Narrative as/and Complex System/s

Marie-Laure Ryan

What does it mean to say "narrative is complex" or "narrative is a complex system"? Complex compared to what: ordering a beer, following a recipe, proving a theorem? Are all narratives complex or only some of them — the ones considered "literary" or, to extend the idea to all media, the ones considered artistic? Is the scientific notion of complexity useful to narrative theory, or should narrative theory develop its own notion of complexity? Such are the questions that I propose to address in this chapter.

In systems theory complexity is associated with concepts such as emergence, nonlinearity, decentralized control, feedback loops, recursion, self-organization, simulation, and distributed intelligence and with formulas such as "the whole is more than the sum of its parts" or "small events can have vast consequences" (the famous butterfly effect). The so-called science of complexity is a loosely defined field that extends into many disciplines and covers a wide variety of phenomena: the organization of ant colonies and beehives; the mode of operation of the brain; the fluctuations of the stock market; the evolution of species; the development of the weather; the organization of the World Wide Web; and the functioning of the immune system, to name only a few. What these systems have in common, according to scientist Melanie Mitchell, is (a) collective behavior: they are made of a number of individual elements that follow relatively simple rules, without the guidance of a leader; (b) signal and information processing: all these systems produce signals, exchanged both internally (i.e., between their components) and externally (between the system and its external environment); and (c) adaptation: these systems change their behavior according to the circumstances, to improve their chance of survival (2011, 12–13). Though there is no unanimously accepted definition of complexity, nor a fixed set of transdisciplinary criteria for measuring and comparing complexity across systems, these conditions lead Mitchell to the following definition: a complex system is "a system in which large networks of components with no central control give rise to complex collective behavior, sophisticated information processing and adaptation via learning or evolution" (13). (Note, however, the redundancy, which points to the difficulty of defining complexity: a complex system is one that "give[s] rise to complex collective behavior" and "sophisticated [or complex] information processing.")

Other scientists dispute the necessity of adaptation: the late John Holland, a leading authority in the field, argued that there are two types of complexity: complex physical systems, or CPS, and complex adaptive systems, or CAS (2014, 6–9). An example of CPS is a cellular automaton, such as John Conway's famous Game of Life: in a cellular automaton, a set of rules is applied over and over again to dots arranged on a grid, causing the global state of the system to change and resulting in unpredictable visual patterns, but the rules remain the same, and the system evolves on its own without input from the designer ("Conway's Game of Life" 2018). In a CAS, by contrast, "the elements, usually called agents, learn to adapt in response to interaction with other agents" (Holland 2014, 8). Moreover, "as the agents adapt to each other, new agents with new strategies usually emerge" (9).

Whether or not the individual components of a system are capable of adaptation, the trademark of complex systems is the lack of a central controlling entity, comparable to the CPU of a computer, that pursues deliberate goals. Out of the simple rules followed by the components of complex systems arise (or more precisely, emerge) patterns and behaviors that could not be predicted by simply looking at the rules or by solving an equation. Therein resides the difference between a simple, linear system and a complex system. With a simple system there is a formula (an equation) that makes it possible to calculate the state of the system at a given time without calculating all the previous states. For instance, if a car moves at a steady speed of a hundred kilometers per hour, we know how far it has gone at a certain time by giving a value to t in the equation x =100t, where t represents the number of hours the car has been traveling and x the distance, without having to calculate how far it has gone at time t-1, t-2, and so on. But with a complex system, such as a cellular automaton, there is no equation that predicts future states. To know what the global state of the system will be like at time t, it is necessary to compute its state at all preceding times, by recursively applying its rules to their own output (or new rules if there is adaptation). In other words, the state of a complex system at a future time t cannot be directly calculated, but it can be obtained through a simulation. There is consequently a close relation between emergence and simulation: we run a simulation to discover emergent properties; conversely, a property is emergent, when it can be discovered only through a simulation.

An often cited example of a complex, emergent system is the ant colony (Johnson 2001). An ant colony consists of large numbers of individuals who follow very simple rules: they forage for food, they fight intruders, they leave chemical signals when they find food, and they respond to the signals left by other ants by joining in the trail. No ant has a global plan for the building and maintenance of the colony — in fact no ant may even be aware of its existence. Yet, as Mitchell explains, "the ants in a colony, each performing its own relatively simple actions, work together to build astoundingly complex structures that are clearly of great importance to the survival of the colony as a whole" (2011, 4). A stunning example of this cooperation is the building of bridges, where ants use their own bodies to allow the passage of the colony over a gap. When the units of one level combine to form units of a higher level, the whole becomes more than the sum of its parts, and the system is hierarchical. This property is regarded by scientists (i.e., Holland 2014, 4) as the trademark of nonlinearity.

Given this all-too-brief survey of the key concepts of complexity theory, how can one address the issue of narrative complexity? The notion of complexity was not invented by scientists, and it is quite possible to develop an approach that owes nothing to them. Moreover, the idea of narrative complexity can be conceived in two ways: as a property of narrative in general or as a property of certain narratives, so the set of all stories can be ranked on a continuum running from simple (minimal) stories to maximally complex ones. By cross-classifying dichotomies we can divide the study of narrative complexity into four categories:

1. An approach, not inspired by science, to the complexity inherent to all narratives

2. An approach, not inspired by science, to the complexity of a subset of all narratives

3. An approach, inspired by science, to the complexity inherent to all narratives

4. An approach, inspired by science, to the complexity of a subset of all narratives

Here I concentrate on points 3 and 4, leaving their distinction to the conclusion. But the criteria for considering an approach as inspired by science are rather loose. Complexity fascinates, and some of the keywords of complexity science, such as emergence, nonlinearity, and feedback loops have infiltrated literary-critical discourse (for instance, see Hayles 2005), but it is very difficult to draw a line between those uses that are too highly metaphorical to deserve serious consideration and those that involve a reasonably close, interesting analogy between two domains that remain far apart in their objects and methods. If we compare the concepts of complexity theory to those of physics, such as force, mass, energy, gravity, entropy, superposition, and so on, we find that they are often not original to the field and that their meaning is extremely vague: there are at least thirty-one different scientific definitions of complexity (Horgan 1995, 107). This vagueness explains, in part, their popularity with literary critics, but it also questions their scientific status — a questioning that actually extends to the whole field of complexity theory. My standard of "scientificity" will therefore rest on the use of two concepts that theorists regard as constitutive of complex systems: (1) decentralized control and (2) emergence. It may seem artificial to study them separately, since decentralized control is a prerequisite for emergence: "Emergence refers to the spontaneous creation of order and functionality from the bottom up" (Page 2009, 20). Yet although emergence presupposes decentralized control, these two concepts have distinct narrative applications.

Decentralized Control

At first sight narrative and complex systems are polar opposites: complexity rejects the idea of a system controlled from the top down by a central authority, but narratives are the work of an author or authorial collective. Complex systems constantly evolve, often in unpredictable ways, but narrative, once inscribed on the page or in celluloid, no longer changes. Yet if we move from the level of author-text relations to the level of the plot, the comparison of narrative and complex systems become much more productive.

For a narrative application of complexity theory, I propose to go back to Mitchell's (2011) definition of complex systems quoted on the first page of this chapter. The elements mentioned in condition (a) can be associated with characters and the rules that control their behavior with the principles that motivate character actions: goals, plans, desires, fears, and values. The exchange of signals of condition (b) correspond to the circulation of information between characters (mostly through dialogue) and between characters and the world at large, and the adaptive behavior of condition (c) is represented by the characters' ability to react to changes in the global state of the storyworld. From the three conditions we can thus derive a reasonably accurate model of plot. Prototypical narratives are about dynamic networks of human (or humanlike) relations, and these networks can be conceived, at least to some extent, as complex systems. The analogy requires, however, some adaptations: the number of elements is far smaller in narrative than in most complex systems; the rules that motivate the behavior of characters are not nearly as simple as those that govern the behavior of ants or of dots in a cellular automaton; the elements of the system follow their own personal rules rather than all behaving according to the same principles; and adaptation means that the rules can change as the characters evaluate new situations. Humans, unlike ants, are able to alter their behavior, which is why human societies are much more susceptible to change (and narratively much more interesting) than ant colonies.

While authors control characters and plots from the top down, using them to pursue certain artistic goals or to demonstrate certain moral or ideological theses, within the fictional world there is no central controlling instance. Characters think of themselves as freely acting human beings, and they know nothing of authorial designs. Plots may be dominated by one character, namely the hero, but this does not mean that heroes are central controlling units, because there are lots of events that they cannot control, such as the machinations of the villain. Good plots emerge out of the conflict between different personal goals: if the goals of the characters were fully compatible, there would be no story worth telling. Narrative is therefore a top-down, centrally controlled system on the level of authorial design, but it must give the impression of an emergent, bottom-up system on the level of plot. A proper balance between these two types of control is essential to the aesthetic success of narrative. Readers must be aware that characters are authorial creations in the service of a global design, but they must also regard them as autonomous agents. When the actions of characters are too obviously dictated by the interest of the plot, this is perceived as a "cheap plot trick" (Ryan 2009), this is to say, as an authorial failure.

Human interactions, the proper stuff of narrative, may be too complex to be generated by cellular automata (at least in their current state of development), but by insisting on a systemic interconnection of elements, complexity theory can inspire a network-based approach to narrative (as developed in Poulaki 2014). One type of relation that lends itself particularly well to network analysis is who among the characters interacts with whom in direct acts of communication. The importance of a character for a plot is at least in part a function of what is called its "degree" in network theory, that is, the number of connections leading in or out of the node that represents this character. In general, the higher the degree, the more important the character. Diagramming interactions between characters makes it also possible to distinguish several types of plot. The network for an episodic or epic narrative (figure 1), such as the "journey of the hero" pattern or a bildungsroman, would show a heavily connected character who encounters many different characters, but connections between groups of characters will be very few or nonexistent: each episode tends to have its own cast. The network of the so-called it-narrative, or circulation narrative (a genre popular in the eighteenth century that retraces the travel of an object, such as a bank note, through multiple owners), would show several reasonably dense networks (corresponding to the various stories), connected to one another through the possession of a common object or through a single connection, corresponding to a character who serves as mediator (figure 2; note that to account for change of possession of the object within a story it would take a different kind of diagram). This structural type also describes narratives with multiple, distinct stories taking place in the same storyworld, such as Cloud Atlas, by David Mitchell. The networks of dramatic narratives (figure 3, for Jean Racine's Phèdre) are more densely connected: characters are strongly interrelated, except for some marginal confidante or messenger types, and the system of connections, rather than individual characters, is the heart of the plot. The degree of connectivity of a character node gives a rough idea of its importance for the plot, but this is not an absolutely reliable criterion, because this would make Thésée the main character of Phèdre. We also need to map the number of scenes in which characters interact with one another, as shown in figure 3. But a computer examining this network would again think that Thésée is more important than Phèdre, because the diagram makes no distinction between dramatically significant exchanges (such as the unique exchange between Phèdre and Hippolyte in which she confesses her love for him) and short, purely utilitarian exchanges where characters transmit news. While a computer could create the diagrams of figures 1 and 2, it takes human interpretation to create a diagram that shows the dramatic weight of exchanges, as I have tried to do in figure 3. This shows the limits of a computational approach to narrative complexity.

Still, plot diagrams based on interpersonal relations are not useless, because they can show a certain type of complexity, which can be used as the basis of a plot typology. By the standard of connectivity, the complexity of a plot is not a function of the number of characters or of events, but a matter of interconnection. The more difficult it is to cut out parts of the network, the more they contribute to a plot's complexity. For instance, it is easy to cut away the episodes of an it-narrative by severing one link. By this standard, the dramatic narrative is the most complex, followed by the epic narrative and the it-narrative.

The network diagram may tell about the complexity of the system of human relations that underlies a narrative, but it does not tell anything about the nature of the interactions that constitute the plot. For such a model, I propose to turn to the metaphor of the landscape introduced by Scott E. Page, a specialist of complex systems. Page distinguishes three kinds of landscapes, which correspond to three kinds of systems of increasing complexity (2009, 6–9). Imagine that with every type of landscape there is a problem to solve, the problem of finding the highest point in the landscape. Now keep in mind that virtually all narratives are about finding solutions to problems, if by the pair "problem/solution" one understands patterns such as the pursuit of desires, the fulfillment of obligations, the restauration of a broken equilibrium, or adaptation to changed circumstances.