existence and cause

Science begins with philosophy, specifically with metaphysics. No one has solved to universal satisfaction the ontological question "why do we believe things exist?" or the cosmological question "what is cause?" Science sidesteps both questions, treating them rather than answering them.

We human animals believe things exist because we receive certain sensory data that are consistent with belief in some world apart from us. When sensory data disagree--data available to two different persons or two different sets of sensory data to the same person--doubt ensues. Thus, one fails to believe in the reality of things perceived in hallucinations and dreams.

Can one tell, however, whether things exist when there are no senses (or extensions of senses such as cameras and tape recorders) to record data? Bertrand Russell, who was fond of the concept of the limit once he had made it precise, suggested a thought-experiment as follows. Does a chair exist when there is no one in the room to perceive it? Leave for an hour and return; the chair is still there. Then leave for half an hour; no change upon return. Then leave for 15 min, 7.5 min, etc. No matter how arbitrarily short an absence is chosen, the chair is always there upon return. In the limit of indefinitely short absences, then, the chair remains: it is "always" there and hence must have a reality apart from our sense data. We really believe in the existence of things not because of "proofs" like Russell's, but because to believe leads to no inconsistencies in our lives and thinking.

Science employs a more satisfactory way to sidestep the problem of cause. If I push on Russell's chair and it moves, I am tempted to say that my pushing caused the movement. I become really convinced of this explanation, however, when I see the chair fail to move when I do not push, and when I see the chair always move when I do push. In short, we infercause from a constant conjunction of temporally ordered variables, such as first I push, then the chair moves. One may, of course, investigate the example in greater detail, measuring lever forces in my arm, frictional drag of the chair's legs on the floor, or even the molecular and atomic events that transpire. In the end, however, one merely emerges with a more complete description of a whole chain of temporally ordered variables.

Science owes to the physicist P. W. Bridgman (e.g., 1927) the doctrine that relationships among measurable variables have no reality apart from the operations by which the data are obtained. Science is concerned with such operational relationships: observing variables and discovering consistencies in their values relative to one another. Whether or not there is anything more fundamental about causation than constant conjunction of temporally ordered, observable variables is a problem for philosophy rather than science.

epistemological cycles

Any representation of how observable variables are related may be called a model. Some special types of models are familiar: physical objects called models are used by marine architects to see how hull design (one set of variables) affects speed through the water, amount of wake created and so forth (another set of variables). Such physical models are widely used in experimental analyses of animal communication (see Hailman, 1977a, esp. figures 14 and 18). Another familiar model in science is the mathematical model, an abstract statement of relationships among variables expressed in certain arbitrary symbols that are, often literally, Greek to many readers. However, the notion of a model need not be restricted to these extremes: a model is any expression of suggested relationships among observable variables.

One may classify models according to the degree of trust placed in them by the scientific community (Walker, 1963). A hypothesis is a well-stated but relatively untested model. A theory is a model that has survived sufficient testing to merit widespread interest. Finally, a law is a model so well tested that it is unlikely to be overturned completely by new data. Further testing may show the law to be a special, included set of relations within some broader model--as the laws of physical mechanics are now a special case of relativistic models. Hypothesis, theory and law mean other things in other contexts, of course, and one should also admit as models those very tentative statements we call ideas, hunches, suggestions and so on.

A model in science accrues trust by surviving empirical tests, but it is rarely possible to test a model as a whole. Rather, one employs deductive reasoning to derive a prediction that must be true if the model is true. It is the prediction from the model that is measured against the meterstick of reality: if it does not match reality, the prediction is false, and so must be the model (assuming the deductive processes were sound). If the prediction is true (matches reality), however, it cannot mean the model is necessarily true: false models can provide true predictions. For example, the model that the sun revolves around the earth from east to west correctly predicts that the sun as seen from the earth rises in the east and sets in the west. (Other predictions from the same model do not survive empirical tests, of course.) Therefore, an empirical test can never prove a model; it can only fail to disprove it.

The prediction from the model instructs the scientist to collect certain data. This she does by obervation, whether or not a formal “experiment” is performed. Niko Tinbergen once said in a quote I cannot relocate that if the observer of animal behavior is patient enough, Mother Nature will perform the experiments. Various combinations of variables occur spontaneously, so one does not necessarily have to insure these combinations by manipulations in the laboratory. Formal experiments do, however, help isolate the variables of interest so that one may observe their relationships with greater confidence that no extraneous variables were influencing the ones of interest.

One is left with the task of comparing the data with the prediction in order to reach a decision as to whether the two are the same or different. If they are the same, one goes on believing for the moment that the model correctly predicts observable reality, and ideally devises some new prediction to be tested. Each time a model correctly predicts reality, one places more confidence in the model. On the other hand, should the data fail to match the prediction, one must conclude that the model itself is false. In that case, a new model must be erected and tested.

The processes for generating models are called inductsion, but these processes are nothing like the deductive logical processes used to derive predictions from the model. My own belief is that the usual college definitions of induction and deduction are gobbledygook. Deduction is not best described as a process of reasoning from genralities to specifics, although that may sometimes be the case if “generalities” and “specifics” can be satisfactorily identified. Deduction is, rather, the rearrangement of knowledge contained in the model and its assumptions. The defining criterion for deduction is that if the model is true in any sense, the prediction deduced from it is true in the same sense. Induction, on the other hand, is poorly described as reasoning from specifics to generalities. It is doubtful that inductive processes are best termed reasoning (which connotes some lawful, logical process), and in any case they do not necessarily begin with specifics and end with generalities. It seems best to state that induction is the creative process of science, a potpourri of mental processes whose end point is a statement about the relationships among observable variables. Induction is as mysterious as the creation of a great painting or symphony, and like artists and composers, scientists differ in their creative abilities.

Fig 1-2. The cycle of scientific epistemology ,with processes indicated by arrows and results of processes indicated by boxes to which the arrows point. Mathematics plays a role in all processes.

Each small step in this cycle of epistemology could be the subject of a book; I condense the summary into fig 1-2. It is my view of how science as a whole evaluates its structure of knowledge, not necessarily how an individual scientist proceeds (Hailman, 1975). Not every individual scientific study will complete the entire cycle. Theoreticians may concern themselves simply with providing models; others may ingeniously manipulate models deductively to provide testable predictions; empiricists may concentrate almost exclusively upon collecting relevant data; and others may devote their lives to careful scrutiny of data to see whether these do or do not match the predictions. In some cases, one scientist will complete the entire cycle, perhaps several times, testing and retesting some model, or continually producing better ones as previous ones fail to predict reality.

This volume on animal communication and light is concerned primarily with creating very tentatively expressed models relating various factors to the characteristics of optical signals. It is a statement of future directions in creating models, deriving predictions from them and testing those predictions.

mathematical roles

Mathematics is not, as often asserted, the language of science--because so many of us have failed to master that tongue. When biology speaks the language with the same fluency as physics, science will be the richer. The emphasis of mathematics in science is, however, often misplaced. Popularizations of science emphasize the precision with which measurements of natural phenomena are made in the observational process of fig 1-2. Scientists themselves often pay most attention to statistical comparisons that lead to the decisions as to whether or not data fit the prediction. Nor can the mathematical thinking of Einstein, Bohr or others be denied an important role in their inductive processes. In my view, though, it is in the deductive processes that mathematics plays the most critical role in science.

The recurring problem with grand schemes purporting to explain animal behavior lies with the impossibility of deriving testable predictions from them. Readers suffer through long arguments about the instincts, motivations or approach-withdrawal processes of animal behavior without discovering how these schemes might be put to empirical test. Scrutiny reveals that most such schemes are incapable of falsification, and hence do not deserve the approbation of “model” at all.

In other cases, potentially testable models remain unevaluated because the deductive processes leading to predictions are difficult or faulty. Here, mathematics plays its crucial role in forcing articulation of the model in symbolic form so that rules of mathematical deduction may be employed to derive predictions from it (fig 1-2). Taken broadly to include symbolic logic or its equivalent of Boolean algebra, mathematical deduction makes scientific epistemology possible.

The characteristic of valid deduction is that if the model be true, the prediction must be true. If the model be false, valid deduction leads to predictions that may be either true or false (in the sense of matching or failing to match empirically gathered data). It is not difficult to appreciate that invalid deductive processes may deliver either true or false predictions, regardless of whether they are derived from true or false models. Therefore, from the viewpoint of the scientist as the decision-point in fig 1-2, there exists a logical truth-table that allows him or her to evaluate the truth-value of the model being tested, as shown in table 1-I.

Table 1-I

Truth-value of Models, Dependent upon Validity of Deduction and Comparison of Predictions with Data

It is evident from table 1-I that if the scientist’s deduction is invalid, or if its validity is in doubt, nothing can be learned about the truth-status of the model regardless of the empirical outcome of testing. For this reason the entire scientific edifice balances on the necessity for valid deduction--validity that must be established by inspection of the deductive steps per se without reference to empirical aspects of the scientific problem. Such validity can be established only when the deductive processes may be inspected for strict conformity with the rules. The best set of rules that exist for unerring deduction are those of mathematics.

Finally, one may note from table 1-I that if the deductive process is valid, one may with certainty falsify a model when there is not a match between prediction and empirical data. When the match exists, the model itself may still be false: there is no logical way to be sure. Repeated testing of different predictions, each in turn failing to reject the model, is the sole method for increasing confidence in it.

behavioral determinants

It is crucial to understand that the question “why does the animal do that?” is actually several different questions. There are four classes of models that explain animal behavior and they are mutually compatible (Tinbergen, 1963; Hailman, 1976a). Because the act of communicating is behavior, there are four kinds of answers to questions about how communicative behavior is determined.

First, one may ask how internal and external factors combine to dictate the behavior that we as outsiders observe. This classical concern of motivational and sensory studies may be called the dynamic control of behavior. The mere ordering of descriptions of behavior into coherent schemes constitutes the first model of dynamic control, later to be refined by sensory and motivational experiments, and perhaps even physiological studies of control model of the communicative act.

Second, one may ask how antecedent events in the life of the animal brought about the control properties of the behavioral end-point one observes. The two factors that control such ontogenetic development are the genetic endowment from the parents and the environmental conditions in which the animal develops. The study of ontogeny ineludes dealing with the classical nature-nurture question that still arises in modern times as an “instinct-learning” dichotomy, although an operational approach to the problem quickly alters its formulation (Hailman, 1976b). To date, rather little is known about genetical and experiential factors in optical communication of animals (Hailman, 1977a), and unfortunately little more can be added in this volume.

The third fundamental behavioral determinant focuses upon the population of animals rather than its component individuals. Many patterns of behavior including communication persist from generation to generation. A high correlation between the behavior of parents and their offspring may be due to the similarities in their genes or to the similarities in their rearing environments, including the cultural environment (ontogeny, above). In most cases, one assumes that such preservation of behavior from generation to generation is maintained by natural selection or closely related processes such as kinship selection (see Brown, 1975 and Wilson, 1975).

Because natural selection is a primary cause of behavioral preservation, this determinant is often referred to as the “adaptive significance” or “selective advantage” of behavior. I myself have frequently referred to the “adaptive function” or simply “function” of behavior (e.g., Hailman, 1967a, 1976a, 1977a). However, the term “function” is unfortunate: it narrowly connotes only processes of population genetics to the exclusion of experiential factors, and it has other meanings in biology (Hailman, 1976a). For example, when one discovers how something works (control, above), one often says it “functions” in a certain way. From the analysis of control one then guesses at the selective advantages a behavioral pattern confers upon its practitioner, so that function refers to two related but distinct behavioral determinants. Finally, some behavioral patterns may persist in populations by mechanisms other than natural selection, so not every behavioral pattern has an assignable adaptive function (Hailman, 1977b).

The fourth and last class of behavioral determination concerns antecedent events in the history of the population, which may be called the phylogeny of behavior. To contemporary biologists the word phylogeny is nearly synonymous with evolutionary history, but I use the term in its older and wider sense of simply the history of the population. Therefore, phylogenetic considerations include the origins of culturally transmitted behavior and hence complete a comprehensive albeit general scheme for investigating behavior.

Table 1-II

Classes of Behavioral Determinants

Table 1-II summarizes the scheme of behavioral determinants in pointing out that at both the level of the individual organism and the population of which it is a part, there is a class of immediate causes of behavior as well as a class of antecedent origins. One cannot, in my view, claim to understand any behavior, including communicative behavior, without understanding all of its simultaneous and interacting classes of causes and origins. Unfortunately, when animal communication by light is scrutinized systematically according to table 1-II, the review of knowledge and understanding turns out to be short (Hailman, 1977a).

comparative method

The cycle of epistemology (fig 1-2) may be applied directly to all the classes of behavioral determinants (table 1-II), but it is not always obvious how this application should be accomplished. One may experiment directly with the dynamic control of communicative behavior by manipulating external factors such as social signals and other stimuli, or by manipulating internal factors such as hormonal balance (see Hailman, 1977a). Similarly, the ontogenetic determinants of communicative behavior may be studied directly by manipulating the genetical endowment or the rearing environment of the developing individual. It is more difficult to study directly the determinants of communicative (or other) behavior at the population level. To do so, biologists frequently employ the less direct strategy of the comparative method.

The comparative method utilizes comparisons among (sometimes within) species of animals to test predictions concerning the preservation and phylogeny of behavior. In its most powerful form four groups of animals are compared by this method: two evolutionary or cultural lines whose populations live in each of two environmental situations (table 1-III). The behavioral characteristics of the groups of populations are then inspected for similarities and differences. If traits are similar in groups (a,a) and (b,a) of the table, but different from traits in groups (a, b) and (b ,b) which are similar to one another, then the differences correlate with differences in the environmental situation. One attributes the correlation to mechanisms of preservation that are tied to the immediate environmental situation. Alternatively, if groups (a,a) and (a,b) show similarities, and groups (b,a) and (b,b) show similarities of a different kind, then the traits under investigation may be attributed to phylogeny: historical determination due to common evolutionary or cultural descent.

Table 1-III

Groups for the Comparative Method

Although evolutionary biologists think almost unconsciously in terms of comparative studies, persons differently trained have difficulty applying and understanding the method. The difficulty is partly due to the idealized scheme of table 1-III: rarely can all four cells of the table be filled. For example, E. Cullen (1957) suggested that the black neck-mark on the chick of the kittiwake gull is an optical signal that inhibits parental aggression. This cliff-nesting species is forced into parental-offspring propinquity for a longer period than in surface-nesting gulls, where chicks are soon able to roam over the relatively large reproductive territory of their parents. Cullen’s suggestion is a model in the sense of fig 1-2: it predicts that forced propinquity of cliff-nesting favors selection for a particular kind of signal. Let the kittiwake represent group (a,a) in table l-III. Then closely related gull species that do not nest on cliffs, represented by group (a,b), should lack the neck-band signal; as Cullen points out, this is true for all known suface-nesting gulls. Armed with her model, I investigated the cliff-nesting swallow-tailed gull of the Galapagos Islands and found the older young possessed a black neckband as predicted (Hailman, 1965), providing another example of group (a,a). The consistent difference between groups (a,a) and (a,b) fails to reject Cullen’s model of preservation: one continues to believe that the aggression-inhibiting signal is selected for by forced propinquity of parent and offspring.

Alternatively, other optical signals of these same gulls have been subjected to considerable ethological scrutiny (N. Tinbergen, 1959). When their signals are compared with those of other groups of birds (line b in table 1-III), consistent differences between gulls and other birds are evident, regardless of the environmental situations in which the species live (columns of table 1-III). Tinbergen rightly concludes that at least certain characteristics of optical displays are due primarily to common evolutionary ancestry.

strategic gameplan

The overall strategy is straightforward and charted in such a way that its outline could be followed in scrutinizing chemical, acoustical or other signals having nothing to do with light. First, a framework of the communication process itself is provided as a synthesis combining some of the notions of ethology, cybernetics and semiotics (ch 2). Then in successive chapters the characteristics of the communication channel (ch 3), the sender (ch 4) and the receiver (ch 5) are scrutinized for factors that constrain the design of optical signals. Next the communication of misinformation, in the traditional sense of concealment and mimicry, is reinvestigated from the viewpoint of optical principles employed in visual deception (ch 6). The design of optical signals to minimize the effects of environmental noise then places communication in the ecological context (ch 7). The analysis cuiminates with considerations of how the qualitative kind of information sent helps to determine the design of optical signals (ch 8). The volume concludes (ch 9) with a few comments on the problems and prospects of studying optical signals.

tactical rules

The book is structured linearly and best read straight through, although various aids are used to facilitate referencing back to earlier chapters. Each chapter follows a similar plan, the plan and mechanics being obvious upon reading. The problems of mathematical notation prove crushing. Because one quickly exhausts the English and Greek alphabets, two compromises were made to avoid esoteric type-symbols: many relationships are stated in words rather than equations and symbols remaining may be used differently in different chapters, although they are used to mean only one thing within a chapter. Equations are listed in the contents for ready reference and the symbols themselves appear in the index. The terminal bibliography has been used as an index to citations of references in the text for further aid.

Some mathematical operators may be unfamiliar. I follow Batschelet’s (1975) sensible suggestion of using the proportionality operator (α) when constants and parameters of an equation arse irrelevant to the point being made. Thus, y = ax+b may be rendered simply y α χ when the values of a and b are not of central concern. There is, alas, no widely used symbol for the monotone operator, so it is not without a touch of irony that one is invented for a volume on optical signals. When the value of y never decreases with an increase in the value of x_, this function may be written y ↥ x, unless the increase is proportional and may be expressed more precisely. Finally, when the value of y depends upon the value of x, but not necessarily in a monotonie relationship, the function is denoted f:x ↦ y (Batschelet, 1975), and one states that “x maps to y.” This means precisely the same thing as the older notation y = f(x). The mapping or functional operator(↦) was devised to prevent confusion with logical implication (⇒) and convergence (→) , all three being denoted formerly by the last symbol. Higher mathematics are not employed, although the reader requires some familiarity with basic notions of set theory, probability, algebra, geometry, trigonometry and logarithms.