Biology Reference
In-Depth Information
again under different selection regimes: Networks in which all nodes start with
equal generative entrenchment will spontaneously break symmetry, generat-
ing differential entrenchment under random mutation (another consequence of
genericity in Kauffman's sense, also illustrated in Fig. 1). Differential gener-
ative entrenchment will also arise spontaneously with the random addition of
modifier loci or with environmental fluctuations differentially affecting different
genotypes (Wimsatt & Schank, 1988; Wimsatt, 2001). But entrenchment and its
conservation under selection make increasing deviations from perfect equality
or symmetry (with no differential entrenchment) inevitable, and therefore self-
amplifying. Loci persisting longer for any of these reasons have a greater chance
of acquiring additional modifier loci, leading to their further entrenchment, and
increasing disparities in entrenchment. Cellular differentiation in metazoan evo-
lution presumably inevitably does the same thing and is crucial to the evolution
of increasing size, as environmental heterogeneities for cells located in differ-
ent places in the cell mass become inevitable and specialized transport and
coordination mechanisms become essential.
So why should this matter? Loss of a node in a network through which many
nodes are reached should cause more disruption than those leading to only a few.
(The same goes for changes in its properties, which are more highly constrained
if those connections are going to remain unchanged.) So, prima facie, more
negative
18
selection coefficients should be assigned to changes in nodes with
more nodes and connections downstream. This property is plausibly generic for
causal mechanisms of all types. It may be realized differently in mechanisms of
different types and appear differently in different representations of their static
and dynamic structure, but it seems unavoidable.
Notice also that selective consequences and intensities emerge directly from
the structural properties of the systems under consideration. This is important:
when this is true, selection coefficients are not external 'add ons' to black-box
models of the phenotype, as was true for population genetic models.
19
So the
complaint that distances selection models from system structure is not valid
for evolutionary models based in differential generative entrenchment. They are
properly part of the subject matter of the NSB.
For deeply entrenched traits, the negative consequences of changing them in
uncontrolled ways are virtually unconditional. The chances of making a change
18
This simple way of putting it makes it look like a monotonic relation between variables, but actually we are
talking about changes in the means and higher moments of distributions.
19
In artificial life simulations a systematic distinction is made between simulations in which fitness measures
are 'intrinsic' to the artificial organisms and those given externally, where the designer of the simulation
specifies the choice rules for what is to be optimized. In the former case, the only way to tell which
morph is fittest is to see if any succeeds systematically. While in principle this could be inferred from
'engineering models' of fitness, the interactions 'in practice' inevitably contain unanticipated dimensions and
new unexpected consequences.
