Re: Directed evolution

From: Robert J. Bradbury (
Date: Fri Apr 28 2000 - 05:45:00 MDT

Oh boy, is this a can of worms....

On Thu, 27 Apr 2000, Technotranscendence wrote:

> On Saturday, April 15, 2000 5:24 PM Anders Sandberg wrote:
> > Normal evolution is not deliberately directed, it progresses in the
> > usual blind variation-replication-selection manner with no memory.
> I disagree. There seems to be evidence of orthogenetic (not a typo; I do
> NOT mean ontogenetic here) processes in evolution. E.g., back and forth
> mutation rates are usually not equal. (Drisophilia antennae mutate into
> legs usually. Very rarely do we see the reverse.) There are other
> examples.

I think you may be misinterpreting Anders statement. I believe he is
saying that there is "no memory" with regard to what previously may
have worked in an environment. Nature can develop something, it can
work for a while, get lost and then be reinvented. The local optimal
adaptions may be reinvented many times from many directions, "eyes"
and "flight" for example.

When discussing mutations you have to be careful to avoid overgeneralizing.
Mutations in specific base pairs, amino acids and gene sequences and
arrangements (such as the homeobox genes) are all governed by different
processes that are environment and genome specific. For example, there
are gene rearrangement "effectors" in Drosophila that do not exist
in other species. There may be hyper-mutation stress responses in
bacteria that do not exist in other species. The entire history
of mammalian genomes is an interesting study in the development and
divergence of retrotransposible elements and random chromosome
rearrangements. The genome and gene duplications that allow the
evolution of increased diversity are not easily "reversed".

There may be no "memory" in genomes, but there is a "history".

> It might be better to say that evolution appears to be "blind" in terms of
> fitting the environment. The internal (to the genomic system or organism or
> species) mechanisms producing diversity do not appear to "know" about the
> environment.

The bacterial "stress" responses would seem to contradict this. Genome
histories seem to retain the fact that when the environment gets difficult, you
want to increase the mutation rate to increase the probability of survival.

> > Intelligent entities are capable of creating systems based on
> > earlier experience and make deliberate jumps in form space that would
> > be unlikely with undirected evolution.
> The use of "undirected" here can be misleading. I would say a lot of
> evolutionary processes are directed, but not in the sense of there being a
> conscious goal.

I would disagree. They are "directed" only in the sense that they try
all possible variations from a particular state. Anders is commenting
on the problem of conciousness determining that a local optima has been
reached that cannot be transcended by simple random variations. You have to
have conscious "sight" to see that there are nearby "peaks" that are higher
than the local situation and determine how to "leap" over to those peaks.
The real problem is that evolution cannot backtrack!

> Instead, what I mean is that the initial conditions of
> biological evolution (for each of us, our initial condition is our genetics
> and cytoplasm at birth; for a population of organisms, the initial
> conditions are its genome, which is typically nonrandom) constrain in which
> directions variation will take place. This sets the ground on which
> environmental fitness (boundary conditions) can act.

Precisely. This "ground" is the limited space in which evolution
can occur and it requires consciousness to see the unoccupied spaces.

> I think in a lot of areas, we could
> benefit from using things like evolutionary computation because minds are
> constrained too.

Clearly the approaches are complementary. All "conscious" approaches
may suffer from blind spots. Evolution suffers from limited "foundations"
(occupation points in the phase space). The question is whether there
are mathematical approaches that can be clearly defined to search the
phase space, avoiding the blind spots and exploring the undeveloped


This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:09:54 MDT