Wednesday 15 September 2021

Day 2 Group selection for Maynard Smith and Sober


 

So yesterday went really well. To my surprise, i did everything i'd planned!  My brain was working, i made some good headway on the talk.

Today - meh. I often find that a good, smart day is followed by a dip. And i didn't even get drunk! I just woke up and couldn't face going back to the notes i'd finished the evening before. It's not the end of the world because i was in and out of EPSA talks and meetings today anyway. 

But then this book arrived in the post and has saved me. It's a book that's kind of hard to get hold of. At least, libraries rarely have it,  and pdfs don't seem to circulate online, so i had to stump up and buy it. Which is another reason why scholars without much cash - say early career folk who pay for a lot of childcare - are disadvantaged in academia. But i don't have eye-watering childcare costs anymore, so worldofbooks got my money.

It was so worth it! For a random edited collection from 1987, its pretty widely-cited, and now i can see why.  Chapter 5 'How to Model Evolution' by John Maynard Smith, and the ensuing back and forth he has with Elliott Sober, is gold.

I think i can build one of my chapters around it (not the one i was meant to be working on, but hey)  so i'm going to put my initial thoughts about it down here.

Maynard-Smith's Chapter 5 is on the group selection, or multilevel selection debates. He starts by describing what Samir Okasha has subsequently termed 'perspective pluralism' in which there are disputes about the best way to model an evolutionary process - using classical or inclusive fitness, for example, or applying the maths of gene selection versus individual selection, for another.  As these disputes do not concern any kind of empirical or predictive disagreement, they are at worst pseudoproblems, and at best tussles about which conceptual scheme best captures the causal structure of the relevant system.

He goes on to give a classy statement of what has become central to my own work - the classic Darwinian model of evolution by natural selection taking place in any population of entities which multiply with heritable variation in fitness. Because Darwin identified the relevant entities with organisms, he was able to offer an explanation for the adaptive design of organisms. 

"Today we are asking whether there are entities other than individuals with the properties of multiplication, variation and heredity, and that therefore evolve adaptations by natural selection." (p.121).

Damn that's pithy - maybe i should start my evolutionary chapter with that.

Maynard Smith goes against Lewontin and most others by calling such entities units of evolution, instead of units of selection. I have every sympathy with his argument - mere multiplication with fitness variation is sufficient for selection, but for an evolutionary response to that selection you need heredity too.  And he notes that "It is important to distinguish between objects we can expect to evolve adaptations and those we cannot." (p. 122). If this were a protest i'd be doing jazz hands right now.

He then describes a scenario in which we'd have group selection but no group adaptation, because of insufficient heredity.

Enter Elliott Sober from stage left. His chapter 6 is a direct response. He claims (but ever so elegantly) that Maynard Smith has made a mistake, because it is neither necessary nor sufficient for group adaptation that groups be units of evolution (i.e. groups which multiply with heritable fitness variance. Damn!! Thank god i have a large bag of popcorn.

So, Sober's reasoning is the following:

  • A group with heritable fitness variation will fail to have group adaptations if it evolves what GC Williams called 'fortuitous group benefits' A.K.A. cross-level by-products. Adopting Williams' example, we can imagine herds of deer that differ in running speed, and that produce 'offspring' herds in proportion to their success in outrunning predators, and even that offspring herds inherit the speed of their parent herd.  all the ingredients for being a group unit of evolution are there, but running speed isn't a group adaptation - its a deer adaptation. the appearance of a group effect is just a statistical artefact. on the other hand.....
  • Group adaptations don't require groups to be units of evolution, because  you could get a group adaptation evolving as a response to group selection that is based on lower-level heredity, not group heredity. In other words, Sober denies that a group adaptation requires group-level heredity, even though it does require group-level selection.

This is interesting, and contra to my instincts, so why does he say so? The example he describes has the group adaptation evolving in a system where there is group-level heritability, in a statistical sense - offspring groups resemble parent groups. But he points out that you can get this to happen via an external mechanism that has nothing to do with the parents. For example, it could be some sort of positive assortment mechanism, acting in a Simpson paradox model with altruists competing against selfish group-mates. "A given daughter colony may have founders drawn from many different parent colonies. My point is this does not mean that altruism - a group adaptation- cannot evolve." (p.136). 

Sober doesn't say this, but in his favour, many philosophers would insist that heredity has to be a causal notion - offspring should resemble the parent because of some causal connection to the parent. In the model he describes, that causal connection is gone. Sober is saying that a sort of greenbeard effect could be enough.

Finally, Sober claims that altruism is a paradigmatic example of a group adaptation, because it cannot evolve by within-group selection, so it must be a response to group selection, aka a group adaptation.

Because Sober takes the explanation of adaptation to be a core priority in evolutionary biology, he concludes that the unit of selection is more important than the unit of evolution. And being a unit of selection requires only that there is selection on x, not that x's traits are heritable.

                                                                       ***

All of this is crucial for me, because i advocate defining the organism as something with the capacity to multiply in a way that produces heritable variance in fitness. This is different again from what both Sober and Maynard Smith are positing, in a few ways i won't go into here*. But its closer to Maynard Smith's unit of evolution than Sober's unit of selection, because i'm arguing that heredity is essential. Has Sober already proved me wrong in 1987?

I think there are a few ways one might respond to his argument. One would be to deny that his case is one of group adaptation. This is where Maynard Smith goes, in Chapter 7's 'Reply to Sober'. He denies that altruism is a group adaptation, basically. His reasoning seems to be that the groups are not adaptive/optimised enough in the altruism case. If they were adapted then there would be no cheaters - no lower-level conflict at all. This seems too strong to me. After all, it would seem to lead Maynard Smith to have to say that house mice do not bear adaptations, in so far as they carry t-alleles for meiotic drive.

Another possibility, which Maynard Smith points to, is to deny that external mechanisms of assortment, such as the one described in the altruism case, will occur at all often in nature. For all intents and purposes then, systems in which there is group adaptation will be ones in which there is group heredity. 

My intuitions on this aren't massively strong, but i know that Pierrick Bourrat has argued that for an evolutionary transition to occur you need group heritability, in the statistical sense. I could be talked into being less strict about group heredity having to be non-fortuitous i think. I'm not as worried about cross-level by-products as some either (but i'll save that for another post!)

I think i'm inclined to say that if you start thinking through the details of what a plausible assortment mechanism would look like in real life, it would probably sufficiently associated with the parent group that it would qualify as true group heredity. I can't imagine 'fortuitous heredity' lasting very long, and certainly not for traits that require a more complex sort of organisation than merely a high frequency of some lower-level type. Germ separation is an oft-used mechanism in nature, but that's a true group trait if ever there was one. 

I think Lloyd's notion 'bearer of adaptation' is perfectly serviceable if we want to start with a current trait and explain it in terms of a selection (and response!) history. But if we want a forward-looking notion, one that can support predictions about the dynamics of evolving systems, then its enough to locate units which are capable of producing adaptations, given sufficient time, stable directional selection pressures and so on. I guess Sober's point is that if we insist on group-level heredity, that isn't a mere statistical summation of lower-level heredity, then our predictions might go wrong, because we'll miss cases like the group altruism model. 

But, it's past 10pm at this point, so i no longer know anything. I'll think about it again tomorrow. Basta oggi.



*Briefly, i talk about a capacity for being selected, rather than actually being selected, because i don't think whether or not something qualifies as an evolutionary individual ought to depend on whether or not selection happens to be acting right now. Apart from anything else, it would mean that a solitary bacterium in a new habitat wouldn't qualify, because it isn't a member of a population. But of course, give it time, and it might be!


No comments: