Bad News

A performance from the player's perspective. A player, right, engages in conversation with our live actor, left, who improvises as a procedurally generated townsperson whose life history was simulated prior to the performance. A model theater separates them, and its curtains are drawn closed between conversations with the townspeople. Before opening the curtain to improvise live, the actor first gets into character—during conversation, he must adhere to the generated personality, life history, and beliefs of the character he is playing (all of which is displayed on a hidden actor interface that updates in real time).

Photo: Slamdance TV

Bad News is an installation work. Though the setup for each performance is unique, our design philosophy stipulates that the model theater be positioned in an isolated area—this makes each performance an intimate experience, and one in which the the player herself does not feel like a performer. Because performances last approximately 45 minutes, most audience members will not have an opportunity to experience it firsthand. In many of our installations, a frontward-facing area has hosted a visually intriguing facade along with our wizard station (described later in the gallery), allowing for onlookers to observe the wizard’s livecoding. This makes behind-the-scenes activity a kind of metaperformance, and one that is more available to exhibition audiences. This installation was built by technicians at Big Pictures Los Angeles.

Photo: James Ryan

While the narrative and setting is uniquely generated by a computer program each time, all Bad News performances center on an unidentified person who has died alone in an American small town in the summer of 1979. The player, cast as assistant to the county mortician, is tasked with determining the identity of the deceased, determining the identity of a next of kin, and finding and notifying that person of the death—she must deliver the bad news. As such, each performance culminates in a scene in which a character, played by a live actor, is notified about the death of a loved one. These scenes are emotionally intense, sometimes culminating in tears from both the player and actor.

Photo: Alex Flores

A hidden actor interface—visible only to him—displays information about the generated personality, life history, and beliefs of the character that he is currently improvising live (here, one Robert Meng). As the subject of conversation shifts between other residents in town, the interface updates dynamically to display the improvised character's beliefs about the new subject of conversation (here, Ernest Borstel) . These beliefs are accumulated during a world generation procedure—this works in the style of Dwarf Fortress—made possible by a custom artificial intelligence framework developed in the course of our research as PhD students at the University of California, Santa Cruz. Character beliefs may be false, which means that the player must vet the sources of acquired information to home in on the truth of the matter.

Photo: James Ryan

Our actor is Dr. Ben Samuel, one of the few individuals in the world whose background cuts across professional acting and artificial intelligence research. Ben's acting credits include Hulu's first original scripted series, Battleground—the New York Times called him "the best reason" to watch the show—and he's pictured here (right) with Jay Hayden (left) in a still from the series. Ben's PhD thesis, Crafting Stories Through Play (2016), positions Bad News and other projects against his theory of "shared authorship", which characterizes projects in which a system and player collaborate, through play, to construct stories that neither entity could have created on its own. Ben is now a professor of computer science at the University of New Orleans.

Photo: Hulu

The player is provided a tablet computer, home to a special interface displaying information about the player's shifting whereabouts (see next photo), along with a notepad.

Photo: James Ryan

A mockup of the player interface. This interface is available to the player via a tablet computer that is positioned in front of the model theater. To take an action in the virtual world, such as leaving the home described in this example, the player simply utters the action aloud—e.g., “I go outside”—and the wizard, out of sight, listens in remotely (via a hidden microphone) and then livecodes to execute the command in real time (causing the interface to update accordingly).

Photo: James Ryan

We encourage players to take copious notes during a performance—frequently these notes feature diagrams capturing family genealogies, town histories, and love triangles, among other kinds of information that may be produced by the computer during the world generation procedure that precedes a performance. These notes were taken by a player at the Slamdance DIG showcase in Los Angeles.

Photo: James Ryan

Our wizard, Dr. James Ryan, works behind the scenes to manage each performance as it is transpiring. Specifically, the wizard listens in on gameplay (via audio captured by a hidden microphone), livecodes to update the player and actor interfaces (and to identify emergent storylines generated by the computer), and communicates with the actor in real time using a web chat. Together, the wizard and actor work to ensure that each 45-minute performance has a satisfying dramatic arc. James’s PhD thesis, Curating Simulated Storyworlds (2018), is a manifesto for emergent narrative—a mode of story creation that relies on the computer simulation of miniature worlds—in which Bad News features heavily. James is now a research scientist at the historic computing firm BBN Technologies, where the internet was first built.

Photo: Ian Stroud

In this shot of the wizard's console during the town generation phase that precedes each performance, a sampling of simulated events are displayed on screen as the town is being generated. This procedure generates the town in which the next performance will take place by simulating its entire history, day by day, from its beginnings as a farming village in the summer of 1839 up until the death of a townsperson in the summer of 1979. As the result of this procedure, which last a few minutes in real time, an entire miniature world is created: a town bursting with characters who have lived out entire lives in the machine. This is all driven by a novel AI framework called Talk of the Town, about which we have written several academic publications. For a detailed overview, see Chapter 9 of James Ryan’s dissertation.

Photo: Alex Flores

After the town in which a performance will take place has been generated by the computer, the wizard explores the data describing its history in search of emergent storylines that he may feed to the actor via a live web chat. As part of this interaction, the two settle on a set of potent storylines—ones that were not written, but rather emerged in the machine by happenstance during its simulation of town life—which the actor will attempt to surface to the player naturalistically, through conversation. In the shorthand of this communication, 'D' refers to the performance’s deceased character. In his dissertation, James argues that such emergent storylines are a form of nonfiction.

Photo: James Ryan

A screenshot of the wizard's view during a performance. On the left is an active Python console with a generated town loaded into memory—using this, the wizard executes player actions that were spoken aloud (lines with a 'pc.' prefix) and searches the simulated town history for emergent storylines (e.g., pertaining to the deceased's person's father). On the right is an ongoing web chat between the wizard and the actor, which allows for dramatic coordination and communication about all aspects of the performance, including player commands that were not picked up in the audio, dramatic intrigue found in the town's simulated history, experience-level authorial goals, and anything else that may need to be communicated on the fly.

Photo: James Ryan

Our third team member is the guide, Dr. Adam Summerville, who onboards players prior to their performances and provides information and a behind-the-scenes perspective to onlookers in exhibition settings. Additionally, he serves as a self-proclaimed den mother, ensuring that the actor and wizard are well-nourished during uninterrupted blocks of performances, which have spanned as long as twelve hours. In this photo, Adam explains the project to a group of museum attendees at the San Francisco Museum of Modern Art. Adam’s PhD thesis, Learning from Games for Generative Purposes (2018), explores the use of machine learning for generating and understanding videogame content. He is now a professor of computer science at Cal Poly Pomona.

Photo: Anastasia Pahules

Sometimes technical issues arise during gameplay—here, the guide assists the wizard in getting things back on track during a performance at the 2016 IndieCade Festival.

Photo: Jim Whitehead

As noted earlier, each performance culminates in a scene in which the player notifies a next of kin (played by the actor) about the death of a loved one. After this scene, the player debriefs with the wizard, who shows them the underlying technology and then generates an epilogue for the town by simulating thirty years into the future. In the style of a “Where Are They Now” epilogue, this allows the player (and also the wizard and actor) to discover what ended up happening to all of the principal characters that the player encountered during the performance. In this photo, two players experience an epilogue at the San Francisco Museum of Modern Art.

Photo: Anastasia Pahules

Bad News is an award-winning project. Here, the team is shown accepting the Audience Choice award at the 2016 IndieCade Festival, sometimes called "the Sundance of videogames". (You can watch us accept the award here.) Earlier, we won the Innovative Game Design award of the Student Game Competition at the ACM Conference on Human Factors in Computing Systems (CHI).

Photo: IndieCade

Leading up to our installation at the Slamdance Film Festival, we conducted a live interview on the Mountain Morning Show on Park City TV. The segment can be viewed here.

Photo: Park City TV

Here, the wizard (left) listens in on a performance (right) at the second Slamdance DIG showcase, held at Big Pictures Los Angeles. Bad News is designed as an experience for an audience of one, but we allow up to two players simultaneously.

Photo: Nina Ryan

Bad News is an intimate experience for an audience of one, though on several occasions we have carried out a kind of metaperformance that we call an augmented radio play. In this configuration, the performance takes place in an isolated room, but audio is piped into an auditorium, where a live audience listens in, in the style of a radio play. Throughout, the player, actor, and wizard interfaces are displayed on a large screen while the wizard performatively livecodes and provides, along with the guide, commentary on what is happening both in the performance and behind the scenes. Additionally, audience members are invited to call out ideas for emergent storylines that may be identified in the town’s history and fed to the actor in real time. For example, one audience member noticed a love triangle involving the deceased character, which the wizard hadn’t noticed, and after being fed to the actor this emergent storyline became central to the performance. The augmented radio play shown here took place as part of the Polytechnic lecture series at the Slamdance Film Festival in Park City, Utah. Out of frame to the left, an audience listens as the performance unfolds to the right.

Photo: Eseel Borlasa

Bad News can be installed anywhere. In this photo, Ben Samuel sets up the model theater in preparation for a performance in the redwood forest on the beautiful campus of the University of California, Santa Cruz, where the team developed the piece and earned their PhDs as members of the Expressive Intelligence Studio.

Photo: James Ryan

A shot of the performance in the redwoods, on campus at the University of California, Santa Cruz.

Photo: Alex Flores

Sometimes the wizard station is situated far away from where the player and actor sit. In this photo, the wizard operates from a command center that is hidden in a back room at the University of California, Santa Cruz, while the player and actor are positioned in the redwood forest outside. In one case, the wizard managed a row of performances in Los Angeles from his home in Santa Cruz, some 350 miles away.

Photo: Alex Flores

A view of the audience-facing facade of our installation at Big Pictures Los Angeles, built by the curation team there. The shade of pink is meant to evoke the time period of gameplay—the late 1970s—and the holes in the board at right allow exhibition-goers to peep in on gameplay without disrupting it.

Photo: James Ryan