5.4: Stigmergy and Superorganisms

To compute solutions to the traveling salesman problem, ants from a colony interact with and alter their environment in a fairly minimal way: they deposit a pheromone trail that can be later detected by other colony members. However, impressive examples of richer interactions between social insects and their world are easily found.

For example, wasps are social insects that house their colonies in nests of intricate structure that exhibit, across species, tremendous variability in size, shape, and location (Downing & Jeanne, 1986). The size of nests ranges from a mere dozen to nearly a million cells or combs (Theraulaz, Bonabeau, & Deneubourg, 1998). The construction of some nests requires that specialized labor be coordinated (Jeanne, 1996, p. 473): “In the complexity and regularity of their nests and the diversity of their construction techniques, wasps equal or surpass many of the ants and bees.”

More impressive nests are constructed by other kinds of insect colonies, such as termites, whose vast mounds are built over many years by millions of individual insects. A typical termite mound has a height of 2 meters, while some as high as 7 meters have been observed (von Frisch, 1974). Termite mounds adopt a variety of structural innovations to control their internal temperature, including ventilation shafts or shape and orientation to minimize the effects of sun or rain. Such nests,

seem [to be] evidence of a master plan which controls the activities of the builders and is based on the requirements of the community. How this can come to pass within the enormous complex of millions of blind workers is something we do not know. (von Frisch, 1974, p. 150)

How do colonies of simple insects, such as wasps or termites, coordinate the actions of individuals to create their impressive, intricate nests? “One of the challenges of insect sociobiology is to explain how such colony-level behavior emerges from the individual decisions of members of the colony” (Jeanne, 1996, p. 473).

One theoretical approach to this problem is found in the pioneering work of entomologist William Morton Wheeler, who argued that biology had to explain how organisms cope with complex and unstable environments. With respect to social insects, Wheeler (1911) proposed that a colony of ants, considered as a whole, is actually an organism, calling the colony-as-organism the superorganism: “The animal colony is a true organism and not merely the analogue of the person” (p. 310).

Wheeler (1926) agreed that the characteristics of a superorganism must emerge from the actions of its parts, that is, its individual colony members. However, Wheeler also argued that higher-order properties could not be reduced to properties of the superorganism’s components. He endorsed ideas that were later popularized by Gestalt psychology, such as the notion that the whole is not merely the sum of its parts (Koffka, 1935; Köhler, 1947).

The unique qualitative character of organic wholes is due to the peculiar nonadditive relations or interactions among their parts. In other words, the whole is not merely a sum, or resultant, but also an emergent novelty, or creative synthesis. (Wheeler, 1926, p. 433)

Wheeler’s theory is an example of holism (Sawyer, 2002), in which the regularities governing a whole system cannot be easily reduced to a theory that appeals to the properties of the system’s parts. Holistic theories have often been criticized as being nonscientific (Wilson & Lumsden, 1991). The problem with these theories is that in many instances they resist traditional, reductionist approaches to defining the laws responsible for emerging regularities. “Holism is an idea that has haunted biology and philosophy for nearly a century, without coming into clear focus” (Wilson & Lumsden, 1991, p. 401).

Theorists who rejected Wheeler’s proposal of the superorganism proposed alternative theories that reduced colonial intelligence to the actions of individual colony members. A pioneer of this alternative was a contemporary of Wheeler, French biologist Etienne Rabaud. “His entire work on insect societies was an attempt to demonstrate that each individual insect in a society behaves as if it were alone” (Theraulaz & Bonabeau, 1999). Wilson and Lumsden adopted a similar position:

It is tempting to postulate some very complex force distinct from individual repertories and operating at the level of the colony. But a closer look shows that the superorganismic order is actually a straightforward summation of often surprisingly simple individual responses. (Wilson & Lumsden, 1991, p. 402)

Of interest to embodied cognitive science are theories which propose that dynamic environmental control guides the construction of the elaborate nests.

The first concern of such a theory is the general account that it provides of the behavior of each individual. For example, consider one influential theory of wasp behavior (Evans, 1966; Evans & West-Eberhard, 1970), in which a hierarchy of internal drives serves to release behaviors. For instance, high-level drives might include mating, feeding, and brood-rearing. Such drives set in motion lower-level sequences of behavior, which in turn might activate even lower-level behavioral sequences. In short, Evans views wasp behavior as being rooted in innate programs, where a program is a set of behaviors that are produced in a particular sequence, and where the sequence is dictated by the control of a hierarchical arrangement of drives. For example, a brood-rearing drive might activate a drive for capturing prey, which in turn activates a set of behaviors that produces a hunting flight.

Critically, though, Evans’ programs are also controlled by releasing stimuli that are external to the wasp. In particular, one behavior in the sequence is presumed to produce an environmental signal that serves to initiate the next behavior in the sequence. For instance, in Evans’ (1966) model of the construction of a burrow by a solitary digger wasp, the digging behavior of a wasp produces loosened soil, which serves as a signal for the wasp to initiate scraping behavior. This behavior in turn causes the burrow to be clogged, which serves as a signal for clearing behavior. Having a sequence of behaviors under the control of both internal drives and external releasers provides a balance between rigidity and flexibility; the internal drives serve to provide a general behavioral goal, while variations in external releasers can produce variations in behaviors: e.g., resulting in an atypical nest structure when nest damage elicits a varied behavioral sequence. “Each element in the ‘reaction chain’ is dependent upon that preceding it as well as upon certain factors in the environment (often gestalts), and each act is capable a certain latitude of execution” (p. 144).

If an individual’s behavior is a program whose actions are under some environmental control (Evans, 1966; Evans & West-Eberhard, 1970), then it is a small step to imagine how the actions of one member of a colony can affect the later actions of other members, even in the extreme case where there is absolutely no direct communication amongst colony members; an individual in the colony simply changes the environment in such a way that new behaviors are triggered by other colony members.

This kind of theorizing is prominent in modern accounts of nest construction by social paper wasps (Theraulaz & Bonabeau, 1999). A nest for such wasps consists of a lattice of cells, where each cell is essentially a comb created from a hexagonal arrangement of walls. When a large nest is under construction, where will new cells be added?

Theraulaz and Bonabeau (1999) answered this question by assuming that the addition of new cells was under environmental control. They hypothesized that an individual wasp’s decision about where to build a new cell wall was driven by its perception of existing walls. Their theory consisted of two simple rules. First, if there is a location on the nest in which three walls of a cell already existed, then this was proposed as a stimulus to cause a wasp to add another wall here with high probability. Second, if only two walls already existed as part of a cell, this was also a stimulus to add a wall, but this stimulus produced this action with a much lower probability.

The crucial characteristic of this approach is that behavior is controlled, and the activities of the members of a colony are coordinated, by a dynamic environment. That is, when an individual is triggered to add a cell wall to the nest, then the nest structure changes. Such changes in nest appearance in turn affect the behavior of other wasps, affecting choices about the locations where walls will be added next. Theraulaz and Bonabeau (1999) created a nest building simulation that only used these two rules, and demonstrated that it created simulated nests that were very similar in structure to real wasp nests.

In addition to adding cells laterally to the nest, wasps must also lengthen existing walls to accommodate the growth of larvae that live inside the cells. Karsai (1999) proposed another environmentally controlled model of this aspect of nest building. His theory is that wasps perceive the relative difference between the longest and the shortest wall of a cell. If this difference was below a threshold value, then the cell was untouched. However, if this difference exceeded a certain threshold, then this would cause a wasp to lengthen the shortest wall. Karsai used a computer simulation to demonstrate that this simple model provided an accurate account of the three-dimensional growth of a wasp nest over time.

The externalization of control illustrated in theories of wasp nest construction is called stigmergy (Grasse, 1959). The term comes from the Greek stigma, meaning “sting,” and ergon, meaning “work,” capturing the notion that the environment is a stimulus that causes particular work, or behaviour, to occur. It was first used in theories of termite mound construction proposed by French zoologist PierrePaul Grassé (Theraulaz & Bonabeau, 1999). Grassé demonstrated that the termites themselves do not coordinate or regulate their building behaviour, but that this is instead controlled by the mound structure itself.

Stigmergy is appealing because it can explain how very simple agents create extremely complex products, particularly in the case where the final product, such as a termite mound, is extended in space and time far beyond the life expectancy of the organisms that create it. As well, it accounts for the building of large, sophisticated nests without the need for a complete blueprint and without the need for direct communication amongst colony members (Bonabeau et al., 1998; Downing & Jeanne, 1988; Grasse, 1959; Karsai, 1999; Karsai & Penzes, 1998; Karsai & Wenzel, 2000; Theraulaz & Bonabeau, 1995). Stigmergy places an emphasis on the importance of the environment that is typically absent in the classical sandwich that characterizes theories in both classical and connectionist cognitive science. However, early classical theories were sympathetic to the role of stigmergy (Simon, 1969). In Simon’s famous parable of the ant, observers recorded the path travelled by an ant along a beach. How might we account for the complicated twists and turns of the ant’s route? Cognitive scientists tend to explain complex behaviours by invoking complicated representational mechanisms (Braitenberg, 1984). In contrast, Simon (1969) noted that the path might result from simple internal processes reacting to complex external forces— the various obstacles along the natural terrain of the beach: “Viewed as a geometric figure, the ant’s path is irregular, complex, hard to describe. But its complexity is really a complexity in the surface of the beach, not a complexity in the ant” (p. 24).

Similarly, Braitenberg (1984) argued that when researchers explain behaviour by appealing to internal processes, they ignore the environment: “When we analyze a mechanism, we tend to overestimate its complexity” (p. 20). He suggested an alternative approach, synthetic psychology, in which simple agents (such as robots) are built and then observed in environments of varying complexity. This approach can provide cognitive science with more powerful, and much simpler, theories by taking advantage of the fact that not all of the intelligence must be placed inside an agent.

Embodied cognitive scientists recognize that the external world can be used to scaffold cognition and that working memory—and other components of a classical architecture—have leaked into the world (Brooks, 1999; Chemero, 2009; Clark, 1997, 2003; Hutchins, 1995; Pfeifer & Scheier, 1999). In many respect, embodied cognitive science is primarily a reaction against the overemphasis of internal processing that is imposed by the classical sandwich.