entropy:
Several important points have been raised in this thread. Unfortunately this
medium doesn't really allow detailed responses in a manner that promotes a
complete understanding. However, please allow me to try a few brief replies.
PHYLETIC GRADUALISM VS. DARWINIAN GRADUALISM
========================================
Tom Maier writes:
>Slow and steady almost always wins the race. And the same is true of virtually
>all forms of evolutionary optimization. Darwinian gradualism is the
>overwhelming rule in nature --
There is documented evidence of punctuation in speciation. Gradualism
may occasionally exist, but I think punctuation is the dominating
rule. When you look at a series of species over long time then
you could say that there is an effect of gradualism, but I think
that is an illustion of human hindsight. I suppose that if you
look at each generation during the punctuation event then
that could also be called gradualism (over a short time span).
========================================
Tom's comments reflect a common confusion between phyletic gradualism and the
gradualism advocated by Darwin. They are not the same same phenomena. But Tom
is by no means alone in this. Indeed, in a great many ways, Gould himself has
fallen prey to the confusion, resurrecting the notion of Goldschmidt's
"hopeful monsters" in several of his Natural History essays ten to twenty
years ago, especially in those essays that concerned themselves with items
such as homeotic mutations (e.g., antennapaedia).
Steven Stanley even more aggressively bought into the interpretation that Tom
wishes to place on the evolutionary process in his 1979 book, "Macroevolution:
Pattern and Process." In it (e.g., Chap. 6), he argues for such saltational
events as quantum speciation, primarily through mechanisms such as the
appearence of developmental defects (achrondoplasia, etc.), modifications in
regulatory gene structures (polydactly, etc.), spontaneous polyploidies, and
abnormal chromosomal rearrangements.
Niles Eldredge has rather vigorously backed away from much of these
exaggerated interpretations. In his 1985 book, "Time Frames, The Rethinking of
Darwinian Evolution and the Theory of Punctuated Evolution," Eldredge wrote,
"Simple and, I think, unobjectionable as the propositions of [PE] are, they
have been widely misinterpreted and, I would say, in some instances
misrepresented" (p. 16). He goes on to write on the same page, "Gould and I
have differed to some extent on the significance, the implications -- and even
on occasion, some aspects of the basic content -- of 'punctuated equilibria.'"
Paraphrasing Eldredge, much of what he says in "Time Frames" is, "Look guys,
you've wildly overinterpreted what we said." Indeed, on page 141, he writes:
"There is something of an antiadaptationist backlash going on within
evolutionary biology these days. The most common misconception about
'punctuated equilibria' -- that Gould and I proposed a saltationist [peak
jumping] model of overnight change supposedly based on sudden mutations with
large-scale effects (macromutations a la Richard Goldschmidt) -- in a way
reflects this altered mood."
However, Eldredge's attempts to set the record straight are as much a
reflection of the differences in fundamental philosophy between he and Gould
as they are differences between he and his critics. I sat through perhaps a
half dozen of Gould's talks ten to twenty years ago -- and Gould put words in
Darwin's mouth that Darwin never said. As many of the readers of this list
well know, much of that "antiadaptionist backlash" is due solely to Gould
himself.
Because Darwin spent much of his adult life with plant and animal breeders,
Darwin knew as well as any biologist alive the extraordinary rapidity by which
animal and plant varieties could be shaped under the extreme pressures of
artificial selection. Completely new breeds of orchids, pidgeons and dogs were
commonly created in Victorian England within the lifetime of a single breeder.
Nonetheless, in spite of that rapidity, or perhaps more accurately, because of
it, Darwin argued vigorously that the only probable route to evolutionary
change was through the gradual modification of the phenotype.
Darwinian gradualism is not phyletic gradualism. At the core of what Eldredge
and Gould were excoriating in their initial articles was nothing more than the
persistence of early 20th Century diagrams in biology texts that continued to
show the gradual modification of a lineage over perhaps millions of years. As
paleontologists, they simply didn't see this form of evolution in the fossil
record.
Moreover, as evolutionary biologists, they knew that they should not have
expected it, either. Novel incursions into new adaptive zones are almost
always accompanied by rapid "genotypic revolutions," to use Ernst Mayr's
terms. Gould is perhaps the most philosophically inconsistent of all modern
evolutionary biologists. On one hand, he will vigorously argue for the non-
optimality of current designs, and on the other, argue for long periods of
stasis in a lineage, punctuated by episodic periods of rapid change.
Similarly, he will argue for a lack of progressivity in evolutionary design
and almost simultaneously explain the reasons for the very rapid evolution of
eye.
Punctuated equilibria has little to do with Darwinian gradualism, per se.
Eldredge, Mayr, and virtually all professional evolutionary biologists,
including Gould more often than not, deeply believe in the accumulation of
micromutations as the principal -- if not sole -- mechanism by which phyletic
evolution occurs.
The observation that lineages stay so profoundly static for such long periods
of time is at once reliable and reasonably powerful evidence, albeit indirect,
of the general optimality of most lineages and the persistence of selection
pressures on those lineages.
DOES EVOLUTION CLIMB HILLS OR DESCEND WELLS?
=======================================
Tom also writes:
Punctuation and equilibrium of species make a lot of sense to me.
Dawkins and others have described evolution with the analogy of
"climbing mount improbable", That seems totally upside-down
in my view and I see speciation more as "falling into valley
probable". Mine is an adaptationist viewpoint with the species
spending most of their time with their gene pool centered in
a valley; the valley being a mode of life that does well for the
population. Genes and genetic combinations that stray too
far from the valley are trimmed out, so the gene pool remains
rather stable and centered most of the time and thus exhibits
equilibrium.
=========================================
On this point, and on this view of evolution, Tom and I very profoundly agree.
One of my former engineering students, David Fogel, wrote in his 1995 book,
"Evolutionary Computation: Toward a New Philosophy of Machine Intelligence":
"Atmar (1979), Templeton (1982), Raven and Johnson (1986, p. 400) and others
suggested that it is more appropriate to view the adaptive landscape from an
inverted position. The peaks become troughs, "minimized prediction error
entropy wells" (Atmar, 1979). Such a viewpoint is intuitively appealing.
Searching for peaks depicts evolution as a slowly advancing, tedious,
uncertain process. Moreover, there appears to be a certain fragility to an
evolving phyletic line; an optimized population might be expected to quickly
fall off the peak under slight perturbations. But the inverted topography
leaves an altogether different impression. Populations advance rapidly,
falling down the walls of the error troughs [where they will remain for
extended periods of time]" (p. 45).
Pictures and images do matter a great deal. There have now been several
generations of evolutionary biologists and geneticists who have sat through an
innumerable number of lectures where Wright's adaptive topography has been
drawn as peaks instead of troughs. The inevitable take-home lesson -- although
no one ever explicitly says it -- is that evolution is a slow and tedious
process, where the end result is a fragile compromise, ready to fall of the
peak at a moment's notice. But nothing could be further from the truth.
It was exactly such misleading imagery that Gould and Lewontin were primarily
objecting to in their initial articles on PE. I feel as strongly about the
"peaks" of adaptive topographies.
Inverting the adaptive topography is only a matter of a minus sign, and thus
is, in one sense, trivial. Any good mathematical evolutionary biologist could
invert the topography in his mind and come to the precisely the same
conclusions. Nonetheless, very few people do.
In the inverted topography, populations fall to the center of the troughs at
speeds proportionate to the steepness of their various walls and remain there,
immobile, essentially trapped, unless a second well lies very close indeed --
or the nature of the topography changes. Evolution on this Wrightian
landscape, inverted or not, is a series of punctuated equilibria, and always
has been.
But there is one more important and profound philosophical reason to invert
the topography than simply to insure proper first impressions in new students:
evolutionary optimization operates to minimize predictive surprise. Every
attribute of an evolving phenotype "predicts" the environment in which it is
most likely to operate. The quality of that prediction is the true measure of
a lineage's fitness (appropriateness to its environment). As that quality is
improved, total behavioral error is driven ever downward towards zero.
On a second point, Tom also writes:
======================================
If another mode of life is nearby and sustaining, then some
of the gene pool of the parent species might stray over into
this new valley and become captive there. The new mode of
life will have a new and re-centered gene pool.
Rather than have evolution climbing mountains, I think
that it's following neighboring attractors.
======================================
That, too, I believe to be quite correct. Please see Fig. 3 in a 1994 paper of
my own that is at the following URL:
aics-research.com
Surprisingly, the height of the intervening barrier between wells is
irrelevant to the process of a population moving between wells. Rather, the
quality of interest is the distance in phenotypic space when ratioed against
current populational variance. Absolute distance allows you to calculate
probabilities, no matter how improbable they may seem. A one-in-a-billion shot
becomes essentially a sure thing if you take ten billion shots.
I believe one of the most philosophically important schematic diagrams in
evolutionary biology is Fig. 1 of Richard Lewontin's 1974 book, "The genetic
basis of evolutionary change." That figure is recapitulated as Fig. 1 in the
HTML paper referenced above. Although I consider Lewontin's diagram to be
essentially equal in importance to Hutchinson's concept of the niche,
Lewontin's diagram has unfortunately been very rarely repeated in evolutionary
biology texts. However, the figure is beginning to be commonly repeated among
the work of the students of evolutionary computation, enough so that it is now
being misunderstood and misrepresented, one of the unfortunate but sure signs
of success.
In Lewontin's diagram, two state spaces exist, a genotypic (informational)
state space, G, and a phenotypic (expressed behavioral) space, P. Most
traditional mathematical optimization techniques work by constraining
themselves to moving solely across the P surface of the adaptive topography
(e.g., "steepest descent" algorithms). Sufficient experience has now been
garnered to suggest that all strongly-determined numerical optimization
methods tend to work acceptably well on some adaptive surfaces, but exhibit a
pronounced tendency to stall in indefinite oscillation, fail to converge, or
become entrapped in local optima on others. These stalls are caused by the
highly-determined correlation between parent and child trial vectors.
But this is not the manner in which evolutionary optimization proceeds. The
mutagenesis guaranteed by replicative error in G generates a continuum of
fine- and large-grained mutations. The change of a single bit or a base-pair
may cause no modification in expression in P at all -- or it may cause
catastrophic change. Or it may be something in between. The correlation
between parent and child trial vectors is often strong, but it is never
absolute. Under random mutation, no combination is impossible, thus global
solutions in a finite state space are guaranteed in infinite time.
Most importantly, the mutagenesis in G is uncorrelated to the current fitness
levels in P, other than in one particular manner: Ultimately competitive
pressures develop such that all well-adapted lineages eventually begin to
evolve intrinsic error-suppression and repair mechanisms to eliminate the most
gross, and thus most fruitless, mutational forms that waste phenotypes and
render the lineage far less competitive than it would otherwise be.
Tom also writes:
=======================================
>simply because probability theory (in the form
>of Boltzmannian thermodynamics) virtually guarantees it to be so.
I don't think that evolution is as simple as thermodynamics.
There are many more degrees of freedom in the system of evolution than
are possible in the systems that Boltzmann studied. One of the
most glaring differences is the step function changes that are seen
in evolution that are not seen in thermodynamics (which I mentioned
above).
=======================================
On this point, Tom and I disagree -- and I don't believe that Tom himself
would defend what he has written. The simple and trivial response would be to
say, by implication, that Newtonian physics does not similarly apply to the
motion of galaxies because Newton worked on much simpler systems. But there is
a much more important response than that.
The shape of the adaptive topography is unknown and unknowable to an evolving
phyletic lineage. It can never be known in its entirety under the physics that
governs a Darwinian universe. The only mode of exploration of the adaptive
landscape is through localized simple trial-and-error, with the selective
retention of the better variants.
If there has proven to be any advantage to the simulated evolutionary
optimization work that is being conducted for engineering purposes, it is that
these simulations allow you to sit outside the universe, god-like, so that you
can at any time stop the process and ask relevant questions. The most common
question is always: "Why are we stuck?" In these simulations, you can very
directly calculate probabilites and ask the question: "What would it take for
us to move into the next well and how improbable would that event be?"
However, as soon as you ask that question, the universe has become
mathematically indistinguishably both Darwinian and Boltzmannian.
And it is exactly at this point that Catherine's original question can be
answered. The next great leap forward, the next exploitation of a "concept"
such as flight, will virtually always go the lineage sitting in the well
closest to that deeper, but as of yet unexplored, trough, given that
sufficient mutagenesis exists in the closest lineage.
Wirt Atmar |
