Riedl’s training-generation system is called “Scheherazade” after the queen in One Thousand and One Nights who saved her life by telling stories. In response to a user-provided topic, Scheherazade uses crowd sourcing technology to capture stories about a topic from multiple individuals and combines the stories into an integrated situational model that allows the user to make decisions and explore outcomes. Mark Riedl, a 2011 YFA recipient, is an assistant professor of Computer Science at the Georgia Institute of Technology who specializes in the intersection of artificial intelligence, virtual worlds and storytelling. As director of the university’s Entertainment Intelligence Lab, he researches narrative intelligence: the ability to organize and explain the world in terms of stories. Narrative intelligence is crucial for people to tell and understand stories, learn from experiences and operate effectively in the real world. Computers with narrative intelligence could theoretically educate, train, entertain and generally interact with humans the way people naturally interact with each other. “One of DARPA’s goals with the Young Faculty Awards program is to find common ground between university-led basic research and defense needs. The approach Mark is taking with crowdsourcing narratives could help DoD to better leverage the experiences of its warfighters in developing new training tools,” said William Casebeer, DARPA program manager. “Narratives can contain a great deal of collective wisdom about how events unfold and how you can shape the course of the story with your actions and reactions. Being able to tap that collective wisdom using crowdsourcing with those who have important training and operational experiences is critical.” Riedl used the hypothetical scenario of a bank robbery as a test case for collecting stories and generating a plot graph. In the example, a would-be bank robber named John (1) drives to the bank, (2) enters the bank, and then, (3) faced with a plot decision, either sees Sally (the bank teller), waits in line or scans the bank. At such decision points, the narrative can split based on which actions the contributors agreed would follow as a result of the player’s choice. John and Sally’s interactions unfold through a series of such decision points until, as all of the narrative lines agree, John ultimately leaves the bank, at which point he is either arrested or gets away, and the story concludes. Riedl has focused his research on understanding basic, cultural situations, such as going to a restaurant, going to a movie theater, or catching an airplane. These scenarios are currently presented to users as a series of text-based questions and sets of answers. In the future, he envisions rapidly constructing training simulations for complex, mission-oriented scenarios, presented as three-dimensional visualizations. Participating warfighters could record their stories into the system within days or even hours after the experience. Their collective knowledge could benefit, for example, a soldier on foot patrol in an unfamiliar culture. Narratives detailing common mistakes in social interaction could prevent that soldier from misinterpreting intent in a tense situation. A branch in such a narrative might detail the unintended consequences—such as a failure to collect important information that might save lives later—stemming from an unintended social slight. To help overcome these challenges and improve the viability of online cultural training, one of the academic researchers receiving mentorship and funding through DARPA’s Young Faculty Awards (YFA) program has developed a computer system that can automatically parse and aggregate people’s stories about a given topic and reconstruct variations of those experiences. The outputs are interactive training simulations similar to role-playing videogames or choose-your-own-adventure books. By Dialogo June 19, 2013 Scheherazade works by collecting human experiences on a specific topic in linear narrative form and building a generalized model about the topic domain using plot graphs. It can handle any topic for which people generally agree on the main events that should occur, although not necessarily on the specific sequence of events. The system instructs contributors to segment their narratives to avoid complex linguistic structures, and form sentences that contain only one event and one verb. The system then analyzes the narrative examples to identify consensus among primitive plot points, and clusters them based on semantic similarity to create plot events that unfold sequentially until a decision point is reached, at which point a new line of plot events and decision points is triggered. The process is described in detail in Riedl’s paper on “Story Generation with Crowdsourced Plot Graphs.” The wars in Afghanistan and Iraq demonstrate the strategic significance of tactical actions by junior and noncommissioned officers who interact with local populations. This kind of interaction benefits from extensive cultural training, but opportunities for such training are limited by the compression of the Department of Defense’s force-generation cycles. Virtual training simulations provide a partial solution by offering warfighters on-demand, computer-based training, but creating such tools currently requires substantial investments of time, money and skilled personnel. Because some plot events cannot logically co-occur in a single narrative, the system identifies mutual exclusion between plot events. In the bank robbery scenario, mutual exclusions include pulling a gun versus handing a note, using a bag versus handling money directly, or escaping versus being caught.