264 Ulrich Krohs
vidual.^8 It is much more plausible to regard evolvability as a by-product of other selective
processes. Following Elliott Sober’s distinction between “selection for” and “selection of ”
(Sober 1984: 97–102), there is merely selection of modularity and evolvability. As Wagner,
Mezey, and Calabretta (2005) concisely put it, evolvability is selected only indirectly.
An argument similar to the one against selection for evolvability also holds with respect
to selection for modularity due to a supposed increased evolvability of modular systems.
This becomes clear from a comparison with the technological case. In engineering it is
well known that modular structure, though enabling for quick modifi cations of the system
within a specifi ed range of related designs, may in the long run decrease the further evolv-
ability of a system. Existing modules, in particular if they are used in parallel in different
contexts, may pose constraints on the evolvability of a class of artifacts (Pahl et al. 2007:
509). A modular design is less fl exible than individual design with respect to the adoption
of changing requirements as soon as a certain design space covered by the modular systems
is left. What can be learned from the technological example is that modularity does not
increase evolvability in every case and each respect, but that any particular modular design
also poses evolutionary constraints on the organism in question, mostly in cases where a
particular module serves different roles in an organism. Any change needs to be compatible
with all of these roles. This limits the range of viable variation and may well limit the
evolvability of modular structures. Consequently selection of modular structures cannot
be explained in general as selection of modularity due to increased evolvability.
Let me come back to arguments in favor of selection of modular systems that do not
refer to a supposed increase in evolvability. What must still be expected from an adaptive
explanation of modularity is an argument for why there is selection of modular struc-
tures—though not of modularity as such. In their discussion of evolutionary explanations
of modularity, Wagner, Mezey, and Calabretta (2005) compile scenarios of eight evolution-
ary paths, with the proviso that the scenarios will have to be adjusted to new fi ndings,
since in most cases empirical data are still missing. Among these scenarios are the stepwise
lowering of integration, occurrence of new modules after duplication of components so
that the additional component may take a new function, or the occurrence of developmen-
tal modules by a sorting process that collects genes contributing to the development of the
same character. Because empirical data are lacking, simulations are crucially important to
support the relevance of the scenarios. Interestingly only some of the simulations the
authors quote yield modular structures, and they do this only under a limited range of
assumed conditions. The degree of modularization that is reached is low in most cases.^9
However, if one looks not to the mathematical implementation but to the conceptual
background of the simulation models—which boils down to looking at the arguments
discussed here as arguments to make plausible why modular structures evolved—there
seems to be no reason for this resistance to full modularization. The conceptual framework
that is applied supports an adaptive view on modularity. But it seems to know of the
advantages of modularity only. The simulations, which show results with even less