Darwin's Mistake

Tom DiBenedetto tdib at OCEANCONSERVANCY.ORG
Mon Jul 15 13:15:28 CDT 2002

Ken Kinman has raised the issue of classification methods again, and I would like to
address, in an explicit manner, the reasons why, though I respect his concerns (and
his efforts), I feel that his approach is misguided. In a response to one of his previous
posts, I pointed out that phylogenetic systematics was not intended to represent
more than one aspect of evolution. Nevertheless, I will here argue that cladistics, in
practice, has actually become a method that succeeds in representing both major
aspects of evolution (relationships and anagenetic change) in a manner that is far
more useful and accurate than either his method or traditional Darwinian methods.

Ken's general approach has its roots in what I am here referring to as Darwin's
mistake. It is the notion that the traditional Linnean ranking system could be utilized
as a variable in a classification, to capture a sense of anagenetic change. This is not
an absurd or stupid notion, it can probably be legitimately considered to be a creative
and interesting proposal. It clearly captured the imagination of systematists for a
century. But it is fatally flawed, and the time has (long since) come to recognize that,
and to move on.

The fundamental problem with using the ranking system for this purpose is that the
ranking system is, by its very nature, hierarchical. "Species" refers to specific
instances of a phenomenon. "Genus" refers to general categories that encompass
the specific instances. "Family" and higher ranks were explictly proposed and used
as higher-level groupings of this type in an overall hierarchy. But the phenomenon
that Darwin wanted to capture - the degree of change - is not hierarchical. It is, if
anything, linear or horizontal. It is the relationships that are hierarchical - in other
words, it is the _other_ aspect of evolution (the relationships) that is properly
addressed by an inherently hierarchical system. And that is, of course, how anyone
looking at a hierarchical classification will see it. The root problem is the attempt to
jam two "reality variables" into one "presentation variable". The result is that the
hierarchy fails to accurately represent the real, natural hierarchy (the relationships)
and the other reality variable, the "degree of change" is not precisely represented
either. Instead of some explicit representation of the degree of change, one is forced
to represent it merely as an adjustment to the hierarchy. So we are presented with a
hierarchy, but must quickly advise everyone that it is not meant to really represent
THAT hierarchy (the one everyone imagines is being represented, the relationships),
but it is a modified version, modified in some subjective "black box" manner that
utterly fails to give us any meaningful measure of the actual degree of change (the
degree of change is not presented as being equal to 5.4, or something like that,
merely "sufficient to raise a family to the level of a class"). What is really sad, and
seemingly desparate, is the attempt to argue that this distorted hierarchy is somehow
a valid representation of historical reality.

Finding a means to represent two different aspects of a situation is not a problem
that is restricted to systematics. As a conceptual example of how to approach this, I
offer the idea behind a GIS - a geographical information system. One has a set of
variables - e.g. ground topography, road systems, cities, railroads etc. The goal is to
have coherent accurate representations of each of these variables, and to combine
them for analysis in a manner that respects their integrity, but allows one to see the
relationships that exist between them. This is achieved by maintaining each data
source as a coherent, separate entity, but then layering them over eachother in a
manner such that the relationships between the layers is visible. When considering a
representation of evolution, we have, at least for now in this discussion, two layers of
information. The method that has emerged in cladistic practice essentially entails
laying out the relationship scheme (the cladogram) in a manner that is analagous to
a base map in a GIS. On top of this layer one can then align the data that represents
the "degree of change". This data can be, and is, quantified as the number of
apomorphic changes, or it could conceivably be quantified in any other manner that
you choose. As a linear measure, one can represent this data using visualization
tools that seem appropriate. For instance, one can change the branch lengths of the
cladogram (without changing the topology) in proportion to the quantification. Or one
can put tick marks on the branches to represent apomorphies. One could also
change the color of different branches, or add a third dimension to the representation
and bend the branches back or forward in order to capture some quantified variable.
Just as in a GIS, there are many variables (many layers) that can be accomodated.
The crucial point though, is that each layer retains its integrity and can always be
seen clearly, in isolation, if one wishes to do so. These type of representations have,
of course, been available to the community in software packages for at least a

One might argue that classifications need also work in a textual format. Whereas
such a format is limited in the number of variables that can be represented in a
coherent manner, it certainly can accommodate the two aspects of evolution we are
discussing here. A cladistic textual classification with parenthetical notation of the
number of apomorphies at each level (or any other quanitified variable) can clearly
present the degree of change against the backdrop of the coherent phylogeny.

Having both followed and engaged in the discussions that make up the "cladist wars"
for many years now, I must acknowledge that Ken is far more sensitive to the real
issues and challenges that cladists have raised than are many of the other defenders
of the Darwinian systematic tradition. Nevertheless, I do think that he is stuck to
some degree with a conceptual approach that is crude, and greatly limited. He has
advanced that system as far as it can probaly go. The notion that different aspects of
a phenomenon can be represented with discrete, but internally coherent, layers of
information is, I think, a great advance over the original Darwinian proposal to
conflate these aspects in a single presentation variable.

Tom diBenedetto

More information about the Taxacom mailing list