Macro Social Simulation
with Michael Mateas and members of EIS
Generating social worlds
About macro social simulation
Fictional characters, like real people, live in social worlds. They have history with other people — family, friends, co-workers, lovers. That history informs what they want to do in the present and how they hope the future will unfold.
There's a limit to how much of a social world it makes sense to hand-author. As authors of a game, we might know who lives behind each door in the main character's apartment building. But we won't know that for every door on the block. As we won't know how these characters' lives intersect with others, perhaps with long-simmering tensions that might enflame based on the player's choices.
To create fully-populated fictional worlds, the only tractable approach is to generate them. They should be generated using hand-authored rules, so that the systems can be guided by artistic intent, evaluated, and improved. (This is in contrast to black-box LLM-based generation.) Going further, once such a system is in place, we can generate a new social world for every player, or every playthrough.
Our most developed project taking this approach is Bad News, developed by EIS members James Ryan, Ben Samuel, and Adam Summerville. This piece is presented as live, interactive, dramatic performance.
Each Bad News session begins with a social simulation of decades of history in a small U.S. town. Through this, characters who live in the town come to know each other, form friendships, families, and animosities, work with and for one another, be born, die, and so on. A unique town is created in this way for each performance, and each performance is done for an audience of one, after which that town is never experienced again. Performances take place on two sides of a curtain. On one side sits the audience and on the other sits Samuel — who in addition to being a computational media researcher is also a professional actor. Samuel draws the curtain aside and greets the audience member as the newly-hired assistant of his character, the local mortician. The audience member's job is to locate the next-of-kin for an anonymous dead body, by talking with residents of the town, but first Samuel talks the audience member through crafting a "cover story" for their work, so that they do not unintentionally inform someone else of the death before the next of kin.
After this, Samuel draws the curtain closed again, and the audience member chooses to go where they wish in the town, and speak with who they wish. As they do, Samuel draws the curtain aside and improvisationally portrays each character. But before he begins, Ryan searches through the simulation record of the town, finding information about the character's personality, the most dramatic events of that character's life so far, connections they might have with characters the audience member has already met, and so on. While no performance can be planned ahead of time, the evocative social simulation that the team has authored inevitably provides intriguing and revealing character experiences for Samuel to work into his performance.
Audiences often engage Bad News lightly at first. But as they come to know the characters and their town more deeply, and as they are drawn into a feeling of relationship through Samuel's performance, over the course of interaction, the tone shifts. By the time the next of kin has been located, audiences are often hesitant. They commonly draw out the conversation, not quite ready to share the bad news for which the piece is named. It is not uncommon for them to shed tears when they do so. In short, Bad News uses social simulation together with interactive performance to explore game players' responsibility of another kind — not for what happened, but simply for being its witness and messenger.
Because Bad News was a completed work, experienced by audiences, we were able to learn things from its reception. One thing we learned was simply how compelling people found the experience, such as when Bad News won the audience choice award at IndieCade 2016, or when Steven T. Wright wrote in Rolling Stone, "this marvel of procedural performance can only be played by a lucky few, and that's a crying shame." The other thing we learned is what it would take to make its kind of experience more widely accessible. While improvisational acting is not an easily-acquired skill, there is a body of knowledge about it, training is available, and even a beginner can get started with it. On the other hand, sifting through the record of the town, finding compelling story situations and character information, is something comparatively new. Creating Bad News and performing it with a variety of people gave us real-world experience with how to perform such "story sifting" in a wide range of interactive situations, each grounded in a different fictional world. (Story sifting deserves a page of its own.)
In parallel, research in "zoomed out" social simulation continued. Bad News was built on a simulation called Talk of the Town (TotT). As part of his dissertation, Ryan created Hennepin, a spiritual successor to TotT. Hennepin built on the foundation of character modeling done in TotT and revised parts such as personality models and event causality to provide a more explainable environment for emergent narratives.
The next phase of research was led by Mateas, with additional advising by EIS alum Mark J. Nelson, and limited involvement from me. Talk of the Town and Hennepin were quite successful for the contexts in which they were created, but as a simulations they were not built to be customizable or extensible — many assumptions about the fictional worlds were baked into the core simulation code. In order to create opportunities for further research — both by ourselves and others — Shi Johnson-Bey created the Neighborly system. Neighborly was a customizable, community-scale social simulation engine, and "rational reconstruction" of TotT. We made Neighborly available to the community (see GitHub link below) but did not use it to build any full storyworlds.
Rather, Johnson-Bey developed Minerva, initially as a fork of Neighborly. This is a dynasty simulator that models procedurally generated characters and families vying for influence and power over a shared map. It is designed for emergent narrative research and data analysis. Minerva's core architecture is based on Neighborly, and its systems and mechanics were inspired by Game of Thrones, the Shogun board game, WorldBox, Crusader Kings III, and the Japanese Clan system. Neighborly's default stories felt like mundane slice-of-life tales about people raising families, working jobs, and moving in and out of romantic relationships. In contrast, Minerva provides a higher-stakes narrative framing about families fighting for power over generations.
