Agent-based modeling and the art of games

For roughly the past century, economics has been dominated by the mathematical modeling of rational human behavior. In particular, the main economics journals are composed nearly entirely of derivations of various equations thought to represent the behavior of individuals (sometimes aggregated into groups), or the use of statistical methods known as econometrics to test whether the dynamics of those models may have some basis in reality.

Such theories have a beautiful simplicity: they are relatively easy to understand, have clearer meanings than their natural language translations, and allow future work to build upon comparatively well-determined foundations. Their limitations, while sometimes overlooked, are similarly well-known: they simplify the diversity of psychologies into relatively few categories based on limited criteria, assume fairly skilled rational thinkers, and have difficulty allowing for the evolution of the underlying rules of a system over time. While particular models have been able to relax one or more of these constraints (Gary Becker’s 1962 paper on markets with randomly acting participants is a classic example), the difficulty of expanding on theories that do so limits their ability to affect the field.

However, as technology has expanded the computational power available to economists, a new form of social modeling has grown in popularity. Known as “agent-based modeling,” it focuses not on solving individual equations to understand the dynamics of a system but rather on setting up a framework in which simple artificial intelligences act and interact and exploring the dynamics of that system directly. This difference is often subtle—after all, agent-based models can simulate classical equation-based systems—but in certain situations, they allow exploration of chaotic social systems that cannot be easily described by a system of equations.

The key property of agent-based models (and computer-based models more generally) is that the structure of the world is a variable. Because algorithms can be branched and looped, an experimenter can alter the fundamental structure of the world on a whim—and the very rules of the game, the equations governing the evolution of the system, can be changed from within, in response to internal stimuli.

Non-interactive art—essentially the only kind available until the last few decades—works like the systems of equations set up by (neo-)classical economists. The artist or author produces the piece, and it is transmitted directly to its recipients, message intact. On the one hand, the integrity of the piece is maintained; the artist is able to imprint it with her intent, even if that intent is to be variously interpretable. On the other, the piece is static: one cannot examine the world of the art from another angle, another lens, without constructing an essentially different work. The effect of this property is limited to a degree in performance art, due to the interpretation of the work by the performers, but books, recorded music, movies, paintings, drawings, and sculptures all possess it.

How are we to see other landscapes during the Starry Night? Had Jocasta not committed suicide, might we infer a different moral from Oedipus Rex? If Rosebud were a pet, would the tragedy of Citizen Kane feel different?

Interactive art—of which I submit video games are an important example—has a fundamentally different nature, in line with that of agent-based models. Rather than presenting a piece that has intrinsically interesting dynamics, developers of agent-based models and interactive art are concerned with establishing a set of rules the outcome of which is at least partially unknown to the artist.

Indeed, while the degree of unknown-ness can range from the relatively linear play of the Mario games or rail shooters to the expansive freedom of the Elder Scrolls series, the core of game design comes down to this: knowing what not to control. With perfect control, the game world is not a game; with perfect freedom, the game world is not a world.

As computational power has increased, so have games moved towards massive, open, minimally rule-guided worlds. Never mind multiple paths to complete a mission; recent games such as Oblivion and Fable allow the player to choose which missions to accept, and even whether to accept any missions at all. An art enthusiast brings her own experiences to the table when considering the import of American Gothic, but a player of the type of games I will describe in future posts as “hypernarrative” is able not just to filter the meaning of the art through a personal lens but in fact to alter the content—and thus transcend the limitations of the artist’s perspective.

Interactive art is not “superior” to non-interactive art. When the details of the content are essential to the message, non-interactive art will be able to more faithfully convey it. The Sistine Chapel is effective in part because of the perspective and limitations of its artist. Nevertheless, interactive art holds a great deal of promise for accelerating the production, integration, and distribution of philosophic meaning in society, by engaging participants in a constant cycle of self-definition through choices made in dynamic systems.

By seeing games as participatory art, we democratize meaning.


Next: a taxonomy of games according to their philosophical characteristics.