Conclusion
Acknowledgements
So: Does it or Doesn't it?
In summary, my claims are:
- The "replicator" terminology, as promulgated by Dawkins and
Hull, suffers from significant, and confusing, internal
contradictions. It would be best abandoned.
- The units of selection in Darwinian evolution are
lineages rather than actors.
- More precisely, the units of selection are S-lineages. Unfortunately, these are somewhat complex and even nebulous
entities. In particular, they are not simply groups of actors
which share some overt, inherited, "trait" ; nor are they simply
"species" in any conventional biological sense. I suspect that
S-lineages can only usefully be identified in a relational sense,
in the context of a particular episode of Darwinian selection. That is,
we identify an S-lineage not so much by any similarity among its
members, as by the competively significant difference between two
(or more) S-lineages.
- Actors may have a von Neumann-like genetic
architecture. This is generally the case for contemporary
terrestrial organisms. In this (special) case, an S-lineage can
typically be identified or marked by a particular segment of the
genetic description, shared by all members of the S-lineage. This
makes it tempting to regard this genetic "tag" (or "gene" ) as
being the S-lineage, or to regard this genetic level of
description as uniquely preferable for describing Darwinian
selection dynamics ( "genic selectionism" ). This temptation
should be resisted. A particular genetic tag can only identify a
selectively important "trait" in the context of the
actors which express or translate it; to put it another way, the
"meaning" or "decoding" of a particular tag can vary (in
principle by an arbitrary amount) from one actor to another, even
in the "same" lineage; furthermore, in principle, the same
"trait" can be correlated, independently, with multiple distinct
genetic "tags" .
- The logic of Darwinian evolution requires replication with
some kind of "heritability" so that coherent S-lineages can
form and compete. However, it does not require any
particular kind of hereditary mechanism, and, in particular, does
not require that the component actors have a well-formed
von Neumann style genetic architecture.
One reader of an earlier draft of this paper wondered what (if any)
pragmatic implications it had for actually designing artificial
evolutionary systems. I can only answer for myself of course. The most
important implication for me is that, once the mystique of "genic"
selectionism is dispelled, we can usefully ask how it is that
von-Neumann style genetic organisation can spontaneously come into
existence at all. As long as "replicators" (in Dawkins' narrow sense
of fragments of genetic material - or even lineages thereof) are
considered as necessary pre-requisites for Darwinian
evolution, we can hardly even formulate this question, never mind
answer it. But once we separate the notion of genetic
organisation from the notion of the units-of-selection, we can
envisage a form of ALife investigation in which we do not already
"wire in" a genetic architecture, but ask how it could arise
and be refined - by a process of Darwinian (but not yet
"genetic" !) evolution. This is, of course, a fundamentally
different direction from "conventional" GA or evolutionary
strategy investigations. I suggest that it is a worthwhile
alternative to pursue.
Acknowledgements
So: Does it or Doesn't it?
McMullin@eeng.dcu.ie
Mon Mar 4 14:08:30 GMT 1996