;

If you ask a group selection proponent the corresponding question (what are the most obvious aspects of the human phenotype that have been influenced by group selection and are coded for in human DNA), I think you would get back a very different list. The group selection proponent might be more likely to mention multi-cellularity - but I think they would be highly likely to back a very different list.

What's up here? Kin selection and group selection are basically the same thing.

I don't pretend to have a complete explanation for this - but I think the group selection enthusiasts were led astray by the urge to distinguish themselves from kin selection enthusiasts. If discussing family groups, it was obvious that kin selection applied to it - and explained it. The group selection advocates therefore focused their attention elsewhere - where it seemed as though there was more chance to explain new phenomena that kin selection failed to cover. This hypothesis explains the migration of group selection enthusiasts to cultural evolution - since that is obviously not down to shared DNA.

Eventually, there was nowhere else to run, and the group selection advocates mostly gave up their claims that group selection was something new and different. Now the rhetoric in the area has mostly shifted to other issues.

Goodnight on kin selection rampage

Charles Goodnight has been explaining why he doesn't like kin selection recently, in a string of articles - most notably Why I Don’t like Kin Selection.

However, this article seems littered with misunderstandings to me. Goodnight claims kin selection can only focus on altruism. That is a mistake - kin selection has also been applied to spiteful behaviour. Goodnight claims that kin selection can't handle cultural relatedness. That's wrong - cultural kin selection handles cultural relatedness just fine. Goodnight claims that kin selection is ¨an optimality approach¨. In fact, evolution is a gigantic optimization process. All adaptations are the result of optimizations. The accusation that that kin selection is an optimality approach just seems totally confused to me. Goodnight claims that multi-level selection models in which selection on different levels acts in the same direction can't be studied using kin selection models. That seems ridiculous to me - of course they can. In fact, neither group-selection nor kin selection models spend much time on this case. For group selection, this is because proponents are still struggling to find evidence for their effects - and this case typically doesn't help do that.

Goodnight winds up publicly explaining where he doesn't understand kin selection. That's fine - but readers should not be persuaded by an article with so many mistakes.

The article closes with:

However, like optimal foraging theory, it appears to mainly be useful in making broad stroke qualitative predictions that can be used in the introduction, or in a laudatory paragraph about how wonderful Hamilton is at the end of a paper. If you want to make quantitative statements about selection in real world populations that will contribute to our understanding of social evolution multilevel selection might be a better choice.

Ironically, this is almost the exact reverse of what kin selection enthusiasts often say about group selection. Kin selection features the coefficient of relatedness - whereas group selection is rarely concerned with the level of relatedness within or between groups. So: kin selection is typically quantitative, while group selection is much more concerned about identifying the level at which a feature is adaptive - which is a more qualitative issue.

Nowak nominates inclusive fitness for retirement

The 2014 annual question at The Edge is: What scientific idea is ready for retirement?

Martin Nowak nominated inclusive fitness. He writes:

Contrary to what is often claimed there exists no empirical test of inclusive fitness theory; nobody has ever performed an actual inclusive fitness calculation for a real population. Inclusive fitness was originally understood as a crude heuristic that can guide intuition in some cases, but not in general. It is only in recent years that inclusive fitness has been elevated—mostly by mediocre theoreticians — to a religious belief, which is universal, unconstrained and always true. Understanding the limitations of inclusive fitness gives us now the opportunity to develop mathematical descriptions of key phenomena in social evolution. It is time to abandon inclusive fitness and focus on a meaningful interaction between theory and experiment in sociobiology.

This seems like unsubstantiated scientific FUD to me.

Nowak claims that the:

dominant and unfortunate impact has been the suppression of meaningful mathematical theories in wide areas of sociobiology.

He means group selection? I think this is unreferenced and unsupported nonsense.

From the 1970s onwards we actually understood that evolution does not permit a single quantity that is always maximized. This fact still has to sink in with many in the inclusive fitness community.

I think that the first sentence here is technically mistaken. You can in fact, model the evolution of any dynamical system using the concept of maximisation of a utility function. Take, for example, the utility function that assigns world events that happen utility 1 and world events that do not happen utility 0. In this case, the evolution of the system can be modeled by "a single quantity that is always maximized". This is the fact of the matter. Nowak's proposed 'fact' is a simple falsehood. We didn't understand this 'from the 1970s onward' - it is simply not true in the first place.

Nowak writes:

On the level of genes there is no inclusive fitness.

Really? Most pairs of genes are either copies of each other (r=1) or not (r=0). Plug these numbers into Hamilton's rule and it works fine. This illustrates the meaning of "The Selfish Gene": genes only care about themselves - or copies of themselves. Nowak doesn't explain what the perceived problem is.

More relating to Nowak's article:

• Marczyk, Jesse (2014) Which Ideas Are Ready to Go to Florida?

• Reuell, Peter (2014) Harvard Gazette press release

Nowak and Wilson are at it again!

You might think Nowak and Wilson would have had their fingers burned in the reaction to their 2010 Nature paper.

However, it seems as though they are at it again. They have a PNAS paper titled: Limitations of inclusive fitness by Benjamin Allen, Martin A. Nowak, Edward O. Wilson.

They deny that evolution can be usefully viewed as an optimization process, saying:

Thus, evolution does not, in general, lead to the maximization of inclusive fitness or any other quantity.

That's complete nonsense: Allen, Nowak, and Wilson don't know what they are talking about. Optimization models are perfectly general - and can model any dynamical system.

The guts of their paper is a criticism of the application of Price's equation to inclusive fitness.

Their main beef is linearity. They are concerned that the 'adding' and 'subtracting' that goes on in the definition of inclusive fitness limits its generality. They write:

Inclusive fitness assumes that personal fitness is the sum of additive components caused by individual actions. This assumption does not hold for the majority of evolutionary processes or scenarios.

However, this linearity doesn't really cause problems - since you can approximate non-linear curves using a series of line segments.

Inclusive fitness is mostly concerned with the issue of whether some helping behaviour will evolve. It calculates whether the net selective effect of a behaviour on the frequency of a gene responsible for it (some of which may be copies in relatives of the actor) is positive or negative (relative to a set of alternatives) in the current environment. The math of inclusive fitness tells you whether the gene is selectively favoured. That's not quite the same thing as whether the behaviour will evolve (maybe it will drift into extinction) - but it's a good start. Nobody ever thought that the behaviours and genes involved had to not interact. Nobody thought that the rate of change of the frequency of the gene would be fixed over time. Maybe the environment will change in crazy non-linear ways in the future - and you'll have to redo your sums.

Basically, it isn't true that inclusive fitness theories assume that personal fitness is a linear sum of action fitness deltas. An action can change the environment (or the actor) in ways that affect the costs and benefits of future actions - and this is perfectly compatible with Hamilton's rule. It means that you may have to apply the rule more than once, is all.

Allen, Nowak, and Wilson's paper is like saying: I've got this crazy non-linear function, there's no way your linear approximation can match it. Except that: yes, there is - a series of linear functions can approximate any other function arbitrarily closely.

If you think that making this sort of objection is childish, I think that you're right - it is childish.

The paper says it was: "supported by a grant from the John Templeton Foundation". It seems like even more fuel for those who think that the Templeton Foundation is systematically distorting science.

Commentary

• Nowak and colleagues skewer Inclusive Fitness
• Now I know I am “El Lobo Solitario”: I don’t even agree with Allen, Nowak, and Wilson
• Inclusive fitness, models, and religious evolution

References

• Allen B, Nowak MA, & Wilson EO (2013) Limitations of inclusive fitness. Proceedings of the National Academy of Sciences of the United States of America
• Nowak MA, Tarnita CE, & Wilson EO (2010) The evolution of eusociality. Nature, 466 (7310), 1057-62