Re: Constraints (was Re: Adaptationism's Lessons)

Paul Gallagher (
18 Sep 1996 20:07:44 -0400

In <51p5r6$> (Bryant) writes:

>In article <51n998$>, Paul Gallagher <> wrote:

>And as I've pointed out before in these parts, that's not the kind of
>prediction I'm talking about. The adaptationist program does not make
>predictions about future selection. It makes predictions about the
>"design" (the historical or evolutionary function) of complex traits.

The Lenski paper shows how important random drift is in moving about the
adaptive landscape in a real population of E. coli. Genetically identical
populations in identical environments end up with many different morphologies.
How has selection predicted the design in this case? (And why don't you
consider ideas about random drift "prediction testing science"?)

At the opposite extreme, Loren Rieseberg recreated the species, Helianthus
anomalus, by breeding the parent species from which it arose as a hybrid.
The parents are the common sunflower and the petioled sunflower. The
hybrids were bred with each other and with each of their parents for four
generations. Then 197 segments of the genome of each lineage were compared.
They were almost identical. Either intense deterministic selection is
occurring (which would help adaptationism) or something else is going on.
Perhaps the mechanical properties of chromosomes allow only certain gene
combinations to occur. If this is the case, selection is besides the point,
since the variation on which it could act is highly constrained. Amundson writes
that selection is a reasonably complete explanation of a trait only when the
variation on which it acts is spontaneous, heritable, abundant, small and
continuous in its effects, and not directed along a particular path. If
the idea about chromosomal mechanical constraints is correct, then variation is
limited along a single path in the case of these sunflowers.

>Since I've quoted the Spandrels paper to negate an earlier mistake of
>yours, perhaps you can take my word for my having repeatedly read through

I don't think I was mistaken. I think there's a difference between accusing
someone of thinking "all traits are currently adaptive, and optimally so,"
and accusing them of thinking "most traits are adaptive, and close to
optimal." Anyway, I think Gould and Lewontin are making more points
beyond a critique of optimization theory. They have written articles
on the subject, but they dealt with other topics here. In particular,
they advocated Seilacher's construction morphology. This is the idea that
morphological analysis requires morer than functional analysis. Just
elucidating the function of a trait does not explain it. You need to
address phylogenetic, structural, ecophenotypic, and chance factors as
well to get a complete and useful explanation. These latter four factors
aren't just something holding back adaptation. They're equal players in
determining the form of an organism.

In particular, I don't understand why you (and Dennett too) think showing
a function for the divaricate patterns on shells in some organisms proves
Gould and Lewontin wrong. They themselves discuss the function of these patterns
in some organisms. But Seilacher's idea is that that these are secondary
adaptations. They are functional now, but their origin may not be adaptive.
Instead, they may be a consequence of the geometrical or mechanical nature
of shell growth, or a chemical property of the shell materials. These sorts
of ideas can be and have been tested using mathematical or physical or
chemical models.

>>Besides, shouldn't you make predictions that will falsify a hypothesis,
>>not just make predictions? If "jealousy" is in the genes, we might
>>predict that jealousy should increase fitness - but if it turned out
>>jealousy didn't increase fitness, that wouldn't prove the hypothesis that
>>jealousy is in the genes, false.

>You're assuming current utility. Bad boy.

I assumed the opposite. The idea is to think up a prediction that if
proved wrong would falsify the hypothesis that "jealousy is genetic."
That's what you want to do: test the hypothesis by thinking up
predictions that would falsify it. If jealousy does not increase fitness,
that would not prove the hypothesis that jealousy is genetic wrong,
precisely because you could rescue the hypothesis by saying that it once
increased fitness, but does not do so currently. So, the fitness test is
not a falsifiable prediction.

If I had said, jealousy isn't genetic because it right now doesn't
increase fitness, then you would be right to say I was assuming current
utility. But I didn't say that. I said X is logically independent of Y.
You someone took that to mean if not X, then not Y.

Talk to Wilkins about this. He's on your side and he knows all about
Popper and the idea of falsification.

>I'll trust Mayr and Maynard Smith to present a reasonable historical
>perspective. Wanna hear what they think about your venerated
>ivory tower revolutionary?

I've heard from Maynard Smith. Including his observation that Gould's
ideas arise from Gould's being Jewish! (from an interview in Skeptic
magazine on-line) How has Mayr bad mouthed Gould?

But I'll let you have the last word. We're too far apart in our positions
to have a useful discussion.


I'll include a brief summary of some of the ideas in Spandrels that I wrote
in an old article I posted.

To review: traits may arise as a correlated consequence of selection
directed elsewhere, or they may be the currently non-adaptive legacy of past
adaptations, or they may be adaptations that reuse past adaptations for
different functions than those for which selection originally favored them,
or they may be adaptations due to selection that are nonetheless no better
or even worse contributors to fitness than other possible adaptations, when
variation exists to allow evolution into multiple coadapted adaptive peaks.
The critique of the adaptationist program is *not* an assertion that the
differences among species are not the result of natural selection.

But if you allow that drift may have played a role on evolution, beneficial
alleles have been lost and deleterious alleles have become fixed. The
probability that a mutation will become fixed is only 2 * its selection
coefficient, while it is probable a deleterious allele will become fixed
if its selection coefficient is less than the reciprocal of the population
size. Moreover, the more genes that contribute to a trait, the smaller
the selection intensity per locus.