Closing the circle between perceptions and behavior: A cybernetic view of behavior and its consequences for studying motivation and development

Abstract

The dynamic aspect of behavior is exaggerated during social interactions such as sex, combat and rough-and-tumble play where the movements of the two animals involved continually influence one another. The behavioral ‘markers’ abstracted from this stream can greatly influence the conclusions drawn about the effects of experimental procedures and how changes during development are interpreted. By using methods of analysis that treat behaving systems as being dynamic and governed by negative feedback processes, the behavioral markers that are abstracted can more accurately reflect the underlying mechanisms. Using examples from rats engaged in play fighting, serious fighting and food defense, it is shown that motivational from non-motivational contributions to behavioral output and changes in that output with age can be discerned. For example, while sex differences in the frequency of initiating play by juvenile rats are shown to reflect differences in the motivation to engage in this behavior, sex differences in preferred motor patterns used during play do not. Rather, they reflect differences in perceptual and motor systems. Although an issue that is often neglected, we show that behavioral description, and the theoretical underpinnings of that description, is critical for the study of the mechanisms that produce and regulate behavior.

1. Introduction

A central tenet of classical ethology is that description precedes explanation (Tinbergen, 1963). Nearly 40 years ago, Lorenz (1973) deplored the prevailing trend for description to be downplayed – a trend that pervaded the disciplines that became known as Animal Behavior (Hinde, 1982) and Behavioral Neuroscience (Teitelbaum, 1986). Yet, how behavior is described has a great bearing on the kinds of explanations that are posited (Golani, 1976, Pellis, 1996), and this applies just as much to issues related to motivation and development (Alberts, 2007, Hogan, 2001). Consider play fighting in rats. Two juvenile rats meet, and, after sniffing one another, they jump in a jerky manner, run away and then run back toward one another, with one jumping on the other, which leads to a protracted bout of wrestling. One stops, the other breaks free, and then the whole cycle repeats itself, again and again (Bolles and Woods, 1964, Poole and Fish, 1975). Depending on what aspects of this behavior are emphasized and measured, three different motivational explanations have been posited.

One approach has been to focus on measuring the distinctive behavior patterns typical of fighting in rats, such as ‘upright posture’, ‘lateral posture’, ‘supine posture’ (Grant, 1963). The conclusion drawn from such studies is that play fighting resembles serious fighting, but differs in some aspects of organization, such as the frequency of use of particular behavior patterns and in how those behavior patterns are sequenced (Poole and Fish, 1976). Moreover, with increasing age, the organization of play fighting increasingly resembles that of serious fighting (Takahashi and Lore, 1983, Yamada-Haga, 2002). Based on this descriptive similarity between play fighting and serious fighting, many researchers have concluded that play fighting is immature aggression and developmental changes reflect the maturation of the motivational system associated with aggression (Hurst et al., 1996, Silverman, 1978, Taylor, 1980).

An alternative view arose from noting that, during play fighting, young rats compete to gain contact with the nape of their partner's neck, which, if contacted, is nuzzled with the snout (Fig. 1) (Pellis and Pellis, 1987, Siviy and Panksepp, 1987). In contrast, during serious fighting, adult rats compete to bite their opponent's flanks and lower dorsum (Blanchard et al., 1977a, Pellis and Pellis, 1987). Given that the nape is the target of attack and defense during play fighting from its earliest emergence before weaning until well after sexual maturity and into adulthood (Pellis and Pellis, 1990, Pellis and Pellis, 1997), playful fighting remains distinct at all ages. Based on this clear separation of play from fighting, some researchers have posited that this behavior is regulated by a play-specific motivational system (Panksepp, 1998).

Fig. 1.

A pair of juvenile male rats is shown engaging in a play fight in which they compete for access to each other's napes.

From Pellis and Pellis, 1987; reprinted with permission of Wiley-Liss, Inc., a subsidiary of John Wiley & Sons, Inc.

By focusing on describing the behavioral similarities and differences in the play fighting of rats compared to other species of rodents, another possible interpretation emerges. It appears that, in many lineages of rodents, the targets competed over during play fighting, such as the nape of the neck in rats, are the same as those in adult sexual behavior. Thus, the play fighting of rats may be seen as an immature version of the sexual motivational system, and indeed, in some rodents over the course of development, the play does come to resemble the sexual behavior of adults more closely, including mounting and thrusting as they approach sexual maturity. In rats, however, even though the play target seems derived from precopulatory behavior, the pattern of competition over the nape remains different at all ages – that is, the play does not developmentally grade into sex. A model that integrates both the species similarities and differences posits that in the lineage that led to rats, play fighting first emerged as a form of immature sexual behavior, but at some stage in the evolution of that lineage, such play gained some independent control mechanisms, including motivational ones, that led to a dissociation between sex and play. That is, play evolved from changes to the motivational system underpinning sex (Pellis and Pellis, 2009).

Clearly, description matters, and does so in two ways. First, the initial description of the phenomenon influences what it is we think it is that needs to be explained. Second, based on the descriptive basis for the phenomenon, what we decide to measure to reflect that phenomenon can greatly influence how the results of experiments are interpreted. Given that description is not a neutral process, it is not only important to define, explicitly the behavioral measurements to be used, but it is also important to make explicit the theoretical basis for choosing those measurements (Golani, 1976).

Behavior, whether it occurs in actions involving one animal or multiple animals (e.g., Buhl et al., 2006, Moran et al., 1981, Pellis et al., 2009), can be viewed as a dynamic system of interacting components, in which the behavior of any one component cannot be understood outside the context of the other components (e.g., Alberts, 2007, May et al., 2006). Moreover, a critical feature of these dynamic interactions is that they involve negative feedback loops (Cools, 1985, Golani, 1976, Pellis et al., 2009). With these two starting assumptions – that behavior is dynamic and that it involves negative feedback loops – an explicit theoretical framework can be developed. Such a framework imposes a rationale for the kinds of behavioral measurements to be chosen for experimental purposes, and because the underpinnings for those measurements are explicit, there is an empirical basis for critique and refinement of measurements proposed (Golani, 1976).

2. Behavior as seen through the lens of cybernetics

A powerful framework within which to explore the role of feedback loops in dynamic systems is cybernetics. The term cybernetics comes from the Greek for steering a ship, reflecting a self-correcting system that maintains a goal – in the case of the ship, a heading – by taking compensatory actions to offset the effects of disturbances, such as countervailing winds or tides. Norbert Wiener and his colleagues were the first to recognize that the logic behind self-regulating machines could apply equally well to the overt behavior of living things (Wiener, 1961), in which the animal varies its behavior in order to maintain some aspect of its relationship to the environment constant. William Powers built on this by proposing an explicitly cybernetic theory specifically designed to explain behavioral and psychological phenomena (Powers, 1973). Powers’ model provides two practical features of direct relevance to the study of animate behavior. First, he reasons that what animals maintain constant is some perception that they have of the world, and that behavior is thus part of the means to do so. For example, when driving, we keep the side of the front of the car oriented a certain distance to the centerline. Potholes, gusts of wind and other cars may disturb that relationship, leading to movements of the steering wheel (by our hands, and hence our behavior) to regain and maintain the perceptual relationship between the car and the centerline. The behavior is variable and that variability is in the service of maintaining a constant perception. Indeed, Powers calls his theory ‘perceptual control theory’ (PCT), to highlight that behavior varies to maintain a constant perception.

Second, Powers’ model envisages a nested hierarchy of cybernetic control systems, in which a system higher in the hierarchy can change the reference signal (i.e., the value of the perception) for a system lower in the hierarchy. To use a non-biological example, consider the thermostat of a house. The thermostat is set at, say, 20 °C, which is the reference signal, and it compares this value with the actual value of the temperature inside the house. If the actual temperature starts falling below 20 °C, the thermostat signals the furnace to switch on, but then, as the furnace pumps warm air into the house, the temperature rises so that the actual temperature rises above 20 °C. The thermostat then signals the furnace to shut off. In this way, the inside of the house is maintained at around 20 °C. This is the circular action of a single cybernetic system. But let's say that the owner of the house prefers a cooler ambient temperature when going to bed. To achieve this, the setting on the thermostat is changed to 15 °C. In this instance, the owner is acting as a higher-level control system, affecting the functioning of the lower order system by changing the reference signal. This hierarchical model solves two problems: first, it accounts for how homeostasis is maintained (constancy in the face of disturbance), and second, how a different value can be achieved and defended. Switching between homeostatic values, known as rheostasis, has been shown to occur in physiological systems, such as when body temperature is elevated during a fever (Mrosovsky, 1990). We will illustrate this principle in behavioral contexts below. Note that, in this model, perception and behavior are linked together in a circular causal network, and so the model is applicable to the view of behavior arising from a dynamic system of interacting components.

The most critical issue for analysis becomes that of identifying what the regulated or controlled perception may be. Once that is identified, the variation in the behavior can be analyzed to determine if it occurs in the service of maintaining that perceptual constancy (Cziko, 2000). To make these points as concrete as possible, we will explore two such controlled perceptions that have been found to be relevant to many social interactions and then how controlled perceptions can be used to evaluate the behavioral content of social interactions.

2.1. Identifying controlled perceptions

2.1.1. Targets as locations in space

In rats, aggressive bites are directed at the lower flanks and dorsum and defensive bites are directed at the side of the face (Blanchard et al., 1977a). Given the dynamics involved in fighting – movements by one animal are countered by those of the other (Geist, 1978) – these body areas may be bitten more often than others simply because of opportunity. Several lines of evidence support the view that these body areas constitute targets for offensive and defensive biting, respectively. First, measuring bite wounds in free-living animals show that the majority of bites are concentrated on the rump, lower dorsum and lower flanks in one cluster and the face in another (Blanchard et al., 1985). Second, in the resident-intruder paradigm – where, in the laboratory, an unfamiliar male rat is placed in the cage (territory) of another male rat – most bite wounds are on the intruder's lower dorsum and flanks, and on the resident's face (Blanchard and Blanchard, 1990). Third, in the laboratory, when presenting a resident male with an anesthetized intruder male, the resident directs its bites to the intruder's lower dorsum. This is the case even when the intruder is placed on its back, leading the resident to bypass the ventrum, reaching around and beneath, to gain access to the flanks and dorsum (Blanchard et al., 1977a, Takahashi and Blanchard, 1982). Fourth, when an anesthetized resident male is held by the experimenter and brought toward an intruder male, the intruder directs bites at the face of the resident, not at its rump (Blanchard and Blanchard, 1994). Fifth, examination of the movements performed by residents and intruders shows that there is a close correlation between lunges toward the lower flanks by the resident (attacker) and lunges toward the face by the intruder (defender) (Blanchard and Blanchard, 1977, Blanchard et al., 1977a, Pellis and Pellis, 1987). Removing the vibrissae of the intruder diminishes its ability to match the movements of its opponent (Blanchard et al., 1977b). Thus, observational and experimental evidence converge, showing that the lower flanks and the face are targets, or in PCT parlance, controlled perceptions, with the animals behaving in ways to access those targets in preference to any other possible body locations.

2.1.2. Distance as a relationship to spatial locations

Another way to think about a target is that, whereas the proximity value for the attacker is zero, for the opponent, it is a value greater than zero. That is, targets can be conceptualized as different distance values on particular loci of the body. Therefore, the regulated perception is some minimum distance to be gained or maintained (Powers, 2009). The role of inter-animal distance in organizing interactions is well illustrated by how rats protect their food from other rats.

When a rat eats a small food item, it holds the item with its forepaws while chewing at it. In a colony of rats, another rat is attracted to the one eating and approaches the eater's mouth, which then leads the rat with the food to swerve away, laterally, from the robber (Whishaw, 1988). This food protection maneuver follows a typical form, with the robber usually approaching alongside the flank of the rat with the food. As the mouth of the robber approaches the mouth of the rat with the food, the defender then dodges laterally – 90° or more (Fig. 2). Indeed, different sized dodges have been reported to be associated with food items of different sizes, quality and hardness, as well as with attacks by different robbers (Pellis et al., 2006, Whishaw and Gorny, 1994). However, as the logic underlying PCT is that behavior varies to control a constant perception, what, in this case, is the perception that is held constant?

Fig. 2.

Food robbing and dodging are shown in two female rats.

From Whishaw, 1988; adapted with permission from Elsevier.

In a PCT-based study of robbing and dodging, the total angular movement by the dodger and the robber was compared, as was the distance between the mouth of the dodger and the mouth of the robber with the total angular movement of the robber (Bell and Pellis, in press). Although the movement of the two rats was correlated (Fig. 3A; r = 0.445, N = 252, p < 0.001), the ending distance was not correlated with the movement of the robber (Fig. 3B; r = −0.059, N = 252, p = 0.360). That is, the approach by the robber is matched by the withdrawal made by the dodging rat, which leads their movements to be correlated. Inter-animal distance is different. Once the dodging rat achieves its ‘preferred’ minimum distance, it is maintained, irrespective of the robber's movements, because of the compensatory movements by the dodger. That is, the preferred inter-animal distance is gained and maintained by the dodger, and so is a regulated or controlled perception.

Fig. 3.

The magnitude of movement, expressed as the total angular displacement of the robber, is compared to the magnitude of movement by the defending dodger (a) and to the inter-animal distance (mouth-to-mouth) at the end of the dodge (b).

From Bell and Pellis, in press; reprinted with permission from Elsevier.

By maintaining a particular controlled perception – a particular distance between the mouths – the dodger's behavior is flexible and adaptable, as it compensates for the movement of the robber. Moreover, this one controlled perception can explain a large portion of the variance in the magnitude of the dodges both within and between subjects. The preferred distances gained and maintained by individual rats can be changed, depending on such factors as the type of food defended, suggesting that high level control systems can modify the function of lower level control systems by altering the value of the controlled perception (Bell and Pellis, in press).

2.2. Explaining the variation in behavior

Many of the actions performed during fighting can be analyzed as tactics of attack and defense (Blanchard and Blanchard, 1994, Geist, 1978, Pellis, 1997) – that is, as maneuvers that compensate for the disturbance created by the other animal in gaining or maintaining the controlled perception (Cools, 1985). As noted above, some researchers have used the presence of behavior patterns, like the ‘lateral display/posture’, in both play fighting and serious fighting as evidence that play fighting is an immature version of aggression (e.g., Poole and Fish, 1975, Silverman, 1978, Takahashi and Lore, 1983, Taylor, 1980). Viewing the lateral posture as a tactic rather than an aggressive display (Barnett and Marples, 1981, Blanchard et al., 1977a, Blanchard et al., 1977b), and analyzing how it is used with regard to the differing targets attacked and defended in playful versus aggressive fighting, the lateral maneuver emerges as being used in a markedly different manner in the two forms of fighting (Pellis and Pellis, 1987).

A common defensive strategy of intruders, in a resident-intruder territorial fight, is to stand upright and track the movements of the attacker, so as to maintain their teeth oriented toward their opponent's face (Blanchard et al., 1977a, Blanchard et al., 1977b). Thus, the attacker has two problems: (1) in order to gain access to its opponent's rump, it must overcome its defenses and (2) avoid being bitten on the face while doing so. As shown in Fig. 4, by moving laterally toward the opponent, the attacker can keep its head clear of a defensive strike (a–d), but, at the same time, can press its flank against the defender's ventrum (e and f). If this pushing manages to off-balance the defender successfully, the attacker can then lunge and bite at the defender's lower dorsum (g). If the defender manages to lunge at the side of the attacker's face before being pushed off balance, the lateral orientation affords the attacker the ability to swerve its head away from its opponent's approaching teeth (Pellis and Pellis, 1987).

Fig. 4.

In serious fighting in adult rats, attackers often use a lateral orientation to approach and push against a defender, shown standing its ground facing its attacker in an upright position. The attacker's fur is raised (i.e., piloerected).

From Pellis and Pellis, 1987; reprinted with permission of Wiley-Liss, Inc., a subsidiary of John Wiley & Sons, Inc.

In play fighting, as both partners typically attack and defend the one target, the nape (see Fig. 1) (Pellis and Pellis, 1987, Siviy and Panksepp, 1987), they do not need to protect their faces as they approach to attack the offensive target as in serious fighting. Consequently, during play fighting, rats do not use the lateral tactic for offense. However, in play fighting, rats do use a defensive version of the lateral maneuver. When contacted from the side on the nape, the defender may swerve its head and nape away from its attacker while simultaneously moving its lower flank toward the attacker, blocking the attacker's further movement toward the defender's nape (Pellis and Pellis, 1987). Therefore, in serious fighting, the lateral tactic is used for offense, whereas in play fighting it is used for defense. Hence, the presence of the lateral tactic in both forms of fighting is not evidence that playful fighting is an immature version of aggression.

In some situations, the variation in the behavior cannot be explained directly as arising from the compensatory actions taken to gain or maintain a controlled perception. For example, in play fighting, once the defending rat has rotated around its longitudinal axis to lay supine, the attacker can stand over its partner, in what is known as a ‘pin’ (see Fig. 1h) (Panksepp, 1981). Typically, the animal standing on top of its supine partner uses this position to advantage by restraining the movements of its partner, and does so by maintaining its hind paws on the ground for postural support and using its forepaws to restrain its partner. That is, from a PCT perspective, the behaviors by the on top rat occur in the service of gaining or maintaining a controlled perception (i.e., delivering attacks to the partner's nape and avoiding counterattacks to its own). However, sometimes, rats, especially as juveniles, stand on their supine partner with all four feet, creating an unstable position and one from which it is harder to prevent the supine partner from delivering successful counterattacks (Pellis et al., 2005).

Given that the on top partner diminishes its own ability to gain and maintain controlled perceptions, a maneuver such as standing on one's partner with all four feet cannot be explained in terms of the movements being compensatory to the movements of the partner. This odd behavior is, however, consistent with another difference between serious and playful fighting. Loss of control during serious fighting can lead to a lethal bite or blow by the opponent, so attack, defense and counterattack are finely-tuned, so as to avoid allowing one's opponent such an opening (Geist, 1978). Play fighting is different from serious fighting because for it to remain playful one animal cannot persistently subdue its partner – rather, the participants need to reciprocate (Dugatkin and Bekoff, 2003). In play fighting, reciprocity often arises when the animal that has gained the advantage takes some action that handicaps that advantage, which then gives its partner an increased opportunity to regain the advantage (Pellis et al., 2010). Thus, in play fighting, many movements that are inconsistent with the functional requirements of attack and defense of targets can be explained by some higher order control mechanisms altering the value of the controlled perceptions (Powers, 1973).

3. Implications for the study of motivation and its development

The first problem is deciding what kind of motivational influence is reflected by any particular measure or sets of measures. For example, in the play fighting of rats, most of the tactics used to defend against nape contact involve the rat turning to face its attacker by rotating around its longitudinal axis, leading to the defender to lay supine (see Fig. 1). However, in about 20% of cases, the defender runs, leaps or swerves away, evading contact (Pellis and Pellis, 1987). Even more rarely (5–10% of defenses), the defender may turn to face its attacker by rotating horizontally around a vertical axis (Fig. 5). Females use evasions and horizontal rotations significantly more often than do males (Pellis and Pellis, 1990, Pellis and Pellis, 1997, Pellis et al., 1994).

Fig. 5.

Pivoting on its hind legs, a defender turns to face its attacker and then launches its own attack at its partner's nape.

From Pellis et al., 1994; adapted with permission from APA.

Given that evasion reduces the likelihood of physical contact (Varlinskaya et al., 1999), and the horizontal turning defense does not lead to the protracted bodily contact that occurs when the defender rotates to supine (Pellis and Pellis, 1987), the greater use of these tactics may reflect females’ reduced motivation to engage in such play. This would be consistent with the lower motivation for play fighting in females as measured either by a lower frequency of play fighting in general, or a lower frequency of launching playful attacks (e.g., Meaney and Stewart, 1981, Pellis and Pellis, 1990, Pellis and Pellis, 1997, Thor and Holloway, 1983). Such an interpretation would imply a fairly high level motivational factor, in which females are regulating a lower level of engagement in play. However, the food protection example shows that, sometimes, much simpler mechanisms can produce complex outcomes (Bell and Pellis, in press).

Female rats begin to respond to their partner's approach sooner than do males – that is, at a longer distance between the attacker's snout and the defender's nape. This permits the execution of the turn to face maneuver (Fig. 5) before the attacker has made full body contact, so blocking the turn. Males, on the other hand, by starting the turn later, often allow the attacker to be in body contact at the outset of the turn maneuver, constraining the defender's ability to turn completely before the attacker is fully on the defender's dorsum; this forces the defender to switch to a different tactic, that of rolling over to supine. Similarly, when attacked from the side toward the neck, females, by beginning to react to the attacker's approach sooner, are more successful in executing an evasive maneuver before contact is made (Pellis et al., 1994). A relatively small sex difference in the distance maintained between pair mates can thus account for sex differences in the frequency with which different defensive tactics are used (Pellis et al., 1997). Thus, higher-level motivational mechanisms need not apply to all sex differences in play. The same may apply for many developmental changes in behavior.

Imagine a litter of 7 day-old rat pups being placed, scattered, on a tabletop. The table has a smooth, flat surface and walls on its four sides. After some time elapses, the pups huddle in a pile in one of the corners. The problem is to explain how and why this aggregation occurs (Alberts, 2007). The functional explanation for why they aggregate is that the pups are not capable of maintaining their body temperature and so aggregating reduces the speed of heat loss. From a mechanism point of view, two relatively simple perceptual mechanisms seem to be sufficient to produce the aggregation pattern – the pups are thigmotactic, preferring vertical to horizontal surfaces, and they are thermotactic, preferring warmer to cooler surfaces. When placed on the tabletop, they begin to move, and when they encounter vertical surfaces they remain in contact, hence the aggregates typically form against a wall, and usually in a corner (i.e., greater vertical surface area). However, once another pup is close, it is warmer than the wall, so the pups are attracted to each other rather than the wall. Consequently, over time, the pups aggregate in one of the corners. But the aggregation pattern changes with age, so that by 10 days old, it appears different to that which occurs when the pups are 7 days old.

The developmental change can be accounted for by the addition of a third simple perceptual factor: the pups become sensitive to the movement detected around them, becoming more motile in the presence of more active littermates and less motile in the presence of less active littermates (Alberts, 2007). Note that, at both ages, the aggregates form from a dynamic interaction of the constituent elements (pups and walls) and that the factors influencing the behavior of the individual pups are cybernetic ones (gaining and maintaining contact with vertical surfaces and warmer surfaces). In this case, motivated behavior seems to arise from relatively simple cybernetic rules operating in a particular environment, and age-related changes in that motivated behavior arises by the addition of another simple rule to interact with the existing ones.

3.1. Validating controlled perceptions

A practical issue arises from taking the dynamic/cybernetic approach advocated in this paper – that of being certain that the controlled perception discerned is really the one that is controlled. The examples above have illustrated a variety of descriptive and experimental approaches used to do so – with the perception that is held constant being the one most likely to be the one that is controlled (Cziko, 2000). However, ascertaining whether the behavior present can be fully accounted for by the identified controlled perceptions can be more difficult (Bell and Pellis, in press, Pellis et al., 2009). Recent developments in computer-based simulations and robotics (Pfeifer and Bongard, 2007, Railsback and Grimm, 2011) provide valuable new ways to identify and test the sufficiency of presumed controlled perceptions in accounting for the organization of behavioral sequences. For example, computer simulations were used to test whether the experimentally discerned rules for huddling were sufficient to create the patterns of aggregation seen in pups (Schank and Alberts, 1997).

The computer simulations showed that virtual pups governed by the two rules formed aggregates that were geometrically and statistically similar to those of 7 day-old pups and, when the third rule was added, the aggregation pattern was more like that of 10 day-old pups. Thus, the computer simulations supported the experimental data in showing that these rules are necessary for huddling, but the simulations also showed that under some conditions, these rules are sufficient to recreate the pattern seen in real animals. Similarly, simulations can be used to test other rules in other behaviors thought to account for those behaviors.

In the process of imprinting, a young animal, such as a gosling, becomes attached to an object of suitable size and mobility (Lorenz, 1935). Once the attachment is formed, the drive for maintaining proximity leads the animal to behave so as to maintain that proximity, and, if prevented from doing so, will exhibit distress (e.g., Panksepp et al., 1980). Many complications remain with characterizing the mechanisms involved, such as if any components of the imprinting process are innate, and whether learning mechanisms, beyond the general purpose ones of classical and operant conditioning, are necessary for the young animal to form the attachment with the ‘parent’ figure (Bolhuis and Honey, 1998). Whatever the exact mechanisms of imprinting may be, they are postulated to explain how an individual gosling comes to follow a mother figure preferentially. Translated into a PCT framework, a gosling imprinted on its mother could be postulated to keep a particular inter-animal distance between itself and its mother as the controlled perception. If each gosling attempts to gain and maintain a particular proximity (distance) from the mother, when the movement pattern of multiple goslings is plotted, they should show a cloud-like swarm with individual trajectories toward the mother (Fig. 6A). However, if you go to Google Images and look for ‘imprinting, geese and Lorenz’, many images of goslings following mother geese or Lorenz himself appear and some of these Images but not the majority – show goslings swarming like a cloud. In most cases, the goslings are following the mother/Lorenz in a straight line. Irrespective of how frequent straight lines are relative to clouds, the question is whether individual goslings following a distance rule relative to its mother is sufficient to produce straight lines, even rarely. Powers (personal communication, 2010) produced computer simulations in which virtual goslings were programmed to behave as expected from the traditional model; that is, gain and maintain a particular distance in relation to the mother. When so programmed, the only formations involving multiple goslings that were produced were cloud-like swarms (see Fig. 6A). However, no matter how many times this simulation was run, it never produced a pattern in which the goslings traveled in a straight line. This is informative, because unlike observations of real geese, computer simulations can be run thousands of times, making even rare events possible. This failure to produce a straight-line pattern suggests that while gosling-mother attraction is necessary, it is not sufficient.

Fig. 6.

Representations of goslings (G) moving toward a mother goose (M), where they orient their movement exclusively toward the mother (A) and where they orient both to the mother and to each other (B). The mother's trajectory (dark solid line) and the trajectory (lighter broken line) of each gosling are shown over three successive phases of following (1–3). The smaller circles in each panel reflect obstacles that the animals have to move around to avoid as they move through space. The diagrams are based on simulations developed by William Powers.

To obtain a straight-line configuration, Powers had to change the parameters of the simulation so that what the gosling kept constant was its position relative to the nearest moving animal. He then obtained simulations that led to the virtual goslings ending their movement in a straight line (Fig. 6B). All that was needed to place the mother at the head of the line was a simple rule: ‘if there is a choice between two moving bodies to follow, follow the larger one’. But then, once the goslings began to move, the subsidiary rule became ‘if there is a choice between two or more bodies to follow, follow the closest one’ (see Powers, 2009, for descriptions and programs for these kinds of simulations). The simulations suggest that a rule that couples the individual gosling and its mother is insufficient to account for all the variations in following behavior observed in nature. It remains to be empirically determined if the particular rules developed in the simulation are the ones that the goslings use, but the simulations provide a useful guide for further research.

Because robots that are constructed to follow the presumed rules governing a particular behavior are more realistic versions of the living animals, they can be even more informative in directing researchers attention to previously unconsidered factors in influencing the construction of the behavior (Holland and McFarland, 2001). For example, robotic rat pups that were programmed to follow the same two or three rules as the computer simulated pups showed that, with the first two rules only, the pattern huddle formation was similar to that of 7 day-old pups, and adding the third rule was necessary to produce huddles like those of 10 day-old pups. However, the robotic simulations also highlighted the importance of body shape and environment in constructing the aggregates (May et al., 2006, Schank et al., 2004).

4. Conclusion

A dynamic, cybernetic view posits that behavior and perception are linked together in a circular manner. This is at odds with the usual sensory input triggers motor output perspective of linear causality that pervades most of the behavioral sciences (Cziko, 2000, Marken, 2009). However, as the various examples used in this paper show, a dynamic perspective of social interactions can yield insights missed by a non-dynamic approach and that when that dynamic perspective is combined with a cybernetic one (i.e., circular causality), novel mechanisms that lead to modified output can be identified.