Download session_proposal_Space_Evo_Exp_Ishpssb2013 general

The space of explanations in evolutionary biology.
Organizer. Philippe Huneman.
Proposal.
Biological explanations have for a long time been thought in terms of two dichotomies:
ultimate vs. proximate causal explanations (Mayr 1961), and developmental vs. selectionist
explanations (Sober 1984). Explanations in evolutionary biology were supposed to display
ultimate causes, and be mostly selectionnist.
However the picture of explanation in philosophy of biology has recently diversified,
both: 1) because of the sophistication of the doctrines of explanation in philosophy of science
(Pearl 2000, Woodward 2005, Machamer, Darden and Craver 2000); and 2) of the increasing
exploration of various fields of evolutionary biology. Even though the main dichotomies have
not been challenged, philosophical issues now rely on a finer-grained understanding of the
explanatory structure of biological theory.
Concerning (1), the mechanistic view of science has been increasingly adopted in
biological subdisciplines because of the accurate picture of explanatory practice it provides – in
molecular biology (Darden 2006) or in the neurosciences (Craver 2007). However it is still
controversial whether it can be applied to the evolutionary sciences: first because natural selection
may not be conceived of as a mechanism (e.g. Millstein and Skipper 2005), or at least requires
specific conditions of abstraction, like statistical mechanics (Strevens 2003), second because
population genetics or ecology seem to use explanations which are not affected by differences in
mechanisms, such as „equilibrium explanations“ (Sober 1983), or „topological explanations“
(Huneman 2010); even more generally, some philosophers vindicated the expalantory role of
mathematics as such (Batterman 2010).
As to (2), the causal structure of evolutionary theory has raised new questions, concerning
the nature of explanation. First, the usual idea that selection, mutation and mirgration are causes
of population change as forces acting on frequencies (Sober 1984) has been vigorously challenged
by some authors (e.g. Walsh 2010, Matthen 2009), because the causal facts at the level of
individual organisms may be sufficient to constitute selection: selection-related properties
therefore would not be causal but „statistical properties“ (Walsh 2007). Hence the task is to
clarify the statistical or causal nature of explanations by natural selection, and prior to that, make
sense of what a statistical explanation should be.
Second, biologists have always been dealing with natural selection under two perspectives:
a change in allelic frequencies – especially in population genetics - , and an optimisation of
phenotypic values – especially in behavioural ecology. Optimality and dynamical explanations
hence coexist, even though the dynamical aspect of allelic selection may not be mathematically
supporting the optimising aspect (e.g. frequency-dependence, heterozygote superiority, etc.) , as
Grafen (2002) emphasised it: therefore this coexistence needs to be investigated.
One must then ask whether explanations of each type, defined according to these
categories, have a role in evolutionary theory; whether equivalences can hold between models
proper to distinct explanatory strategies; and how a specific explanatory project of one kind can
be related to others. Therefore this symposium will investigate evolutionary explanations along
such three axes - mechanisms and structures, statistical and causal, dynamical and optimising - ,
and finally shed light on reasons for advocating explanatory pluralism in the field.
Abstracts.
Statistically autonomous explanations.
André Ariew. University of Missouri.
Ian Hacking (The Taming of Chance, 1990) calls Francis Galton's account of the "reversion to
mediocrity" phenomena found in the distribution of hereditary genius an early instance of a
"statistically autonomous explanation". To Hacking, "autonomous" is opposed to "causal".
Galton's explanation does not cite possible underlying causal features of a population. Rather, it
involves demonstrating that regression is a mathematical consequence of the Normal Law. In my
talk, I will present Galton's early demonstration of reversion through his use of a quincunx, a
shot-dropping machine. The aim is to strengthen and elucidate Hacking's view that statistically
autonomous explanations are an important form of explanation in the history of science. I will
show that the key features of statistically autonomous explanations in Galton's reversion are
found in early 20th century population genetical accounts of evolution by natural selection.
Consequently, population genetical accounts of natural selection are also instances of statistically
autonomous explanation. One key feature is found in the explanandum: the phenomena to be
explained is an instance of a large-scale regularity that would not be explained by reference to
individual level effects. The other key feature is found in the explanans. The aim of a statistically
autonomous explanation is to demonstrate that the large scale regularity would happen if suchand-such were the case even if such-and-such could never actually happen in real world
populations (Gibbard and Varian, "Economic Models", 1978). Finally, I regard statistically
autonomous explanations as an alternative to Michael Strevens' fully causal "kairetic account" of
explanation (Strevens, "The Causal and Unification Approaches to Explanation Unified—
Causally", 2004).
Multilevel Mechanisms of Evolutionary Change.
Lindley Darden. University of Maryland.
Theodosius Dobzhansky in his1937 Genetics and the Origin of Species claimed that "the mechanisms
of evolution as seen by a geneticist" consist of mechanisms at three levels. This multilevel analysis
still captures the key mechanisms of evolutionary change. First, mechanisms produce the
variations that are the raw material for change, including mutation mechanisms of imperfect
copying of DNA (including repair mechanisms), as well as larger scale chromosomal changes and
recombination. The second level includes mechanisms that change populations, genotypically
and phenotypically. The most important is the mechanism of natural selection, which is the only
known mechanism for producing adaptations. In the natural selection mechanism, the crucial
joint activities of variant organisms and a critical environmental factor produce populational
changes in subsequent generations. Finally, isolating mechanisms give rise to new species that are
reproductively isolated from previous conspecifics. This paper argues that natural selection is,
indeed, a mechanism (despite recent claims to the contrary) and places the natural selection
mechanism into the context of the multilevel mechanisms of evolutionary change
Articulating mechanisms and topologies as mutually required in explanatory strategies.
Philippe Huneman. IHPST, Paris.
Evolutionary biology displays topological explanations, namely explanations which,
instead of considering underlying mechanisms, explain by pinpointing topological properties of
abstract spaces (such as food-webs) associated to the system, which entail the explananda
(Huneman 2010). This is often used to explain the stability of an ecological community (e.g.
Montoya, Solé 2002). I will investigate how mechanistic and topological explanations are
connected in evolutionary theory.
Prima facie they are answering different questions: mechanisms are used to investigate the
production of these topological structures (e.g. scale-free food webs) which possess the
properties used as explanantia in topological explanations (e.g., specific mechanisms of predation
may yield food webs of a given topological nature). Here, mechanistic explanations explain what
makes topological explanations possible.
However, I here consider two reverse relations, less self-evident. 1. The mechanisms of
allele frequency changes in populations are modeled by population genetics; yet such modeling
has to make assumptions, about especially the possibility of bracketing development, and about
considering the dynamics of a few loci as capable of representing the whole dynamics in the
genomes population. Such assumptions will be characterized in terms of topological properties of
genotype/phenotype maps, because what is required is the mappings to be conservative of some
kinds of neighborhoods. 2. Modeling selection over fitness landscapes may involve mechanisms
of trait optimization, but only under some conditions of local invariance of the landscape, which
are topological conditions.
Hence a two-ways relation: topological (resp. mechanistic) explanations can provide
conditions of validity for mechanistic (resp. topological) explanations.
Maximum Entropy Explanations in Biology
Aidan Lyon, Department of Philosophy, University of Maryland, College Park
[email protected]
There are many robust and simple patterns in biology that arise out of the aggregation of
a myriad of chaotic, and complex processes. A phenotypic trait such as height is the
outcome of a huge complex array of interactions between genes and the environment.
However, as Galton (1889) and many others have noticed, such traits often exhibit a very
simple pattern: they are often normally distributed. Similarly simple patterns occur all
throughout nature: the log-normal, power law, and exponential distributions––just to name
a few.
Recent work, particularly by Frank (2009), has shown that these distributions can, in
some sense, be explained by their maximum entropy properties. Frank argues that by
maximising the entropy of a probability subject to a few informational constraints, one
obtains a framework that neatly unifies and explains many of the robust and simple
patterns that we observe in biology.
However, it’s not at all clear how entropy and it’s maximisation can explain anything
in nature. Entropy, in this context, is usually understood in terms of information, or the lack
of information, that some ideal epistemic agent has (e.g., Frank 2009, Jaynes 2003). But
how can the information that some agent has explain why, for example, heights are
normally distributed? Intuitively, it can’t. In this paper, I survey a number of interpretations
of the entropy of a probability distribution and examine how they can be embedded in a
theory of explanation so that we can make sense of such maximum entropy explanations
in biology.
References
Frank, S.A. (2009). The common patterns of nature. Journal of Evolutionary Biology, 22, 1563–1585.
Galton, F. (1889). Natural Inheritance. MacMillan, London/New York.
Jaynes, E.T. (2003). Probability Theory: The Logic of Science. Cambridge University Press, New-York
Asymptotic Idealization in Evolutionary Explanation.
Michael Strevens
Philosophy Dept., NYU, NY
[email protected]
Batterman and others have argued that many explanations have an asymptotic form: they
explain a state of affairs or behavior by showing that it emerges “in the limit”. Explanatory
models of this sort have two properties that, on the DN account, explanations are supposed to
lack: (a) the “premises” of the model do not entail the explanandum, and (b) the model is
idealized in certain ways having to do with the infinitude introduced by the use of limits, and so
false of the real world. My talk will present a simple example of asymptotic explanation in
evolutionary biology; it will show how to understand the limiting aspect of the relevant model,
and in particular to relate it to the conventional wisdom among biologists as to what the model is
doing; and it will attempt to fit the model into a modern theory of scientific explanation.
Varieties of Invariance
Denis M. Walsh
University of Toronto
[email protected]
An explanation serves two functions: metaphysical and cognitive. It must identify a feature of the
world, the explanans, that relates in the appropriate way to the explanandum, and it must
describe the relation in such a way as to provide understanding. One of the principal virtues of
Modern Mechanism, I argue, is that it provides a compelling, generalizable model for the
structure, or anatomy, of an explanation. Explanations are bipartite. They cite a mechanism—an
entity undergoing an activity—and an elucidating description. The hallmark of the relation
between a mechanism and the effect it explains is invariance. Invariance is a robust
counterfactual relation, but it is not the only one. I claim that statistical and teleological
explanations conform to the same bipartite structure, of robust invariance relation and
elucidating description. Some phenomena are susceptible of more than one complete
explanation. Explanations of the same phenomena in different modes neither supersede nor
exclude one another. I argue that they are ‘miscible’.