Traits, Genes, and Coding 379vehicles of causal information content, since all manner of other physical media,
such as fluctuating currents in wires and sound waves, are standardly thought to be
good for the job. A more serious barrier to the use of causal information concepts
in genetics is that, given the standard conception of the genome as specifying
phenotypic outcomes in a disjunctive manner (i.e., develop likethisunderthese
environmental circumstances, likethisundertheseenvironmental circumstances,
and so on), the causal information view licences us to speak about genetic coding
in ways that biologists don’t. For example, to use an example due to Griffiths
[2001], on the basis of a purely causal notion of information, the human genome
would encode the instruction “when exposed to the drug thalidomide grow only
rudimentary limbs.” But biologists are unlikely to be tempted by such a claim.
What this tells us is that the notion of causal information fails to capture the
standard usage of informational terms in biology.
The most substantial problem confronting claims that appropriate causal co-
variation is sufficient for genetic representation (or for genetic information), how-
ever, is one ofexcessive liberality. It is indeed a familiar point from the literature
that genes are not the only factors in the developmental system that might be
identified as causally co-varying with traits. Of course, it seems clear enough
that if one could hold non-genetic causal factors in the developing body and the
environment constant, while varying the genotype, then one would find causal co-
variations between genes and phenotypic traits. However, if one could hold the
genotype and the non-genetic causal factors in the developing body constant, while
varying environmental factors, then one would find causal co-variations between
environmental variables and phenotypic traits. Similarly, if one could hold the
genotype and the environment constant, while varying non-genetic causal factors
in the developing body, then one would find causal co-variations between those fac-
tors and phenotypic traits. But now if causal co-variation is a sufficient condition
for a developmental factor to be representational, and if non-genetic causal factors
in the developing body and the environment can causally co-vary with phenotypic
traits, then those extra-genetic elements will sometimes count as coding for traits.
This spells trouble because, given that many of the non-genetic factors here will
be illegitimate ones, it falls foul of the weakened uniqueness constraint. In short,
as a sufficient condition for coding, causal co-variation is excessively liberal, in
that it licences explanations in whichtoo muchof some extended developmental
system might emerge as coding for traits. So while it is eminently plausible that
appropriate causal co-variation is necessary for genetic representation, it cannot
be sufficient; genetic representation must be appropriate causal co-variationplus
something else.^5
What might that something else be? Here is a suggestion: genes code for traits
because they (additionally)set certain parametersfor the developmental systems
that generate phenotypes. Perhaps then we can say that while genes do not fully
(^5) The fact that any systematic causal co-variation account of genetic coding will be excessively
liberal (in the sense identified in the main text) is widely appreciated; see, e.g., [Griffiths and
Gray, 1994]; [Maynard Smith, 2000a]; [Griffiths, 2001]; [Sarkar, 2005].