Incomplete Nature Part 14

In everyday experience, we often use the redundancy of interpretive consequences as a means for detecting representational error whenever this is available. In practical terms, this is the widely employed method of fact checking. Comparing multiple independent reports of the same event can help to reduce interpretive error. For example, multiple witnesses to a crime who may have only observed some of the relevant events, and who may have poor memories of the details, or who may be withholding or falsifying evidence, can provide accounts that can be compared and cross-checked to reconstruct the most probable course of events. Those accounts that have concordant reference are taken to provide the most likely and most accurate representations of what occurred.

This is also the essence of the method employed in the empirical sciences. When an independent researcher replicates the results of another researcher's experiments, it reinforces confidence in the prior claim. To do this, the second researcher provisionally a.s.sumes the accuracy of the prior claim and operates accordingly. There is an interesting asymmetry to this process. Failure to replicate prior results can lead to serious theoretical revisions, while discovering consistency between results is only a minimal guarantee of the correctness of a theory. Using many independent methods, a.n.a.logous to obtaining reports from multiple independent witnesses, and finding consistent results between them all, is even more convincing. This is why developing new tools for investigating the same empirical phenomenon in different ways provides for a considerable increase in the representational confidence that is generated.

The logic of fact checking differs in an important respect from introducing signal redundancy, to distinguish signal from noise in the Shannonian sense. This is because it actually involves a means for increasing the potential entropy of the signal, not decreasing it, as happens when signal redundancy is increased. To detect representational error in this way, it is necessary to compare different and to some extent independent sources of representational information, and to instead take advantage of their otherwise uncorrelated diversity to overcome error. Each source of information will have its own idiosyncrasies to contribute, a.n.a.logous to noise, but all will share in common being generated with respect to, and under the influence of, the same extrinsic events. So correcting representational error entails both an increase in the entropy of the signal-e.g., by multiplying interpretive processes-and taking advantage of redundancies in the constraints imposed on these processes by the common object of interest which is the extrinsic factor they all share in common.

This is a higher-order variant on the same theme, because it effectively treats each interpretive process as though it is a replica of the signal to be compared to the original, and relies on the probability of independent sources of variation to highlight redundancies of reference. Alternative interpretations that share prominent features in common despite being independently generated are likely to have been influenced by the same extrinsic cause. Unfortunately, a.s.suming a common source of redundant constraints in independent interpretations is never an infallible inference, because it involves an a.s.sessment of similarity and difference, which is itself an interpretive process. There can be many reasons for not detecting difference, particularly when the Shannon entropy-reflected in the number, complexity, and diversity of interpretive sources-is not large.

To explore this more carefully, consider the example of the detective who compares many sources of information and uses their correlations to infer a common event, which they may or may not each indicate. Over time, as more interpretive techniques have become available for this purpose-such as DNA sequencing, materials a.n.a.lysis, and trace-elements detection-both the effective signal entropy and the sources of interpretive redundancy available to law enforcement have increased, with an attendant increase in interpretive confidence. The detective's problem, or that of a jury listening to a welter of potentially untrustworthy evidence, is to reduce the uncertainty of interpretation-to get at the "truth." They must generate an interpretive response to the whole ensemble of sources of evidence and counterevidence that best corresponds with what actually occurred beyond direct observation. The consistency (redundancy) and inconsistency (non-redundancy) of the evidence is not itself a guarantee that a given interpretation is accurate. Faced with the problem of comparing alternative interpretations of the same events, one is often forced to a.n.a.lyze other features of the source of the information to determine if there are systematic biases that might be introducing spurious or intentionally skewed levels of redundancy. Creating the false appearance of independent sources of information is, for example, a major tool employed in propaganda and confidence schemes.

Error checking is most effective when the interpretive challenge involves objects or states of affairs that remain available for subsequent exploration. In these cases, the redundancy a.n.a.lysis can be proactive and fairly straightforward. As we have already noted, testing a particular scientific hypothesis involves behaving as though a given trial interpretation (the initial report) is accurate, and observing the consequences of continuing to act in accordance with that interpretation (attempting to replicate it) to see if the consequences remain consistent with it. By acting in accord with a given interpretation, causal consequences of this process can be generated to act as virtual new interpretations, each of which can be compared. This approach is also relevant for criminal investigation. So, for example, on suspicion that a given business is corrupt, a law enforcement agency might set up a sting operation that will proceed as though their suspicion is true, and observe the consequences. If the hypothesis is true, the actions of the perpetrators will parallel the prediction.

The concept of producing actions to test for interpretive error was also hinted at in Bateson's aphorism about information, and again involves the performance of work: acting to change circ.u.mstances to produce predictable results. All of these approaches to the problem of representational error checking reinforce the claim that interpretation is a dynamical process, which inevitably involves the generation of new information in the form of new signals and new interactions that do work with respect to those that were generated previously. Although this s.h.i.+fts our a.n.a.lysis upwards to a higher order of the information generation process, the same core logic that we have seen at work in Shannon's cla.s.sic a.n.a.lysis still applies: the information conveyed is determined with respect to the alternatives eliminated-whether about reliability of the signal, reliability of reference, or reliability of interpretation.

DARWINIAN INFORMATION.

In many respects, this process of error detection is crudely a.n.a.logous to the logic of natural selection, with an hypothesis as the a.n.a.logue of a variant phenotype, and the selective exclusion of certain of these based on their non-concordance with others as the a.n.a.logue of selection. Indeed, many theorists have compared scientific research and other truth-seeking enterprises to Darwinian processes,3 and a number of contemporary philosophers have developed theories of function and mental content based roughly on the logic of natural selection.4 But a number of problems with these approaches have been uncovered, mostly having to do with information only being defined with respect to past conditions, not current conditions. There are also questions about whether these processes can account for and detect error.5 To sort out these problems of the interpretive generation of information, then, it may be helpful to consider a.n.a.logous problems posed by evolutionary theory.

In the standard Darwinian account of evolution by natural selection, many individual organisms with variant forms const.i.tute a pool of options from which only a subset is able to successfully reproduce, to pa.s.s on its characteristics to the next generation. This subset succeeds because of its comparatively better fittedness to prevailing environmental conditions; and as a result of genetic inheritance, the new pool of variant individuals that is produced inherits features from the parent generation which functioned best in that environment.

By a.n.a.logy to Shannon's model of the transmission of information, the initial variety of genotype and phenotype forms in the prior generation provides a measure of the potential entropy of the lineage, and the extent of the reduction in "transmitted" forms due to differential reproduction and elimination provides a measure of the information generated by that process. So, in theory, one should be able to quantify this entropy reduction for a given population of organisms for a given number of generations, and estimate the amount of Shannon information produced per time in the evolution of that lineage. It is this parallelism that warrants talking about evolution in informational terms, and ultimately for describing evolution as a process that produces new information. But what is it information about? (Figure 13.1) In the general case, reduction in signal entropy is evidence of work imposed on the signal medium from some extrinsic source, and the basis of what that signal can be about. Can we a.n.a.logously conclude that "genetic" and/or phenotypic entropy is about the environment of a given lineage? In biological evolution, however, the outside source of influence may not be the environment within which this process takes place. This may be a pa.s.sive context, a boundary condition such as available sunlight, ambient temperature, humidity, and so on.6 So the work performed to "make a difference" in the generation of this Darwinian information cannot be a.s.sumed to be extrinsic to the organism. It is instead mostly intrinsic to the source of the "signal"-the work required of organisms in order to persist and reproduce. Constant work is required to maintain the far-from-equilibrium conditions that characterize life. This makes organisms constantly dependent on successfully extracting resources from their environment. But such dependency also makes them selectively sensitive to the availability of these extrinsic resources. So the information that organisms transmit to future generations will reflect the relative fit between the specific forms of work they tend to engage in, the inherited constraints that make this possible, and the features in the environment that are critical to this process. In this way, the constraints implicit in this organism-environment relations.h.i.+p can become represented in the selective preservation of some living dynamics and not others.

FIGURE 13.1: A cartoon depiction of the process of natural selection, showing its parallel with the logic of information generation. However, in this case the work responsible for reducing the variety is generated not by an outside influence but by the individual organisms within the population. Each organism's teleodynamic work is ultimately responsible for utilizing the constraints of the organism-environment dependency. The variety of organism traits in succeeding generations is thus reduced in ways better fitted to the environment.

This inversion of the locus of work and source of constraint in the Shannon-Boltzmann information relations.h.i.+p is also a characteristic shared by many scientific instruments that serve as detectors. By incessantly generating a far-from-equilibrium process, a device's intrinsic instabilities do work that can be used to exemplify their highly sensitive reactivity to certain contextual factors. A process that must continuously do work to maintain an unstable state requires specific conditions in its environment; its state can therefore be used as an indication of the presence, absence, or change of these conditions. A change in these conditions can thus generate a large difference in signal constraint because the far-from-equilibrium state of the device affords a significant potential to do work. So it can amplify even a tiny difference in some critical parameter into a large difference in the dynamics of the signal medium.

Though highly specific sensitivity reduces the diversity of what can thereby be conveyed by a change in this dynamic signal, it can instead provide high precision of reference. The canary in the mine can tell the miner that although he is not yet gasping for air, that possibility is incrementally close. Similarly, a metal detector is highly sensitive to the presence or absence of an object capable of being attracted to a magnet, but little else, because only conductive metals can disrupt the detector's constantly generated magnetic field. Despite the low entropy of the signal, this specificity is what the treasure hunter or minesweeper wants. While this is intuitively obvious, it often gets ignored in technical discussions that do not distinguish these two levels of information. It again demonstrates the difference between a.s.sessments of information and entropy at these different levels of a.n.a.lysis.

Because an organism is the locus of the work that is responsible for generating the constraints that const.i.tute information about its world, what this information can be about is highly limited, specific, and self-centered. Like the treasure hunter with his metal detector, an organism can only obtain information about its environment that its internally generated dynamic processes are sensitive to-von Uexkull's Umwelt, the constellation of self-centered species-relevant features of the world. For the most part, these are statistical features and general tendencies, like the average density of food resources, the probability of predation, or the cyclicity of the seasons, and specific molecular or microdynamic features of the immediate physical context. They are typically either directly or indirectly relevant boundary conditions, which support the far-from-equilibrium processes const.i.tuting the various components of the organism's metabolism with a variety and indirectness of effect that correlates with organism complexity. So, although the information embodied in organism adaptations can be understood to be about its environment, this aboutness is always and only with respect to the constraints of organism dynamics, and not with respect to just any arbitrary properties of things.

FROM NOISE TO SIGNAL.

Before we can turn our attention to the implications of this a.n.a.lysis of information and work for understanding the evolution of life and mind, we need to consider one further feature exemplified by evolutionary processes: their ability to create new forms of information. This is due to a fundamental difference between the abstract logic of communication theory and the evolutionary process: the s.h.i.+fting status of signal versus noise in evolution. If we liken the transmission of traits from generation to generation via reproduction to signal transmission over a communication channel, then mutation in biology is the a.n.a.logue of noise introduced into a communication channel. In most communication processes, noise is a nuisance. It degrades the information by introducing new, uncorrelated entropy into the signal, and this increases uncertainty about what is signal and what is not, thereby potentially corrupting the message. But whereas the introduction of noise decreases the potential Shannon information capacity of a channel, it paradoxically increases the capacity for reference, because it increases total Shannon entropy. It is as though an additional information channel is available, because noise is also a consequence of the openness of the physical system that is being used as a sign medium, and so it too reflects some source of signal modification besides that which the sender provides. Of course, noise is just noise if you are only interested in what was originally sent, and not in the cause of the degradation of that signal. And yet, this normative decision depends on the interpretation process. Noise can be signal to a repairman.

From the perspective of Shannon information, noise is a source of equivocation or ambiguity in the signal. A noisy signal, like a text containing typos, contains signals replaced by uncorrelated alternatives. Shannon's a.n.a.lysis showed that it is possible to compensate for equivocation between signal and non-signal if the transmission and interpretation processes can take advantage of signal constraints or redundancies. In the evolution of adaptive phenotypes, however, there is no such shared expectation to go on. Understanding how this is accomplished when there is no context of introduced redundancy to rely upon is the critical clue to explaining how evolution ultimately transforms noise into signal.

What if there is no information available in the signal to help discern transmitted from randomly subst.i.tuted bits? Consider the case of a set of instructions in which there are word subst.i.tution errors, but in which there is no violation of meaning, spelling, or grammar to indicate that it is inappropriate. This sometimes happens with foreign-made devices which come with a.s.sembly or use instructions that have been poorly translated from an unfamiliar language. In these circ.u.mstances, we often provisionally a.s.sume that the instructions are accurate and attempt to accomplish the task described. If in the process we find that something doesn't work out as described, we may suspect error in the instructions, and careful attention to the task described can often provide clues to the locus of this error.

This trial-and-error approach is also a form of hypothesis testing, as described above. The redundancy being relied upon is between the referential information in the communication and the constraints of the application context. If the information accurately represents features of some physical system (e.g., the instructions about operating some mechanical device), its interpretation in terms of the actions performed on (or interactions with) that system will correlate well with physical constraints required to achieve a given expected result. Thus the reference of a sign or signal is also susceptible to error correction via redundancy at a higher level than signal organization. The reference of a signal implicitly makes a prediction about certain extrinsic causal possibilities. Physical or logical interactions with these extrinsic conditions will be constrained either to conform or not conform to this prediction, and if not, this will disconfirm the represented state.

So the logic of natural selection is a.n.a.logous in many ways to a trial-and-error process, except that in natural selection there is no extrinsic source of representation to check against. Success or failure to reproduce is all that distinguishes representational accuracy of the information embodied in the genotype and phenotype. But reproduction allows for further iterative testing of these interpretive consequences: the succeeding generations effectively stand in for the outside source of comparison necessary for error checking. If one's genetic inheritance contributes to producing a body with appropriate adaptations, it is because the constraints it embodies are in some degree of correspondence with constraints of the environment. Unfortunately, dead men tell no tales, as the cruel aphorism suggests. So it would seem that there is no recording of the many failures-to-correspond; no independent representation of which were the errors and which were not.

What counts as useful genetic information in biological evolution is determined after the fact with respect to its ability to pa.s.s through the functional error correction mechanism of natural selection. The significance of the inheritance signal is both "tested" and refined by the way the far-from-equilibrium dynamics of organism development, maintenance, and reproduction conform to environmental constraints and opportunities. Evolution is thus a generator of information for the organism and a process that rectifies this information with its reference. Thus, although the evolutionary process is itself non-normative (i.e., is not intrinsically directed toward a goal), it produces organisms which are capable of making normative a.s.sessments of the information they receive. In this sense, evolution generates and rectifies referential information. In this way the evolutionary process can progressively increase the functional correspondences between genetics, organism dynamics, and contextual preconditions. To the extent that the constraints of this dynamics conform to environmental constraints that are consistent with its continuation, these intrinsic constraints embody this correspondence in the ongoing dynamics. In this way, natural selection exemplifies the logic of error correction.

Students of evolution have not usually insisted that the absence of the lineages that go extinct is what determines the functionality of the traits that persist. One could see the surviving lineages and their adaptations through the lens of engineering design in terms of identified functions that were designed to achieve a previously specified purpose. But although this a.n.a.logy has a superficial attractiveness, it is undermined by the fact that few if any biological structures can be said to have only one distinguis.h.i.+ng function. Their fittedness, internally and externally, is irreducibly systemic, because adaptations are the remainders of a larger cohort of variants selected with respect to one another and their environmental context. There is no simple mapping of genetic-phenotypic information and adaptive function. That which const.i.tutes the reference of the inherited information is ultimately defined only negatively (i.e., by constraint). Biological function is not, then, positively constructed, but is rather the evolutionary remainder that occupies the constrained s.p.a.ce of functional correlations that have not been eliminated. This is the basis for novel functions to emerge in evolution, as well as the possibility for evolutionary exaptation-the s.h.i.+ft from one adaptive function to another. In this respect, genetic information is neither merely retrospective-about successful adaptation in the past-nor does it antic.i.p.ate future novel adaptations. It is not an aspect of a static relations.h.i.+p, but emerges in process, as its interpretive consequences perform work that may or may not turn out to support this process continuing.

INFORMATION EMERGING.

Although the account so far has been framed in terms of the information involved in biological evolution, this model is generalizable to other domains. The nested dependencies of the three levels of entropy reduction-here characterized by Shannon's, Boltzmann's, and Darwin's variations on this theme of entropy reduction-define a recursive architecture that demonstrates three hierarchically nested notions of information. These three very roughly parallel the cla.s.sic hierarchic distinctions between syntax (Shannon), semantics (add Boltzmann), and pragmatics (add Darwin). They also roughly parallel the relations.h.i.+p between data, content, and significance, though to understand how these semiotic levels are interrelated, we must carry this a.n.a.lysis out of the realm of biology and into the domain of communication (Figure 13.2).

The appeal to Darwinian selection as the ultimate mechanism for the generation of new information relations.h.i.+ps might suggest that we should take a strictly etiological view of information. In other words, we might be tempted to argue that information is only discernable post hoc, after selection, and after entropy reduction. But this is misleading, both for biology and for information relations.h.i.+ps in general. As with biological function, the specific selection history of a given representational capacity may be necessary to explain present usage, but past correspondences are not what it is currently about. Past correspondences have improved the chances for reliable and precise predictive correspondence, but it is a relations.h.i.+p to the present condition that matters. The very fact that information and noise are not intrinsically distinguished, and that mutational noise can become biological information in the course of evolution, exemplifies this property.

FIGURE 13.2: Three nested conceptions of information. Shannon information is the most minimal and the most basic. Referential information is emergent (e) from Shannon information, and significant-or useful-information is emergent from referential information.

This is the problem with simple etiological explanations of adaptive function and representation, which treat information and function as retrospectively determined by their selection history. Because information-generating processes emerge in systems const.i.tuted by a pragmatic selection history, the ground of the correspondence between information and context is determined negatively, so to speak, by virtue of possible correspondences that have been eliminated, but it leaves open the issue of correspondences never presented. No specific correspondence is embodied with full precision and present correspondence is not guaranteed. With functional correspondence underdetermined, novel functions can arise de novo in unprecedented contexts, and incidental properties of the sign or signal may come to serendipitously serve emergent functions. In short, while the possibility of information generation and interpretation depends on a specific physical selection history, the present influence of this information on the persistence of the system that enables it may be serendipitously unrelated to this history. This is the basis for the evolution of new function, but it is also why information is always potentially fallible.

The evolutionary process is not, however, a normative process. Conditions can be good or bad for an organism, or for life in general; an organism's responses to the world can be effective or ineffective in achieving its intrinsic ends, and its adaptational dynamics can accurately or inaccurately link changes in organism activity with changes in extrinsic conditions contributing to the persistence of this dynamical organization; but evolution just occurs. So, although the evolutionary process can further the pragmatic convergence between interpreted content and extrinsic reference, information is not in any sense available to evolution, only to the organisms that are its products. Evolution generates the capacity to interpret something as information. This capacity is intrinsic to a self-perpetuating, far-from-equilibrium system, which depends on its environment and does work to modify that environment in a way that reinforces its persistence. Information is a relational property defined with respect to this persistently unstable dynamical regularity; or, as the philosopher Charles Sanders Peirce would have said, with respect to a "habit"7-understood in its most generic sense; specifically, a self-perpetuating self-rectifying habit.

So genetic information is about cellular chemical-reaction possibilities, their roles in const.i.tuting the organism, and how this relations.h.i.+p between genes and their effects also correlated with extrinsic conditions that supported the maintenance of these possibilities in the past. It is information about organism design and function because it introduces critical constraints into the non-equilibrium processes which may ultimately contribute to the perpetuation of that relations.h.i.+p. It is interpreted by the persistence of the self-perpetuating process that it contributes to. But although it is not information about the present world-only extrinsic signs can fill this role-the ability to make use of environmental features as information nevertheless depends upon this closed interpretation of genetic information for the ability to obtain information from extrinsic sources.

This ability to use extrinsically generated events and objects as information derives from the special dynamics of living processes. Because organisms are const.i.tuted by specially organized, persistent, far-from-equilibrium processes, they are intrinsically incomplete. In this regard, they are processes organized around absence. Not only are biological adaptations evolved and defined with respect to features of the world extrinsic to the organism, but in many respects these are only potential features which may also be absent from the current environment. Thus adaptations of an organism that have to deal with unusual conditions, like high alt.i.tude or extremes of heat or cold, may never be expressed in a lifetime. For this reason, the maintenance of intrinsically unstable, far-from-equilibrium conditions entails mechanisms that effectively antic.i.p.ate the possible variations of environmental conditions by simply not excluding them. They do so with respect to a living process that is at the same time incessantly asymmetrically directed contrary to high-probability states in multiple ways: they do work (a) to maintain their far-from-equilibrium state (which supports persistence of the ability to do work); (b) to generate specific organic forms (i.e., they constrain dynamical processes and generate structures which have highly constrained, low-probability features); and (c) to achieve the specific outcome of maintaining themselves long enough to reproduce the global organization supporting processes a, b, and c. So, with respect to these three improbably asymmetric dynamics, there are many critical extrinsic factors that may be relevant. This combination of absence and necessary relevance to an asymmetric process, incessantly interacting with and modifying the world, is what projects the property of information into otherwise merely physical states and events.

Consider a non-mentalistic example: a deciduous tree which alters its metabolism in response to decreasing day length and cooling temperatures in the early months of autumn, resulting in the eventual withdrawal of metabolic support for its leaves, so that they dry up and eventually become severed from the branches they grew from. This adaptation to the difficulties of winter involves a mechanism which treats these environmental changes as information about likely future events that would have an impact on survival and effective reproduction. Insofar as this response has, in previous generations, resulted in persistence of the lineage compared to others lacking it, the mechanism has acquired interpretive reliability. The reliability of the seasonal changes in these factors provides constrained variation to which the constraints of the tree's metabolic mechanisms have become tuned. But it is not merely these correlations that const.i.tute the informational property of these seasonal changes for the tree. The day length and mean temperatures are also correlated, but one is not intrinsically information about the other nor about the change of season. It is only with respect to the end-directed improbable dynamics of the tree's metabolic processes that one or the other of these is informative-and specifically informative about boundary conditions potentially affecting that dynamics.

At one point, I worked in an office near a number of trees of the same species that had been planted as part of the landscape design for the campus. A few of these trees, which were planted close to an automated streetlamp and next to the exhaust from the building's ventilation system, always were very late to change the color of their leaves and drop them, compared to the others. On the one hand, one might argue that these few trees were misinterpreting these artificial signs, because they don't accurately represent seasonal changes. On the other hand, to the extent that these artificial conditions were nevertheless reliably predictive of local factors affecting the trees' metabolism, one would be justified in arguing that the interpretation was correct, because it promoted the dynamical outcome by virtue of which the mechanism exists. This s.h.i.+fts the focus from the evolved function to the immediate incremental consequence of the evolved mechanism as the ground for referential information. The evolved mechanism constrains the dynamics of possible interpretation, but doesn't determine it. Each moment of interpretation is in some way supportive or disruptive of the self-maintenance of this dynamical trend. This means that not only is there an historical origin for the normative property of this interpretive process, there is also an ahistorical and immediately efficacious normative property as well. And this need not be consistent with its evolved function. In fact, this possibility is a necessary condition for evolution, since essentially every adaptation has evolved from prior forms and mechanisms that often served very different adaptive functions (such as feathers originally evolving as a form of insulation).

Function and representation are made possible by the way living processes are intrinsically organized around absent and extrinsic factors, and the Darwinian process inevitably generates increasingly convoluted forms of dependency on such internal absences. Information is a relational property that emerges from nested layers of constraint: constraints of signal probability (Shannon), constraints of the dynamics of signal generation (Boltzmann), and the constraints required for self-maintaining, far-from-equilibrium, end-directed dynamics (Darwin). Because information is a relations.h.i.+p among levels of constraint generated by intrinsically unstable physical processes, it is also normative with respect to those processes. But constraint is a negative property, and thus neither something intrinsic nor determinate. This means it is intrinsically incomplete and fallible. Yet it is these very properties that make it evolvable and indefinitely refinable.

REPRESENTATION.

We can conclude that a representational relations.h.i.+p cannot be vested in any object or structure or sign vehicle. It is not reducible to any specific physical distinction, nor is it fully const.i.tuted by a correspondence relations.h.i.+p. But neither is it a primitive una.n.a.lyzable property of minds. Instead, even simple functional and representational relations.h.i.+ps emerge from a nested interdependence of generative processes that are distinctive only insofar as they embody specific absences in their dynamics and their relations.h.i.+ps to one another. These absences embody, in the negative, the constraints imposed on the physical substrates of signals, thoughts, and communications which can be transferred from one substrate to another, and which thereby play efficacious roles in the world as inherited constraints on what tends to occur, rather than acting as pushes or pulls forcing events in one direction or another. Constraints don't do work, but they are the scaffolding upon which the capacity to do work depends.

This is only the barest outline of an information theory that is sufficient to account for some of the most basic features of functional and representational relations.h.i.+ps, so it cannot be expected to span the entire gap from biological function to conscious agency. But considering that even very elementary accounts of biological function and representation are currently little more than a.n.a.logies to man-made machines and human communications, even a general schema that offers a constructive rather than a merely descriptive a.n.a.logical approach is an important advance.

In this exploration of the relations.h.i.+p between information theory, thermodynamics, and natural selection, we have unpacked some of the unrecognized complexity hidden within the concept of information. By generalizing the insight captured by Claude Shannon's equation of information with entropy reduction and constraint propagation, and tracing its linkage to a.n.a.logues in thermodynamic and evolutionary domains, we have been able to address some of the most vexing issues of representation, reference, and normativity (i.e., usefulness). By removing these inadequacies in current definitions of information, we may at last overcome the seemingly insurmountable obstacles to formulating a theory of representation that is sufficiently rich to serve as the basis for biology and the cognitive neurosciences, and sufficiently grounded in physics to explain representational fallibility, error checking, information creation, and the relations.h.i.+p between informational and energetic processes.

14.

EVOLUTION.

Natural selection disposes what self-organization proposes.

-STANLEY SALTHE, 2009.

NATURAL ELIMINATION.

The term evolution literally refers to an unrolling process, like the unrolling of a scroll to reveal its contents. Although it can be used in a generic sense to describe a process that develops in a particular direction, as in the "evolution of a chemical reaction" or "stellar evolution" (changes that occur during the lifetime of a star), since the time of Darwin the term has become closely a.s.sociated with the biological process by which living species have come to differentiate and diversify over geological time. In biological discussions, its meaning is a.s.sumed to be synonymous with natural selection and related processes such as genetic drift; and in non-technical discussions, it is generally a.s.sumed to refer to some version of the Darwinian perspective on the origin of species. The Darwinian connotation is helpful in distinguis.h.i.+ng what amounts to a negative or subtractive conception of change, in which certain forms are progressively eliminated or culled, from a positive conception such as we see in design processes where new modifications are added. In the standard model, both the preservation of the "more fit" and the elimination of the "less fit" are understood as the result of surviving to reproduce, or not, respectively.

The processes underlying biological evolution are not, of course, limited to a subtractive effect. Even natural selection in its simplest form requires the production of variations of form and multiplication of offspring from which the most successful are preserved. In most traditional accounts of Darwinism, the source of novel variations is considered to be the primary positive factor. The introduction of novel variation into the process, according to strict neo-Darwinism, however, is presumed to be the result of a form of damage-genetic mutational-which is essentially an expression of the second law of thermodynamics, and thus an order-destroying effect. But, as the epigraph to this chapter hints, there may be a source of increasing orderliness available as well: self-organizing processes. The recognition that there needs to be such a "positive" (order-introducing) factor, and not merely a multiplicative factor, in order to explain biological evolution is becoming more widespread. This requirement is echoed also by Peter Corning, who argues that "a fully adequate theory of evolution must encompa.s.s both self-organization and selection."2 A failure to recognize this need has been the source of persistent theoretical problems for evolutionary theory. So precisely demonstrating how self-organization and selection processes are functionally intertwined may help resolve some of these riddles.

It might be more accurate to say that natural selection theory is explicit about its subtractive aspect but agnostic about its additive aspect. Although natural selection offers a powerful logic that can account for the way organisms have evolved to fit their surroundings, it leaves out nearly all of the mechanistic detail of the processes involved in generating organisms, their parts, and their offspring. Indeed, it is one of the virtues of this theory that it is entirely agnostic about the specific mechanisms responsible for the growth and regeneration of structures and functions, for reproducing individual organisms, for pa.s.sing on genetic information, and for explaining the many possible sources of variation that affect this process. But it is precisely the process of generating physical bodies and maintaining metabolism that const.i.tutes the coin of the natural selection economy. Variations do not exist in the abstract; they are always variations of some organism structure or process or their outcome. Organisms must compete for resources to build their parts and to maintain the far-from-equilibrium dynamics, which is the source of this self-maintenance and reproduction. It is the efficiency and context-appropriateness of these processes which determine the differential reproduction that determines what persists and does not. Or to put this more simply, what natural selection eliminates is determined with respect to the effectiveness of what gets generated in a given context.

This decoupling of function from the processes responsible for its origination means that the specific mechanisms involved don't matter. Only the consequence matters, irrespective of how it was achieved. So natural selection is a cla.s.sic case of the ends justifying the means, by preserving them, retrospectively. This decoupling from specific substrates and mechanisms was the source of Darwin's most revolutionary insight: that adaptation in biology only becomes realized ex post facto. It helped him to recognize that variations of structure and function that arise by accident can nevertheless be functional. More important, this decoupling of consequence from cause allows for the widest possible diversity of mechanisms to be available for recruitment to serve a given adaptive function. For example, the means by which flight, photosynthesis, or thought can be accomplished in a given organism is irrelevant so long as it produces useful results. Whether flight is achieved by the fluff of dandelion seeds caught by a light breeze, the leathery sheets of Pterosaur or bat wings, or the lightness and wind resistance of feathers, what counts is getting airborne. Even though these functions depend upon underlying mechanisms that converge on certain specific mechanical constraints, it is the realization of these constraints and not their specific embodiment that matters.

In this way, natural selection is a process defined by multiple realizability, and is the paradigm exemplar for defining functionalism. This complete openness to substrate and mechanism is what opens the Pandora's box of evolvability, and makes the explosive creativity of life and mind possible. So it is with the emergence of the process of natural selection that true functional generality comes into the world. Adaptations and organisms are in this sense general types that exist irrespective of the specific details of their embodiment.

The vast power of evolvability thus is a consequence of the fact that natural selection is a process that regularly transforms incidental physical properties into functional attributes. An adaptation is the realization of a set of constraints on candidate mechanisms, and so long as those constraints are maintained, other features are arbitrary. But this means that with every adaptation, there are innumerable other arbitrary properties potentially brought into play. Although a given structure or process must be embodied by specific substrates with innumerable properties that are incidental to current adaptive usefulness-for example, the sound of the heartbeat or the redness of blood-any of these incidental properties may at any point themselves become substrates for selection, and thus functional or dysfunctional properties. Precisely because the logic of natural selection is mechanism-neutral, it also is minimally restricted in what kinds of properties can be recruited, except that they must be immediately usable given the constraints of organic chemistry (which is why, say, nuclear energy has never been recruited by natural selection to serve organism functions). The relevance of this for emergence arguments should be obvious. This capacity to transform the incidental and accidental into the significant and indispensable radically minimizes the causal role of any specific intrinsic physical tendency or immediate antecedent condition in determining what will or will not be the physical substrate for an evolved function.

Because an adaptation is mechanism-neutral, it is a bit like an algorithm that can be implemented on diverse machines, and in this sense it has something of the character of a description. It is a general, in philosophical terms. But an adaptation is not identical with the collection of properties const.i.tuting the specific mechanism that embodies it. It is far less than these! Only a very small subset of physical attributes of a given physical implementation of some function, such as oxygen binding, are ever relevant to the success of that adaptation. An adaptive mechanism is also something beyond any of its properties as well. It is the consequence of an extended history of constraints being pa.s.sed forward from generation to generation-constraints that have perhaps been embodied in many different substrates along the way. Like the process of addition that can at various times be embodied in finger movements, the s.h.i.+fting of beads on an abacus, or the changing distributions of electric charge in the memory registers of a computer, the specific physical links in this chain have become incidental. It is only the transmission of conserved constraints that is critical, and the constraints are not the stuff. Of course, these constraints must be embodied in some specific physical substrate at every step, and the transfer from substrate to substrate must always be mediated by a specific physical mechanism that precisely determines the material overlap and the work of transferring these constraints. But it is the conservation of constraint, not of energy, material, or specific mechanism, that matters.

Precisely because constraint is not something positive, but is rather something not realized, only some of the physical details of the mechanisms recruited as adaptations are of functional relevance, even though all are part of the physical processes involved. Only those functionally relevant physical attributes-ones that guarantee the preservation of these constraints-are likely to contribute to future generations' body composition. Other physical attributes are susceptible to eventually being eliminated if the same constraints can be transferred to other substrates, or may just get degraded by variations acc.u.mulated over the generations. It is in this sense that we are justified in arguing that life's mechanisms are general types.

The power of evolutionary theory to explain much of the complexity and diversity of biological forms derives from this physical abstractness. This decoupling of material generative processes from functional consequence is responsible for the persistence of an adaptation across many generations, but also what allows for the transfer of function to progressively more suitable substrates. In a more mundane sense, it is also responsible for one of the most basic attributes of life: the successful bottom-up development of a vastly complex and well-integrated organism, regulated by a comparatively small set of constraints pa.s.sed on in the genes and cytoplasm received from a parent organism.

Because the intrinsic material components and dynamical properties that const.i.tute a given adaptation are not essential, but are determined by constraints on properties, not some finite set of specifically relevant properties, we need to think about the origin of organic mechanisms quite differently than designed mechanisms. In the evolutionary history leading up to a given form of adaptation, those constraints on materials and dynamics that were retained were merely those that were not eliminated. As scientists and engineers, we tend to focus on the properties that we discern to be most relevant to our abstract sense of a given function; but life is only dependent on excluding those that are least helpful.

This is the critical break in causal symmetry that makes the concept of biological function an emergent general physical property that is not determined by its lower-level const.i.tuent properties, even though we may find it useful to try to list and describe them. Those const.i.tuent properties that we biologists may find helpful in understanding and categorizing biological functions were not themselves the targets of "selection"; they represent merely part of the residue of what was not eliminated. Evolution is not imposed design, but progressive constraint.

"LIFE'S SEVERAL POWERS"

So it is essential to the power of natural selection theory that it quite explicitly declines to address how the relevant mechanisms and their variant forms are generated. The Darwinian logic quite correctly treats these issues as separable from the explanation of the determination of adaptation. Evolutionary theorists have therefore been justified in both accepting the plausibility of accidental generation of variant forms (e.g., via "chance" mutation) and rejecting the necessity of actively acquired (e.g., Lamarckian) and functionally biased variations (aka "directed mutations"). These features are neither a problem nor a necessity for the theory because the details of these mechanisms are irrelevant for the functional explanation it provides. In fact, though textbooks often suggest otherwise, Darwin himself was open to many possible sources of form generation and genetic variation, including Lamarckian mechanisms. He just recognized that the logic of natural selection was sufficient irrespective of how these generative, reproductive, and variant consequences were achieved.

Recognizing that natural selection logic is agnostic to these factors doesn't make the problem of mechanism go away, however. As we have stressed throughout much of this book, the problems of how regular dynamical processes and physical forms are generated in the face of the second law of thermodynamics, and how biological information is generated, preserved, and transmitted, are far from trivial components in biological explanation. Natural selection may occur irrespective of the specific details of form generation and reproduction, but it only occurs if these processes are reliably present. Some specific mechanisms are required, and the processes that are capable of producing such regularities are limited and must have very specific dynamical properties in order to thwart the ubiquity of local thermodynamic tendencies. We are not, then, freed from answering questions about how specific means of achieving these processes might influence evolution.

Organisms are highly complex systems and, as many critics of natural selection have been fond of pointing out, the number of possible variant configurations of a complicated multipart system like an organism is truly astronomical. Even in the case of the simplest organisms, an undirected, unconstrained process for sorting through the nearly infinite range of combinatorial variants of forms, and testing for an optimal configuration by the trial and error of natural selection, would fail to adequately sample this s.p.a.ce of possible variations, even given the lifetime of the universe. In contrast to this vast s.p.a.ce of possible combinatorial variants, the process of biological development in any given lineage of organisms is highly constrained, so that only a tiny fraction of the possible variations of mechanism can ever be generated and tested by natural selection. While this limitation may at first appear to impose an added burden on the theory, it turns out to be the most important aid to the solution of this conundrum.

Just as organism "design" by evolution must be understood negatively, development of organism form is also not a construction process, in the sense we might imagine from an engineering point of view. It is the expression of the interaction of many morphodynamic processes. Like the reciprocally end-and-means logic of the morphodynamic processes that generate the "body" of an autogen, a vastly more complex and interwoven fabric of morphodynamic processes is responsible for the generation of an organism body from its initial state following conception. The inherited constraints that we identify as genetic information are specifically constraints that make morphodynamic processes highly probable at the chemical, macromolecular, cellular, and intercellular levels. Evolution in this sense can be thought of as a process of capturing, taming, and integrating diverse morphodynamic processes for the sake of their collective preservation.

Recognizing the morphodynamic origin of organism form is a critical factor in this a.n.a.lysis, because it means that variations appearing at the genetic level only get expressed through a dynamical filter with highly specific tendencies to dampen and amplify constraints. The variant forms that eventually become subject to natural selection have in this sense been vetted by self-organization (as the epigraph to this chapter suggests). Even if the mechanism of the generation of variations is independent of and irrelevant to the consequences of natural selection, this does not mean that variations get expressed irrespective of the systemic integration of the organism. These quite restrictive intrinsic constraints on developmentally expressed variations significantly bias variation to be minimally discordant with respect to existing ontogenetic processes, even if they are generated irrespective of any potential adaptive outcome.

Although these morphogenetic constraints radically restrict the exploration of all possible advantageous variations of form, they nevertheless vastly increase the probability of generating variations with some degree of intraorganism fittedness-that is, concordance with the constraints and synergies that maintain the integrity of organism functions. As a result, natural selection ends up sampling phenotypic options in just a tiny fraction of the possible variation s.p.a.ce; but these variants are much more likely to exhibit features that are not too discordant with or disruptive of already existing functions.

The significance of morphogenetic constraints on the generation of variations is only one aspect of a more global set of conditions that must be considered as antecedently relevant to natural selection theory. The importance of the constraints introduced by morphogenetic processes is that they contribute a positive counterpart to the negative logic of natural selection. Because it only provides an ex post facto culling influence, natural selection is only so generative as its supply of optional forms is profligate. The constrained generation of variant forms is only the most minimal expression of this presupposition of a positive process in evolution. As James Mark Baldwin argued with respect to "organic selection"-his so-called New Factor in Evolution: Natural selection is too often treated as a positive agency. It is not a positive agency; it is entirely negative. It is simply a statement of what occurs when an organism does not have the qualifications necessary to enable it to survive in given conditions of life. . . . So we may say that the means of survival is always an additional question to the negative statement of the operation of natural selection.

The positive qualifications which the organism has arise as congenital variations of a kind which enable the organism to cope with the conditions of life. This is the positive side of Darwinism, as the principle of natural selection is the negative side.3 A full specification of "the positive side of Darwinism" requires significantly more than merely a theory of variations, however. It requires organisms and all of the complex properties we ascribe to them (or at least their surrogates, e.g., in simulations), as well as the a.s.sociated physical processes that enable organisms to preserve their critical features and to reproduce. In general terms, this means unit systems exhibiting such critical properties as self-maintenance in the face of constant thermodynamic breakdown, continual far-from-equilibrium chemistry, mechanisms sufficient to generate complete replicas of themselves, and sufficient organization to maintain these properties in the face of significant variations in the conditions of existence. But if this is a requirement for evolution by natural selection, it cannot itself have initially arisen by natural selection. Natural selection presupposes the existence of a non-h.o.m.ogeneous population of individuated systems with these thermodynamically rare and complex properties.

Darwin thus appropriately ends his "Origin of Species" with the following majestic sentence: There is grandeur in this view of life, with its several powers, having been originally breathed into a few forms or into one; and that, whilst this planet has gone cycling on according to the fixed law of gravity, from so simple a beginning endless forms most beautiful and most wonderful have been, and are being, evolved.

To perhaps state an obvious fact-one that is nevertheless underappreciated and glossed over by the majority of texts on biological evolution-natural selection a.s.sumes the existence of processes of persistent non-equilibrium thermodynamics, self-maintenance, reproduction, and adaptation. It cannot therefore be the complete explanation for their origins, particularly for the origins of their teleodynamic character. Natural selection can only improve the fit between these dynamical processes and the various environmental conditions on which they depend or must defend against; it cannot generate them.

The first teleodynamic systems emerged, they did not evolve. They emerged from specific patterns of interdependency that happened to arise between morphodynamic processes. Previously, we have used the concept of an autogen to exemplify the requirement for this emergent transition. As we have seen, self-organization is a distinct and vastly simpler process than natural selection. Morphodynamic processes form the ground on which teleodynamic and thus evolutionary processes have been built. In simple terms, self-organization is the expression of the intrinsic dynamics of a system that get expressed within certain non-equilibrium boundary conditions, whereas natural selection is a function of the organization of a system's internal non-equilibrium dynamics with respect to external conditions. In this way, natural selection involves a specific mode of intrinsic self-organization dynamics that maintains itself by virtue of the generation of effects which specifically counter certain external changes of conditions away from what is conducive to the persistence of those intrinsic processes.

ABIOGENESIS.

So the first organism wasn't a product of natural evolution. The constellation of processes that we identify with biological evolution ultimately emerged from a kind of proto-evolution, supported by a kind of protolife, that ultimately must trace back to the spontaneous emergence of the first molecular systems capable of some minimal form of evolutionary dynamic. Earlier, it was shown that even a molecular system as simple as an autogen can give rise to a form of natural selection. The emergence of this constellation of properties enabling evolution, even in quite minimal form, marks a fundamental s.h.i.+ft in the dynamical organization of the natural world. In the terms we have been developing here, this is a s.h.i.+ft from thermodynamic and morphodynamic processes to teleodynamic processes. Wherever in the universe this occurs, it is the emergence of the first and simplest lifelike process in that region. All other teleodynamic processes and teleological relations.h.i.+ps to develop in those environments will likely trace their origin to that crucial transition.

For this same reason, the study of the origin of life has a paradoxical status compared to the rest of biology. It violates a crucial and hard-won dogma of biology: the denial of spontaneous generation. From early Greco-Roman times, it was thought that some, if not all, life arose from inanimate matter by way of spontaneous generation. Evidence to support this theory of the genesis of life was seen in the way maggots would emerge spontaneously from rotting beef or molds would form on stale bread or overripe fruits, seemingly without these life forms being intrinsically present in these materials. In 1668, in one of the world's first controlled biological experiments, Francisco Redi challenged the theory by demonstrating that maggots did not emerge from meat when put under gla.s.s. But belief that life emerged spontaneously from non-life nevertheless died hard. Indeed, it is ultimately (but only ultimately) a necessary a.s.sumption of scientific materialism that the essence of life does not arise from a realm outside of the physical substrates of its const.i.tution. In these first centuries of the European Enlightenment it was, in this respect, the scientifically fas.h.i.+onable alternative to an otherworldly spiritual origin theory, and it was one of the central fascinations of alchemical lore. Alchemical "recipes" for generating life (even homunculi, as we saw in chapter 2) were not uncommon, and considered the mark of truly perfected alchemical methods. In many respects, the central preoccupation of alchemy-transformation of matter from one form to another, including from non-life to life-was the prescientific precursor to emergentism.

Support for spontaneous generation theories of life persisted until, more than two centuries after Redi, it was famously repudiated in 1889 by Louis Pasteur.4 Pasteur showed that sterilized rotting meat, placed in a flask with an S-shaped neck, never gave rise to maggots so long as outside contamination was prevented. Indeed, the method of sterilization, coupled with hermetic isolation of the sterilized food (e.g., in a tightly sealed canning jar), did far more than prevent new organisms from forming; it also prevented them-even those microbes still unknown to nineteenth-century science-from aiding the breakdown and putrefication of food. In this way, long-term preservation of food might be considered the most significant spin-off of origins-of-life research.

With the negative results of Pasteur's experiments, it quickly became biological dogma that the emergence of life from non-life was effectively not something to worry about. Indeed, the now ubiquitous methods of sterilizing, canning, freezing, bottling, irradiating, and yes, pasteurizing food, depend on it. The truth of the maxim "Only life begets life" is tested untold billions of times in the modern world with each can or bottle of food that is produced and consumed. But of course to accept this fact, unconditionally, leads to a conceptual paradox. Either life has been around forever in a universe without beginning, or else it originates from some other non-physical realm (e.g., of spirit) in which its a.n.a.logue is somehow preformed, or it emerges from non-living materials, abiogenetically. The first two of these are reductio ad absurdum claims in some form or other. And so we are left with the problem of proving the spontaneous generation theorists right, at least in some limited form.

Charles Darwin clearly understood that this problem was outside of his purview when at the end of the Origin of Species he describes evolution commencing only after life has been "originally breathed into a few forms or into one." But this hardly set the issue aside. In fact, the debate with respect to spontaneous generation was actually quite intense among Darwinians in the latter part of the nineteenth century. Perhaps the most notable battle lines were exemplified by Alfred Russel Wallace-the co-discoverer of the principle of natural selection-and Thomas Henry Huxley-Darwin's "bulldog," as he was sometimes described. Wallace sided with theorists like the famed neurologist Henry Charlton Bastian, who supported experimentation to demonstrate the spontaneous generation of life from non-life, and who wrote three influential books on the subject in the decades following the publication of the Origin of Species. Bastian's interest was in part a function of his medical interest in the basis of microbial disease, but also of his critical stance against supernatural theories of life.

Although Darwin's theory appeared to specifically undermine supernatural sources for the forms and adaptive design of organisms, its silence with respect to the origins of life and the initiation of biological evolution left a gaping ho