1. Introduction

Genetics and developmental biology treat DNA as containing information about aspects of the organism and its development. It is contentious exactly which aspects these are, but anything from protein structure (e.g., Crick Reference Crick1958; Stent Reference Stent1981) to complex phenotypic traits (e.g., Jacob Reference Jacob[1970] 1974) has been suggested at some point. Whatever its precise content, information is thought to be stored in DNA and to be capable of being expressed in a right or wrong way. This idea is challenging: Aboutness, misinterpretation, and storage are semantic properties, but such properties are not posited by ordinary biochemistry. How, then, can semantic properties be integrated with biochemistry and molecular biology? If semantic properties are not normally attributed to molecules and chemical processes, why should DNA and developmental mechanisms be exceptions? Is genetic information an empty or misleading metaphor as some have alleged (e.g., Sarkar Reference Sarkar and Sarkar1996; Kay Reference Kay2000; Keller Reference Keller, Beurton, Falk and Rheinberger2000)?

To a large extent, the recent philosophical discussion has been framed in terms of the claim that genes contain information about phenotypic traits (Sterelny et al. Reference Sterelny, Smith and Dickison1996; Maynard Smith Reference Maynard Smith2000; Griffiths Reference Griffiths2001; Godfrey-Smith Reference Godfrey-Smith2003; Moss Reference Moss2003; Sarkar Reference Sarkar and Hitchcock2003; Wheeler Reference Wheeler, Rojszczak, Cachro and Kurczewski2003). This is probably the most far-reaching and controversial idea from a biological point of view. Surely, many biologists endorse it (e.g., Jacob Reference Jacob[1970] 1974; Alberts et al. Reference Alberts2002; Davidson et al. Reference Davidson2002; Wolpert et al. Reference Wolpert2002), but many others deny it (e.g., Stent Reference Stent1981; Kauffman Reference Kauffman1993; Noble Reference Noble2002). One of the most prominent critics of this view has been Sydney Brenner (see de Chadarevian Reference de Chadarevian1998). He and other skeptics emphasize that phenotypic traits are the most distant of genetic effects and those requiring the largest and most diverse set of additional causal factors to bring them about.

Here I put this claim about phenotypic traits to one side. For the more fundamental question seems to be whether DNA contains semantic information about any traits or effects. If it does, then it is likely to be information about, at least, the effects marked out by the least ambitious and least controversial claims regarding genetic information. The topic of this paper is, therefore, the legitimacy of these modest claims. These claims center around the idea that DNA carries information about some of its proximate effects, and they feature processes like DNA synthesis, RNA synthesis, and protein synthesis. Assertions like DNA contains information about proteins seem acceptable even to philosophers rejecting semantic information about phenotypic traits (Godfrey-Smith Reference Godfrey-Smith2000b; Griffiths Reference Griffiths2001; Moss Reference Moss2003). However, with the exception of Godfrey-Smith's work (Godfrey-Smith Reference Godfrey-Smith2000b, Reference Godfrey-Smith2003), the adequacy of such limited claims has rarely been evaluated.

The debate about genetic information is closely linked to debates over genetic determinism and the concept of a gene.Footnote ¹ The argument in this paper, though, is restricted to the question of whether molecules like DNA, which may or may not be (parts of) genes, contain semantic information. I will argue that they do: As templates for the synthesis of macromolecules, nucleic acids determine their products in a way that is constitutive for instructions in general. It is therefore legitimate to attribute instructional content to molecular templates. Thus, my proposal falls broadly into Godfrey-Smith's (Reference Godfrey-Smith and Hardcastle1999) category of ‘developmental role theories’ of genetic information.

I start by outlining the most limited claims regarding genetic information. I analyze the way in which molecular templates determine the distinguishing features of their products. I then argue that everyday examples of instructions share this way of determining their outcomes, and further, that the controversial semantic features of DNA can be understood in terms of the properties of instructions. The last sections consider possible objections and two thought experiments from the recent literature. But first, a few words about what seem to be three major desiderata for any theory of genetic information.

2. Desiderata

It is common ground among philosophers of biology that satisfactory accounts of genetic information must capture its semantic nature (for an exception, see Sarkar Reference Sarkar and Hitchcock2003) and that accounts based on the mathematical theory of information fail to do so.Footnote ² Therefore, explaining the following three semantic features is the first desideratum. First, the informational entities must be ‘intentional’ in a broad sense: they must be about some other entities. Second, the idea of misrepresentation must make sense (Sterelny and Griffiths Reference Sterelny and Griffiths1999); for example, molecular outcomes must be capable of being erroneous in a stronger sense than just being an uncommon effect (Godfrey-Smith Reference Godfrey-Smith and Hardcastle1999). (A strong sense of error does not imply aboutness, for something can malfunction without being about anything.) Third, there must be an account of what it means for genetic information to be stored in DNA. One aspect of this idea is the persistence of a semantic content over time; another is the view that genes contain information that may not currently be used or expressed (Sterelny and Griffiths Reference Sterelny and Griffiths1999). By contrast, it would not make sense to distinguish between cases in which smoke stores, rather than expresses, its (natural) information about fire.

The second desideratum for justifying the attribution of genetic information is a successful naturalization of its semantic features. It is difficult to see how semantic information could be legitimately attributed to any biological entity if it were not possible to account for it in nonsemantic terms. A glance at some recent theories of genetic information confirms that this desideratum is widely shared. According to Godfrey-Smith (Reference Godfrey-Smith2000b), DNA contains genetic information in virtue of its peculiar causal role in development. On teleosemantic accounts (Sterelny et al. Reference Sterelny, Smith and Dickison1996; Maynard Smith Reference Maynard Smith2000), a stretch of DNA is about a certain developmental effect in virtue of this effect being the (etiological) biological function of this piece of DNA: It has genetic information about an effect just in case it has been selected to produce this effect in the past. And Jablonka (Reference Jablonka2002) argues that an allele is a source carrying information (about a state for a receiver), because the receiver's reaction depends on the source's structural properties, and because it tends to adapt the receiver to that state.Footnote ³

The third desideratum proves less straightforward. Biologists typically think that genetic information is contained only in a rather small set of biological macromolecules, i.e., in nucleic acids and (perhaps) proteins. It seems plausible, therefore, that a satisfactory philosophical account should restrict the attribution of genetic information to these molecules. At first sight, this adequacy condition appears to underwrite a major complaint against teleosemantic accounts: If all that is necessary for containing information is to have a biological function, any trait selected for a particular function would contain information about its function (Godfrey-Smith Reference Godfrey-Smith and Hardcastle1999; Griffiths Reference Griffiths2001).

This criticism is justified, but I think the desideratum should be relaxed. A theory of genetic information is not intended as a mere explication of the biologists’ actual use of this term, but rather as a substantial enquiry into the presence of semantic features. It may turn out that other, nongenetic, developmental factors share the very properties in virtue of which DNA is attributed semantic information. Indeed, Griffiths’ (Reference Griffiths2001) parity thesis states that this will be the case on all satisfactory accounts. From his perspective, it seems that accounts allowing for such a symmetric distribution of semantic information are desirable, and Jablonka's (Reference Jablonka2002) proposal appears to follow this approach. Regardless of whether one accepts these accounts, one ought to acknowledge that the limitation of semantic information to a few kinds of biomolecules may be the result of a theory of genetic information, but should not be its precondition. On the other hand, one need not conclude that the theory should apply to every biological trait. Rather, it seems sensible to refine the third desideratum: The biologists’ paradigm cases of informational processes should be preserved and the theory's applicability may be modified and extended beyond these cases within reasonable limits. Attributing semantic information to legs because they were selected for walking is beyond reasonable limits. This refined standard retains, I believe, the force of the criticism against teleosemantic accounts.

3. Modest Claims about Genetic Information

Limited claims about genetic information center around the idea that DNA contains information only about some of its most proximate effects, but not about phenotypic traits. An early and influential version of this idea postulates that nucleic acids contain the information for determining the linear order of both amino acids in proteins and bases in nucleic acids (Crick Reference Crick1958). Proteins and nucleic acids acquire their linear order when they are synthesized from their components by using an already existing macromolecule as a template (‘template-directed synthesis’. This process will be explained below).

Genetic information in Crick's sense exhibits the three semantic features outlined above. First, genetic information involves a content, i.e., it is information about the linear order of the product's components. This is apparent from Crick's suggestion that the “sequence [of nucleic acid bases] is a (simple) code for the amino acid sequence of a particular protein” (Crick Reference Crick1958, 152). Another phrase suggesting the presence of a content is “hereditary information required for sequentialization” (144), where ‘sequentialization’ denotes the ordering of amino acids in proteins during protein synthesis. Furthermore, the following definition shows that genetic information is not only about the linear order of proteins, but also about the base sequence of nucleic acids: “Information means here the precise determination of sequence, either of bases in the nucleic acid or of amino acid residues in the protein” (153).

The second semantic feature concerns the possibility of error. Crick (Reference Crick1958) remarked that “mistakes” (140) may occur during protein synthesis and that the order in product molecules can be “right” (144) or “correct” (155). This suggests that genetic information can be expressed correctly or incorrectly in a sense not adequately captured by uncommon effects.

With respect to Crick's (Reference Crick1958) sense of genetic information, it makes intuitive sense to distinguish between information being stored and expressed. DNA can be said to express its information whenever it serves as a template for synthesizing another nucleic acid or a protein. When it does not, it may still do so later, and then it is natural to say that DNA stores its information.

Thus, Crick's (Reference Crick1958) claims about genetic information are modest insofar as DNA is thought to contain information only about the synthesis of its products. Nevertheless, this is information in a semantic, rather than a purely correlational, sense. Moreover, there seems to be something in the process of template-directed synthesis itself that motivates the attribution of semantic information. This process does not appear to be described semantically only in a derivative sense (as would be the case if synthesis were described as the copying or transmission of a message that is not about the product itself).

Godfrey-Smith thinks that, if template and product belonged to the same molecular class, they would bear a “coding relation” to one another only in a special, stipulative sense of ‘coding’ or ‘representation’, and that it would be “very unusual” to use such terminology in those circumstances (Godfrey-Smith Reference Godfrey-Smith2000b; cf. Godfrey-Smith Reference Godfrey-Smith2000a). However, this interpretation is not supported by views such as Crick's (Reference Crick1958), according to which DNA replication (an example of template-directed synthesis) is an informational process in its own right, although both template and product are composed of deoxyribonucleotides. Crick even considered as informational a hypothetical process in which, unlike in replication, the four kinds of deoxyribonucleotides pair up “like-with-like” (Keyes Reference Keyes1999). (Note also that replication and transcription involve direct specification of incoming nucleotides and still count as information transfers.) Perhaps, though, Godfrey-Smith's substantial point is correct, i.e., that ascribing semantic information to such processes is unwarranted (Godfrey-Smith Reference Godfrey-Smith2000b). I leave this issue open for the moment.

To this day, Crick's (Reference Crick1958) claims about genetic information are probably the most uncontroversial, and they were never idiosyncratic. When speculating that the base sequence of nucleic acids is a code for the amino acid sequence in proteins, he added that “[t]his hypothesis appears to be rather widely held” (152). As Crick said, he sought to state explicitly what he took to be a common understanding among his colleagues at the time.

Moreover, information about the order of products (“sequence information”) was the kind of information featuring in his “central dogma of molecular biology” (Crick Reference Crick1958, Reference Crick1970). The ‘dogma’ is concerned with the directions in which genetic information is supposed to be transferred. According to Crick (Reference Crick1970), genetic information is transmitted from DNA to DNA, RNA, and proteins, as well as from RNA to DNA, RNA, and proteins. Although information transfers from proteins to DNA, RNA, and proteins are “possible transfers” (562), they do not actually occur in nature.Footnote ⁴ As Keyes (Reference Keyes1999) emphasized, Watson et al. (Reference Watson1965) were more explicit in associating the three frequent information transfers with three molecular processes: replication or duplication (DNA → DNA), transcription (DNA → RNA), and translation (RNA → protein).

Although the ‘dogma’ has been repeatedly and severely criticized (reviewed in Thieffry and Sarkar Reference Thieffry and Sarkar1998), it is still portrayed as a fundamental biological principle in current textbooks (e.g., Nelson and Cox Reference Nelson and Cox2000; Alberts et al. Reference Alberts2002). More importantly, these criticisms do not challenge the status of replication, transcription, and translation as paradigms of informational processes. For example, Temin and Mizutani's (Reference Temin and Mizutani1970) discovery of reverse transcription (DNA synthesis from a RNA template) added an information transfer to Watson's version of the ‘dogma’.Footnote ⁵ Similarly, RNA-editing and RNA-splicing (modifications of RNA sequences prior to protein synthesis), do not undermine the (alleged) informational nature of transcription and translation. For although these processes limit the predictability of amino acid sequences from DNA sequences (Thieffry and Sarkar Reference Thieffry and Sarkar1998), they occur only after a primary RNA transcript (hnRNA) has been produced. So DNA may still contain information for the primary transcript, and information may be transferred from mature (edited and spliced) mRNA to protein.

I will argue that something like the limited claims expressed by Crick are justified because of the way in which templates determine their outcomes. I therefore now consider template-directed synthesis in more detail.

4. Template-Directed Synthesis

Linear macromolecules like DNA, mRNA, and proteins consist of specific, and often nonrepetitive, sequences of basic components. They are instances of ‘ordered sets’ or ‘n-tuples’ whose elements are the chemical components making up (parts of) the molecule. For example, the DNA-stretch A-C-C-T-G is an instance of the ordered set 〈A, C, C, T, G〉 where A is adenine, T is thymine, C is cytosine, and G is guanine. Ordered sets are defined by three properties: the number, the kind, and the linear order of its elements. Let us see how these properties are determined.

Template nucleotide bases are chemically active, i.e., they can attract and form hydrogen bonds with incoming nucleotides. Hydrogen bonds tend to form only between chemically and spatially complementary bases, i.e., Watson-Crick base pairing is highly specific in a given context. The chemical properties of a template adenine, for example, are such that only an incoming nucleotide ‘carrying’ thymine pairs with it and is then integrated into the growing product chain. Such a pairing is an instance of chemical regularities of the sort ‘If X, then Y* is integrated’ (for brevity, ‘is integrated’ will be omitted in the following), where X and Y* are placeholders for types of template and product components, respectively (the latter indexed by ‘*’), and ‘integrate’ means both the binding of a free product component Y* to X and the linking of Y* to the rest of product molecule. In replication, the pairing regularities between template and product bases satisfy the one-to-one function {〈A, T*〉, 〈T, A*〉, 〈C, G*〉, 〈G, C*〉}, where the first position of each 2-tuple represents a template base, and the second position represents an attracted free base, i.e., a component of the nascent product molecule.

In translation, there is no direct interaction between mRNA codons and amino acids but rather there is a mediation through the specificities of molecules like tRNAs and aminoacyl-tRNA-synthetases. One kind of codon specifies one kind of amino acid, and one kind of amino acid can be specified by more than one kind of codon (‘degeneracy’). As observed by Gatlin (Reference Gatlin1972), these specificities satisfy a many-to-one function. The set of specificities is commonly known as the ‘genetic code’.

I take it then that the identity of a product residue is determined by two factors. One factor consists in the ‘If X, then Y*’-type reactions between template and product components. As long as the identity of a component at a specific template position is unspecified, any of the ‘If X, then Y*’-reactions may occur at this position. Consequently, the corresponding product component may be any Y*. A second factor is therefore required to determine the identity of the product component. This is the presence of a particular component, say A rather than G, at a given template position. Given an A-token in the template and the regularity ‘If A, then T*’, the presence of A restricts the number of reactions that may otherwise occur at this site to one of the type ‘If A, then T*’. In this sense, a particular template base specifies which kind of ‘If X, then Y*’-reaction will occur. Since the reaction leads to the integration of one particular product component (e.g., T*), the template base also specifies the identity of the product component (given a set of pairing regularities). Furthermore, the number of product residues is determined by the number of template residues that serve as ‘substrates’ for reactions of the sort ‘If X, then Y*’.

The factor determining a product's linear order is the template's linear order.Footnote ⁶ This idea was first stated, with respect to protein synthesis, as the “sequence hypothesis” (Crick Reference Crick1958): “the sequence of bases along the nucleic acid determines the sequence of amino acids of the protein being synthesized” (158). The product elongates through a series of reactions that result in one component being linked to the next. The order in which these reactions occur depends on which template components are next to each other. For example, given a section A-C-G in the template, one reaction will be of the sort ‘If A, then T*’, the next will be ‘If C, then G*’, and so on. The result is the product-section T*-G*-C*.

I now want to draw attention to the seemingly trivial fact that the factors determining a product are present before the product is synthesized. That is, the nucleotides composing the template are present before the synthesis starts and they are arranged in a linear order. Consequently, once a section of nucleic acid has been chosen to be a template, it is fixed prior to synthesis which ‘If X, then Y*’-type reactions will occur and in which order. As Pontecorvo (Reference Pontecorvo1958) observed, templating is a process in which “a collection of building blocks (say, of the four main types of nucleotide), which could arrange themselves into any one of all the possible sequences, becomes arranged into only one sequence because of the prexistence of such a unique sequence.”

This fact distinguishes templating from other biochemical processes. Consider biochemical pathways in which different organic molecules are produced or transformed in a certain order in time. In the Krebs cycle, for example, oxalacetate (O) accepts an acetyl residue yielding citrate (C); subsequently, citrate is isomerised to isocitrate (I); then isocitrate is oxidized to 2-oxoglutarate (G), etc., until the last step yields oxalacetate again and the cycle starts anew. Like linear macromolecules, we can describe this pathway as an ordered set 〈O, C, I, G, …〉 and then ask which factors determine the identity and the order of the elements of this 8-tuple.

The identity of a molecule is determined like a product component in template-directed synthesis. For example, the fact that O is followed by C (rather than by G) is determined both by the presence of O and the regularity ‘If O, then C’ (the regularity here concerns a transformation, not a pairing).

However, the order of elements in the Krebs cycle is determined differently. In the Krebs cycle, each reaction product becomes the substrate of the subsequent reaction. C, for instance, is present as the result of the reaction transforming O into C, and C then becomes the substrate of the subsequent reaction that transforms it into I. Hence, the order of the 8-tuple 〈O, C, I, G, …〉 results from a series of ‘If A, then B’-type reactions in which the presence of one molecule (A) determines the identity of the next (B), and so on. The order is not determined by the properties of a molecule present before the reactions occur. It is not even necessary that the set of substrates is present before the cycle starts, because the substrates are produced as the cycle unfolds.

5. Naturalizing Instructional Content

So far, I attempted to show, in nonsemantic terms, that templates determine their product in a distinctive way. From here, we can get to the semantic features by looking at the properties in virtue of which items—like cooking recipes, computer programs, and the like—are instructions.

Recipes and programs provide specifications of the kind and order of operations, which if carried out, produce an outcome. For example, a recipe for a cake consists of a list of ingredients and a number of specifications that determine the kind and order of actions, which if carried out, produce the cake; and the recipe is provided before it is acted upon. Similarly, a computer program consists of a list of interconnected commands that specify the kind and order of operations that a computer will perform if it runs this program. Programs usually contain specifications of conditional form, and therefore, they rely on inputs to specify which operations to execute. However, the range of possible operations is specified by the program.

It seems, then, that programs and recipes share a peculiar way to determine their outcomes. They specify the kind and order of operations that will result in a certain outcome. Importantly, they specify this before the operations are performed: With a certain program loaded or a particular recipe in place, it is determined which operations (among a set of alternatives defined by other programs or recipes) will occur. The idea that operations are specified before they are performed appears to be the basis for our practice to distinguish between merely specified operations and those that are, in addition, executed.

I suggest that this way of determining an outcome is constitutive for instructions like recipes and programs. It provides a way to analyze, in nonsemantic terms, what it is for something to be an instruction. P instructs the making of Q just in case P determines Q by specifying, in advance, the kind and order of operations that produce Q if the operations are carried out. The proposal is to naturalize instructional processes as a certain way to determine an effect. If this picture is correct, then the content of an instruction consists in the specifications of the kind and order of operations that will result in a certain outcome.

Templates, I suggest, determine their products in just the same way as do recipes and programs. For we saw that a nucleic acid serves as a template for the synthesis of a product just in case it determines the kind and order of product components in the following way: The nucleic acid is present before the start of synthesis and it determines, through the kind and linear order of its components, the kind and sequence of ‘If X, then Y*’-type reactions that will occur. The nucleic acid section reduces the number of possible pairing reactions to one at each of its sites. Thus, the section can be said to specify, before the start of synthesis, the kind and order of reactions that will result in the product if the reactions occur. Hence, templates determine their products in the same way in which recipes and programs determine their outcomes.

If this mode of determination is indeed constitutive for instructional processes, then it is justified to say that molecular templates contain instructional content for the synthesis of their products. On this view, the instructional content of a molecular template consists in those of its (nonsemantic) properties that determine a product in the characterized way. These properties are the number, kind, and order of its components, i.e., the n-tuple instantiated by the template. As I want to show now, the three alleged semantic features of informational molecules can be construed as features of instructions in general.

6. The Semantic Features

7. DNA and Protein

If accepted, the proposed account justifies the idea that templates contain semantic information about the synthesis of their immediate products. This conclusion still falls short of Crick's (Reference Crick1958) claim that DNA contains information about the linear order of amino acids in proteins, because the immediate products of DNA templates are not proteins but (primary) RNA transcripts. I now want to put forward an argument to show that, under some conditions, DNA does contain information about proteins.

What would the argument need to show? On the present account, a particular mRNA's instructional content for synthesizing a protein is the n-tuple it instantiates. The n-tuple's linear order is part of this content, and it determines the linear order of amino acids in the protein. Now suppose that the RNA's order has been preserved (in some sense) from the DNA template, such that the mRNA would possess the same kind of order as the DNA. Then, the linear order of the DNA template would ultimately determine the protein's order in an instructional way. For the DNA's order would determine the linear order of both the mRNA and the protein. Therefore, what needs to be shown is that there is a sense in which the DNA's linear order is preserved in the mRNA's order.

Consider the section C-T-G-A of a DNA template. The bases in the template section stand in neighbor relations to each other (i.e., C is next to T, T is next to G, and G is next to A), and these relations constitute this template's linear order. For each kind of DNA base there is exactly one ‘corresponding’ RNA base, because the biochemical complementarities between DNA and RNA bases satisfy the one-to-one mapping {〈A, U*〉, 〈T, A*〉, 〈C, G*〉, 〈G, C*〉}. Now suppose the section C-T-G-A is transcribed, and we compare the template's linear order with the order of its product. We can then ask whether the neighbor relations among the bases of the DNA template hold among the corresponding RNA bases of the transcript. For example, the neighbor relation ‘C is next to T’ (in the DNA template C-T-G-A) holds in the RNA product just in case the base corresponding to C is next to the base corresponding to T, i.e., just in case the product base G* is next to product base A* (the transcript is G*-A*-C*-U*). If the relations hold in RNA, we say that the template's linear order (its structure) has been preserved in the RNA's linear order.Footnote ⁸ If, by contrast, G* would be next to, say, U* in the product, then the neighbor relations among DNA bases would not hold among RNA bases (the transcript might be G*-U*-A*-C*). As a matter of empirical fact, the transcript will be G*-A*-C*-U*. Therefore, transcription is structure-preserving (as is translation).Footnote ⁹

Having shown that the DNA's linear order is preserved in the mRNA's order, I conclude that the linear order of the protein is ultimately determined by the DNA's order. That is, the DNA's linear order is the instructional content, and hence, the genetic information that the DNA template provides for protein synthesis. Of course, DNA provides the instructional content for protein synthesis only if the RNA transcript is not altered before translation. Since the primary transcripts of eukaryotes are usually modified by RNA-splicing and RNA-editing, it may only be in organisms like bacteria where DNA actually does contain the information about the order of amino acids.

8. Disanalogies

Conceptualizing gene function in terms of instructions is an approach dating back to the early 1950s and taken up repeatedly ever since (see Kay Reference Kay2000). For example, François Jacob (Reference Jacob[1970] 1974) interpreted the genetic material as instructions. Just as in the present account, a part of the justification for his claim was the fact that the template's linear order determines its product's arrangement. However, the idea that has not been worked out previously is, I suggest, that molecular templates share a certain way of determining outcomes with recipes and programs, and that it is this kind of determining an effect that makes them instructional. Nonetheless, given that previous instructional accounts of genetic information did not succeed in demonstrating semantic properties of nucleic acids, one might well wonder why the account defended here should fare any better. In particular, one might point to disanalogies between templating and anthropogenic instructions and argue that they forbid attributing content to molecules.

Recipes and programs do not just bring about a particular outcome; they are designed to do so. Perhaps what makes them be about an outcome is the fact that they were designed to produce it. If so, then either templates are about their outcomes because they are biologically designed to do so, or if they are not so designed, then they cannot be about their products. In the latter case, even if templates determined their products in ways structurally similar to recipes, they would not possess semantic content. (Weber Reference Weber2005 directs a similar line of argument against Maynard Smith Reference Maynard Smith2000.)

Recipes and programs are usually formulated with a purpose, and many natural templates have the biological function to contribute to causing a certain outcome. But I deny that recipes and programs are instructions in virtue of having a purpose, and hence, that templates are instructions in virtue of their biological functions. Suppose the steps of the recipe or the program had been arranged randomly in advance. Then the ‘cake’ may hardly be edible, and the program may not perform anything sensible at all. Nevertheless, these outcomes would have been determined by specifying all individual steps and their order of occurrence such that the steps produce the outcome if they are carried out. I take it that we would still regard the computer output as the result of executing a pre-specified series of operations, and the inedible lump as the result of carrying out some (nasty) sort of instruction. Therefore, the attribution of instructional content is not derived from the function of a template, even if it has one.

Another concern is that recipes and programs seem to carry meaning or instructional content because they are linguistic entities, at least at some level. They are written commands and they instruct what they instruct in virtue of being meaningful sentences. This point can be developed into two objections. First, one may argue that only linguistic entities can carry instructional content in any proper sense. A less strict position would allow for an analogous sense of content in DNA. If this line is taken, DNA would convey information to the extent that it is akin to a language (as some say, DNA is like a string of words). Moreover, its instructional content should then be analogous to the meaning of words. The second objection is that my account fails on the latter score, because it equates instructional content with a template's n-tuple. And certainly, whatever a word's meaning, it will never be the n-tuple given by its letters or something similar. ‘Dog’ does not mean something like 〈d, o, g〉. From this point of view, I simply seem to confuse semantics with syntax.

Many manmade instructions involve verbal descriptions and they have linguistic meaning. It is even plausible to assume that they convey their instructional content by linguistic means. But this does not show that their instructional content somehow arises in virtue of them being linguistic entities. Suppose I happen to find someone's phone number together with a wallpaper index that assigns one digit to each wallpaper color (0-blue, 1-green, etc). I also unpack my bag of colored marbles and arrange them on a thread. I do so by matching each phone number digit to a colored marble according to the digit-color index. Surely, the phone number is the instructional content determining which kind of marble to put on the thread and in which order. Yet phone numbers do not mean anything in a linguistic sense, nor are they designed as instructions to produce chains of marbles.

This example, one may object, does not show that instructional processes can be individuated in purely physical terms. From a physical point of view, the phone number is just ink on paper. Similarly, nothing about the physical properties of the wallpaper index determines that it should be adopted as a rule. These items feature in an instructional process only after someone has decided to use them in this way.Footnote ¹⁰

I agree that in this case, a purely physical description does not capture the instructional nature of the process. But this is unproblematic for the following reason: On the present account, an instructional process is meant to be a process in which an item bears a certain structural relation to an outcome. The relata may be linguistic, chemical, or other entities. But it is right to insist that, if they are chemical entities, then they must bear this structural relation just in virtue of their physicochemical properties. I believe that this is the case in template-directed synthesis. The template determines the features of the product because of the physicochemical properties of nucleic acid bases and their arrangement. The process is driven by the causal powers of the template and product components (plus those of enzymes and so on).Footnote ¹¹ This marks an important difference with recipes and chains of marbles, where human action is required to produce the outcome. But I do not see that it undermines the idea that, in the essential structural aspect, both molecular templates and recipes have a common way of determining their outcomes.

9. Implications

The recent literature provides two interesting thought experiments about the conditions of attributing semantic information to molecules. Let us see how to evaluate them on the present account. Godfrey-Smith (Reference Godfrey-Smith2000b) argues that information requires, among other conditions, that template and product belong to two different chemical classes (such as to require a nontrivial rule of specification, Godfrey-Smith Reference Godfrey-Smith2000a). In his thought experiment, proteins are replicated by synthesis from a protein template. A given template amino acid specifies, via a connector molecule, the same kind of amino acid in the growing protein. He suggests that this process would not involve a “coding relation,” because there would be “no biological reason” for postulating such a relation (Godfrey-Smith Reference Godfrey-Smith2000b), other than in the weak sense of information theory (Reference Godfrey-Smith2000a).

However, on the account defended here, there are biological reasons for claiming that templating is informational even if it does not involve two different molecular classes; for, attributing genetic information to a template is to claim that it determines its product in a certain way, i.e., through the step-by-step realization of operations specified in advance. And, templates so determine their products whether or not the components belong to the same molecular class. Moreover, the different-classes requirement has problematic consequences. Neither replication nor transcription involve different chemical classes, and hence, paradigm cases of informational processes, such as heredity, come out as noninformational.

Note also that my account is not tailored to rescue the primacy of DNA over other developmental factors. Template components may belong to any chemical class. As long as a template specifies its products’ arrangement by determining the kind and the order of individual ‘If X, then Y*’-type reactions, the process is one of instruction. This explains why, in a thought experiment by Neumann-Held (Reference Neumann-Held, Neumann-Held and Rehmann-Sutterforthcoming), a chain of enzymes codes for its product (assuming that ‘x codes for y’ here means ‘x has information for y’). Neumann-Held (Reference Neumann-Held, Neumann-Held and Rehmann-Sutterforthcoming) envisions a process in which a molecule is synthesized from several different components in a series of reactions catalyzed by different enzymes. The enzymes are linearly fixed in a membrane, and the order of the enzymes determines the order in which the components are joined together. Thus, each enzyme can be assigned to the component it helps to join. The conclusion seems to be that this process is not relevantly different from translation and that it is, therefore, as informational as translation. I agree that it is informational, and I suggest it is so because the chain of enzymes determines the product by determining the kind and the order of individual ‘If X, then Y*’-type reactions.

10. Conclusion

The instructional account of genetic information makes a reductive claim about what it is for an entity to instruct something: Instructing consists in an advance specification of the kind and order of steps yielding a certain outcome if the steps are carried out. It claims, further, that molecular templates determine their products in this way. For in a process like replication, one molecule specifies, prior to synthesis, the kind and order of chemical interactions that determine the kind and linear order of the product's components. If this is accepted, then it is legitimate to describe the template's properties, which so determine the product, as the instructional content (or information) for the synthesis of the product.

This account does not apply to other likely paradigm cases of informational processes in molecular genetics. The ‘start signals’ for transcription and translation provide one example. It seems that signaling the start of transcription is regarded as a transfer of genetic information that leads to, but is not based on, template-directed synthesis. However, the instructional account is not meant to cover all paradigm cases. The expression ‘genetic information’ seems to be used differently in different contexts, and it is a separate issue to decide whether it is legitimately used in all those contexts. My aim here was to show that there is at least one class of molecular processes exhibiting the semantic properties of aboutness, error, and information storage.