12 Making Sense
In principio erat finis.
A mechanical device strikes a match and a candle is lit or there is an explosion. The striking of the match (+M) and the presence of oxygen (+O) are the same in both scenarios. The difference that makes the difference as to whether or not there is an explosion (±E) is the presence or absence of hydrogen (±H). A more sophisticated device strikes a match contingent on input from a hydrogen sensor. If the sensor fails to detect hydrogen (–H), the match is struck (+M). If the sensor detects hydrogen (+H), the match is not struck (–M). The first device is an effector of an explosion in the presence of hydrogen but does not “choose” the explosion because it does not “use” information. It couples a state of the world (±H) to an outcome (±E). The second device “prefers” darkness to an explosion in the presence of hydrogen. It couples one bit of information about the world (±H) to one degree of freedom in action (±M). It “responds” to +H with –M and to –H with +M. It is “undecided” until observation of what was uncertain (information) is interpreted in definite action (meaning).
The first difference in the phrase “a difference that makes a difference” is cause, the difference-maker or independent variable, and the second is effect, the difference-made or dependent variable. But whether the second difference is an interpretation of the first depends on the evolved or designed function of an interpreter. The first device does not interpret. Things just happen (±H, ±E). For the second device, the first difference is information (±H) and the second is meaning (±M). For an outside interpreter of the second device, ±H and ±M contain mutual (or redundant) information: either can be inferred to “mean the other.” The observer can predict whether or not the match will be struck (meaning) by observing whether or not hydrogen is present (information) or could infer whether or not hydrogen was present (meaning) by observing whether or not the match was struck (information).
On what reasoned grounds can I claim that the second device interprets the presence of hydrogen as a reason not to light a candle but reject the seemingly parallel claim that the first device interprets the presence of hydrogen as a reason to cause an explosion? My argument makes an implicit appeal to what Ruth Millikan (1989) has called proper functions. The proper function of the first device is lighting candles. Explosions are unintended consequences. The proper function of the second device is the use of information to decide whether or not to strike a match.
The responses of an interpreter with one degree of freedom of interpretation may seem mechanical and uncomprehending, hardly deserving the “meaningful” label, but any truly sophisticated interpreter will have many degrees of freedom with multiple levels of internal interpretation in which the interpretation of one part of the system is news (and hence information) to other parts of the system. Consider multiple rewirings of a complex device such that each new device responds to the same inputs with different outputs. There is only a shallow sense in which the inputs of these devices are causes of the devices’ outputs. An understanding of how inputs are interpreted as outputs requires an understanding of a device’s inner workings. An understanding of why a device interprets particular inputs as particular outputs requires an understanding of the device’s function and history. We cannot invoke an omniscient homunculus within the system that has an overview of what the system “knows” (Dennett 1991). An interpreter cannot “know” what it will choose until it chooses—if it “knows” it has chosen—but an observer can often predict a consistent interpreter’s choices with confidence.
John Dewey recognized that “stimulus and response are not distinctions of existence, but teleological distinctions, that is, distinctions of function, or part played, with reference to reaching or maintaining an end” (1896, 365). A response implies a purpose. One cannot simply draw an arbitrary boundary around part of a complex web of processes, and then describe all causes crossing the boundary from outside to inside as stimuli, all causes crossing from inside to outside as responses, and all processes within the boundary as interpretation. Interpreters are intentional mechanisms that have evolved or been designed to use information in choice. Under this version of behaviorism it is not the input that determines the output but the relation between input and output that determines the mechanism.
Teleology of Interpretation
Meaning and function are intentional terms. The previous chapter grounded talk of biological function in the unintended teleology of adaptation by natural selection (following Dennett 1987, 1995; Millikan 1989; Neander 1991; Papineau 1984). Final causes were presented as efficient, even indispensable summaries of complex concatenations of efficient causes. A token effect cannot precede its token cause. But when one generalizes from causes of tokens to causes of kinds, effect-tokens both precede and succeed cause-tokens in recursive processes. A full causal account of an egg or a chicken contains long series of past chickens and past eggs. An egg is both an effect of a chicken-that-was and a cause of a chicken-to-be.
Natural selection subsumes all processes by which the environment selects a subset from a set of actual things. Reproduction replenishes numbers of the diminished subset before the next round of selection. Although nature’s “choices” are unintended, some of her “choices” leave genetic records that allow repetition of that which was “chosen” as intended choices of living beings. Reproductive recursion is rescued from eternal recurrence of the same by the input of new variation via mutation and the shuffling of genetic texts in sexual reproduction. By these processes, recursively selected subsets accumulate information about what worked in the past. And what worked was the interpretation of information from the environment in “real time” rather than evolutionary time. As a consequence, the world now abounds with biological interpreters that select which differences will make a difference from the myriad potential causes in their environment and choose actions from sets of possible actions on the basis of observations that could have been different. The mapping of possible inputs to outputs is embodied in the interpreter’s fine structure, with the fit between information and meaning—the efficaciousness of interpretation—derived from past natural selection refined by developmental processes during the interpreter’s life.
To “intend” is to choose for anticipated effects. We can distinguish two kinds of intentionality. Primary intentionality is the repetition of causes that worked in the past. This is the intentionality of adaptation by natural selection and of conditioned reflexes. Past effects are anticipated to occur again. Secondary intentionality is choice of action after simulation of possible choices and their effects. Simulated effects are anticipated to occur when the action is performed. Secondary intentionality requires imagination, an ability to “hold in mind” and evaluate virtual outcomes. Primary intentionality is “primary” in the sense that anticipation evolved before imagination.
Before the Interlude, I wrote that information has meaning for an interpreter when it is used to achieve an end. The present chapter simplifies and clarifies by explicitly equating meaning and interpretation. The action or thing chosen is the meaning of the observation for the interpreter. Information resides in differences among things that remain “possible” until observation of an actual thing. Meaning is the response of the interpreter to the observation and is itself an actual thing that can be observed and used as information by another interpreter. By these definitions, “semantic information” is a contradiction in terms.
Information and Meaning
An interpreter can be viewed as an input–output device that uses observations to choose actions (figure 12.1). Interpretation subsumes all internal processes that couple observations (information) to actions (meaning).1 The number of independent things a device could observe can be considered a measure of its uncertainty (entropy of observation). The number of independent actions in its repertoire of response can be considered a measure of its indecision (entropy of action). Uncertainty is resolved by observation, indecision by choice. Uncertainty and indecision are measures of potential things. Observations and meanings are actual things. The same observation can mean different things to different interpreters, and different observations can mean the same thing (i.e., be interpreted as the same choice of action).
Figure 12.1
An interpreter is a computational mechanism for which information is input and meaning is output.
An interpreter’s possible inputs are the things to which it could respond. Its possible outputs are the ways in which it could respond. These capabilities are competences of the interpreter, not properties of its world. They are subjective rather than objective. Observations inform, whether the thing observed is ontologically uncertain (undetermined until observation) or epistemically uncertain (determined but “unknown” until observation). Observation of what is epistemically uncertain provides information about prior events. Meanings may be “mistaken” because of malfunction, unanticipated inputs, or because what was once adaptive is now maladaptive. Unintended meanings may be used as information by other interpreters or by the same interpreter in self-reflection.
My purpose in this “behaviorist” account is not to belittle the complexities of interpretation but to argue that there is no ghost in the semantics. Information resides in distinguishable things in the interpreter’s world, and the meaning of a particular input for the interpreter is simply whatever physical thing (print on paper, sound vibrations, neural states, etc.) is the output of information processing by the interpreter. The complexities reside in how the inner workings of the interpreter map observations onto actions. There is no nonmaterial domain in which meanings reside outside of physical interpretation. If you protest that this paragraph means more than ink marks on paper or pixels on a screen, then those ink marks or pixels have been input to a very sophisticated interpreter—your good self, and I thank you for reading.
Consider a paradigmatic small dark something moving across a frog’s visual field that causes the frog to stick out its tongue to intercept the thing. If we treat the frog as a black box, photons falling on retinas are information (input); sticking out the tongue is meaning (output). If we were to peek inside the frog’s brain, we would find multiple interpretations of interpretations between sensory excitation and motor action. My claim is that each physical state can be considered the meaning of prior information processing and that these states inform subsequent neural states that are themselves new meanings. The frog’s visual system interprets incident photons as information about distance, direction of movement, and speed of the speck. These meanings inform subsequent interpretation as motor action. The frog minimizes immediate interpretation of a speck’s nature so as not to give a fly time to interpret the frog’s intentions (a small dark object in the mouth is worth ten flies that got away). Once the moving something has been intercepted, the frog has ample time to interpret whether the thing is food and what kind of food (using oral rather than visual sensors) and to modify its sensory criteria for future protrusions of the tongue.
Any spoken or written claim by a philosopher about what internal states mean to the frog—whether “fly,” “food,” or “small moving thing”—is an interpretation of the philosopher. It is thus the philosopher’s meaning, not the frog’s meaning. If we were to peek inside the black boxes of philosophers’ minds there would undoubtedly be many interpretations of interpretations, meanings of meanings, before keys were struck on keyboards or words spoken or shouted. If a literate frog were to write a memoir of her experience, she might report that she saw the speck as a fly but was mistaken. Her interpretation would be of similar kind to the philosophers’ interpretations. An interpreter, even an interpreter of itself, never has direct access to things in themselves but only to information about things.
The claim that meaning is whatever physical thing an interpreter interprets information to mean is a definition, not a claim that all interpretations are equally useful. Some interpretations are “better” than others because they inform more subsequent interpretations or enable meaningful interpretation of what was previously uninterpretable. Our perceptions have evolved to present useful information about the world to guide our actions, and our interpretations of words have evolved, from childhood, to make sense of what others are saying. Although information and meaning are defined relative to an interpreter as subject, interpreters may aspire or have evolved to interpret information objectively (Lindley 2000).
Interpretation of Interpretations
The nonliving world is a repository of unintended information useful for living interpreters. Unintended information is also present in interpretations of other interpreters. When an interpretation is reinterpreted, one must distinguish the intentions of the first interpreter (the producer) from those of the second interpreter (the consumer). The evasive movements of a gazelle pursued by a cheetah are intended to make the gazelle harder to catch. The cheetah observes and interprets these movements with the intention of catching the gazelle. These interpretations of the cheetah are unintended by the gazelle.
When a healthy gazelle sees a hunting dog rather than a cheetah, it interprets the situation as an occasion to stot (jump up and down). Hunting dogs preferentially chase gazelles that do not stot or stot more feebly. A vigorous gazelle and a hunting dog both benefit from the hunting dog chasing a feebler gazelle. The hunting dog’s decision to chase a nonstotting gazelle is intended by the stotting gazelle. The evolutionary rationale of stotting is thought to be that stotting “signals” to the hunting dog that the gazelle has a good chance of outlasting the hunting dog in an extended pursuit and is therefore not worth chasing (FitzGibbon and Fanshawe 1988). But this is an interpretation of behavioral ecologists, not, as far as we know, of either gazelles or hunting dogs. Their interpretations are simply stotting and not chasing. (Gazelles do not stot to cheetahs because cheetahs lack endurance but are capable of short bursts of great speed. Gazelles interpret cheetahs as reasons to get far away as quickly and unpredictably as possible.)
Interpretations are actions chosen to achieve ends. Some interpretations are intended to be used as information by subsequent interpreters or by the interpreter itself at some later time. I will use text to refer to an interpretation intended to inform subsequent choice. A text is an output of an author (producer) intended to be input to a reader (consumer), but it is the reader who chooses how to interpret the text. A text anticipates interpretive competence of intended readers. It may be a static object or dynamic performance. By this expanded definition, written documents, works of art, DNA and mRNA, neural activity, and the tape of a Turing machine are all texts. My spoken words are an ephemeral text “written” in sound intended to be interpreted by listeners. A painting is a persistent text “written” in pigment intended to be interpreted by viewers. Parallel white lines crossing a road are texts intended to be interpreted by pedestrians as places to cross and by motorists as places where pedestrians cross. A peacock’s tail is a text intended to arouse the admiration of peahens. Stotting is a text intended to discourage hunting dogs.
An author’s intended interpretation of a text should be distinguished from the actual interpretation of the reader. A hunting dog who detects unintended evidence of weakness in a gazelle’s performance may chase a stotting gazelle. The author’s intentions are also distinct from how the author intends a text to be interpreted by readers. Some texts are deceptive. The folded wings of a camouflaged moth are intended to be interpreted as “not a moth” by moth-predators, but the flash of “eyespots” as the moth unfolds its wings for flight are intended to be interpreted, for a crucial split-second, as “eyes of a moth-predator predator.” If these texts are interpreted by moth-predators as intended by the moth, then they have served their purpose and have been interpreted as intended by the author by being misinterpreted by the reader.
A completed nest contains clues about its construction. If birds model their own nest on the nest in which they hatched, then the parental nest informs the construction of the offspring nest. If parents constructed nests in ways that were easily interpreted by offspring and this method of construction was repeated because it enhanced the survival of the parents’ grandoffspring in offsprings’ nests, then the nest is a text of the parents with an intended interpretation by offspring. This example shows that a thing may function both as a tool (for holding eggs) and as a text (for instructing offspring). When the Mafia leave the body of an informer in a town square, the murder is both a direct means to an end (removal of an informer) and a text (a warning to potential informers).
The broken-wing display of a ground-nesting bird is a text intended to be interpreted as “here is easy prey” by a predator. The function of the bird’s “helpless fluttering” is to lead the predator away from hidden eggs that are truly easy prey. If the predator recognizes the display as a text, concludes that “a nest is near,” and searches for the nest, then the text has failed to be interpreted as intended but the predator has correctly interpreted the situation. If, on the other hand, the bird truly has a broken wing, then its seemingly helpless fluttering would indeed be helpless fluttering, and an attempt to escape, rather than a text with an intended interpretation. The predator would truly interpret the bird’s actions if it concluded “here is easy prey.” If the predator mistakenly interpreted helpless fluttering as “a broken-wing display” and started looking for a nest then the predator would have misinterpreted the situation.
One could imagine a bird with an actual broken wing “unconvincingly” attempting to escape with the intention that the predator interpret its actions as “this is a broken wing display; therefore, the bird is not easy prey but a nest is near.” If the predator started to search for a nest, then the predator would have recognized that the bird’s actions were a text, would have interpreted the text as intended by the bird, but would have misunderstood the bird’s intentions. The text would have achieved the bird’s intention but foiled the intention of the predator who had been “deliberately” misled.
Ultraviolet photons damage unpigmented skin. Some skin responds to ultraviolet exposure by deposition of melanin. For skin that does not respond in this way, ultraviolet photons are not used as information and simply cause unintended damage. For skin that is able to darken, ultraviolet light is interpreted by the skin as melanin to prevent further damage. An observer of tanned skin can infer that the skin has been exposed to the sun and that, depending on context, the skin’s owner works outdoors or has sufficient leisure to spend time at the beach. Because possession of leisure has social value, some people choose to expose their skin to solar lamps so as to be interpreted as possessors of leisure. In this case, tanned skin is a text intended to inform the interpretations of intended observers.
An interpretation (a meaning) can be considered to “represent” the interpreted information and to be an ascription of content to the information by the interpreter. (The interpretation stands as a metaphor of the observation.) Representation raises intriguing questions about how internal processes of an interpreter derive meaning from information and why the interpreter has evolved or been designed to interpret information in the way that it does. A text not only “represents” information used by its author in the text’s composition but “presents” information to intended readers. Presentation raises additional questions of how authors anticipate readers’ responses.
Texts act not directly but indirectly, through the interpretations of readers. Texts are not agents. They do nothing, in and of themselves; but one need only consider the Bhagavad Gita, the Declaration of Independence, or The Protocols of the Elders of Zion to appreciate the differences texts can make in the world.
Private and Public Texts
The inner workings of complex interpreters involve delegation of tasks among subinterpreters, each presenting texts for use by other subinterpreters. These private texts, intended for internal use, all have a physical form. Some are ephemeral. A percept is an interpretation of sensory input intended to inform subsequent interpretation. Others are persistent. A memory is a textual record to be consulted when needed. Consciousness is a private text, our mental desktop, a very short-term memory written in we know not what medium, consulted by other subinterpreters to know where to “look” for relevant information. The scene that is seen functions as a simplified and constantly updated interpretation of the world that is compared against incoming percepts for detection of relevant differences.
I view a landscape by Claude Monet in which artfully arranged splodges of color on canvas present a scene on the Seine with sailboats and a group of five ducks to one side. Not only do I see the play of light on water, I hear the clink of rigging as the boats rock at anchor. These meanings are my interpretations. My companion sees gulls rather than ducks. As I move in toward the canvas, the things that I saw dissolve into ill-formed blobs. The ducks are revealed as no more than five dabs of white paint. The scene that I saw was underdetermined by input from my eyes, but Monet’s genius was to suggest, with minimal means, internal sources of information that filled in the picture. He created the illusion of detail that I expected to find when I looked closely. This paragraph does not contain all the information I want to inform your interpretation. By the artful placement of ink on paper I want to suggest paintings and scenes you have seen in order to evoke an “Aha! I see what he means.” I particularly want you to see that an author always relies on rich sources and resources in the private texts of his readers for them to make sense of his public text.
My text is the product of multiple drafts of an evolving text. In the process of reading and rereading, writing and rewriting, I came to understand what I meant and mean. My meaning is the public text that you see, not some nebulous sense in my mind to which the text points. As my aging mind becomes less nimble, I rely more and more on public texts of previous selves as aide-mémoire of what I wish to mean. What persists in my brain are reworked memories of earlier drafts and regrets about what I wrote poorly. Once a text is published, its meanings for readers are untethered from its author’s intentions. My meaning becomes your meaning as you read my text.
Meanings of Genes
What do genes mean? The short answer is whatever physical thing an interpreter interprets them to mean. Cells contain three very important interpreters that evolved long ago to interpret aperiodic polymers: DNA polymerases complement sense-strands of DNA with their antisense-strand; RNA polymerases transcribe sense-strands of DNA as RNA; ribosomes translate messenger RNAs (mRNAs) as proteins.
For a DNA polymerase, the meaning of a DNA strand is its complement. For an RNA polymerase, the meaning of a DNA strand is an mRNA. For a ribosome, the meaning of an mRNA is a protein. Thus, the same DNA sequence means different things to a DNA polymerase and an RNA polymerase. The information (input) is the same, but the meaning (output) is different. DNA and mRNA are texts intended to be interpreted. RNA polymerases also transcribe transfer RNAs (tRNAs) and ribosomal RNAs (rRNAs). These are tools to be used in translation of mRNAs, not texts to be further interpreted. Among the texts interpreted by RNA polymerases and ribosomes are instructions for the assembly of RNA polymerases and ribosomes. Interpreter know thyself.
These molecular machines are general-purpose, mindless interpreters of specialized texts. DNA polymerases, in particular, resemble monkish scribes in the scriptorium of the Total Library of all possible texts (Borges 2000), faithfully copying DNA sequences that are both unilluminated and unilluminating. One can posit a Library of Mendel that contains all possible DNA texts of some finite length (Dennett 1995). Only an infinitesimal subset of texts the length of the human genome can ever have existed. Natural selection has acted in this much more restricted, yet unimaginably vast Library of Darwin that contains all past and present DNA texts. In this library, differences between texts that are still read and texts that are no longer read convey information about what has and hasn’t worked in the past.
A DNA strand is an interpretation of its antisense, via one round of DNA replication, and of its sense, via two rounds of DNA replication. But a gene can also be interpreted as an mRNA, via transcription. Thus, a gene means itself for a DNA polymerase but means an mRNA for an RNA polymerase. Complex interpreters can be built out of simpler interpreters. A DNA strand is interpreted as a protein by the combined interpretative system of an RNA polymerase and ribosome via two steps of interpretation, first transcribed as mRNA, then translated as protein. Thus, a gene means a protein for this compound interpreter.
A more contentious claim is that an organism’s genes collectively mean the organism. Past organisms have been responsible for the replication and transmission of present genes, which are interpreted via complex processes of development to produce present organisms that are responsible for the replication and transmission of future genes. Organisms and their genomes are thus recursively related via primary intentionality. As such, organisms can be considered to interpret their genomes as themselves. This bald statement should not be overinterpreted. Every interpretation of a genomic text as an organism is unique because texts are always interpreted in the context of other sources of information. Organisms interpret their genomes in environmental context and not every detail is intended.
Minor genetic differences may have major organismal effects. The 282nd amino acid of human factor VIII proteins is arginine, specified by the codon CGC on the sense-strand of the factor VIII gene. The complementary codon on the antisense-strand GCG is the template for transcription of CGC in the mRNA which is translated as arginine by the ribosome. Cytosine (C) preceding guanine (G) can be chemically modified by attachment of a methyl group to produce 5-methylcytosine (5-meC). Spontaneous deamination of 5-meC converts 5-meC into thymine (T), creating a heritable mutation that changes antisense GCG to GTG, which is interpreted as sense CAC by DNA polymerase. RNA polymerase transcribes GTG (DNA) as CAC (mRNA), which is translated as histidine by the ribosome. Factor VIII protein, with histidine rather than arginine as the 282nd amino acid, fails to clot blood causing life-threatening hemophilia in males.
Mutations in DNA that change an amino acid in protein are described as nonsynonymous. DNA polymerases, RNA polymerases, and ribosomes faithfully interpret nonsynonymous mutations without regard for the functionality of the resulting protein. For a DNA polymerase, CAC means GTG (antisense) and CAC (sense); for an RNA polymerase, GTG means CAC; for a ribosome CAC means histidine. These are the intended meanings of interpreters that have evolved to represent whatever text they are presented. But, histidine and hemophilia are unintended from the perspective of the organism: none of the fetus’s male ancestors possessed the mutant protein or suffered from hemophilia. A genetic counsellor reads a printout of the DNA sequence of the factor VIII gene present in amniotic fluid of a male fetus whose mother is a carrier of a mutant factor VIII gene on one of her X chromosomes. The difference that will make a difference is G versus A in the middle position of the 282nd codon of the factor VIII gene. Before viewing the printout, the counsellor is uncertain about what the data will show. If she reads G, then she can confidently inform the parents that the child will be unaffected; but if she reads A then she must inform them that the child will be affected.
Mutations in DNA that do not change the amino acid translated by ribosomes are described as synonymous. An evolutionary biologist might use the ratio of nonsynonymous to synonymous differences between two DNA sequences to infer whether the sequences have been subject to natural selection. In this case, meaningless differences for an intended interpreter (a ribosome) are meaningful differences for an unintended interpreter (an evolutionary biologist).
Telegraphing One’s Intentions
An engineering communication theory is just like a very proper and discreet girl at the telegraph office accepting your telegram. She pays no attention to the meaning, whether it be sad or joyous or embarrassing. But she must be prepared to deal intelligently with all messages that come to her desk.
—Warren Weaver (1949)
The Zimmermann telegram of 1917 conveyed an offer of a military alliance between the German Empire and the Republic of Mexico if the United States of America entered the Great War. In the event of such an alliance, Germany promised to provide financial support to Mexico and recognize Mexican reconquest of Texas, New Mexico, and Arizona.
Great Britain cut all German telegraphic cables to the New World on August 5, 1914. Therefore, the German message to Mexico had to be conveyed by an indirect route. A plaintext of the message in German was composed in Berlin, under the direction of the secretary of state for foreign affairs (Arthur Zimmermann), and then encrypted using code 7500. The encoded message was passed to the American Embassy in Berlin for transmission by US diplomatic channels to the German Embassy in Washington. The message was sent by cable from Berlin to Copenhagen, from Copenhagen to London, and then by transatlantic cable from London to the State Department in Washington, DC. The encrypted message in code 7500 was passed to the German Embassy in Washington where it was decrypted into German and then re-encrypted in code 13040. Re-encryption was necessary because the German Embassy in Mexico lacked a code book for the more secure code 7500. The message in code 13040 was sent by telegram from the Western Union office in Washington to the Western Union office in Mexico City where it was printed as a typescript and delivered to the German Embassy in Mexico on January 1, 1917. The typescript was decrypted into German by the ambassador’s secretary, before the offer was communicated by the ambassador to the Mexican President in Spanish.
Unbeknownst to the governments of the United States and Germany, the telegraphic signals that included the message in code 7500 had been intercepted by British intelligence in London and the message partially decrypted. The British then requested their agents in Mexico obtain a copy of the telegram in code 13040. The major reason for this step was to hide from the US government that British intelligence was reading their diplomatic cables. British intelligence decrypted the telegram in code 13040 (it contained little unexpected for them) and then the British government provided the telegram to the American government in its coded form and as versions decrypted into German and translated into English. The message was then released by the pro-war party in the United States to the American press. The popular outcry hastened the entry of the United States into the war on the side of Great Britain and its allies. The British government had decided that the benefit of revealing the telegram to the United States would outweigh the unavoidable information provided to Germany that Britain could read code 13040.
The Zimmermann telegram passed through many texts in many media, with interpretation occurring at each conversion from one medium to another. Sometimes “the message” was in German, sometimes in code 7500, sometimes in code 13040, sometimes in Morse, sometimes in Spanish, and sometimes in English. For the various clerks in telegraph offices in Berlin, Copenhagen, London, Washington, and Mexico City, the meaning of the text was simply the “mechanical” transcription of a stream of dots and dashes into an otherwise uninterpretable string of numbers and spaces or the equally mechanical conversion of a string of numbers and spaces into a sequence of dots and dashes. These clerks were employed to interpret all texts mechanically and lacked the context to interpret encrypted texts in any other way. The receipt of an encrypted text by a clerk conveyed a very large amount of information, because the clerk had few expectations about what the text might be, but the encrypted text had no intelligible meaning for the clerk who lacked the key to decrypt the message.
The staff of the German embassy in Washington obtained less information from receipt of a typescript in code 7500 than did the clerks through whose hands the message had passed, because the staff had much stronger expectations about the message. For example, the embassy staff expected the message to consist of numbers and spaces only, whereas the absence of letters was not a prior expectation of the telegraph clerks. But the message had much richer meaning for the embassy staff because they possessed code book 7500 and were therefore able to interpret the message as words in German to be re-encrypted in code 13040 using a different key. Because the staff were fluent in German, they undoubtedly formed a mental interpretation (a memory) of the telegram’s content that could be called upon in the subsequent investigation by German intelligence of how the message had been intercepted.
Now consider the communication of the message to Mexican president Venustiano Carranza. The German ambassador to Mexico, Heinrich von Eckardt, possessed a typescript (perhaps a manuscript) in German from which he made a verbal offer to Carranza in Spanish. Eckardt’s interpretation of ink marks on paper as words in German involved complex interpretation into a neural text that underwent an equally complex process of translation into Spanish in the speech-production centers of his brain. The neural text from the speech production centers was then conveyed to Eckardt’s vocal apparatus, where it was composed as a phonic text (sound vibrations) that was received and interpreted by Carranza’s ears and reinterpreted at various levels in Carranza’s brain. At every step in the process, the constantly changing meaning was nothing more than the physical text produced by an interpreter, whether this consisted of ink marks on paper, neural representations in brains (whatever thoughts may be), or vibrations in the air.
Information resides not in things but in the reduction of uncertainty (entropy) of an interpreter who observes a thing. That which is known does not inform. If we suppose that American intelligence kept copies of German diplomatic telegrams sent from the Western Union office in Washington, then the American government learned nothing new from the Mexican telegram provided to them by the British government beyond that the British had a copy. What provided new information was the decryption of the telegram by British intelligence into written texts in German and English. American uncertainty about whether these texts contained British disinformation was dispelled when Zimmermann conceded that he had sent the telegram and that the decryption was accurate.
The coded texts of the Zimmermann telegram were intended to be uninterpretable (meaningless) for unintended readers. This was the only reason Germany would have sent such a message via the intermediary of the US State Department. The various code books were the private keys intended to be used by intended readers to interpret the strings of numbers and spaces of a typescript as an intended text in German. British intelligence sought to reconstruct this intended interpretation without possession of the private key. Unintended information in the arrangement of code groups (“similar” words had “similar” codes) provided clues that allowed British intelligence to construct their own key and interpret the message. The task of British intelligence was made easier by an unintended difference: Germany used numeric codes, the United States used alphabetic codes. German interpolations in the American cyphertext stood out like a sore thumb.
My account of the convoluted transmission of the Zimmermann telegram is reconstructed from Friedman and Mendelsohn (1994), von zur Gathen (2006), Freeman (2006), and Boghardt (2012). My interpretation of what happened is made difficult by intentional disinformation and unintentional misinformation in the historical record. And I may have misinterpreted the texts I have read and so may have unintentionally misinformed you about the details. My words will nevertheless have served their purpose if your interpretation of what I have written approximates how I intend you to interpret my text.
Mutual Information and Meaning
The concept of “semantic information” views meaning as present in information prior to its interpretation. On this view, interpreters repackage preexisting meaning in new media. I propose instead that meaning be considered the output of the interpretive process of which information is the input. On this view, answers to questions of meaning should be sought in the study of mechanisms of interpretation and the origins of interpretive competence. Once these difficult questions are answered, there will be no remaining ghost in the semantics. A preference between “semantic information” and “meaning as interpretation” is a choice of how “meaning” is defined, not a judgment of fact. “Meaning as interpretation” sits comfortably with the belief that information is not an objective property of things in the world but represents the epistemic uncertainty of an observer.
Consider two sequential transformations of Arthur Zimmermann’s message to Heinrich von Eckardt (figure 12.2). In the first, Zimmermann’s plaintext was translated into cyphertext by one of his secretaries using the textual prosthesis of codebook 7500. In the second, the cyphertext was re-encoded as an electrical signal by a telegraph clerk using the mechanical prosthesis of a transmitter. If these transformations were performed as intended, then the cyphertext could be reconstructed from the signal by a telegraph receiver and the plaintext could be reconstructed from the cyphertext by a reader in possession of codebook 7500.
Faithful transmission of the “Zimmermann telegram” depended on mutual information (statistical dependencies) among multiple coding systems that allowed messages to be encoded and decoded by senders and receivers with the appropriate codebooks. Proponents of “semantic information” often equate meaning with mutual information or consider meaning to be derived from mutual information. In their view, the various texts of the “Zimmermann telegram” are vehicles of a common meaning conserved through the many transformations of transmission. From the perspective of “meaning as interpretation,” mutual information allows an observation of one thing to be usefully interpreted as about something else. Meaning resides not in the immediate observation but in the interpretive synthesis of observation with background information (context), in order to act effectively in the world.
Figure 12.2
The interpretation of the plaintext of the Zimmermann telegram as a telegraphic signal.
The world contains many unrecognized associations between values of one thing and values of another. These associations are “recognized” once an interpreter embodies a mechanism that couples observation of one to useful action on the other. Such mechanisms can be conceptualized as a codebook (text) that enables meaning to be “read” from observation. Codebooks codify mutual information in useable form. The intended extension of “codebook” in this paragraph includes all forms of embodied knowledge revealed in effective action. (Consider the physical implementation of the “genetic code,” which is represented nowhere in tabular form except in recent texts of biologists.) Under this general definition, “codebook” and “interpreter” become synonyms. There are two broad ways of obtaining a codebook. The simplest is to be given a copy (or to steal one). By this means, copies of codebook 7500 facilitated secret communication between the Foreign Office in Berlin and the German Embassy in Washington. The second is to construct a codebook by statistical inference from observations in context. This was the more difficult task achieved by British intelligence and by a child as she learns a language.
A definition of meaning as interpretation simplifies many semantic problems. These problems include polysemy, how the same observation can have different meanings for different interpreters; synthesis, how information from multiple sources can be combined to generate new meanings; and the status of perverse but sincere interpretations. Consider divination by tea leaves. A fortune-teller observes the configuration of leaves in the bottom of a teacup and then uses the observed pattern to answer a question posed by a client. The pattern informs the fortune-teller’s advice if a different pattern would have resulted in different advice. Do all possible interpretations exist as “semantic information” in the configuration of the tea leaves, or does the fortune-teller mistakenly believe she has identified “semantic information” that is absent? The redefinition of meaning as output avoids the horns of this dilemma and the need to distinguish “true” from “false” meanings. If different fortune-tellers (or the same fortune-teller at different times) interpret similar patterns as similar fortunes, then mutual information exists between tea leaves and fortunes; but these are meanings of the fortune-tellers, not of the tea leaves.
Information Theory and Meaning
The fundamental problem of communication is that of reproducing at one point either exactly or approximately a message selected at another point. Frequently the messages have meaning; that is they refer to or are correlated according to some system with certain physical or conceptual entities. These semantic aspects of communication are irrelevant to the engineering problem.
—Claude Shannon (1948)
Claude Shannon (1948) and Warren Weaver (1949) developed information theory as part of a broader theory of communication. Weaver conceptualized communication in “a very broad sense to include all of the procedures by which one mind can affect another” (1949, 11). He recognized technical, semantic, and influential problems of communication: technical problems were “concerned with the accuracy of transference of information from sender to receiver”; semantic problems were “concerned with the interpretation of meaning by the receiver, as compared with the intended meaning of the sender”; influential problems were “concerned with the success with which the meaning conveyed to the receiver leads to the desired conduct on his part.” Thus, Weaver used “information” in the context of technical problems but “meaning” in the context of semantic and influential problems. Shannon’s “fundamental problem of communication” was the technical problem of reproducing a message accurately at another place. He adopted the design stance to address this technical problem, and his mathematical theory ignored problems of intentionality. Weaver, by contrast, adopted the intentional stance in talking about the semantic and influential problems (“intended meaning,” “desired conduct”), but these problems eluded mathematical treatment (see Dennett 1987 for discussion of the design and intentional stances).
In Shannon’s theory “the actual message [was] one selected from a set of possible messages” (1948, 31). He used logarithmic entropy—the number of independent ways in which the message could have been different—as a measure of information content. “If the number of messages in the set is finite then this number or any monotonic function of this number can be regarded as a measure of the information produced when one message is chosen from the set, all choices being equally likely” (379). For Weaver, this meant that “the word information relates not so much to what you do say, as to what you could say. That is, information is a measure of your freedom of choice when you select a message” (1949, 12). Information measured in this way had “nothing to do with meaning.” For Shannon, a message possessed meaning if it referred to, or was correlated with, physical or conceptual entities “according to some system” (379). This conceptualization of meaning had two components: the first was the correlation between messages and things in the world (mutual information), and the second was the “system” that embodied the correlation.
Figure 12.3 is a variant of a famous figure from Shannon (1948) in which a message is selected by an information source and passed to a transmitter that translates the message into a signal that is transmitted to a receiver that back-translates the received signal as a message that is passed on to the destination. The signal sent and signal received differ because of inputs of unintended information (noise). The central focus of the figure is the channel between transmitter and receiver. The technical challenge was to ensure that the message passed to the destination by the receiver was as close as possible to the message passed to the transmitter by the information source. In Weaver’s example, “When I talk to you, my brain is the information source, yours the destination; my vocal system is the transmitter, and your ear with the eighth nerve is the receiver” (1949, 12). Weaver and Shannon had relatively little to say about how information sources selected a particular message or about how destinations interpreted the message after its receipt (nor did they have much to say about the internal workings of transmitters and receivers). In terms of my interpretation of meaning, information source, transmitter, receiver, and destination are all interpreters.
Figure 12.3
Shannon’s (1948) diagram in which a message selected by an information source is transmitted to a destination.
Weaver and Shannon recognized the interpretation of information to be an important question, but their focus was on the transfer of information between interpreters. This chapter addresses how information is used rather than how it is transmitted and is concerned with the general problem of interpretation (use of information) of which communication—the production and interpretation of texts (things intended to be interpreted)—is a special case. The domain of interpretation includes the use of unintended information from the environment in addition to the interpretation of texts.
Difference Making and Mechanisms
Preference is an attribute of relations (differences and samenesses) rather than things (objects and events). To be told that someone chose x tells us nothing about his preferences unless we are also told what he rejected. When we choose one thing rather than another, we do not unconditionally endorse the thing chosen but express a preference for that thing relative to the other. We might be “making the best of a bad bargain” or “choosing the lesser of two evils.”
A dichotomous choice between x1 and x2 can be usefully expressed as 
where 
= (x1 + x2)/2 is a sameness and ∆ = (x1–x2)/2 is a difference. Being told the relations of things, 
conveys the same information as being told their identities, x1 and x2, but the relational form clarifies what is at stake. We choose from what is different (±∆) in the context of what is the same (
). At the moment of choice 
is the point to which we have come and ±∆ a divergence of possible futures: a future to live and a future to die. Once a choice has been made, the future-chosen is past, the latest turn in the history of paths taken, and the future-rejected is what could have been. How come?—the path taken—preserves a trace of Why not?—the reasons for the rejection of paths not taken (if reasons there were).
Ponder again the flame that causes or does not cause an explosion. On one hand, the striking of the match and the presence of oxygen do not make a difference. It is the presence versus absence of hydrogen that makes the difference. On the other hand, the striking of the match and the presence of oxygen are essential parts of the mechanism that causes the explosion. When a scientist performs a controlled experiment, with and without hydrogen, she converts a possible difference into two actual series of events. (If she varied the presence of oxygen or striking of the match, then these become experimental variables and potential difference makers.) Observations are of actual things, not differences among possible things. Actual things, not differences, participate in mechanisms. But we study mechanisms to make a difference.
Causation as difference making is what could have been different. It is a history of paths not taken. Causation as mechanism is what could not have been otherwise. It is one damn thing after another, a single path. The relation between these two concepts of cause is much debated (Hall 2004; Strevens 2013; Waters 2007). Different actions could have different outcomes, but particular actions cause particular outcomes. We readily interpret the world as identities or relations and slip facilely between these perspectives.
An interpreter is an evolved or designed mechanism that couples possible inputs (entropy of observation) to possible outputs (entropy of action). These degrees of freedom are capabilities of the mechanism (what it could observe and what it could do). The interpreter is uncertain and undecided until an actual input is interpreted as actual action. Information is what could have been otherwise before observation. Meaning is what would have been otherwise had the observation been different. Interpretation couples information (difference maker) to meaning (difference made). For an interpreter that has evolved to intervene in its own fate, the only useful information is about differences that could make a difference.
Causation as difference making projects our epistemic uncertainty onto our mechanisms of choice. In a singular universe in which token events happen once, why should “could have been otherwise” be less respectable than “could not have been otherwise”? What difference does it make? How can one decide? Fisher (1934) flipped a bit in my mind, and I now choose “could have been otherwise,” but I understand neither the difference nor what is at stake. The bit could flip back. From the perspective of “could have been otherwise,” our choices change the world. We have evolved to choose because past choices have made a difference and future choices will make a difference.
The Parable of the Bathtub
If you make yourself really small, you can externalize virtually everything.
—Daniel Dennett (1984)
The study of mechanisms is frequently connected with reductionism, the ideology that larger things are properly explained by the properties and interactions of their smaller parts. It is incontrovertible that events at a small scale can have large effects. A mutation in a factor VIII gene of a single cell resulted in at least ten of Queen Victoria’s male descendants suffering from hemophilia, including heirs to the thrones of Russia and Spain (Ingram 1976). All genetic differences that distinguish armadillos from aardvarks and zebras from zebus originated, in this way, as unintended mutations in single cells.
It should also be incontrovertible that large things affect small things. My paternal grandfather was gassed in the Great War. As he sat upright for inspection in his hospital bed, he probably raised his right hand to his head. (He performed this action when dying sixty years later in a different hospital bed.) The origin of the hand salute goes back centuries, perhaps millennia, lost in the fog of time. Among many theories, the salute is the ritualized removal of a hat in the presence of a superior, the raising of the visor by a knight in armor, or a demonstration that the weapon hand is empty, but this is conjecture. John Stewart Haig’s salute would have been an almost automatic gesture in the presence of an Australian or British officer but would not have been elicited by other outwardly similar persons; definitely not by a German officer. From a mechanistic perspective, his salute was caused by release of acetylcholine at neuromuscular junctions triggering actin filaments to slide past myosin filaments in muscle fibrils of his arm. But what molecular mechanism caused his hand to be raised in the presence of a person of an abstract kind? How does a military tradition move an ion across a membrane?
The detritus from collisions between protons moving in opposite directions at 0.99999999c with a combined energy of 13 TeV has been interpreted as evidence for the existence of the Higgs boson. Such events are created and detected at the Large Hadron Collider, housed in a circular tunnel 27 kilometers in circumference, spanning the Swiss–French border. The even more powerful Superconducting Supercollider was canceled by the United States Congress in 1983 because of its immense cost. The differences that made the difference between the Higgs boson being detected in Europe rather than North America are to be sought in alternative political arrangements. No interpretable subatomic mechanism explains why two protons collide at these enormous energies in France but not in Texas. Events at the level of interpersonal relations and trans-Atlantic rivalries impinge upon, and predict, the movements of fundamental particles. Choices are the means by which big complex things control small simple things to make big differences.
An event’s causal influence can wax or wane. A “butterfly effect,” whereby the flapping of wings in Marilia causes a downpour in Sydney, expresses the intuition that small differences can have large effects. The existence of dynamic attractors has the opposite implication. Attractors cancel differences. Consider a bathtub. Whether water enters the tub from a spilled cup, the shower head, the cold faucet, or the hot faucet, it goes down the drain. The tub is literally a basin of attraction. All the degrees of freedom of water molecules to move “as they will” come to naught because the predispositions of the tub are imposed upon them willy-nilly “against their will.” Each water molecule has a unique narrative of how it came to the drain, but these disparate histories have no future consequences. These are differences that do not make a difference. It is the form of the tub that cancels past differences, not the matter from which it is made. The bathtub enforces its will.
The vortex that forms as water exits a tub is a recurrent attractor. The living world abounds in recurrent goal-directed attractors because “endless forms most beautiful and most wonderful have been, and are being, evolved” (Darwin 1859, 490). Convergent structures are attractors in adaptive space over evolutionary time. Adult forms are attractors in morphological space over developmental time. Innate and learned actions are attractors in performance space over behavioral time. Cultural conventions are attractors in social space and the means by which a group’s members converge on common meanings. Minor variations of font or pronunciation make little difference in how words are understood. Word tokens are rocks of stability that facilitate freedom of expression. A salute is a strange attractor indeed.
Organisms embody elaborate hierarchies of homeostatic attractors that ensure basic functions are unperturbed by causal fluctuations at multiple spatial and temporal scales. Bodily attractors at all levels—from the molecular, to the cellular, to the individual—buffer organismal fates from the unforeseen. This cancelation of irrelevant differences allows “focus” on what is relevant. From the myriad potential causes in its world, an organism selects those causes that are applicable for its needs to intervene adaptively at decision points. The organism is an “unmoved mover” moved by self-selected information in pursuit of intended ends. It determines which differences will make a difference. The regress of responsibility stops here. Organisms pull their own strings.
The Meanings of Life
Life is eternal recurrence. That which hath been shall be and that which was done shall be done again. But when a sperm fuses with an egg, the nascent interpreter is something new under the sun. Our genes convey information from the deep past to be used with information from current events to inform our choices in a rapidly changing world. Why a zygote receives one set of genes rather than another can be considered chance, a concatenation of random events that picks one from the numberless possibilities that could have been sampled from its mother’s and father’s genomes. The cards are ancient, but each hand is new. The fall of the cards cannot be anticipated, but we attempt to do the best with the cards we are dealt. We play our hands and the cards are reshuffled.
The meaning of a life is the life that is lived. Your body is an interpretation of your life. Cartesian dualism divided the indivisible body into tools to be used once a choice had been made (res extensa) and parts that were texts used in choosing (res cogitans). When an organism acts, it is the organism’s evolutionary and developmental history that determines to which information it responds and how it responds. The organism exerts its will as the lead actor in its own drama, as a canceler of irrelevant causes and of competing narratives. Yet, despite this autonomy of action, many actors succumb to the slings and arrows of outrageous fortune, to factors beyond their control, foremost among which are the roles played by other actors.
Self-reflective organisms (selves) respond to their world with internal changes that rewire connections between inputs and outputs to make more effective future choices, learn from experience which inputs to attend and which to ignore, perfect performance by practice with feedback from past action, and possess rich memories to inform future choice. Highly sophisticated selves augment their behaviors by observation of what works for other actors, learn from the instruction of parents and other tutors, and choose principles by which to live in pursuit of self-chosen goals. These internal changes comprise an embodied memory of the self’s life experience (the meaning of its life). This intricate and intimate private text, an interpretation intended to be self-interpreted, seamlessly melds ancient wisdom of genetic and cultural texts with news from the senses. It is responsive and responsible. It is the material and mortal soul that dies with the body.
In the beginning was mechanism. Things just happened. The origin of meaning was the origin of intentional difference making. Choice became free as degrees of freedom of observation and of intended action broadened and deepened. To understand a free choice, one must understand an interpreter’s soul.
In fine est principium.
1 Pearl’s (2000) “causal models” can be considered a class of interpreters in which the do-operator performs the role of an observation by fixing the value of an input variable. Tononi’s (2004) “integration of information” refers to internal causal processes of interpreters.