Introduction

The US Department of Defense seeks to expand its research in modeling and simulation to include not only traditional physics-based models but also models of individual and societal behaviors. This endeavor has been well-documented in a recently published National Research Council study, “Behavioral Modeling and Simulation: From Individuals to Societies” (Committee on Human Factors, Division of Behavioral and Social Sciences and Education, National Research Council, National Academies Press, Washington, D.C., 2008, ISBN 0-309-11862-X). The study also notes that a technological infrastructure and analysis capability must be developed for behavioral modeling that the DoD can properly select and then deploy.

Communication & Training

Verified and accredited behavioral models are necessary for the DoD’s new mission areas of Stability, Security, Transition, and Reconstruction (SSTR), where the main goal focuses not on the physics of breaking things, but rather on communication and interaction with individuals and groups. Achieving this goal requires developing behavioral models for both analysis and training. Analysis seeks to understand the space of potential outcomes and the possibility spaces with respect to planned operations in areas in which we are not culturally native. Training focuses on providing a real-time interactive environment in which personnel can experience executing operations in nonnative environments. Carrying out such analysis and training requires testing proposed behavioral models in an environment in which it is possible to probe and virtually operate against them sufficiently to verify and accredit those models. If we can do this, we can build training and analysis systems to provide military personnel with an on-ground understanding of their mission areas before their deployment overseas, potentially saving lives and reducing misunderstandings.

Developing a Game-Based Analysis Capability

Ongoing efforts at the USC GamePipe Laboratory focus on developing the technological basis for the analysis capability called for in the NRC study, including a massively multiplayer online game (MMOG) infrastructure that can be used as a testbed for models of individual and group phenomena. One of the first tasks of this project was the selection of an appropriate open source or openly available MMOG. Using our background and strength in developing online games, we built our own infrastructure on top of Microsoft’s XNA. That MMOG now includes a set of US force, opposing force, and civilian animations and behaviors, a network game infrastructure, and a well-defined software interface for the insertion of externally developed individual and group models.

An infrastructure for analyzing online gameplay in real-time that is also under development provides visualizations of what is happening in the game, with the eventual goal of providing a real-time analysis and understanding of the space of potential outcomes for the ongoing game and embedded model computation. An architecture for obtaining game inputs from real-world news feeds is also under development, with the intention of providing variability and emergent behaviors in our game populations based on extracts from each day’s events.  The overarching goal of all of this work is to develop the technological basis for providing the analysis capability currently missing from the behavioral models being developed for SSTR.

State of the Art

With respect to behavioral modeling, the current state of the art indicates that models are developed in government, corporate, and academic environments far from the modeled individuals and cultures. Those models provide what looks like a great story using highly abstracted and distilled inputs, but there is no certainty regarding how those models will perform when deployed where lives might be at stake. Rather, it is preferable to provide real-world inputs for such behavioral models, then examine their outputs. Ideally, such models would be embedded into an online game infrastructure where live players can change the inputs into the models, with those models reacting and then changing the visual display, which live players can then analyze. Currently, the majority of behavioral models are built and understood only by the developer, thus they are not developed to be interoperable with any standard for display and interactivity. Part of our work studies how to define an interface for such models and their connection into an online game infrastructure that provides the visualizations and interactivity necessary for the behavioral models to be well understood.

Key Challenges

Some key scientific and technical challenges to our project mesh tightly with developing the following component tasks.

MMOG Infrastructure

We have developed an MMOG infrastructure that can be used as a game testbed for models of individual and group phenomena. An open source or openly available infrastructure is critical for developing the MMOG analysis capability and for unencumbered distribution of our results.

Cosmopolis

The MMOG under construction bears the working title Cosmopolis. Set on present-day Earth, it has a two-level structure with an outer game and several subgames.

The outer game is a city- and world-building simulation that includes player- and guild-level conflicts. Inner games—or subgames—can be housed in any of the world’s buildings or areas, and can belong to any genre, such as sports, first-person shooter (FPS), or gambling. Each player can choose to partake of the outer game, the subgames, or both levels. In the outer game, players can choose to compete in a city’s subgames, for their own gain or on behalf of their guild. The guild whose members win the most subgames in a given city over a given period will rule that city. Doing so lets them levy taxes, govern new construction, and attempt to take over neighboring cities.

The outer game and subgames format supports our aim to attract as many players and player types as possible, which translates into more research data. Players can freely interact in peaceful venues, fight it out in small venues such as FPS subgames, or engage in guild-level wars with up to an entire city as the prize. Alternatively, players can explore new subgames in various distinctive locales, find and complete task assignments, and achieve top subgame ranking, guild rulership, or even world domination.

The GamePipe Experience

The USC GamePipe Laboratory has developed interfaces to its online game infrastructure for behavioral model integration. We are working with models from Carnegie Mellon University’s Center for the Computational Analysis of Social and Organizational Systems as exemplars (www.casos.cs.cmu.edu/). We are also reaching out to other organizations that create and deploy behavioral models.

CASOS created Construct, the first behavioral model that is being used to analyze our game’s data (www.casos.cs.cmu.edu/projects/construct). This social networking model essentially tracks the dissemination of knowledge and the growth of communication networks. Game data extracted for Construct’s analysis includes player statistics—guild membership, abilities, battles, tasks, and adjacency—and communications records such as who talked to or physically approached each other, and when. CASOS also tracks knowledge dissemination for Construct via formal, in-game player-to-player training in various skills, such as combo fighting moves or guild-based emotes and gestures. Data logging also captures various other aspects of the game, including city construction, guild growth, and subgame rankings. All data logs can be presented either in real time or after the fact, either as a graphical snapshot or as a trend over time.

A specific set of data to be presented for CASOS analysis includes the record of conflicts from FPS subgames and guild wars: team formation, weapons used, communications patterns, casualties, and city takeovers. CASOS also tracks the effects of these conflicts, as well as the effects of rare “terrorist attacks” and “natural catastrophes,” as staged by the game administrators, on the behavior of noncombatant populations.

Real-Time, Online Gameplay Analysis

We are developing an infrastructure for analyzing online gameplay in realtime that will create visualizations of what is happening in the game. Its eventual goal will be to provide real-time analysis and understanding of the space of potential outcomes for the ongoing game and embedded model computations. This task’s core challenge is to generate meaningful visualizations in the game space with respect to the state of behavioral model computations. Much experimentation and prototyping must be carried out to generate meaningful visual results. We are working closely with behavioral model developers for this task.

All data-logging parameters—such as timeslice granularity and the tracking of players, guilds, attributes, and presentation format—will be adjustable via a Web-based research interface. Further, our game will offer a critical degree of research flexibility beyond the standard MMOG’s data-logging capability. Our overall design composes a federated model architecture: each subgame provides a potential lab for a different social and behavioral model, maintaining interoperability with the outer game. Subgames can be added, and outer world gameplay can be tweaked, all to meet the analytical needs of any behavioral model creators who use our game as a research testbed.

Incorporate Real-World RSS Feeds

To allow behavioral models to track the effect of public information on social groups, we include real-world RSS feeds on in-game billboards, news screens, and similar media. The first application of this technique is to address in-game resources. Each region in the game world has a distinct level of various natural resources to exploit via mining, harvesting, and so forth.

The value of a given resource rises and falls in accordance with the price of its real-world counterpart. Thus, the daily price quote for gold according to the Wall Street Journal governs the in-game cost of gold for a commensurate period. These real-world commodity prices appear in the game’s various news sources. Eventually, disparity in resource value between areas might lead to more people moving from a poorer to a richer region, or to a hostile takeover attempt from a neighboring region.

Michael Zyda is a Director at the USC GamePipe Laboratory. Contact him at zyda@usc.edu.

Marc Spraragen is a PhD student at the USC GamePipe Laboratory. Contact him at spraragen@usc.edu.

Balakrishnan Ranganathan is a PhD student at the USC GamePipe Laboratory. Contact him at brangana@usc.edu.

Keywords: Behavioral models, Online games, MMOGs, Entertainment computing

A pdf of an earlier version of this article is here.