3For an algorithmic approach to adaptation see Holland (1975). For a comparison of Holland’s adaptive theories and selection theories, see Darden (1983).

110

LINDLEY DARDEN AND JOSEPH A. CAIN

Our task now is to proceed to develop an abstraction for selection type theories. We will draw on recent discussions in the philosophy of biology literature to develop an abstraction for natural selection. The construction of that abstraction will be constrained by our desire to make it applicable, possibly at a slightly higher level of abstraction, to clonal selection and selective theories of brain function. After stating the abstraction for natural selection, we will instantiate it in the other two analogous cases.

3. The Process of Natural Selection. A selection process may be broken down into a series of steps from which a more abstract characterization can be developed.

Our task now is to discuss these steps in more detail and develop an abstraction with appropriate variables.

A. Preconditions for Natural Selection. Natural selection operates on a population of varying individuals in a given location, such as the classic example of the peppered moths in a region with trees darkened by industrial pollution. This case illustrates the following preconditions for selection: the set of individuals making up a population, the presence of the variation, and the environment. Although variation is often discussed as a precondition for selection (for example, Lewontin 1970; Sober 1984), the other preconditions that we list have received less attention. In addition to varying, the individuals

111

SELECTION TYPE THEORIES

on which selection acts must share some features. If the population is an interbreeding population, then the organisms, of course, share many genes and their associated characteristics. We will assume that a set of individuals is delineated by specifying features that they share, not by enumerating its members. Thus our abstraction has as a first precondition that a set of individuals exists. "Group" in Hull’s (1980, 1981) sense might be a better word that "set"; groups tend to be spatiotemporally localized entities and their members are considered part of the group because of their location (Hull 1980, p. 314). But using "group" would introduce confusion with the different issue of whether "group selection" exists. The term "class" might avoid introducing perhaps extraneous connotations from set theory, but Kitcher (1987, p. 186) argues that it introduces confusions of its own.

Hull stresses that selection "can act only on spatiotemporally localized entities" (Hull 1980, p. 314). Sober makes a related claim when he argues that a "common causal influence" operates in a selective process:

too.) Cain (unpublished) stresses the role that critical environmental factors play in creating conditions for a selective interaction. The components of the environment that affect the nature of the struggle for existence will he referred to as "critical factors". What factors in the environment are critical determines what variant properties are relevant to surviving (or otherwise benefiting from) the selective interaction. Thus, the fit between variant and environment (or in a finer-grained analysis, the fit between the variant property and the critical factor) is the key aspect of the next step, the selective interaction.

B. Selective Interaction. Much of the early discussion in philosophy of biology about the nature of selection focuses on the effects of selection, namely differential survival and reproduction. If no account can be given of selection except in terms of the effects, then the specter of tautology is raised: "survival of the fittest" becomes "survival of those that survive".

Lewontin, in his now classic 1970 article discussing the steps in selection and the properties of evolving entities, states: "different phenotypes have different rates of survival and reproduction in different environments" (Lewontin 1970, p. 1). But, we may ask, why is reproduction differential? Lewontin lists three properties of entities that evolve: variation, reproduction and heritability. A fourth component, one involving interaction during the selection step, must be added to this list in order to explain how variation gives rise to differential reproduction. In an earlier attempt at producing an abstraction for natural selection, Darden characterizes the selection step common to artificial selection and Darwinian natural selection. She writes: selection of a subset of variants occurs by an agent selecting according to a criterion (Darden 1982, p. 16). But "agent" brings inappropriate negative analogy from the artificial to the natural case.

A better characterization of the selection step can be constructed using ideas from Hull (1980) and Sober (1984). Especially important are the ideas of interaction and the causal role of the variant property. Their ideas have to be augmented by more detail about the role of the critical factor in the environment. Hull recognizes the nexus of the selective event:

Sober (1984) stresses the causal nature of the selective interaction even more than Hull did. Sober advances us on our way to producing an abstraction for natural selection by providing somewhat helpful character-

113

SELECTION TYPE THEORIES

izations of group and genic selection:

X is the hierarchical level considered as the level of selection. Y's are the objects that vary. Z is the level that Sober calls the "benchmark". A benchmark is an object that benefits or suffers in virtue of having properties that are selected for (Sober 1984, p. 270). However, the relation of Y to X and Z is unclear. For the genic selection example Y = Z = organisms. But in the group selection case, Y is not equivalent to Z: Y = groups, but Z = organisms. Y's are the variants, but Sober is unclear about whether the variation is to be located at the level of causal influence (Y) or at the benchmark (Z). For simplicity, initially we are going to assume that the variant, the interactor, and that which directly benefits or suffers can be represented by the same variable. Alternative, lower-

114

LINDLEY DARDEN AND JOSEPH A. CAIN

order abstractions can be constructed that differentiate these three variables. Sober’s analysis introduces complexities because he is grappling with the issue of effects at various hierarchical levels. Since Lewontin’s (1970) analysis of selection at different levels, others have also tried to analyze selection and its effects at different levels in structural or functional hierarchies (for example, Hull 1980: Arnold and Fristrup 1982). In the abstraction we are constructing, we wish to avoid most of the complexity the hierarchical perspective on selection introduces. However, we will use one distinction used by Vrba and Gould (which parallels a similar distinction in Sober): sorting and selecting.

Vrba and Gould’s definition of selection makes use of their distinction between sorting and selecting and also shows their localization of the selective interaction at the hierarchical level of the variant property:

The single most important omission from the previous discussions is a neglect of the role that environments play in selective interactions. Although Hull and Vrba and Gould allude to the interactions with the environment, no analysis is given of conditions in an environment that are relevant to the selective interaction. Sober doesn’t explicitly mention the role of the environment in his schematic discussions of group and genic selection (Sober 1984, p. 280). Consideration of his steps shows that he packs the role of the environment into "common causal influence" and into his analysis of how a variant property can be a "positive causal factor" in selection. We analyze the environment in terms of the critical

115

SELECTION TYPE THEORIES

factor that makes a given property causally relevant to the selective interaction. The critical environmental factor and the interaction of the variant property with it are the keys to the subsequent effect of the selective interaction.

In summary, the main points about the selective interaction are the following: first, an explicit recognition that a step in the temporal process occurs between the formation of variants and the effect of differential reproduction; second, that this step involves interaction; third, the interaction can be analyzed either in terms of an individual and its environment or, in a finer-grained way, in terms of the variant property and the critical factor in the environment; fourth, the variant property plays a causal role in the interaction; and, finally, effects can be manifested at levels other than the level of the interaction.

C. Effect. Sober uses the language of "benefit and suffer" to characterize the differential effects of the selective interaction. Although such language sounds somewhat anthropomorphic and will require a value judgment as to what is a benefit, it is appropriately abstract for our purposes for a number of reasons. In natural selection, the interaction may produce an effect that only later results in differential survival and reproduction. Thus, the process should have a step for specifying such an intermediate effect. Also, the benefits in the many different instances of selection events will be of many different types, for example, escaping predators, obtaining food, defending territory. "Benefit" is a neutral term for characterizing numerous types of positive effects.

Most importantly, the neutral language of "benefit" is desirable for constructing an abstraction that can apply at different levels in an organizational hierarchy within an evolutionary theory context. It is even more desirable for our current purpose of constructing an abstraction that can be instantiated in other selection type theories. We would be happy with less anthropomorphic and value-laden sounding term than "benefit", but we haven’t yet found one at an appropriately abstract level.

D. and E. Longer-Range Effects. We have carefully delayed adding differential reproduction to our steps in a selection process until these later steps. As others have argued, defining selection as differential reproduction masks many of the important features of selection. For artificial selection, it may be very important to separate the benefit from differential reproduction. The breeder might select on a criterion of low seed output in grapes; hence, decreased reproductive capacity actually results in more grape plants being grown. Nonetheless, for natural selection, differential reproduction as a result of the differential interaction is the important consequence that makes a difference in evolutionary change.

116

LINDLEY DARDEN AND JOSEPH A. CAIN

Differential replication is an integral part of Hull’s analysis of selection, as is the even longer-range consequence of the production of a lineage. Hull defines a lineage as, "an entity that changes indefinitely through time as a result of replication and interaction" (Hull 1980. p. 327).

We have discussed steps A-E and will turn to a construction of a more abstract characterization of them, attempting to be more precise and use variables in place of many of the words of the looser preliminary list.

4. Abstraction for Natural Selection. In considering what aspects of the various discussions should be preserved in an abstraction for selection theories, the following components are important. Darden's idea of an "array of variants" focuses attention on the fact that one needs to specify a set of things that vary. Implicit in that idea is that the individuals in the set share many properties but vary in one or more. For simplicity, Sober suggests a single variant property P. Hull stresses causal interaction. Sober's primary emphasis is on the "common causal influence" operating in the selective event and the role of the variant property as a "positive causal factor". He provides appropriate neutral language of the effect being some sort of "benefit". Vrba and Gould provide the distinction between selection at a given level in a hierarchy and sorting at other levels. To these discussions we have added the concept of a critical factor in the environment.

Drawing on this previous work, we now will attempt to develop our abstraction. A typical selection process can be analyzed into a group of preconditions, such that if they obtain, then a selective interaction occurs, resulting in a series of effects. A selection process typically occurs if:

A. Preconditions

i. A set of Y's exists and

ii. Y's vary as to whether they have property P and

iii. Y's are in an environment E with critical factor F.

B. Interaction

iv. Y's, in virtue of possessing or not possessing P, interact differently with environment E and

v. critical factor F affects the interaction such that

vi. the possession of P causes Y's with P to benefit and those without P to suffer.

117

SELECTION TYPE THEORIES

(vi’. This causal interaction may have the concomitant effect of sorting Z's.)

D. Longer-range effect

E. Even longer-range effect

viii. D may be followed by longer-range benefits.

To give some substance to these abstract statements, let’s consider the classic case of the peppered moths. (i) A population or set of peppered moths exists. (ii) The moths vary in the property of color, dark or light. (iii) Moths are preyed upon by birds and live in woods near coal-burning plants where trees are darkened by soot. (iv) Moths, in virtue of possessing or not possessing dark color, interact differently with birds. (v) The darkness of the trees affects the interaction such that (vi) the possession of dark color causes moths with dark color to benefit and those without dark color to suffer. (This causal interaction may have the concomitant effect of sorting genes in the moth population.) (vii) Those moths with dark color tend to produce more offspring, which inherit dark color.

The abstraction is a simplification in a number of explicit ways. In condition (i) the existence of a set is specified, but common properties that define the set are not explicitly given, though they could be. Condition (ii) is the simplest case in which Y's differ by only one property. If the way in which Y's differ is unknown, or Y's differ in more than one property, this condition could be altered accordingly. Condition (ii) considers the range of the variant property, which here is stated in the simplest possible way, that is, having or not having P, could be altered to reflect other possibilities, such as a continuous gradation of some sort or quantitative differences. (See Darden (1982) for a discussion of instantiating details about variation.)

Specifying an environment and a single factor within it that is relevant to the selective interaction (conditions iv and v) may sometimes be difficult. How to delineate the relevant "environment" from everything else may be a formidable task. Limiting discussion to a single environmental factor F is a simplification. In the moth example, the environment includes the nature of the predator as well as environmental conditions, for example, color of trees. Thus sometimes the effects of the critical factor are mediated through an agent, such as a predator.

Condition (iv) of interaction is separated from the causal condition (vi) because it is possible to have different interactions without having dif-

118

LINDLEY DARDEN AND JOSEPH A. CAIN

ferential benefits. For example, if toads have tongues of different lengths, then those with longer tongues can catch insects at a greater distance. But if insects are sufficiently plentiful, then all toads get adequate food despite this differential interaction; no relative benefit or suffering results. Some critical factor in the environment, condition (v), must produce a situation that makes P causally relevant to benefiting.

With some trepidation, we have chosen to follow others and include causal language in our abstraction. We are not here providing an analysis of the "cause" in our abstraction. We are merely stating that in a typical selection process the interaction between variant property P and the environmental factor F causes those variants with P to benefit. Explaining the way differences in interaction cause differences in benefit will depend on case-specific information. We realize the statistical nature of the process and the implicit ceteris paribus clause: sometimes those with P will be in appropriate conditions to benefit, but other times something else will befall them and they won't actually benefit. "Benefit" and "suffer" are vague, but this vagueness has the advantage of allowing instantiation in numerous ways in different specific cases. Benefit could be interpreted as contributing to survival and reproduction (with inheritance of the beneficial property), but it need not. Benefit can be measured instead in terms of immediate effects on Y's. Theories lacking a reproduction step might be called "election" rather than "selection" theories.⁴

119

SELECTION TYPE THEORIES

formation. Jerne’s theory was subsequently modified by Burnet (1957) to the theory of clonal selection.

The problem is to explain how the immune system forms antibodies that are able to deactivate many different types of foreign substances (called antigens) that invade the body, while not attacking the body’s own substances. Jerne, in reflecting on the components of his theory, says:

Thus Jerne proposes several mechanisms: one for producing variants, another for negative selection against any antibodies that would attack the body’s own substances, and a third positive selective mechanism for increasing production of antibodies against an invading antigen. The positive selective mechanism works, according to Jerne, by a specific, circulating antibody attaching to an invading antigen, the complex of antigen-antibody being engulfed by a cell, and that cell (or perhaps another one that is signaled) beginning to produce more of the specific antibody molecule. Jerne, writing before the details of protein synthesis had been worked out in molecular biology, admits that the idea that a protein molecule entering a cell could signal the cell to produce more molecules of that kind is an "unfamiliar" notion (Jerne 1955, pp. 849-850).

Although Burnet (1957) endorses much of Jerne’s theory, he objects to the mechanism of production of antibodies and proposes an alternative at the level of the cell rather than the molecule. A similar change in level from molecule to cell was proposed (apparently independently) by Talmadge (1957). (For further discussion of Talmadge’s role see Burnet (1957, p. 67); Talmadge (1986, p. 7); and Ada and Nossal (1987, p. 65). Jerne (1966, p. 308) in a retrospective on the theory in 1966 gives credit to Burnet and does not mention Talmadge.)

Of Jerne’s theory, Burnet says:

120

LINDLEY DARDEN AND JOSEPH A. CAIN

protein which can replicate only when taken into an appropriate cell. (Burnet 1957, p. 67)

Thus Burnet changes the level of organization of the unit that replicates from molecule to cell. Furthermore, he claims that the cell is the unit of variation; one cell produces one type of antibody. He proposes somatic mutation as the mechanism of the production of variants. In addition to making the cell the unit of variation and of reproduction, he also makes it the unit of interaction; the lymphocyte cell has a reactive site that is "equivalent" to the antibody it produces. The reaction between the reactive site and antigen serves as a signal to the cell to produce a clone of cells like itself. The cloned cells then release tree antibodies of the specific type to attack the invading antigen. (For additional discussion, see Lederberg 1959; and Schaffner 1980, pp. 69-71.)

Jerne and Burnet’s theories illustrate additional instances of selection type theories that may be analyzed to yield an abstraction. The differences in their theories show that the variables can be instantiated in different ways either to improve a theory or to yield a competing alternative. In this case, the level of organization–molecule or cell–provides an important point of difference in the ways the variables are instantiated.

Jerne’s and Burnet’s theories attempt to solve an adaptation problem: how is the immune system’s defense against an invading antigen produced, or, more specifically, how are antibodies against an antigen formed? Fighting invaders is an adaptive response of the immune system.

By comparing Burnet’s theory to our previous abstraction, we can develop an abstraction for Burnet’s clonal selection theory.

A selection process occurs if:

121

SELECTION TYPE THEORIES

The selection event involves a differential activation. The effect is a differential amplification of cells that has the concomitant effect of producing more antibodies of the given type.

6. Selection Theories for Brain Function. A number of rather speculative theories for explaining various aspects of brain functioning have been proposed that are selection type theories. These include theories by Conrad (1976); Changeux (1985); and Edelman (1978). Here, we will focus on Edelman’s theory. The adaptation problem is to explain increasing efficiency of the brain in processing a given type of signal. First, how does the brain accommodate new types of signals that have never before been encountered? Second, how does the brain refine a response pattern when it encounters a type of signal repeatedly? (Edelman 1978, p. 55.) Edelman proposes a theory to solve this adaptation problem; he calls it the "group-selective theory of higher brain function":

The units of variation are groups of neurons. Groups having nearly identical functions are part of a repertoire, in Edelman’s terminology (p. 62), or a set of variants, in our terminology. Functionality of groups of neurons is defined by their response to signals. Although all the groups in a repertoire share the property of being able to respond to a given signal, these groups vary in their internal structural organization. In Edelman’s terminology (and used similarly in molecular biology) the groups are degenerate, not redundant: groups with different intrinsic connections respond to identical stimuli. "Degenerate groups are isofunctional, but

122

LINDLEY DARDEN AND JOSEPH A. CAIN

nonisomorphic . . . redundant groups are isofunctional and isomorphic" (Edelman 1978, p. 59, Fig. 3).

The variant groups are produced as a result of both genetic factors and "epigenetic" factors, which occur during the embryological and early development of the brain. Such factors include "selective stabilization" of synapses (connections between neurons) discussed by Changeux (Edelman 1978, pp. 88-89). But even after these selection events, which give rise to structurally different neuron groups, much degeneracy remains, according to Edelman. This degeneracy provides the variability for the selective interactions that occur between the neuron groups and the signal input throughout the life of the organism.

By some means that Edelman does not specify, the variation in internal structure of each particular group determines how well it will respond to a given stimulus. As a result of the variation in structure, some neuron groups are more easily stimulated or more often stimulated than others. That stimulation has the effect of either reinforcing a particular neuron group or somehow selecting against the other groups in that repertoire, thus altering their potential for subsequent stimulation by similar signals. The accumulation of such alterations has the longer-range effect of building up a secondary repertoire of groups. In other words, the reinforced groups have the potential to respond more efficiently in the future.

Edelman’s work is obviously only in the early stages of theory construction. Many details remain to be specified. Constructing an abstraction following the pattern from the natural selection and clonal selection theories is a method that Edelman himself recognizes as valuable. (See his Table 1: "Some Characteristics of Selective Systems", Edelman 1978, p. 91.) The following is our attempt at instantiating our abstraction with Edelman’s theory.

A selection process occurs if:

i. A set of degenerate neuron groups exist that respond to a given signal (that is, an isofunctional repertoire

exists), and

ii. neuron groups vary in intrinsic connections and thus in their abilities to respond to a given signal, and

iii. neuron groups are in regions of the brain that receive a given sensory signal with a particular spatiotemporal

configuration such that

iv. neuron groups, in virtue of their different intrinsic connections, interact differently with the sensory signal,

and

123

SELECTION TYPE THEORIES

v. the particular spatiotemporal configuration of the signal affects the interaction such that

vi. the possession of certain intrinsic connections causes neuron groups with such connections to be stimulated

and/or those without such connections to he inhibited.

(vi’. This causal interaction has the concomitant effect of sorting individual neurons.)

vii. Stimulation of neuron groups causes reinforcement of those groups.

viii. A secondary repertoire of committed neuron groups develops for responding more efficiently to signals

similar to those previously encountered.

125

SELECTION TYPE THEORIES

payoffs can be found. Table 1 shows ways of instantiating the variables in the different cases. Figure 1 shows the hierarchy of relations among the various theories.

Consider once again an imaginary scientist who has a problem to solve. Our results can now provide guidance in theory construction. First, a problem must be identified as an adaptation problem. Making such an identification may not be an easy task. It involves determining that a process occurs over time such that something becomes better adapted to another. One explanation of something’s becoming better adapted is that it is a consequence of a selective process.

The term "adaptation" has received a lot of attention in the philosophy of biology literature and we wish to skirt most of those issues here. However, it is useful to relate what has been called "engineering fitness" (see for example, Burian 1983) and our usage of "adaptation" in an adaptation problem. A deer may be fit in the engineering sense because it can run swiftly: it is adapted to escaping predators. Running swiftly is property P in our abstraction; predator speed is the critical environmental factor F. The interaction between P and F may cause the deer to benefit. Many such interactions lead to the population of deer becoming better adapted: they run more swiftly and escape predators. Thus, two senses of adaptation are present: the one called "engineering fitness" that focuses on the relation of the organism being adapted to its environment and the sense of population becoming better adapted by the accumulation of such adapted properties.

Both senses of adaptation are found in the immune system and brain cases as well. The fit between a given antibody and an antigen is the "engineering fitness" that contributes to the body’s becoming adapted to fight that antigen. The recognition of a signal by a neuron group is the "match" (Edelman’s term) that corresponds to "engineering fitness"; reinforcement of that group with numerous such matches produces adaptation in the brain, which becomes better able to respond more efficiently to such types of signals.

126

LINDLEY DARDEN AND JOSEPH A CAIN

127

SELECTION TYPE THEORIES

features with adaptation processes and must be distinguished from them in order to determine what kinds of theories to evoke. In induction, a very specific preprogrammed response is initiated; it is a less creative process than are selective and instructive processes, which can give rise to new types of adapted forms.

Once an adaptation problem has been identified, then our analysis suggests considering a selection theory as an explanation for it. Instantiating the variables in our abstraction is one way of constructing a selection theory. Table I provides the blanks to be filled and the abstraction in section 4 provides the outline of the theory. Some variables may be instantiated in different ways (for example, at different levels of organization) to produce alternative hypotheses, which can then be tested to confirm or disconfirm them.

Using a selection theory in a new case provides a measure of "plausibility by analogy" (Darden 1976, p. 145). By invoking a type of process that has been confirmed to apply in other biological cases, a measure of plausibility is provided for the new hypothesis–it is like processes known to occur elsewhere. But "plausibility by analogy" does not substitute for direct empirical evidence. Of course, a new theory will require testing against the data in its own domain. Instantiating a type of theory, even a previously very successful type, is no guarantee that a theory will be confirmed in a new case.

This work demonstrates that at least one abstraction for an important type of theory can be constructed. The extent to which other types of problems and types of theories can be found and abstracted is an important area for future work for historians and philosophers of science. Such work holds the promise of providing "compiled hindsight" (Darden 1987) from the history of science, which may prove useful to current researchers in constructing new theories and, perhaps, providing insights into types of processes common in the natural world in which we live.

REFERENCES

Ada, G. L. and Gustav, N. (1987), "The Clonal Selection Theory", Scientific American 257: 62-69.

Arnold, A. J., and Fristrup, K. (1982), "The Theory of Evolution by Natural Selection A Hierarchical Expansion",Paleobiology 8: 113-129.

Ayala, F. J., and Dobzhansky, T. (eds.) (1974), Studies in the Philosophy of Biology. Berkeley: University of California Press.

Beatty, J. (1980), "What's Wrong with the Received View of Evolutionary Theory", in P. D. Asquith and R. N. Giere (eds.), PSA 1980. vol. 2. East Lansing, Michigan:Philosophy of Science Association, pp. 397-426.

Brandon, R. (1980), "A Structural Description of Evolutionary Theory", in P. D. Asquith and R. N. Giere (eds.), PSA 1980. vol. 2. East Lansing, Michigan: Philosophy of Science Association, pp. 427-439.

Cambridge: Cambridge University Press, pp. 287-314.

Burnet, F. M. ( l957),"A Modification of Jerne's Theory of Antibody Production Using the Concept of Clonal Selection", The Australian Journal of Science 20: 67-69.

Cain, J. A. (unpublished, 1987), "The Role of the Environment in Selective Interactions", presented at Summer Conference on the History, Philosophy, and Social Studies of Biology, Virginia Polytechnical Institute and Stale University, Blacksburg, Virginia, June 16-20, 1987.

Changeux, J. P. ( 1985), Neuronal Man: The Biology of Mind. New York: Pantheon, Random House.

Conrad, M. (1976), "Complementary Molecular Models of Learning and Memory", BioSystems 8: 119-138.

Jerne. N. K.(1955). "The Natural-Selection Theory of Antibody Formation", Proceedings of The National Academy of Sciences 41: 849-857.

Shapere, D. (1974), "On the Relations Between Compositional and Evolutionary Theories", in Ayala and Dobzhansky (eds.) (1974), pp. 187-201.