[FRIAM] (no subject)

Wed Mar 24 20:10:55 EDT 2021

Thank you EricS for this long and thoughtful answer.  I hope The Gang will spend a lot of Friday, mining it out.  What I, of course, appreciate the most is your acknowledging that The Problem is in fact A Problem.  That you give me some sources to read if I am ever to understand it better, is gravy … good thick dark gravy. 

Nick 

Nick Thompson

 <mailto:ThompNickSon2 at gmail.com> ThompNickSon2 at gmail.com

 <https://wordpress.clarku.edu/nthompson/> https://wordpress.clarku.edu/nthompson/

From: Friam <friam-bounces at redfish.com> On Behalf Of David Eric Smith
Sent: Wednesday, March 24, 2021 5:46 PM
To: The Friday Morning Applied Complexity Coffee Group <friam at redfish.com>
Cc: David Eric Smith <desmith at santafe.edu>
Subject: Re: [FRIAM] (no subject)

Nick, hi,

I won’t be able to do anything like even a good-faith effort to reply to this thread.  The time it would cost me to compose something I wouldn’t either disagree with or regret having written will cost me more shame and punishment for delinquency than I can budget at the moment.

Your outline flags all good things.  I am not sure whether it provides a framework from which to make progress in a community that likes to fight (not FRIAM, the evolutionary silverbacks).

I do not claim, at all, to address an encompassing picture or propose a strategy that could be worked through to some satisfactory resolution.  I will claim that part of the problem is a bad problem in conceptual delineation in much (not all!) of the community, which the terminology canalizes and makes it hard to escape from.  For part of that I do have something I think is a corrective.  I don’t know if it will ever be accepted anywhere, so I put it on the BioRxiv here:

https://www.biorxiv.org/content/10.1101/2021.02.09.430402v1.abstract

Don’t be concerned yet (or maybe ever) with what I build, but perhaps take it as a scaffold through which to see more clearly part of the existing problem, which I think is not too hard to correct.  

A good starting foil, which illustrates several of the points in your outline with its own internal contradictions, is this one by Lynch:

https://www.pnas.org/content/104/suppl_1/8597

His whole thesis is that combinatorial factors aren’t heritable and so can’t be selected (the arch-form of which is Williams/Dawkins genic selection; the endlessly-nebulous form of which is the Units of Selection chautauqua that will never end; I guess Samir Okasha gives as good a version of it as can be found).  Yet Lynch ends by saying (somewhere near the bottom of the paper) “so you see, we are stuck with admitting that genotypes are the units of selection”, after just having said that the arrangement that makes a genotype different from a gene isn’t selectable.  

So I say go back a few steps and take a different tack.  Various parts of the following are all well handled by George Price, Warren Ewens, or Steve Frank, cited by me in the above:

1. Fisher says, distinguish the definition of fitness as a summary statistic, from the causal models of sources of fitness that you are trying to estimate from samples of that statistic.  Then fitness will never be a tautology.  

2. Fisher’s definition is apportionment of offspring to ancestors by type.  Immediately that summary statistic doesn’t exist for anything that doesn’t have a simple forward branching structure, so we have an unbridgeable conceptual inconsistency with Darwin’s differential reproductive success, which treated organisms (Steven Gould in SET argues very hard) and didn’t worry about how the mechanic of heredity would make that hard to quantify.

3. The only place the two don’t come into conflict are replicators, and voila! We have the corner into which the modern synthesis painted evolution, which led to excess emphasis on replicators and a lack of system to handle anything that isn’t a replicator.

4. Fisher of course mostly worked on diploid organisms, which are not replicators, but having introduced differential apportionment precisely to get out of the tautology “fit are the fit”, he said “well, if it doesn’t have one parent, do additive regressions”.  Shubik would go on an epic rant that if your definition of a process depends on which case you are trying to model, it was never a definition and you didn’t have either a theory or a formal system.  (What does “+” mean in this thing you call “addition"?  Well, depends on whether you are a republican or a democrat. etc.). All that is entirely right, and should be held to forever.  Lewontin would point out that regressions are not explanatorily or causally sufficient, since the thing you call fitness depends on idiosyncrasies of the sample of the population state.  So you call it a property assignable to genes, while ensuring that It never is, except in the cases for which you didn’t need it, and even those it mis-handles (in my paper; trivial but didactic example from an earlier paper of Ewens and Lessard, cited.)

5. So where are we now?  Population genetics has tried to do the best it can with this history, and aspires to the quite valid wish for a compact formal abstraction that will project out infinite detail but keep a usable computational system.  The way it does this — says I — is to call “evolution” a Polya’s Urn problem that hands _names_ of entities (genes or other “units of selection”) to The Rest of Nature, and nature hands back “fitnesses” for those names from a black box.  This separation is essential to their reduction, and it then dictates that fitness be the only channel for information flow.

6. So now I can say briefly what my claim is:

6a: Standard pop-gen is committed to separating gene counts from _what genes do_, the latter going into the black box of the genotype+environment -> phenotype -> fitness map.

6b. If you do that you have cut out infinitely many information channels that selection is actually using.

6c. Price recognized part of this, as later have Steve Frank, David Queller, and Ben Kerr and Peter Godfrey-Smith in varying ways.  Price kept with the idea he calls “corresponding sets” to define fitness, but didn’t make a draconian projection to replicators, because he recognized that definition could be applied at more than one stage in complicated lifecycles.  Frank, Queller, and Kerr have done other things, which I will let you find in the paper if you care; they are all good, and all stop short of an actual solution.

7. So I say, we can keep Fisher’s separation of summary statistics from causal model inference, but include a far richer set of information channels, by formalizing the idea of a lifecycle, and looking at the reproductive cycles within a collection of related possible realizations of the lifecycle, running regressions on those rather than on replication events.  This is no more radical than what we do in chemistry already, recognizing that a molecule is more than a bag of its atoms, even though the relations (bonds) get changed by reactions.  There are good tools for such modeling; it is still compact (though less so than the replicator abstraction), and in the sense of paradigm expressiveness and computational complexity, it is a qualitatively richer class.  

7a.  I would say that the way to say this is that we all throw out infinitely much of “what genes do”, or black-box it somewhere, but the lifecycle approach allows us to take in a category of things genes do, into the selection model itself, that the replicator abstraction requires us to exclude.

7b.  Really REALLY I do not mean I am introducing “other fitnesses”, or “generalized fitness” or any “modifier-fitness”.  Fitness is fitness; Price defined it well, it is a restrictive concept tied to either replicators or the modestly richer “corresponding sets”, and it omits infinitely much.  Don’t think about elephants.  What we want to do is recognize everything else in the universe that aren’t elephants.  There are infinitely many other summary statistics that are not fitness and that mean and cover different regularities.  We have machinery to use many of them; we should.

7c.  Another way to say this is that we start with an object semantics that is richer than the replicator and better able to capture features of the different character of biological objects and the transformations that relate them. 

8. So from then one can do a few elementary things that nobody will care about, to convince the author that there is something to say in all this.  That and $5.50 will buy you a cup of coffee at Starbucks.  

8a. The Price Equation exists for lifecycles, and a variety of things that Fisher’s and Price’s form get wrong, in the sense of causal or explanatory insufficiency, are got correct by the lifecycle version.  It doesn’t mis-specify models without necessity, and doesn’t introduce population-state dependence where it isn’t needed.

8b.  The interpretation of Fisher’s covariance term as “causal”, which was wrong since the beginning and Warren Ewens correctly shows has always been wrong, becomes correct (at least for a much wider suite of phenomena) in the lifecycle model.

8c. If you formalize the model-inference problem in the theory of large deviations for trajectories (one should formalize everything in the theory of large deviations for trajectories if at all possible), you can extract the information channel associated with each lifecycle flow and regression coefficient, and in that way quantify _what_ information is coming in through the coordinated action of objects of different types at multiple stages in the lifecycle.  One also recovers all the usual stuff — least-squares criteria, additive projections if you want — in suitable small-error limits.

In all the above — apologies — I didn’t actually answer the theme of the thread, which is how developmental complexity relates to modularization either in the principle components of environment forcing, or in the statistics of heredity.  I fully think in D’arcy Thompson terms here, but I think the others on the list already do too, so there isn’t anything I can say that adds there.

I think the modularization question will turn out to be hard (sense of multi-faceted and rich, and branching into several principles that will become evident as we start to solve more cases of it), and will be related to the power of reinforcement acting on populations as a regularity-extractor.  I like what Leslie Valiant has to say in PAC, as a way of framing the issues, even though I know he is a punching bag for the geneticists because of various things they understand are important that he isn’t trying to understand deeply or deal with.  I don’t care that he had limitations; the question for me is whether there is a good insight there that could be developed further.

What I intend in the above is that the lifecycle/hypergraph abstraction is a more expressive class of formal models within which one can pose such problems, and we should be able to generate more interesting answers by using it.

Anyway, long, sorry, but….

Eric

On Mar 25, 2021, at 6:52 AM, <thompnickson2 at gmail.com <mailto:thompnickson2 at gmail.com> > <thompnickson2 at gmail.com <mailto:thompnickson2 at gmail.com> > wrote:

Dear EricS,

It occurred to me that you, of all people, might have something to say about the question of how natural selection came about.  I assume it was “scaffolded” by some physical process or constraint in a manner analogous to the “scaffolding” of life by white smokers at the bottom of the sea.  Most accounts of evolution I have encountered start with the assumption of sufficient modularity for selection to go forward.  But given all the inevitable trait-entanglements in the developmental process, this assumption seems wildly gratuitous to me.  Any thoughts you might share with me (us) would be great appreciated.   I attach a decade-old outline of an essay I never wrote (because I wasn’t really comptent to do so) to explain the problem

Hope you are well and getting vaccinated soon. 

Nick 

Nick Thompson

ThompNickSon2 at gmail.com <mailto:ThompNickSon2 at gmail.com> 

https://wordpress.clarku.edu/nthompson/ <https://linkprotect.cudasvc.com/url?a=https%3a%2f%2fwordpress.clarku.edu%2fnthompson%2f&c=E,1,2RrnbOD8pAIB7OFrnOwfqQ7sY7OY8vFbobl3mQjNBfSUN8-y26AW2dc0pD5o2oic2uW41fDw_9ilyeoXdRsowLdvy9Dn18acOzDH2ladP9PlRgfZ1jq1et5h&typo=1> 

<Warring Darwinians for Glen Steve (003).doc>- .... . -..-. . -. -.. -..-. .. ... -..-. .... . .-. .
FRIAM Applied Complexity Group listserv
Zoom Fridays 9:30a-12p Mtn GMT-6   <http://bit.ly/virtualfriam> bit.ly/virtualfriam
un/subscribe  <https://linkprotect.cudasvc.com/url?a=http%3a%2f%2fredfish.com%2fmailman%2flistinfo%2ffriam_redfish.com&c=E,1,oAxSEUSoltvXsWGfGJcX45p2qScq04DSpheJpq-yigWdnySQeeBq9RS-fRSxaf-buPGLv1DCz_0tgTEiaBfZCL004bbJMRputfesc_EvOO9Q2g,,&typo=1> https://linkprotect.cudasvc.com/url?a=http%3a%2f%2fredfish.com%2fmailman%2flistinfo%2ffriam_redfish.com&c=E,1,oAxSEUSoltvXsWGfGJcX45p2qScq04DSpheJpq-yigWdnySQeeBq9RS-fRSxaf-buPGLv1DCz_0tgTEiaBfZCL004bbJMRputfesc_EvOO9Q2g,,&typo=1
FRIAM-COMIC  <https://linkprotect.cudasvc.com/url?a=http%3a%2f%2ffriam-comic.blogspot.com%2f&c=E,1,MQGkQfEWfHzFuLRzlN-ZIJm5ziEBcGH-atm4lb2HPSL4_SLbF2ixMtt7rZ8tjI_zFKMBrBYfrpAfQZqLOHT5ewulspe16jG2U7gxlnfU82o,&typo=1> https://linkprotect.cudasvc.com/url?a=http%3a%2f%2ffriam-comic.blogspot.com%2f&c=E,1,MQGkQfEWfHzFuLRzlN-ZIJm5ziEBcGH-atm4lb2HPSL4_SLbF2ixMtt7rZ8tjI_zFKMBrBYfrpAfQZqLOHT5ewulspe16jG2U7gxlnfU82o,&typo=1
archives:  <http://friam.471366.n2.nabble.com/> http://friam.471366.n2.nabble.com/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20210324/1d39a614/attachment-0001.html>