Here’s a summary of Peter Godfrey-Smith’s argument that genes do not encode information for phenotypic traits.
Peter Godfrey-Smith argues that genes do not encode information about phenotypic traits; but they do encode information about proteins. First we need to be clear on what is meant by information and phenotypic trait.
In a broad sense, a phenotypic trait might include anything other than genes. So the more causally immediate products of genes like proteins would be included. But the sense used here will be more causally distal things like an organisms size, shape, color, structure, behavior patterns etc.
There a many different senses of information. In a Shannon sense, any two processes that are correlated can be described as having information. For example, dark clouds have information about rain, and tree rings have information about age. In a more controversial semantic sense, information is about messages, representations, meaning, or codes. Tree rings are indicators of age, but they do not represent age. By contrast, words on this page are a collection of symbols that convey meaning.
In the Shannon sense, there is information when a reduction of uncertainty in one state is correlated with a reduction of uncertainty in another state. For example, the state of the sky provides information about whether it will rain or not. This is sometimes called the “causal” sense of information, but, strictly speaking, correlation is all that is required. When information is used in this sense, some conditions are fixed as background conditions. For example, if we consider genes as carrying information about phenotypic traits, the environment will be the background condition. But we can also consider the environment as carrying information if we treat the genes as the background condition. In addition, the information relation is symmetric: phenotypic traits carry information for genes as well as genes carrying information about phenotypic traits.
So it turns out that saying genes carry information in the Shannon sense is fairly trivial, since Shannon information is everywhere. This is not to say that Shannon information isn’t useful in biology; it’s that it’s not something particularly special to biology.
Sometimes biologists use the term information in the Shannon sense, but other times they don’t. There is a sense in which there is a genetic code but not an environmental code. The question Godfrey-Smith wants to answer is if there is a sense of information in which we can say that genes code for phenotypic traits.
My approach will be to see whether there are properties of genes that have an important analogy with familiar and central cases of languages and symbol systems. The test is not so much whether these analogies look striking, but whether the analogies are important within biological theory. To what extent does describing genes in informational terms help us to understand how genetic mechanisms work? … My argument is not directed against the idea that genes have an important causal role in producing the phenotypes of organisms; no one can deny that. My argument concerns how they exercise that causal role. I claim that this causal role does not involve the interpretation of an encoded message in which genes specify phenotypic traits.
We can think of the path from genes to phenotypes as taking two steps: the first from gene to protein, the second from proteins to phenotypes. While there is a fairly unified process from genes to proteins, the process from proteins to phenotypes is much more varied. For example, some proteins act as a part of the structure of organism; some act as enzymes; some act as hormones; etc. Sometimes a protein can be used to identify a trait (e.g., the sickle-cell hemoglobin for the sickle-cell trait); but most of the time this won’t be the case. Consequently, Godfrey-Smith concludes that genes do not encode information for phenotypic traits.
The next question is: Do genes encode information for proteins?
Here is the approach that I will take to the problem. I ask: Which attributions of coding and/or informational properties to genes have a useful theoretical role within biology? I will argue in this section that there is a real, but restricted, domain in which the attribution of coding properties to genes does help us to solve problems and understand how organisms work. This domain is the explanation of protein synthesis itself.
There are three reasons why we can describe genes as coding for proteins.
- The building blocks of a nucleic acid sequence (DNA and RNA bases) are different from the building blocks of proteins (amino acids) and there is a (largely) fixed rule linking the two.
- The specification for proteins is combinatorially structured on two levels. First, triplets of nucleic acid bases are specific for particular amino acids. Second, during translation a given triplet (largely) specifies an amino acid independent of its neighbor triplets.
- The rule linking base triplets with amino acids is largely arbitrary. Meaning, “nothing about the chemistry of a particular amino acid is responsible for it corresponding to a particular base triplet. Contingent features of the tRNA molecules, and the enzymes that attach the amino acids to tRNAs, determine which triplets go with which amino acids.”
Godfrey-Smith concludes that these considerations justify describing genes as coding for the primary structure (the linear sequence) of proteins, but nothing above that including the tertiary structure of proteins (the shape).
Hitchcock, Christopher. Contemporary Debates in Philosophy of Science.