On Mar 30, 2026, at 6:53, Santafe <[email protected]> wrote:
I was on the point of trying to answer Nick a few days ago, while also
acknowledging that since I haven’t had time to read his paper, I should keep it
short.
It is helpful that EricC has posted in the meantime, because I often believe I
can understand what he means by something, whereas with Nick I often read
things that seem kinda familiar, but then I find some sentence that Nick writes
following a “therefore”, and I am baffled by what logic he thinks that follows
from, making me realize I probably didn’t understand what he “actually meant”
by the earlier sentences either.
But I will try a short in-line here, with EricC’s version, which is what I was
guessing Nick might have meant a few days ago when he said:
This literature you cite often fails to make a distinction between evolution as a theory and the theory that explains evolution. Evolution as a theory holds that organisms have descended through the ages they have become modified to the circumstances of their existence. It is a thesis about the causes of descent with modification. Natural selection is a theory offered in explanation of that pattern of descent. The literature you describe has often fused these two metaphors into one and therefore is fundamentally unfalsifiable as a theory that explains evolution.
Here I was completely mystified what argument Nick understood himself to be
making. Particularly referring to “the literature I cite”, since I sent him
George Price’s The Nature of Selection, which (in the actual print of the
article) says nothing of the kind.
Put aside that I couldn’t track what the “two” were, or in what sense either
was a “metaphor”, other than that Nick labels them (and all other words; man
with a hammer) as such.
Anyway, moving on as promised:
On Mar 30, 2026, at 0:41, Eric Charles <[email protected]> wrote:
I think this (EricS's reply) speaks to the "formalism" issue, and the value of the
"flavor text".
One of the big problems in evolutionary theory over the last century came from the desire for a very particular type of formalism, in which terms are all inter-defined, and as such, statements made within the system can't be "wrong" in an empirical sense. The classic example (so Nick and I assert in the incipient book) is the equation {\displaystyle s=1-W} where s stands for "selection" and W is "fitness". This has been in biology textbooks of a certain sort for well over half a century, and leads to Dawkins's "Selfish Gene" talk, among other things. It's hard to criticize on the surface, but, unfortunately, once you do that, you cannot usefully study the relationship between fitness and selection, because that-they-are-related becomes a truism!
I _would_ have asked Nick: "What do you mean: is it that the sentence
`adaptedness is an outcome of natural selection' is a tautology, amounting to
nothing more than `fit are the fit'?” I was guessing that might have been Nick’s
meaning, and it sounds like EricC’s meaning here.
I will take your word for the presence of bad writing and nonsense in
evolutionary textbooks a century ago (probably today, too; the ability of many
scientists to render coherent logic in ordinary English is often not something
to be proud of). And I think I can remember somebody with a biology degree (so
journals are required to publish him), who writes something like this (was it
Pigliucci? If not, I apologize to him for bringing in his name.)
But if that is the complaint, it seems to me that that ship had sailed already
with Fisher, though he was so cryptic and imperious in his writing and his
personal style, that it had to wait 40 years for George Price to explain to the
rest of the community what Fisher was saying (and per Glen, much more
importantly, _doing_). Steve Frank has written fairly well on this; or at
least, that is where I got it. (I don’t know which of a half-dozen papers
might be the cleanest source to recommend here.) In any case, if people didn’t
know it had sailed in 1930 or 1958, they certainly should have known it had
sailed by 1972 or 1990 (when Frank found Price’s 1972 manuscript and published
it). de facto, most of the working literature that has content already makes
use of its having sailed, even if they don’t remark on it, because they use
Price Equations as accounting identifies (among other organizing systems).
If one is going to assert that there is nothing but tautology in what
population geneticists compute, and therefore they don’t ever predict anything
— a tautology can’t — then I would say that requires ignoring all the main
content of what one is reading in a very large number of papers. There is an
entire causal/predictive structure, and it can very much be wrong in its
imputations. Often, the sources of being wrong can come from things biologists
don’t front as important elements of cause, where I argue they should, so the
things that are wrong can also be interesting.
What then am I claiming is the non-tautological content, which is there over
and over and over in tons of papers, which Fisher gestured at unhelpfully in
English, but partly did a very important thing with in math, which Steve Frank
acceptably describes, and which Price made didactic?
1. First, obviously, we get rid of horoscope-style and Humpty-Dumpty-style
language. We know that “fitness” can’t be doing much work in a causal account
if all it means is “I look at them, and they seem to be happy”, or if it is
meant to be a synonym for the even worse attributive: “design”. (Remember the
important part of the Vampire lore: they have to be invited in!) We have to
actually say _what we mean a predictive theory to be predicting_. So the
attempt to formalize fitness as “reproductive success, absolute or relative,
according to parent type”, is the declaration of the observable to be the
anchor for empiricism. If you like, the “instrument” that is to be employed.
1a. Now, declaring an observable can’t in itself be “wrong” in Popper’s sense,
because it isn’t a logical conclusion. What it can be, however, is incoherent,
and thus a gesture at a definition, which can’t ever be made into an actual
definition. This is the important thing that Fisher did correctly and
explained horribly: making the point that “fitness” must be formalized as a
summary statistic (that’s a technical term, not a metaphor!) of your
high-dimensional observable. The question of when you are _even saying
anything_ turns on when your summary statistic is coherently definable.
Wherever it is coherently defined, it _cannot_ be “falsifiable”, because now we
clearly have a definition that is a completely different kind of thing than a
putative derived conclusion.
1b. Next, since heredity plays out in diverse ways in different groups,
summary statistics defined for one won’t generally be defined for others.
Fisher, being a King, in many of the worst senses, and being out to capture
territory (negatively spun) and solve prediction problems (positively spun),
defined a variety of summary statistics from histories that are sequences of
population states connected through reproduction. They aren’t all the same,
and they are not all capable of carrying the same “meanings”. (I’ll get to
what that word means in a moment.)
1c. For objects that are suitably described as “replicators” (technical term,
not metaphor!), there is a fitness summary statistic that can also be assigned
as an attribute (just, a marking) of individual organisms. For organisms that
reproduce sexually, or via a variety of complex lifecycles, the summary
statistics that Fisher defined to carry the label “fitness” _cannot_ be
assigned as attributes to any organism, because they are computed from
whole-population counts. So quite apart from what you think you gain by
“marking” an individual organism with an attribute (the bacterial cells
carrying this allele have left 14.32 live daughters per starting cell in 6
generations in this laboratory instantiation of a population dynamic), for
these sexual-lifecycle summary statistics, you can’t even do that much.
1d, Just btw (a thing nobody has any reason to care about), this is where it
is productive to go after Fisher, and the generations that don’t extend beyond
him in their methods. His summary statistics are well-defined, so one can’t
deny that. But they are not _all_ the summary statistics one might try, and
there are others (also well-defined) that can be constructed, which do a better
job of marking organisms and life-stages than Fisher’s fitnesses do, and
support all the same Price-like decompositions. That’s the kind of thing I do.
2. So at the end of all the “1” points above, we have an actual observable that
we have committed to, and that we intend some “theory” to explain or predict in
some way. The whole content of a predictive theory is the assertion that
organisms have attributes in the here-and-now (variously termed “traits”),
which support prediction of the distributions of our fitness summary statistics
as time unfolds. Again, those attributes aren’t vague gestures like “seems to
be doing well”; they should be things one can commit to somehow: it is dark or
light in color; it has softer or harder shells or teeth or claws; etc.
2a. Geneticists try to capture the idea that traits in the here-and-now are
predictive of fitnesses as realized histories unfold, by defining sample
estimators for those dependencies in the form of regressions. Again, if a
regression is defined, there can’t be anything “true” or “false” about it; it
is just a definition.
3. The Evolutionists, for some time, have built up terminology around all this.
Samir Okasha’s book uses it (among other things) as a skeleton. I don’t enjoy
the way they say it, but if one is trying to follow their arguments, the terms
are enough to do that. The summary-statistic definition of the observable is
what they call “realized fitness”, and the regression model on traits in the
here-and-now is what they call the “propensity interpretation”. Yuck; but
don’t shoot the messenger. I probably wouldn’t have coined those terms.
3a. It seems pretty inescapable to me that realized fitness and a regression
model on traits cannot be tautologically related, as one requires diachronic
observation even to assign, while the other is assignable (for any fixed choice
of regression coefficients) from a population state at a single time. (It
treats the population composition as a Markov state.)
3b. Note that EricC’s s = 1-W above never entered any of what I said. If a
selection coefficient is defined in some way as an algebraic function of
offspring numbers, that is just a layer of notation. The existence of both
realized values W (convertible to realized values s), and model-predicted
expectations for W, derived from model-incorporated coefficients s, means that
the algebra isn’t meant to do other than convenience work _within_ either
layer, and the causal argument concerns the fitting of models to sample data.
If textbook writers render this reality badly, then fine: go after them. But
the flu designers who estimate what should go into an annual vaccine don’t
deserve to be tarred with that brush.
4. Finally we can get to the point of what they claim to be doing, and in
competent papers, are doing: Are their regression models predictive, and are
they robust against things that the modeler didn’t notice, but that could have
been included? Hashing out that question is the whole content of the
scientific literature and program.
5. Where is conflation an actual hazard in the world I live in, what-exactly is being conflated, and how might it get something wrong? (So, we have established falsifiable and in that sense meaningful; now I am talking about false and thus interesting.) Well, people might identify a trait that they think confers a tendency to leave more offspring. They might even be right in their identification. All that is on the “propensity” side. The place they can err is in supposing that, if they have identified a cause for a tendency, then they can be assured of an outcome in the same direction (a wrongful binding of the “propensity” term to the “realized” term), and the only uncertainty will be the scale of the outcome effect. This is an error I think evolutionary biologists often do commit in their informal language, even though the math in any paper wouldn’t let them get away with it, so they don’t do it there. Others, not even evolutionary biologists, but a kind of groupie
of the biologists from other domains in science, commit this error of supposition and express it frequently in bar-quality conversation (I have a certain conversation in a NASA working group in mind as I write this.) The slip between the cup of propensity, and the lip of realized outcomes, which biologists don’t account for as fully as I think they should, is illustrated nicely by magnetism: The “tendency” for spins to align in solid-phase Iron is identical at all temperatures. However, when the iron is too hot, the spins don’t macroscopically align “just a little”; rather, they don’t align macroscopically at all. Then, upon cooling just a little below the critical temperature, they do macroscopically align, and the alignment rapidly becomes “a lot” with separation of temperature below the critical value. All this is the lesson about counting that we learned from thermodynamics. And while the things being counted in evolution (completed lifecycles) are different from
the things counted in equilibrium thermo (pairs of aligned spins), the counting argument applies equally well.
Anyway…. Sorry for so many words. I hope that I have given enough here to
explain why I am baffled when I hear modern people claim that “adaptation
occurs through natural selection” is “unfalsifiable”.
The thing about where Dawkins goes wrong is another story, best told by not
quoting his own words, because he is such a rhetorician (even though his
earlier work suggests that he is capable of arguing properly). But it branches
out of the points above, just on the point of mis-representing summary
statistics that cannot be object-attributes, as if they were object attributes,
and then, having set up that strawman, arguing that the only recourse is to
throw the whole thing out above the level of the gene, and ignore that all the
same problems continue to occur for genes, becoming quite serious in the deeper
past. All that can totally be sorted out; there are no fatal confusions, and
metaphor isn’t tripping up anybody who wants to speak and argue carefully.
Eric
