Positivism in evolutionary science
tdib at UMICH.EDU
Thu Dec 4 11:52:50 CST 1997
James Francis Lyons-Weiler wrote:
[I rearranged things a bit to impose coherence]
>On Thu, 4 Dec 1997, Tom DiBenedetto wrote:
>>... discrepancies. Popper goes on at length about the desirability to
>> depart from ad hocness; it is directly related to the
>> "outlandishness" of the hypothesis (actually it might be better to
>> simply refer to the information content of the hypothesis).
> No; as I read Popper, an ad-hoc hypothesis is one that is
> tacked on to save a failed theory, and that is, in principle,
"Ad hoc" refers to the generality of a hypothesis. An ad hoc
hypothesis is one which is merely descriptive; it doesnt attempt to
generalize at all ("THIS swan is white"). It has minimal information
content, and yes, there are no relevant tests. Science is about
moving beyond ad hocness, and actually saying something. A bold
statement ("ALL swans are white") has great (potential) information
content (it purports to speak about a trait in all swans), it is
easily testable, and is maximally distant from ad hocness. When a
hypothesis is falsified, it can be abandonded, or it can be salvaged
by appending certain ad hoc hypotheses to account for the
(ALL swans are white except this one which is a mutant). That is what
we do in cladistics. Homoplasy represents the ad hoc hypotheses we
are forced to accept in order to salvage our phylogenetic hypotheses.
Given the nature of evolution, there are no phylogenetic hypotheses
which can be unanimously corroborated by the evidence from character
distributions, so we prefer the hypothesis which requires minimal ad
> One transformation says more than three? Information content
> a la Popper is the new information that a hypothesis provides
> about nature IF is has passed a test, beyond that which
> we already expect given the background knowledge.
Thats right. One transformation represents a claim that all these
taxa (8 in my example) have a more recent common ancestor than any
other taxa; and that implies many predictions regarding the
distribution of further characters. Many transformations implies a
set of relationships with lesser generality, or no genrality at all.
> Popper's formalized relationships among h (hypothesis), e(evidence, or
> test statement), b (background knowlegde), I (information
> content), C (corroboration) , and empirical content
> (testability) form a consistent calculus that has
> repeatedly been corrupted in its application to
No it hasnt James. I think you simply dont understand the
> IMHO, your
> understanding of falsficationism has been tainted by
> misunderstanding and misapplications by others.
Well, that makes for a nice symmetry in our view of eachother! :)
>> To relate this back to phylogenetics, the most parsimonious tree is the
>> hypothesis which has the greatest information content; it can account
>> for the data more efficiently than any other hypothesis.
> The limitation here is, as we've discussed at length before,
> the utility of parsimony as the test. Other criteria can be
> used as more sensitive, more exact, and more demanding
> (more critical) tests than provided by parsimony.
And as I have tried to explain to you several times before, the
"test" that is implemented in a parsimony analysis is not parsimony
per se, it is a test of *congruence*. We gather together a set of
(hopefully) independant homology hypotheses and test them for
congruence. The parsimonious solution is the solution which reports
the results of that test; it reveals the congruence that exists in
the data, and those discrepancies for which we are required to
postulate ad hoc hypotheses. Less parsimonious trees are simply
inefficient applications of the test; they claim that we need make
recourse to more ad hoc hypotheses in order to account for the data
than is really the case.
> simply because one has postulated
> the simplest hypothesis(and knows it) doesn't mean that the
> hypothesis is afforded any greater corroboration than, say,
> the most complex hypothesis. That's because corroboration
> is a function of boldness.
But James, cant you see that the more complex hypothesis (for the
same phenomenon) is inherintly less bold? It is more complex because
it generalizes less; it tries to account for all the observed
variation by dealing with all those instances with particular
refinements of the theory. A simple hypothesis (e.g. e=mc2) attempts
to make a broad general statement with sweeping application. That is
what being bold is all about.
> If, for a set of organisms,
> it could be shown experimentally that the longest tree was
> in fact the true tree, we would all be very much surprised;
> as a result, the longest tree (in this case) would be
> afforded a greater amount of corroboration than any other.
> We would not be surprised if the shortest tree was true,
> and thus parsimony is not an sufficient test.
Why? I think you are taking this off in the wrong direction. Popper
relates corroboration to boldness to the degree of our surprise, but
he does not mean to erect a psychological standard of corroboration.
The line of argument goes, rather, from corroboration to boldness to
information content. Boldness is a function of how much you are
trying to say; i.e. the information content of your hypothesis.
Read "Realism and the aim of Science" (I seem to recall that the last
chapter has a good summary of this and a discussion of the calculus
>>Simpler hypotheses are better
>> corroborated because they inherintly say more; they have higher
>> information content; they are inherently more bold.
> No, no, no. That's the caricature of Popper than has
> grown up inside phylogenetics, and it is ALL backwards.
I beg to differ,,,,,,,
Tom DiBenedetto http://www-personal.umich.edu/~tdib/
Fish Division tdib at umich.edu
University of Michigan Museum of Zoology
More information about the Taxacom