Generative Entrenchment and the Possiblity of Inheritance.

[Nick Thompson]

Dear Loose group of Correspondents,  
 
For a couple of years now, you all have been suffering with my inability to state what seems to me a fundamental paradox arising from the Developmental Systems Theory: that natural selection is impossible without inheritance and inheritance seems increasingly impossible given the complexity and chaos of developmental systems as we are coming  to know them.   I have just come across a clear statement of this paradox in Wimsatt, W. C. and Shank, Jeffrey C. (2004).  Generative entrenchment, Modularity, and Evolvability; When Genic Selection Meets the Whole Organism.  In, Schlosser, G. and Wagner, G. P.  Modularity in development and evolution.  Chicago: University of Chicago Press.   The title would seem to suggest that the problem they identify relates ONLY to the relation between the organismic and the genic level, but in fact it is potential troublesome at all levels of selection.  
 
The generative structure of the system (including the organism plus relevant aspects of its environment) has a characteristic set of causal interactions which could be variously represented.  One of the simplest representations is a directed graph, where nodes are parts, processes, or events, and arrows are consequences of the presence or operation of notes on other nodes.  For each node, consider how many other nodes can be reached from it by following the arrows.  This indicates how much of the phenotype is downstream of, causally dependent on, or affected by a given node.  We define Generative Entrenchment as the magnitude of this downstream dependence.  [Page, 360, Caps and italics by nst]
 
Darwinian processes should almost inevitably give rise to generative structures (Wimsatt 2001).  However, we are still left with two perplexing questions:  How can complex adaptive systems evolve and continue to evolve in any other than a predominantly accretionary way if their generative elements become increasingly entrenched with increasing complexity (Shank and Wimsatt 2000). How does this permit continued modular evolvability? It is no surprise, therefore, that fundamental research focus of the evolutionary sciences is to figure out how complex systems can continue to evolve when evolutionary processes generically give rise to entrenched structures.  We call this the G[enerative] E[ntrenchment] paradox.  [Page 363
 
In these passages, Wimsatt and Shank lay out with perfect clarity the problem I have been fumbling with.  However, by focusing on Generative Entrenchment, they conceal one startling implication that I see in their view (possibly because they don’t believe it).   Generative Entrenchment threatens Natural Selection because Natural Selection requires some sort of inheritance, and, so far as I can see, any trait that is Generatively Entrenched cannot be inherited at the level at which it is entrenched.   I think I perhaps have a solution to this problem, but I will hold off offering it until I have convinced anybody of the existence of a problem to be solved.  

I apologize for intruding on your otherwise Peaceful saturday.

Nick


Nicholas Thompson
nickthompson at earthlink.net
http://home.earthlink.net/~nickthompson
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://redfish.com/pipermail/friam_redfish.com/attachments/20060520/0b74e050/attachment-0001.htm
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Doc2_.doc
Type: application/msword
Size: 23040 bytes
Desc: Doc2_.doc
Url : http://redfish.com/pipermail/friam_redfish.com/attachments/20060520/0b74e050/Doc2_-0001.doc

[Jochen Fromm-3]

I am not convinced that there is a problem here.
Can you explain the problem in simple words without
using the term "generative entrenchment" ? This
should be possible if it is more than a nebulous idea.

"Generative Entrenchment" is a concept proposed by
William C. Wimsatt, a Professor of Philosophy from the
University of Chicago. He uses "Entrenchment" in the
sense of encapsulation. We can investigate a system
and its parts because not everything is connected to
everything else. If there is a high interdependency
between many dependent modules in a complex system,
how can a complex system be controlled and organized if
all components of a system are closely linked together ?
Obviously it cannot work if parts of the system
are not isolated and encapsulated from each other.
I guess encapsulation and codes are the key here.
The phenomenon of strong emergence comes to mind.
Once a new system evolves in an old system, the old
system has reached the point of maximal complexity
and apparently stops to evolve, because a change in
the base would topple the whole system. This means the
system at the base and the connection between both
systems (in form of the corresponding code) are frozen.

However is there a need for a new complicated buzzword ?
I would rather consider philosophers and sociologists
as experts in "Generative Entrenchment", because
they ENTRENCH (or encapsulate) our ignorance so well
behind newly GENERATED complicated terms.

-J.

[Martin C. Martin-2]

This may be wrong, but I think the idea is that, if the mapping from
genotype to phenotype is very complex, mutation/crossover will have
essentially arbitrary effects.  It will be hard for any such change to
have a coherent effect on the individual.

Things like development make the relationship between a gene and it's
effect on the organism very complex.  In other words, the emergent
behavior of a complex system can be very surprising, and the
interactions between development rules can lead to many surprises.

In a way, it's a question of representation.  As AI has discovered many
times over, "representation is king:" the hardest part of applying any
technique is getting the representation correct.

Which suggests that an interesting question is: what representation is
life on earth using?  Not just how DNA maps to proteins, but how
mutations affect the final genotype.  Perhaps variation and selection
are the easiest part of understanding evolution; perhaps understanding
the representation will give us much more insight.

This may have nothing at all to do with what the original poster
intended, but I think it's an interesting idea on it's own.  :)

- Martin

Jochen Fromm wrote:

> I am not convinced that there is a problem here.
> Can you explain the problem in simple words without
> using the term "generative entrenchment" ? This
> should be possible if it is more than a nebulous idea.
>
> "Generative Entrenchment" is a concept proposed by
> William C. Wimsatt, a Professor of Philosophy from the
> University of Chicago. He uses "Entrenchment" in the
> sense of encapsulation. We can investigate a system
> and its parts because not everything is connected to
> everything else. If there is a high interdependency
> between many dependent modules in a complex system,
> how can a complex system be controlled and organized if
> all components of a system are closely linked together ?
> Obviously it cannot work if parts of the system
> are not isolated and encapsulated from each other.
> I guess encapsulation and codes are the key here.
> The phenomenon of strong emergence comes to mind.
> Once a new system evolves in an old system, the old
> system has reached the point of maximal complexity
> and apparently stops to evolve, because a change in
> the base would topple the whole system. This means the
> system at the base and the connection between both
> systems (in form of the corresponding code) are frozen.
>
> However is there a need for a new complicated buzzword ?
> I would rather consider philosophers and sociologists
> as experts in "Generative Entrenchment", because
> they ENTRENCH (or encapsulate) our ignorance so well
> behind newly GENERATED complicated terms.
>
> -J.
>
>
> ============================================================
> FRIAM Applied Complexity Group listserv
> Meets Fridays 9a-11:30 at cafe at St. John's College
> lectures, archives, unsubscribe, maps at http://www.friam.org

[Carlos Gershenson]

Dear Nick,

I also don't see a problem for inheritance, but from a different  
perspective, that of discrete dynamical systems (such as cellular  
automata and random Boolean networks).
In the early 90's, people like Langton (using CA) and Kauffman (using  
RBNs) suggested that life and computation (and also evolvability)  
must lie somewhere close to the phase transition between ordered and  
chaotic (the popular "edge of chaos"). This is because ordered (or  
frozen) dynamics do not allow novelty in evolution nor information  
transfer in computing. On the other hand, chaos (given by too many  
dependencies or links) loses useful traits already acquired by  
evolution or stored information. Thus, you need a bit of both:  
stability to keep what you already evolved or computed, but with some  
variability to allow the exploration of new solutions and information  
transfer...

There are other subtleties that have been coming up in the years  
since, but I think this is enough to explain why there is no problem  
of inheritance: natural selection selects inheritable systems...

(If anybody is interested, I have a short tutorial on random Boolean  
networks at http://homepages.vub.ac.be/~cgershen/rbn/tut/index.html )

Best regards,

     Carlos Gershenson...
     Centrum Leo Apostel, Vrije Universiteit Brussel
     Krijgskundestraat 33. B-1160 Brussels, Belgium
     http://homepages.vub.ac.be/~cgershen/

   ?Tendencies tend to change...?