The Ants Chapter 9

CHAPTER 9. SOCIAL HOMEOSTASIS AND FLEXIBILITY

What makes social systems most appealing intellectually is the existence of hierarchies. When organization of this kind occurs, there is often more to a whole society than the sum of its parts. Yet even the most emergent properties of its behavior remain strictly ordered by devices of communication and regulation operating at the level of organism. This perception is not a mere truism. There is no way to predict which devices have evolved in particular social species, even though general principles can be adduced that affect all of the species together.

A hierarchy is a system “composed of interrelated subsystems, each of the latter being in turn hierarchic in structure until we reach some lowest level of elementary subsystem” (Simon, 1981). Hierarchies are said to be decomposable, which means that the linkages within the components at each level are much stronger than the linkages between the components at the same level. Thus an ant colony can be decomposed into the occupants of different nests when the colony is polydomous. That is, the occupants of each nest are more tightly connected to each other within each short interval of time than are the occupants of different nests. Moving on down, the occupants of a nest can be decomposed into various castes, which can then be decomposed into individual ants. An individual ant is in turn a hierarchy of organs, tissues, cells, and so on. Often hierarchies are run by a central command structure, in other words a “boss” at the top, but this is not necessary and in fact does not occur in ant colonies. All that is needed to create a hierarchy is the property of decomposability. In two respects the ant colony is a special kind of hierarchy, which can be usefully called a “dense heterarchy” (Wilson and Hölldobler, 1988). The colony is dense in the sense that each individual insect is likely to communicate with any other. Groups of workers specialize as castes of particular tasks, and their activities are subordinated to the needs of the whole colony. However, they do not act by a chain of command independent of the other groups of workers. They are open at all times to influence by most or all of the membership of the colony. An ant colony thus differs in basic organization from the “partitioned” hierarchies of human armies and factories, in which instructions flow down parallel, independent groups of members through two or more levels of command. The colony is also a heterarchy, which is defined as a hierarchy-like system of two or more levels of units with activity in the lower units feeding back to influence the higher levels. Finally, the highest level of the ant colony is the totality of its membership rather than a particular set of superordinate individuals who direct the activity of members at lower levels.

An excellent example of organization by a dense heterarchy is the pattern of food flow into a colony of the fire ant Solenopsis invicta as demonstrated in the experiments performed by Sorenson et al. (1985). When foragers were first starved as a group, they collected disproportionately more honey. When they were well fed but the nurse workers or larvae were starved, the foragers collected more vegetable oils and egg yolk. Previous research had shown that sugars are used mainly by adult ant workers, lipids by workers and some larvae, and proteins by larvae and egg-laying queens (Wilson and Eisner, 1957; Echols, 1966; Lange, 1967; Abbott, 1978; Howard and Tschinkel, 1981). Hence the foragers, which are older workers, respond to the nutritional needs of the colony as a whole and not just to their own hunger. How do they monitor this generalized demand? The answer is that they rely on a combined system and individual decisions joined on a massive scale (Sorenson et al., 1985). A large additional group of workers, the reserves, receive most of the food as it comes into the fire ant nest and then pass it on to other colony members, including the nurses. When the demand they encounter declines in any sector of the colony, the reserve workers accept the corresponding food less readily from the foraging workers. The foragers are unable to dispose of their loads as quickly as before, and they reduce their efforts to collect more food of the same kind. As a result they shift as a group in their emphasis from carbohydrates to oils or proteins or in the reverse direction--according to the needs of the colony. A similar mediating effect based on the rate of acceptance of newly harvested food has been reported in honeybees and other ant species (Wilson, 1971; Brian and Abbott, 1977; Seeley, 1985).

A second example from the same species is found in the phenomenon of mass communication during recruitment (Wilson, 1962b, 1971). Individual workers arriving at food finds such as an animal carcass lay odor trails composed of secretions from the Dufour's gland (as reviewed in Chapter 7). They decide whether to recruit in this manner according to the nutritional needs of the colony and the richness and appropriateness of the food. Individual workers also vary the amount of pheromone emitted according to these variables. The quickness with which a forager inspects the food and returns to the nest laying a trail depends on these same variables, as well as on the amount of crowding by ants already at the food site. Food retrieval proceeds as the number of recruiting workers sets the amount of recruitment pheromone in the trail, which in turn sets the number of workers answering the call. As the colony approaches satiety, or the food site becomes saturated, the likelihood of trail-laying by individual workers diminishes. The flow of food correspondingly stabilizes and finally declines.

A third and very different form of dense heterarchy occurs in the regulation of egg laying by fire ants. The queen is the sole progenitrix of new colony members, and much of the colony's activity is devoted to her protection and nurture. It is reasonable to suppose that her production of eggs is related in some manner to the capacity of the colony to rear additional members. When too few eggs are laid, the growth of the colony is slowed below capacity. When too many are laid, energy is wasted, again reducing colony growth--at least in small colonies. How then does the queen monitor the condition of the colony in order to manufacture approximately the correct number of eggs? It might be more accurate to ask how the nurse workers monitor the condition of the colony and adjust the feeding of the queen to this desirable end. In a series of ingenious experiments, Tschinkel (1988c) showed that the control is based on mass communication. Workers and pupae do not by themselves stimulate queen oviposition. However, the rate of egg laying by the queen increases in a log-log relationship as the number of fourth (and final) instar larvae and earliest pupal stages increases. Pharate pupae (“white larvae”) are the most stimulating of all, becoming maximally so during the 24-hour period after passing the meconium. Tschinkel found that vital dyes placed in food were transferred rapidly from the fourth-instar larvae to the nurse workers and thence to the queen's eggs, indicating a substantial passage of food along this route. From general observations on the behavior of Solenopsis invicta and other kinds of ants, it seems likely that such larvae serve as a metabolic caste, and they are often active in processing protein for egg production by the queen. The queen and fourth instar larvae are therefore linked in a positive feedback loop.

It is becoming evident that much of the structure of the ant heterarchies is based upon order parameters, defined as the proportion and physical distribution of individuals existing in one physiological state or another. Thus the frequency of fourth instar fire ant larvae relative to young larvae determines the rate of queen egg-laying and consequently the number of additional young larvae due to appear during the following month. The number of hungry larvae in comparison with hungry workers back in the nest determines the emphasis placed by foragers on the kind of food they collect, independently of their own hunger. In a second example, the proportion of major workers in the adult population of Pheidole bicarinata affects the ability of final instar larvae to transform into majors themselves. As the proportion rises, the level of inhibiting pheromone produced by the majors also rises. The pheromone evidently has the effect of making the larvae less sensitive to juvenile hormone, so that they enter metamorphosis more quickly and are more likely to become minor workers (D. E. Wheeler, 1986a).

In conclusion, the organization of ant colonies into dense heterarchies permits relatively efficient social regulation with the loosest of command structures. The most striking social effects turn out to be the holistic outcomes of mass communication combined with the rise and decline of pheromones and food stuffs.

Series-parallel operations
The dense heterarchy also provides the colony with the ability to act locally and blanket a large area in quick response. The colony does not depend on the transmission of information up and down chains of command before decisive local action can be taken. It thus meets important contingencies as they arise, such as the appearance of an enemy or the hunger signal of a larva. The full system is also more secure, because when one colony member fails to complete a task another is likely to succeed. In the design of control devices, engineers commonly utilize such series-parallel operations because the breakdown of one unit will not cause the failure of the whole device. This procedure is more reliable than parallel-series operations, the simpler procedure in which sequences of tasks are performed in isolation. When performance in one line of a parallel-series operation breaks down, it is not fixed but merely replaced by another, parallel sequence (see Figure 9-1).

This conception of behavioral acts and sequences, introduced by Oster and Wilson (1978) and refined by Herbers (1981a), sheds new light on the origins of insect social behavior. It can be shown that the reliability of the colony as a whole (P) is related to the reliability of its individual components (p) by a sigmoidal curve of the kind illustrated in Figure 9-2. When the competence of individuals is low, the performance of the group is lower than if the individuals acted independently. However, there is a threshold (p*) of individual reliability above which group reliability exceeds that of individual reliability. In this region, colony-level selection favors cooperative behavior.

The concept of a threshold of individual competence may seem paradoxical, but an example will make the relation intuitively clear. Suppose that several ants were attempting to move larvae from one part of the nest to another part with a more favorable temperature. If each ant tries the entire operation alone, the group result will equal the average individual result. Now suppose that two workers start working together, as in fact they often do when the larvae are large (Figure 9-3). If their competence is low, they will tend to cancel out one another's actions. They may pull in opposite directions, or start and stop independently so as to stall the operation indefinitely. The probability that incompetent ants will make the right combination of correct movements (k/n), such as gripping the larva in the right position, lifting it to the proper height, and selecting the right direction to walk, will be even lower than the chance of doing one thing correctly. In the end fewer larvae will have been transported. However, as individual competence increases, the group will reach a point in which they are able to put together the right combination of actions and speedily move a larger number of larvae. The improvement will be greater still if they divide the labor. For example, some might concentrate on laying odor trails or leading tandem runs while others concentrate on transporting the larvae.

The potency of cooperation in a dense heterarchy is enhanced in another way predicted from reliability theory. This second theorem states that replication at the subunit level is more efficient than redundancy at the system level. Put another way, if a designer is given two sets of parts, it is better to build a single system with redundant components than to build two separate systems. The reliability of the redundant system, hence the chance that it will work at all, is greater than the reliability of the separate systems. This relationship is illustrated in Figure 9-4.

Multiplier effects
The reliability theorems illustrated in Figures 9-2 and 9-4 lead us to believe that the responses of the colony as a whole are generally amplified. In other words, not only do the parallel operations of redundantly acting workers increase the chance that a given task will be performed, but the speed and effectiveness of the colony will be improved.

An example is provided by mass communication during trail laying in the myrmicine Tetramorium impurum, as described by Verhaeghe (1982). Like scouts of fire ants, the foraging workers of Tetramorium impurum vary the amount of pheromone paid into the trail according to the quality of the food find and the degree to which food is needed by the colony. The probability that a follower ant succeeds in following the trail to the end increases as the logarithm of the amount of pheromone deposited. As a result, variations in the quantity of pheromone voided by individual workers translate into large differences in the number of ants traveling back and forth to the target. As more scouts lay trails during their parallel efforts, the segments they individually contribute are more likely to unite into a complete trail. The result is a rapid exploitation of food and new food sites. This particular form of amplification is likely to be widespread in trail-laying ants. In most cases scouts act more or less independently in laying trails. Also, in at least the species of Acanthomyops, Myrmica, Solenopsis, and Tetramorium, workers individually vary the quantity of pheromone (Hangartner, 1969b; Cammaerts-Tricot, 1977; Verhaeghe, 1982).

Other mutiplier effects can arise as a consequence of the statistical properties of caste systems. The tournaments of the honeypot ants of the genus Myrmecocystus, for example, are conducted by major workers. This caste is not a discrete group but merely the largest individuals in a population of workers whose size-frequency curve is unimodal and skewed toward the larger end. The size-frequency curve expands as the colony grows in size, so that larger colonies have a higher percentage of majors and a much larger absolute number of individuals belonging to the extreme size class. As a result they possess an overwhelming advantage over smaller colonies during tournaments, and they are more likely to win territorial disputes (Hölldobler, unpublished).

Rules of thumb
The total behavioral repertory of an individual ant worker is relatively simple, consisting according to species of no more than 20 to 45 acts. Yet the behavior of the colony as a whole is vastly more complex in all but the most primitive species. As we showed in Chapter 8, some of this superstructure is built on the physical caste system. Small and large workers have different behavioral repertories, which are fitted together to form larger wholes. Of even greater consequence are the social feedback loops operating through mass communication. The individual ant need operate only with “rules of thumb,” elementary decisions based on local stimuli that contain relatively small amounts of information. The examples of rules of thumb we have reviewed thus far in this chapter could be expressed as follows: “continue hunting for a certain foodstuff if the present foraging load is accepted by nestmates”; “follow a trail if sufficient pheromone is present”; “feed the queen more if final-instar larvae are present”; and “attend the larvae and other immature stages if regular nurse workers are absent.”  Each of these rules is easily handled by individual workers, even allowing for brains as small as a tenth of a cubic millimeter. Each action is also performed in a probabilistic manner with limited precision. Yet when put together in the form of dense heterarchies involving large numbers of workers, the whole pattern that emerges is strikingly different and more complicated in form, as well as more precise in execution.

The superorganism
The new perception of heterarchical organization has revived the venerable idea of the ant colony as superorganism. William Morton Wheeler, like many of his contemporaries, was guided by this concept. In his celebrated essay “The ant colony as an organism” (1911) he stated that the animal colony is really an organism and not merely the analog of one. Of course, one has to pay close attention to his definition of organism: “An organism is a complex, definitely coordinated and therefore individualized system of activities, which are primarily directed to obtaining and assimilating substances from an environment, to producing other similar systems, known as offspring, and to protecting the system itself and usually also its offspring from disturbances emanating from the environment. The three fundamental activities enumerated in this definition, namely nutrition, reproduction and protection, seem to have their inception in what we know, from exclusively subjective experience, as feelings of hunger, affection and fear respectively.”

These malleable criteria are easily applied to the ant colony and other insect societies. Wheeler saw several important qualities of the ant colony that qualified it as an organism:

1. It behaves as a unit.

2. It shows some idiosyncrasies in behavior, size, and structure that are peculiar to the species.

3. It undergoes a cycle of growth and reproduction that is clearly adaptive.

4. It is differentiated into “germ plasm” (queens and males) and “soma” (workers).

In The Social Insects (1928), Wheeler began to call the insect colony a “superorganism” rather than the more obviously metaphorical “organism.” His ideas on the subject of colonial organization had actually changed little by that time, although he had begun to conceive in a vague fashion of the phenomenon that was later to be called homeostasis:  “We have seen that the insect colony or society may be regarded as a super-organism and hence as a living whole bent on preserving its moving equilibrium and integrity.”

Wheeler's imagery grew from the Zeitgeist prevailing during his most productive years. From about 1900 to 1950, many biologists and philosophers besides Wheeler developed a keen interest in holism and emergent evolution. While a considerable amount of mysticism was generated, the most famous example being Maurice Maeterlinck's “spirit of the hive,” the core of scientifically oriented writing concentrated on analogies between the organism and superorganism. Trophallaxis (the exchange of liquid food among colony members) was cited as the equivalent of circulation, while soldiers were thought the parallel of the immune system, and so on. Although initially stimulating, even inspirational, the entire elaborate exercise eventually exhausted its possibilities. The limitations of a primarily analogical approach became increasingly apparent when investigators turned more to the fine details of communication and castes. By the early 1960s the expression “superorganism” had all but disappeared from the technical vocabulary (see the historical review by Wilson, 1971: 317).

However, old ideas in science never really die. They only sink to mother Earth, like the mythical giant Antaeus, to rise again with renewed vigor. The time may have arrived for a revival of the superorganism concept. We see two reasons, both stemming from the increase of information and technical competence of its analysis since 1960. The first is the beginning of a sound developmental biology of the insect colony, and the second is the rapid improvement of optimization analysis in behavioral ecology and sociobiology.

The new developmental biology has provided an understanding of the way in which castes are determined, their ratios regulated, and their actions coordinated through heterarchical forms of communication. There is a need now for drawing analogies at a deeper level, in which the organizational processes of societies are more precisely measured and compared with their equivalents in the growth and differentiation of tissues of organisms. It is not too much to suggest that insect sociobiology will contribute to a future general theory of biological organization based upon quantitatively defined principles of feedback, multiplier and cascading affects, and optimal spatial arrangements. The new effort will prove additionally useful in calling attention to poorly understood organizational processes and the techniques by which they are more precisely analyzed. For example, Meyer (1966), Hofstadter (1979), Markl (1983, 1985), and Minsky (1986) have cited similarities between the organization of neurons in the brain and of workers in an insect colony.

In The Insect Societies (1971), Wilson observed that faith in reductionism rides on the ability of researchers to use many piecemeal analyses to reconstruct the full colonial system in vitro. Such a reconstruction would mean the full explanation of social behavior by means of integrative mechanisms experimentally demonstrated and the proof of that explanation by the artificial induction of the complete repertory of social responses on the part of isolated members of insect colonies. Three achievements appear to be central to the realization of this dream. None seemed within reach in 1971, but now, nearly twenty years later, all have been partly attained. The three are as follows:

- The activation of social behavioral responses, including the more intricate and delicate aspects of brood care, by exposure of isolated colony members to synthetic pheromones, sounds, and other stimuli emanating from lifeless dummies or else presented wholly in vacuo. These procedures are now routine in research on ants (see Chapter 7).

- The rearing of isolated honeybee larvae or termite nymphs and the determination of their caste at the will of the investigator by appropriate manipulation of food, pheromones, hormones, and other caste-biasing factors. Although immature forms have not yet been raised in isolation, the experimental conditions have been adequately controlled and the key factors of caste determination identified in some species of ants (see Chapter 8).

-The cybernetic simulation of nest building, incorporating into the model only those behavioral elements and sequences of elements actually observed in individual workers, leading in turn to the successful prediction of the responses of isolated workers presented with synthetic nests in various stages of construction. Preliminary steps have been taken in this direction during studies of Nasutitermes termites by Jones (1979, 1980), but we are still a long way from in vitro duplication for any species of social insect.

The historical cycle of research in sociobiology has run from an imagery of emergent properties in complicated societies, to the closer scrutiny of these properties by experimental studies, and finally to the reconstitution of the facts into a new whole. The reconstituted imagery will inevitably differ in important details from the original, and it will provide still more distant perceptions that invite experimental investigation.

A partial explanation of ant colonies might be built in such a manner, but it requires in addition an evolutionary perspective in which the whole system is examined as an adaptation to the environment. The adaptationist explanation is not a dogma, such that the investigator sees each newly discovered process as increasing the fitness of the society. Rather it is a mode of framing hypotheses, a heuristic device used to design field studies and laboratory experimentation.

To take a concrete example, the relative frequencies of the castes are expected to be adjusted by natural selection, and so the caste system is conjectured to be adaptive. In other words, the numbers of minors, majors, callow workers, and so forth are postulated to compose a mix enabling the colonies to survive and reproduce better than would be the case if they possessed a different mix. The optimal mix is attained by adaptive demography, a genetically programmed schedule of birth, growth, and mortality that yields standing crops of caste members in the appropriate numbers. Moving down to the next level of explanation, we are challenged to ask why certain mixes are superior. Lumsden (1982) has pointed to three determining properties of the colony members: (1) the ergonomic gaps between the castes, in other words those tasks not exactly covered by particular castes which have evolved to serve them; (2) the short, Markovian memory of colony members, requiring virtually moment-by-moment perception and adjustment of the colony as a whole to environmental exigencies; and (3) interactions among the colony members. In theory at least, the organization of the colony can be described as the matrix of interaction of the members of the colony both within and across castes, subject to the constraints of ergonomic gaps and limited memory. Some of the workers will interact at very frequent intervals, others seldom. Most of the interactions will be cooperative and productive for the colony, while others will consist mostly of interference and reduce production. The ergonomic matrix presumably evolves toward higher fitness states by the genetic alteration of the relative frequencies and behavioral patterns of the castes and the details of their interactions. The matrix is therefore the scaffolding of the superorganism. Our growing knowledge of caste determination, communication, and feedback loops will make complete sense only when these individual-level processes are correctly placed on the scaffolding.

To summarize, the new holism subsumes explanations of three kinds that can be aligned with one another: the relative adaptiveness of the colonies as superorganismic operating units within their natural environments, the ergonomic matrix that determines the optimal or at least evolutionarily stable mix of castes and communication systems, and the details of the castes and communication systems themselves.

Homeostasis and flexibility
The concept of the colony as an organized whole also implies a norm of reaction in each of the categories of social organization. That is, the genotype prescribes a particular caste system and a particular communication system that emerge in response to the environment in which the colony grows up. In some instances the norm of reaction is narrow, so that only one social response is to be expected regardless of the environment. We then speak of the response as rigid or stereotyped. In other instances the response differs from one environment to another but in a consistent and evidently adaptive manner. In the latter case the norm of reaction is broad, based upon a flexibility in worker behavior or colony development. A useful corollary distinction can be made between “noise,” the irregular fluctuation around the norm of reaction caused by changes in the environment and accidents in development, and the regularized shift in social organization that adapts the colony to new circumstances. In the case of noisy perturbations, the colony tends to return to the original state by means of negative feedback loops in physiology and behavior. In the case of adaptive shifts to new positions in the norm of reaction, the colony uses physiological and behavioral responses to move to alternative modes of social organization.

Both procedures constitute homeostasis in the broad sense. Whether a social process constantly readjusts itself back to the status quo ante or shifts to a new adaptive state, it permits the basic life functions such as regulation of temperature and care of the larvae to proceed with little interruption. To use the traditional expression, the life functions are homeostatic, and the social responses have evolved into either stereotyped or flexible as enabling devices to achieve their end.

An example of social homeostasis is provided by the caste systems of the ant genus Pheidole. Each species of this large cosmopolitan genus has a characteristic ratio of small-headed minor workers to large-headed major workers (see Figures 7-38 and 8-21). When the ratio is altered in a particular colony by an excess of birth or mortality in one of the castes, the colony converges back toward the original ratio within one or two worker generations, extending across one to three months. The feedback loop is an inhibitory pheromone produced by majors that lowers sensitivity to juvenile hormone, so that larvae surrounded by an excess of majors curtail growth and tend to become minors. Those present during a shortage of majors become more sensitive to juvenile hormone, extend growth, and turn into majors.

A second form of social homeostasis is displayed by the behavior (as opposed to the physiology) of the Pheidole major workers. When the relative frequencies of adult minors and majors are anywhere near the species norm, the majors are inactive most of the time. When they do work they keep to the specialized roles of nest defense, ingluvial storage of liquid food, and milling of seeds. In contrast, the minors take responsibility for virtually all of the nursing of immature forms, nest construction, and other quotidian tasks. However, when the minor/major ratio is lowered to below 1:1 (from between 3:1 to 20:1, the norm according to species), so that minors are much scarcer than usual, the repertory of the majors suddenly and dramatically expands to cover most of the tasks of the minors. Within less than an hour, majors can be seen attending brood, handling nest material, and in general acting like rather clumsy, overweight minors. Under these transformed circumstances, the number of kinds of acts has been observed to increase by 1.4X to 4.5X according to species. The rate of activity, measured in number of acts of all categories performed in each unit of time, increased by 15X to 30X (Wilson, 1985a; see Figure 9-5). By this means the majors restore about 75 percent of the missing minors' activity. Even when all of the minors are taken away, the majors are able to rear brood to maturity and otherwise keep the colony going until a substantial minor force is restored.

In such cases as the stand-by role of the Pheidole majors, a distinction can be made between the programmed elasticity in the repertory of individual workers and the resiliency of the colony as a whole, which depends upon the pattern of caste-specific elasticity. The homeostasis achieved is in the vital functions of the colony, sustaining a near-normal load of brood care and other tasks until the normal caste ratios are restored. The signal that informs the Pheidole majors when to remain quiescent and when to spring into action is a ritualized form of aversion on the part of the majors (Wilson, 1985a). They actively avoid minors while in the vicinity of the larvae and other immature states. However, majors do not turn away from each other near the immature stages, and they avoid neither minors nor majors while in other parts of the nest. For their part, minors do not avoid either minors or majors anywhere in the nest. When minors are scarce enough, majors move onto the piles of eggs, larvae, and pupae and commence attending to them. At the same time they commence other tasks and greatly increase their rate of self-grooming. The result of this seemingly simple rule of procedure is a striking division of labor when colonies have a normal caste composition, followed by a homeostatic shift of behavior by the majors when minors become unusually scarce.

A similar aversive phenomenon occurs in the leafcutter ant Atta cephalotes. Most of the foragers engaging in active cutting of leaves and other vegetation are in the media size group, with head widths between 1.8 and 2.2 mm (see Figures 8-29 and 8-30). When 90 percent of these medias were removed experimentally from laboratory colonies by Wilson (1983a), the rate of harvesting remained unaffected, due to the fact that excess workers in the adjacent size classes were already present on a stand-by basis in the foraging area. In addition, the survivors in the 1.8-2.2 mm group increased their individual activity rate by approximately five times. In a manner reminiscent of dominance by the Pheidole minor workers, the prime foragers in the Atta 1.8-2.2 mm group tended to displace the others from the edges of the leaves, where most of the cutting takes place. When these individuals were removed, the auxiliaries participated more freely.

Cases of social homeostasis that return the colony repeatedly to the status quo ante have been discovered with increasing frequency in recent years. When a colony is disturbed, to cite another category, only a small percentage of the workers respond with all-out attacks on the intruder. Le Roux and Le Roux (1979) found that only about 20 percent of Myrmica laevinoda workers attacked an approaching worker of Myrmica ruginodis. Another 30 percent approached it without violence, and the remainder did not visibly respond. When the ants were then rearranged into groups more nearly homogeneous with respect to aggressivity, the behavior of enough ants changed so that the percentage of aggressive workers approached that in the original ensembles. Finally, when the ants were restored to their original groups, there was a tendency for individuals to retain the level of aggressivity they had assumed in the altered groups. In the end the proportions of violent and nonviolent ants were not greatly changed.

The speed with which workers alter their individual “idiosyncratic” behavior in response to stress varies according to both the behavioral category and the context. Queens and very young workers of Myrmica rubra, like those of most other ant species, do not participate in brood transport during colony emigration. However, they readily do so when isolated from the older workers that normally take responsibility for this task, and they accomplish the task in only a little less time. When the older workers are separated into elite and non-elite groups with reference to this task, the two ensembles converge in their levels of activity, but only after a period of three months (Abraham et al., 1980, 1984). Moderate flexibility of a similar kind has been documented in Camponotus vagus by Bonavita-Cougourdan and Morel (1987).

The degree of flexibility can change with the size or age of the colony. Young workers of Lasius niger in founding colonies were found to be more flexible in behavior and slower to specialize than those in more mature colonies (Lenoir, 1979a). Workers of young colonies of Pogonomyrmex barbatus are variable in their response to perturbations, but this greater flexibility--or more precisely variability--results in a less rapid return to the status quo ante (Gordon, 1987). In other words, individual workers of young colonies appear to be more versatile, but the colonies as a whole are less homeostatic. Thus the phenomenon of adaptive demography, entailing programmed changes in the size and age of workers as the colony grows older, is augmented by programmed changes in the pattern of behavior of the individual workers.

Adaptive shifts in behavior
The most conspicuous form of adaptive shift in social organization, as opposed to changes in behavior that merely restore preexisting social organization, is in the degrees of polydomy (number of nests per colony) and polygyny (number of queens per colony or local nest). Repeated studies of Iridomyrmex and Leptothorax  have revealed how variation in these traits within single species arises from flexible procedures in the modes of colony founding and subsequent dispersal of colonies to multiple nest sites by means of budding (Buschinger, 1967, 1968c; Alloway et al., 1982; Greenslade and Halliday, 1983; Hölldobler and Carlin, 1985). Most of the analysis has depended upon direct observation and the correlation of polygyny and polydomy with environmental variables in the field. However, Herbers (1986b) also used field experimentation in her studies of the North American Leptothorax longispinosus. She seeded a forest tract with artificial nest sites consisting of wooden rods with holes drilled in them. During the ensuing two years, the number of both queens and workers per nest declined in comparison with unseeded plots, reflecting an expansion of existing colonies into the newly available nests and hence a thinning effect of the overall population. At the same time, the total population appeared to increase in the experimental plot, suggesting that it was undergoing growth as the colony fragments multiplied.

Some ant species adjust the time of principal foraging activity dramatically in response to weather, the presence of food, and the availability of foragers at any given time (see also Chapter 10). In tropical habitats daily and seasonal changes in foraging activities of ants are related both to temperature (Torres, 1984a,b) and moisture (Levings, 1983; Levings and Windsor, 1984). Documented examples include Paraponera clavata (McCluskey and Brown, 1972; Harrison and Breed, 1987), Acromyrmex octospinosus  (Therrien et al., 1986), and Atta cephalotes  (Cherrett, 1968). Other species, such as Monacis bispinosus, have been observed to shift their foraging schedules in response to invasion by workers of a competing ant species (Swain, 1977). Pheidole titanis, an ant that hunts termites in the deserts and deciduous thorn forests in the southwestern United States and Mexico, changes the diel timing of its raids to avoid a parasitoid phorid fly (Feener, 1988).

Several cases have been reported in which ant species modify their nests in response to parasites or predators. Workers of Formica subsericea plug their nest entrances with soil, pebbles, or grass after being raided by the slavemaking ant Formica subintegra. They also remove the traces of excavated soil and discarded cocoons that otherwise typically litter the surface of Formica subsericea nests (Talbot and Kennedy, 1940). Similarly, colonies of Myrmecocystus mimicus, endangered by raids from neighboring colonies of the same species, close their nest entrances and cover the surroundings with sand. In so doing they cover up colony specific nest markers (Hölldobler, 1981a; and unpublished observations). Workers of the harvester ant Pogonomyrmex rugosus under persistent attack by the western widow spider (Latrodectus hesperus) respond by closing their nest entrances and decreasing or halting their foraging activity, even when less than 0.2 percent of the population is being lost per day (MacKay, 1982b). An identical behavior has been recorded in the Florida harvester ant, Pogonomyrmex badius, which is beleaguered by the theridiid spider Steatoda fulva (Hölldobler, 1971e; Figure 9-6). Workers of the desert genus Cataglyphis in North Africa also close their nest with soil when attacked by hunting spiders (Oxyopis), which are among their most important enemies. The spiders often overcome the obstacle by removing the earth, particle by particle (Harkness and Wehner, 1977).

Variation among colonies
Given even a small amount of such flexible behavior, it is not surprising to find considerable differences in the overall social behavior of ant colonies belonging to the same species. The variance is enhanced by differences in queen fertility, amount of brood present, and other qualities that vary from one week to the next within the same colony. Nor can genetic differences be discounted, as demonstrated by the results of experiments on the caste ratios of Pheidole dentata (Johnston and Wilson, 1985). Strong intercolony variation in the “daily rounds” of Pogonomyrmex badius was discovered by Gordon (1983a), including the numbers of workers engaged in patrolling, nest maintenance and foraging, as well as the time of peak activity. Glunn et al. (1981) found striking differences in the dietary preferences of colonies of fire ants in both the field and laboratory. Herbers (1983) found so much variation in the ethograms of colonies of Leptothorax ambiguus and |L. longispinosus  respectively that the two species could be only tenuously separated on the basis of the behavioral data alone. Research on such variation and its multiple causes is obviously in its earliest stages and will no doubt provide new surprises in the year immediately ahead.

Positive feedback and runaway reactions
Positive feedback loops, in which the product of a process increases the rate of reaction leading to more of the same product, are relatively rare in biology. Where negative feedback loops decelerate processes in such a way as to regulate the product at a more or less constant level (hence --homeostasis), positive feedback tends to produce a runaway reaction that ends only when supervenient negative loops are activated or else the materials needed to create the product become exhausted.

An example of positive feedback is the stimulation cited earlier of queen egg-laying by the presence of mature larvae in the fire ant Solenopsis invicta (Tschinkel, 1988c). The presence of mature larvae causes the queen to lay more eggs that lead to still more mature larvae, who stimulate the queen to lay even yet more eggs, and so on. But the system does not run away, because when large larvae become numerous enough, it is expected (although not yet proved) that further production will be slowed by the workers' inability to feed them. Also, the stimulative effect per larva decelerates as more larvae are added.

A possible positive feedback leading to a runaway reaction does occur in the nuptial flights of the carpenter ant Camponotus herculeanus. When males become sufficiently aroused, they release a pheromone from their mandibular glands that stimulates both other males and the virgin queens, causing these reproductive forms to fly away from the nest during a short interval of time (Hölldobler and Maschwitz, 1965). Whether or not detection of the pheromone itself causes males to release still more of the pheromone remains to be established.

Still another form of a runaway reaction appears to occur during absconding by colonies of the myrmicine ant Pheidole dentata under attack by fire ants, which are among their principal enemies (Wilson, 1976b). When Pheidole dentata minor workers encounter no more than one or a few fire ants near their nest, they recruit major workers to the site. This reaction usually results in an early destruction of the invaders. But when the minor workers run into large numbers of fire ants, a second phase of defense is initiated. A few grapple with the intruders but most scatter outward in all directions. Only a minority of those fleeing return directly to the nest, and proportionately few odor trails are laid even by these individuals. Consequently the fire ants are able to move quickly to the vicinity of the Pheidole nest. There they encounter stiffening resistance as more minor workers are encountered, a larger number of trails are laid, and a growing force of major workers is assembled. Thus the second phase of defense is only an extension of the first conducted in another location. It may seem paradoxical that when faced with more enemies the colony should recruit majors less efficiently. But this weaker response applies only to the territory farthest from the nest. The result of importance is that the majors are committed to battle close to the nest, where in fact they are now most needed.

If the second phase of defense begins to break down, the colony absconds, which constitutes the final, desperate, yet very effective maneuver (see Figure 9-7). The following events lead to absconding. As the fire ants crowd closer to the nest entrances, an increasing number of Pheidole minor workers lay odor trails all the way back to the brood areas. As a result the excitement of the nest workers increases, and many begin to pick up pieces of brood and carry them back and forth. Meanwhile, the defense perimeter continues to shrink as Pheidole majors are immobilized in combat and more Pheidole  minor workers are either immobilized or flee from the nest area after contacting fire ants. Activity within the brood area builds up, sometimes at an apparently exponential rate, and minor workers then begin to run out of the nest, many laden with pieces of brood. At first there is a strong tendency for the refugees to loop back and reenter the nest, but if they continue to encounter fire ants they break away entirely and flee outward. No particular direction is taken by individuals during the exodus, so that the colony ends up scattered over a wide area. The queen also departs under her own power. Later, after settling down, she attracts minor workers who cluster around her. No direct physical transport of one adult by another has been observed. When the fire ants are removed, the Pheidole adults slowly return to their nest and reoccupy it.

In summary, there appear to be two processes contributing to the relative suddenness and en masse quality of absconding:

1) As fire ants press in, more and more Pheidole of both castes are killed, while other minor workers simply desert the area. This leads to a steadily decreasing ratio of defenders to invaders and a shrinking of the perimeter of defense.  A point is reached in which the nest workers are contacting fire ants at a sufficiently high rate to cause them to seize brood pieces and leave.  This final critical level is approached steeply.

2) Simultaneously, there is an exponential increase in the rate at which minor workers run back and forth between the vicinity of the nest entrance and the brood chambers. Since many lay odor trails, there appears to be a buildup in the concentration of the pheromone.

In short, absconding is an explosive social response created by the reciprocal acceleration of alarm and contact with the invading ants. Each process enhances the other until they attain the threshold level at which absconding occurs. This final action dissolves the buildup that created it.

Caste and behavioral flexibility
In Chapter 8 we argued from the theory of the allometric space that the number of castes in the optimal mix can be expected to increase as the logarithm of the number of tasks performed by the colony. It is further true that a very sensitive relation exists between the width of the repertory of individual workers, in other words their moment-to-moment behavioral flexibility, and the number of castes in the optimal mix. The greater the flexibility, the fewer the castes. Moreover, the transition is readily made from a polymorphic state, in which multiple physical castes are specialized on sets of tasks, to a monomorphic state, in which a single physical caste addresses all tasks. In some instances all that is required is a slight increase in either behavioral flexibility or cooperation among the caste members (Oster and Wilson 1978; Traniello, 1987; West-Eberhard, 1987b). The theory can be briefly summarized as follows, using the version of the allometric space displayed in Figure 9-8. The shaded area is the “reach” of an ant belonging to a particular caste, where the caste is defined by some pair of physical traits x and y on the allometric plane. As behavioral flexibility increases, and worker anatomy becomes less specialized, or else cooperation increases, the shaded area expands. This means that workers of that caste can perform more and more tasks with at least the minimal required amount of efficiency.

Now suppose that the tasks are distributed at random over the allometric space. As behavioral flexibility or cooperation increases, there is a surprisingly abrupt transition from an optimal state of polymorphism, comprising multiple physical castes, to monomorphism, in which only a single physical caste is optimal (Figure 9-9). The principal inference from this argument is that for a given degree of physical specialization, a small change in behavioral flexibility or cooperativeness can result in a shift in the colony optimum from polymorphism to monomorphism and back again. The phenomenon is reminiscent of “phase transition” curves in physics, which characterize sudden condensations, shifts from order to disorder, and other abrupt changes of important magnitude. As suggested in Figure 9-10, the trade-off between physical and behavioral castes is not proportionate: the strong selective pressures for morphological differentiation should disappear rather quickly as behavioral complexity increases. The same is true as the ability of members of the same caste to cooperate increases, enlarging the social coverage of the allometric space.

If such an evolutionary “phase transition” truly exists, we should expect genera and other sets of phylogenetically closely related species to display striking variation in the degree of worker polymorphism. This proves to be notably true in the case of ants. Of the 45 genera known to have polymorphic species, at least 11 also contain some monomorphic species (Azteca, Crematogaster, Formica, Iridomyrmex, Monomorium, Myrmecia, Neivamyrmex, Pogonomyrmex, Solenopsis, Strumigenys, and Tetraponera). Some other genera that are fully polymorphic also possess strong interspecific variation in the degree of polymorphism. The species of Camponotus, for example, range over the entire gamut from feeble monophasic allometry and modest size variation to complete dimorphism.

It is possible to envisage three very different means by which the transition occurs in the direction of monomorphism. First, the behavior of workers in particular age-size cohorts can simply become more complex and flexible. The second means is by the addition of temporal castes; the members of a given size cohort now specialize according to age. This second mode, which is followed to some degree by virtually all of the ants and other higher social insects, has the virtue of allowing reversible specialization. Individuals are not frozen within any particular repertory. When the needs of the colony demand it, workers can switch to a new behavioral regime and even, given a few days time, reactivate dormant exocrine glands. Finally, members of the same age-size cohort can add to their repertory by cooperative action, such as moving large prey items in a coordinated manner or defending the nest against enemies too formidable to be subdued by a single member.

Tempo
An important property first analyzed by Oster and Wilson (1978) is the tempo of activity, which varies enormously among different species of social insects. The workers of some ant species walk slowly and with seeming deliberation. As they make their rounds among the brood or forage outside the nest, they appear to waste few movements. Examples of such “cool” species include many members of the Ponerinae, most species of the myrmicine tribe Attini, Basicerotini, and Dacetini, and the large Neotropical members of the dolichoderine genus Dolichoderus. In contrast, the colonies of army ants, fire ants, and species of the dolichoderine genera Conomyrma, Forelius, and Iridomyrmex literally seethe with rapid motion. The workers appear to waste substantial amounts of time canceling one another's actions. One ant in such “hot” colonies may run in one direction with a pupa during colony emigration, another in the opposite direction. Additional time seems to be wasted by colony members who run back and forth to nest sites empty-mandibled.

Although the matter has not been subjected to appropriate measurements, a positive correlation appears to exist between tempo and the mature colony size of species, even when evolutionary “inertia” due to phylogenetic similarities is taken into account. In Chapter 8 we established that a loose correlation exists among species between the mature colony size and the complexity of the physical caste system. Does a causal relation then exist between the three variables--between tempo, colony size, and degree of polymorphism? Perhaps species with small colonies must consist of workers that are more “careful”. In order not to deplete their numbers to a level fatal to the colony as a whole, they must act with greater deliberateness and precision, even at the expense of less productivity. Such species live in circumstances in which colony size is necessarily small for other reasons, such as the low density of food items or the small size of the preferred nest site. Under these conditions colony growth is expected to be relatively slow and worker longevity greater. Such colonies are the analogs of K-selected species. The r-selected colonies of other species are better fitted to a niche in which the premium is on large mature colony size, rapid colony growth, and a high turnover of workers. Such colonies can afford some degree of inefficiency as the price of their large size and high rate of exploitation of the environment. They are “labor-saturated” in the sense that each task is attended by many individuals. With such increased redundancy, the reliability theorems described earlier ensure that the system reliability will be high even if each individual is performing erratically.

Pasteels et al. (1982) and Deneubourg et al. (1983) have pointed out that “errors” committed by ants during recruitment actually have other adaptive advantages when colonies are labor saturated. Because the responses of individual ants are probabilistic, with a large random component during each moment of time, the mass response of foragers can be adjusted by altering probabilities of individual response through changes in recruitment signaling. In a tightly deterministic system, the same mass adjustment would require a much more sophisticated system beyond the brain capacity of insects. Also, errors during recruitment could allow the ants to discover nearby food sources. Finally, a scattering of ants in this manner allows the colony to follow and capture moving prey more efficiently than if the recruitment were highly precise (Wilson, 1962b).

One coadaptation favored by high tempo in colonies is a more differentiated caste system: specialized castes replace the “careful” generalized workers of the low-tempo colonies. Also, since workers of such high-tempo societies live for shorter periods of time, and often will not even survive to the reproductive season, they will be more likely to lack ovaries. Their differentiation into anatomical castes further increases the likelihood of their becoming sterile. These circumstances might account more precisely for the association between large colony size, worker polymorphism, and the lack of worker ovaries that has come to light in separate studies.

Two environmental conditions can lead to the maximization of energy yield at high tempo. The most likely condition is the existence of a relatively rich food source that occurs unpredictably in space. The rate of performance in items retrieved per worker will be the multiple of the frequency of occurrence of the items, the probability of encounter, and the probability of a successful retrieval following each encounter. An increase in tempo will increase the probability of encounter but is likely to decrease the retrieval rate. Such a strategy can be expected in species that capture a wide array of arthropod prey and hence form larger, more polymorphic colonies. In contrast, a species specialized on a few rare, difficult species of prey is likely to adopt a low tempo: the search for the items is careful and aided by special sensory devices, and the assault is deliberate, achieved either by stealth or recruitment of nestmates. Foraging, in other words, is more specially tailored to the prey. A species adapted to predictable and rich food sources, such as aggregations of honeydew-producing aphids, is also likely to operate at a low tempo. The workers need to invest very little in reconnaissance but a great deal in the protection and exploitation of the resource.

The second circumstance leading to a high tempo, in theory at least, is a relatively low loss of energy due to the activities of competitors and predators. Some energy expenditure is of course unavoidable due to metabolism, and the expenditure will accelerate as a function of tempo. If the additional cost due to injury and mortality inflicted by enemies is high so that the energy curve rises even more steeply as a function of tempo, the optimum tempo of the colony will be correspondingly lower. In other words, there will be a tendency to decrease tempo in order to compensate for energy lost to the colony through competition and predation.

Learning
A great deal of the flexibility observed in the behavior of individual ants can be attributed to learning. There is no hard and fast line between innate patterns and learning. It is true that some complicated responses, such as self-grooming and regurgitation, appear to be wholly programmed so that the insect performs them more or less expertly with no prior experience. Other responses are altered according to experience. But even when learning is required, it is genetically “prepared”--that is, certain responses are learned more readily than others. The degree of preparedness in learning varies greatly according to category of behavior. Some forms of behavior, such as the differential aggression toward certain kinds of enemy ants, represent little more than shifts in the intensity of otherwise wholly programmed responses (Carlin and Johnston, 1984). Others, including visual orientation by landmarks, incorporate a sophisticated memorization of details (Jander, 1957; Wehner et al., 1983). Still others entail a complex integration of information on the colony's nutritional status and available food in the environment (Deneubourg et al., 1987; Traniello, 1987b, 1988).

The simplest form of learning in ants, like that in other animals, is habituation, the diminution of the intensity of the response as a result of experience. When an ant colony is disturbed by the intrusion of a glass rod into the nest, the workers launch an attack. If the disturbance is repeated at close intervals, say every hour afterwards, the ants grow steadily less responsive. In this fashion ant workers can be “tamed,” especially if they are kept away from the nest. Even large, venomous species can be picked up, allowed to walk over the hand, and fed (carefully!) with sugar water. Habituation appears to be an adaptive process in which the individual comes to recognize less dangerous stimuli as such, enabling it to adjust at a calmer and more efficient tempo. In Schneirla's experiments on maze learning in Formica (1941), the workers first went through a stage of habituation (“generalized stage”). During this time progress occurred through a simultaneous decrease in excited and erratic behavior and an increase in the tendency to continue running when obstacles were encountered.

Ants are also notably capable of associative learning. In this second major category of behavioral modification, the ants acquire conditioned responses by the association of rewards (such as sugar water or attractive pheromones) with previously meaningless stimuli. This is the phenomenon used by Karl von Frisch to measure the sensory capabilities of honeybees. He and subsequent investigators recognized that if a worker bee can be trained to associate a given wavelength of light or odor with food, the linkage is ipso facto evidence that the bee can detect that particular stimulus. The great majority of cases of learning that have been demonstrated in ants can be loosely classified as associative learning. Some of the effects also constitute imprinting, which is learning that occurs mostly or exclusively during brief “sensitive periods” in the life cycle and can be reversed later only with difficulty, if at all. The most striking example of imprinting by simple association is the learning of colony odors (Chapter 5). The period of greatest sensitivity is usually in the first days after eclosion of the young adult from the pupa, although Isingrini et al. (1985) have provided evidence in the case of Cataglyphis cursor that the learning can begin during the larval stage. Carlin and Schwartz (1988) obtained a similar result in Camponotus floridanus. There is additional evidence that when no contact with nestmates is permitted during the sensitive period, later social behavior can be seriously disturbed. Lenoir (1979a,c) found that young workers of Lasius niger not allowed to contact larvae after eclosion tended to become foragers at an early age. A few hours of later exposure to larvae were enough to cause them to transport larvae in a normal manner. However, workers kept in complete social isolation were aggressive and could be integrated only with difficulty into their colony of origin. In other experiments, cocoon care was abolished altogether by keeping workers of Formica polyctena or Formica rufa from cocoons for three weeks after eclosion (Jaisson, 1975; Jaisson and Fresneau, 1978; Le Moli, 1978; Le Moli and Passetti, 1978). An imprinting-like mode of learning might also play a role in a different socioecological context. Experimental results obtained by Jaisson (1980) indicate that in certain ant species early experience can induce an environmental preference in colony-founding queens. Jaisson suggests that this might explain specific associations that can be found between ants and plants. Similarly, Goodloe et al. (1987) found that workers of the North American slavemaking ant Polyergus lucidus prefer to raid nests of the same species of Formica as that already represented by slaves in their nest.

Some associative learning appears to be little more than the enhancement of innately programmed responses. Colonies of Pheidole dentata react more strongly to invasions by fire ants (Solenopsis invicta) than to invasions by most other kinds of ants, even in the absence of earlier experience. After repeated exposure to Solenopsis invicta and Tetramorium caespitum, the Pheidole increase their response to both species, but much more strongly in the case of the Solenopsis (Carlin and Johnston, 1984). A similar enhancement occurs in species of the “Novomessor” group of Aphaenogaster after experience with their own arch enemies, army ants of the genus Neivamyrmex (McDonald and Topoff, 1986). In a parallel category, workers of Myrmica rubra kept in the presence of queens from the time of eclosion exert more control over the growth of larvae than do those kept apart from queens. A critical period exists in which the young workers become sensitized, after which they are more likely to use aggression and neglect to suppress development of larvae into the queen caste (Evesham, 1984b).

Ants are capable of feats of learning considerably more sophisticated than simple association when finding their way through the environment. Workers of Formica pallidefulva studied by Schneirla (1953a) learned a six-point maze with relative ease at a rate only two or three times slower than that achieved by laboratory rats. Workers of Formica polyctena can remember their way through mazes for periods of up to four days (Chauvin, 1964). Those of Formica rufa, operating under more natural circumstances, can simultaneously memorize the position of four separate landmarks and remember them well enough for use in orientation for as long as a week (Jander, 1957). The latter result was greatly extended by Rosengren (1971), who discovered that the ants can remember specific locations outside the nest over winter.

A complex process also takes place in the brains of ants during exploratory foraging trips. An outward bound worker winds and loops in tortuous searching patterns until it encounters food. But then it takes a direct route, the “bee-line,” in its return trip to the nest. This feat is made possible by sun-compass orientation, that is, using the sun as the reference point for guided movement. Felix Santschi, who discovered the phenomenon in 1911, was attracted by the problem of how workers of desert ants in Tunisia manage to leave their nests, forage at a distance, and then find their way back home over the featureless desert terrain, even when strong winds make odor trails impracticable. He found the answer by means of his now famous “mirror experiment.” When workers of Aphaenogaster, Messor, and Monomorium returning home with booty were shaded from the sun on one side and presented with the image of the sun by means of a mirror held on the opposite side, they reversed their direction 180° and headed confidently away from home. When the shade and mirror were removed, they again turned about 180° and ran homeward. In other words, it was apparent that the ants were reckoning the angle subtended by the sun and the nest and holding it constant as they returned home. Later von Buddenbrock (1917) established that this form of orientation occurs widely among insects, and it has since been discovered in many other invertebrate groups as well. For nocturnally foraging ants a moon-compass response is equally feasible, and has, in fact, been demonstrated in Monomorium by Santschi (1923) and in Formica by Jander (1957).

Santschi also tested the desert ant Cataglyphis bicolor, a common species in Tunisia. The workers forage individually and often travel more than 100 meters away from the nest over featureless terrain. Surprisingly, however, the mirror experiment did not work very well with this species and Santschi (cited in Wehner, 1982) concluded that different cues in the sky other than the sun must serve Cataglyphis for celestial navigation. Almost 40 years later Karl von Frisch discovered these cues: he found that insects can perceive the polarized light in the sky and use its pattern for orientation. Subsequently Wehner and his collaborators, using exacting experiments, proved that the Cataglyphis use the specific pattern of polarized light from the sky to navigate across a featureless plain (Wehner, 1982, 1983a; Fent and Wehner, 1985; Wehner and Müller, 1985). Previously Vowles (1950, 1954) also demonstrated orientation to polarized light in the ant Myrmica ruginodis. Other investigators established its existence in Tetramorium, Tapinoma, Lasius and Camponotus (Gertraud Schifferer in von Frisch, 1950; Carthy, 1951a,b; Jander, 1957; Jacobs-Jessen, 1959).

Since the sun moves through an arc of approximately 15° every hour, what will happen if an ant is trapped on its way home and not permitted to see the sun or portion of the sky for a substantial period of time? If it always keeps a constant angle to the sun regardless of the passage of time--what German investigators call winkeltreue orientation--then the error the trapped ant makes on being released again should equal the arc through which the sun has passed in the interim. Brun (1914) performed just such an experiment with workers of Lasius and found that they did behave as though complying with the winkeltreue rule. For example, a worker was confined in darkness from 4:00 to 5:30 one afternoon, during which time the sun moved through an arc of 22.5°. When released, it set off in a direction that attempted to follow the original angle to the sun and consequently deviated from its original, true path by 23.5°, approximately the amount the sun had traveled. This would seem to be a relatively poor way for an ant to get around, especially if it spends hours at a time away from its nest, and Brun's result never seemed to provide a full explanation of visual orientation. In 1957 Jander showed that experienced Formica rufa workers are really able to do much better. They duplicate the feat of the honeybee whereby they keep track of the sun's movement and constantly adjust the angle of their return journeys. Newly eclosed rufa workers and those which have just emerged from overwintering must learn the sun's movement; until they accomplish this, they orient to the sun in the winkeltreue manner. It is not known whether Brun's Lasius were similarly naive, but at least his results are not inconsistent with the time-compensated orientation demonstrated in Formica by Jander.

It is a matter worth further reflection that ants do not need to see the sun in order to perform time-compensated celestial orientation, in other words true compass orientation. Wehner and his collaborators recently worked out the mechanisms by which insects analyze the conspicuous pattern of polarized light displayed on the celestial canopy and use it for compass-orientation (Rossel and Wehner, 1982, 1984; Wehner and Rossel, 1985; see Figure 9-11a,b). Wehner (1983a) describes it in the following way:

In analyzing skylight patterns insects do not behave like human astrophysicists. Neither do they perform spherical geometry in the sky, nor do they come programmed with a set of skylight tables which would inform them correctly about all possible e-vector patterns occurring in the daytime sky. Thus, they do not seem to rely on some abstract knowledge about the laws of atmospheric optics, but get along quite well without such mathematics. What they use instead is a simple celestial map based on the most remarkable feature of skylight polarization: the intrinsic symmetry line of the e-vector patterns which is identical with the line of maximum polarization [Figure 9-11c]. In the real sky, this line is confined to that half of the sky that lies opposite to the sun, and so is the insect's celestial map [Figure 9-11d]. Furthermore, while the actual skylight patterns vary during the course of the day, the insect's celestial map does not change, but is used in the same way at every time of the day. It only rotates about the zenith as the sun moves across the sky (Wehner and Lanfranconi, 1981).

Using the honeybee as an example, but aware that identical results have been obtained with the ant Cataglyphis, Wehner continues:

In operational terms, the insect's celestial map is analogous to what can be called a neural template. Imagine that while steering a particular compass course the bee aligns the symmetry line of this template with the anti-solar meridian in the sky. All it must do in later trying to reestablish the former compass course is to match the template as closely as possible with whatever e-vector pattern it experiences in the sky. Let us recall, however, that the insect's template is a generalized rather than detailed copy of the actual skylight patterns. Using such a generalized map based on a simple rule [Figure 9-11c] implies that under certain skylight conditions navigational errors must occur. On the other hand, such a rule provides the insect with a rather simple strategy of reading compass information from complicated and ever-changing skylight patterns. An intriguing hypothesis is that decisive features of the map are already determined by the geometry of the receptor array within the insect's retina (for the geometry of these arrays in bees and ants see Wehner, 1982).

On the basis of his experiments with Formica, Jander suggested that the foraging worker performs a continuous series of calculations analogous to the simplest mathematical operation. As it runs outward, the ant continuously perceives the location of the sun and remembers the angle it takes relative to the sun during each of its twists and turns. For each new direction taken, the product of the angle to the sun times the duration of the outward leg of the run is calculated, and the sum of all these products is divided by the total running time to produce the average (weighted) movement angle to the light. When the insect is ready to come home, it need only reverse this mean angle by 180°.

In studies of the desert ant Cataglyphis bicolor, Wehner and his co-workers found an even more subtle interplay of cognitive processes (Wehner et al., 1983; Fent and Wehner, 1985; Wehner and Müller, 1985). When colonies are located on flat desert terrain, with no distinguishing landmarks in the vicinity, the workers employ dead reckoning in a manner similar to that found in Jander's laboratory studies of Formica. That is, they depend on vector navigation by celestial compass. Fent and Wehner also discovered that the Cataglyphis employ both the compound eyes and ocelli in this form of orientation. When the nests are located in terrain with rocks, bushes, or other emergent features, the ants rely substantially on these objects as landmarks. Experiments with artificial landmarks by Wehner and his co-workers showed that the workers use two-dimensional memory images of their surroundings as seen from the nest. When moving back and forth between the nest and foraging areas, they employ a sequence of such memory images. There is no evidence that the ants can integrate a true topographic map in the human manner. Nevertheless, each Cataglyphis worker is able to remember several such sites simultaneously and to travel to each one of them on separate occasions.

Memories of visual images also play a role in the “canopy orientation” of Paltothyreus tarsatus, a large ponerine widely distributed in African forests south of the Sahara (Hölldobler, 1980). The foraging workers memorize particular configurations of branches and leaves in the foliage overhead and use them to find their way back to the nest. To reveal this remarkable phenomenon, Hölldobler took photographs of a Kenyan forest canopy with a wide-angle lens and suspended large prints of it over the foraging arenas of laboratory colonies. When the prints were rotated, homeward bound workers shifted their compass direction in corresponding degree.

Ants learn more than visual orientation cues during orientation. Many species are arboreal and forage, often exclusively, in a three-dimensional maze of branches and twigs in the tree canopy. In this situation orientation based on gravity perception is at least as important as visual orientation. In 1962, Markl discovered a gravity receptor system in ants. It consists of rather simple-looking clusters of external sensory bristles, one pair on the neck (or, more accurately, lobes of the episternum that protrude into the neck region), two pairs on the petiole, and similar organs on the legs and antennae (Figure 9-12). As the ant shifts her position, the head and gaster swing on their respective articulations with the thorax and petiole, the segments of the antennae flex, and the body bends slightly on the upper leg articulations. These minor movements press the sensilla of the hair plates to one side or another with the degree varying according to position. The shift causes the underlying neuron clusters to free differentially. Markl (1962) was able to train workers of Formica polyctena to walk up slopes of as little as 2° from the horizontal. He also succeeded in training them to maintain a constant angle with reference to gravity while running up and down a vertical surface.

The cases of associative learning that have been documented in ants thus far have been passive in nature, requiring only that the individual experience a new, unconditioned stimulus in conjunction with an unconditioned, innate stimulus that carries either a positive or a negative reinforcement. Another form of associative learning that contains greater possibilities for behavioral evolution is operant conditioning. The process begins when the animal performs a new act by accident or exploratory movement that “pays off” in providing reinforcement. Thereafter the animal either repeats the act or avoids repeating it, depending on whether it was rewarded or punished the first time. Operant conditioning is important because it allows individuals to realize the potential of their basic innate programs of behavior more fully. It also enlarges the possibility of new behavioral evolution through genetic assimilation. In the latter phenomena, the new act provides an advantage in natural selection, so that genotypes with the capacity to perform the act--or still better, the tendency to perform it --spread through the population and alter the future pattern of responses of the member organisms. Operant conditioning appears to occur in foraging honeybees and bumblebees as they perfect their harvesting techniques on different species of flowers (Heinrich, 1984). To the best of our knowledge no examples have been reported in ants, but this lack could be due merely to the failure of past investigators to search for them. Behavioral skills likely to be improved by operant learning include the search and pursuit of prey, the handling and transportation of booty, and the manipulation of nest material and brood.

By the same token it is not known whether ants partition learning into short-term memory and long-term memory. Experiments with honeybees have shown that workers visiting flowers first store information in short-term memory and then transfer it to long-term memory. However, if a short-term memory is not reinforced by additional experience within minutes, it is lost. Erber (1976) and Menzel (1985) speculated that this rapid decay prevents the bee from registering the signals from uneconomical types of flowers into its long-term memory. Menzel (1979) further concluded that short-term memory permits a refinement of incoming information, allowing the bee to continue testing flowers until each type has been encountered enough times to judge its relative productivity. In other words, a single experience is not sufficient to commit the forager permanently to a less desirable flower type. It seems likely that a similar strategy is employed by ants during tasks such as nest excavation, larval feeding, and liquid food transfer, but the possibility has not been experimentally tested.

At least a few ants are capable of learning to perform tasks at particular times of the day. Harrison and Breed (1987) succeeded in training workers of Paraponera clavata, a giant ponerine of Central and South America, to come to sugar baits at selected hours during the day or night. Most of the arrivals were accurate to within 30 minutes. When the Paraponera were no longer rewarded at particular places or times in the forest where the experiments were conducted, they ceased responding by the third day. The upper bounds of their ability were not measured. Hence Paraponera cannot yet be compared with honeybees, which have been shown to learn up to nine different times and places within a 24-hour cycle with a high degree of accuracy (Koltermann, 1974).

In fact, temporal learning is likely to prove to be a phylogenetically restricted phenomenon in ants. It appears to be involved in the rapid shifts of foraging times by colonies of Atta, Monacis, and a few other ants in response to competition and food availability (Cherrett, 1968; Lewis et al., 1974a; Swain, 1977). The shifts occur within one or a few days and persist for days to weeks thereafter. They cannot be easily ascribed to the daily occurrence of cues extraneous to the colony, as distinguished from temporal learning. On the other hand, with the exception of the Harrison-Breed data, evidence of temporal learning in ants has not been forthcoming. Grabensberger (1933) claimed to have trained workers of Myrmica rubra to search for food at intervals of 3, 5, 21, 22, 26, and 27 hours respectively, as well as 24 hours. This is an extraordinary result, especially in view that no such non-circadian rhythms have ever been taught to honeybees. It appears that Grabensberger was in error. Reichle (1943), working mostly with Myrmica rubra, and Dobrzanski (1956), working with fourteen European ant species, were unable to detect temporal learning of any kind. Both Reichle and Dobrzanski suggested that none of the species examined is capable of such learning because none is adapted to feeding substantially on nectar. Indeed, Paraponera clavata is different from most other ants in that it does regularly collect nectar (Janzen and Carroll, 1983).

Although the documented cases of associative learning imply an impressive cognitive ability, a closer examination of the actual behavior of ants reveals some peculiar shortcomings. Above all, their responses are severely tailored. Learning of any magnitude is limited to the kinds of tasks in which flexibility and fine adjustment are at a premium, such as orientation to and from the nest and the recognition of colony odors. Even in these cases learning is strongly prepared by innate tendencies to lead the ant to quick, accurate responses. Other constraints have been discovered. Formica pallidefulva workers “follow their noses” during maze learning. If the passage in a maze turns to the left, the ant tends to follow it around on the outer, or right-hand, side. If the next turn brings the ant to a T-choice, it will usually proceed around the right-hand corner and into the right-hand arm of the choice. Consequently ants can learn a maze much more quickly if the arms they tend to follow by momentum have been set by the experimenter to be the correct ones. Also, when the Formica workers do make a wrong turn in the early stages of learning, they usually follow the blind alley to its very end before turning back. Only in later stages do they develop the ability to turn back shortly after a mistake has been made. In short, it takes ants a long time to “realize” they were in error (Schneirla, 1943).

It also appears that ants are not capable of insight learning, one of the more advanced categories in the classification by Thorpe (1963). This means that they cannot duplicate the mammalian feat of reorganizing their memories to construct a new response in the face of a novel problem. A dog can, if given time, work purposely around a transparent barrier instead of trying to push its way through or climb over. With no coaching, a chimpanzee can deduce how to pile boxes in order to reach a banana previously out of reach. No behavior approaching this level of sophistication has been observed in ants, or any other kind of social insect.