CHAPTER 20
Simulation Stories and Free Play

Nick Iuppa & Terry Borst

In the previous chapter we showed how the Leaders team constructed a branching simulation storyline from anecdotal information that was gathered through interviews with experienced soldiers. The information was condensed and re organized into teaching points and turned into a complete learning experience.

Because of the rigors of the branching structure, the participants in such simulations could feel that they are locked into the flow of events in the story, and for that reason have no sense of their own freedom of decision making or their ability to affect the outcome of the simulation. Without this sense of freedom, the simulation becomes less believable, less interesting, less compelling, and generally less effective. Simulations that limit freedom of choice too extensively often negate any sense that the participants are actually controlling the world, and consequently they are distracted from applying the skills that they have learned.

There are a variety of ways to give participants a sense of free will within simulated stories. Two techniques apply directly to branching storylines such as the ones we used in the Leaders simulation. We review those techniques in this chapter.

The more advanced and difficult approaches are really experience management systems that move away from branching storylines and aim to give the participants a greater sense of free will.

But let’s go back to our Leaders simulation and the need to give the participants a sense of free will within the branching story structure. How can we do that?

SIMULATED CONVERSATIONS

In the Leaders simulation, participants progress through the story until they have to make a decision. The elements of the decision are presented to them in the form of questions that are asked by one of the nonplayer characters (NPC).

© 2010 Elsevier Inc. All rights reserved.
DOI: 10.1016/B978-0-240-81343-1.00001-7

As noted in the example from the previous chapter, First Sergeant Jones asks the participant (Captain) if he should move the location of the food distribution site as suggested by Command Sergeant Major Pullman.

The participant can type in a response to this question using any words he or she chooses, and the natural language interface of the Leaders system will interpret the statement and choose a prerecorded response from the many that are available. These responses are audio with attached animation of the characters.

To give the participants a greater sense of free will, they are also allowed to ask questions of the nonplayer character. So, in the case of the question about moving the location, the design team defined as many questions and responses as they could think of. They also wrote and prerecorded the answers to those questions. If a participant asks, for example, how moving the location would affect the morale of the troops who came up with the original plan, First Sergeant Jones has a prerecorded answer.

To make the interaction as smooth as possible, we tested the prerecorded responses by creating a Web site with still pictures and a narrative story setup, but with the same dialog and choices as in the actual simulation. The site was posted on various military Internet sites and soldiers were asked to participate in the exercise in order to validate the questions and answers. We also tried to discover any new questions that we had not thought of. The answers to these new questions were recorded, added to the system, and contributed to a more robust version of the final simulation. For example, one additional question that was discovered and added to the simulation asked, “What do you think about the probability of rain in this location?” To which we wrote and recorded this additional response:

In addition to adding appropriate questions that the participants might ask, we also recognized that soldiers might try and game the system, or try and have fun with the conversations by typing in questions that were out of context or inappropriate. We adapted various devices to counter these actions. The simplest one was to assume that the system just did not understand what the participant was saying. So we recorded a series of responses that presented the possibility that they Captain’s remark was unintelligible. These remarks also worked in those cases where the participants’ comments were off the point. For example:

Other responses dealt directly with statements that seemed to be off the topic. These remarks would be highly original or unusual and therefore unlikely to be anticipated. Our responses dealt directly with the fact that the participant was getting off the track. So, there might be words or phrases in the participant’s question that made sense, but if the question was not related to the issue at hand it was judged to be a non sequitur. Here’s a typical response to a non sequitur question that includes an attempt to get the participant back on track.

The response that the Sergeant gives if the participant asks to be reminded of the original question could also be used as a response to a non sequi-tur question. In our example the response to being asked “What was the question?” was:

INTELLIGENT TUTORING

Intelligent Tutoring is a unique feature of the Leaders simulation that allows instructional statements to appear on the screen to help move the simulation along or get the participant back on track. This response is not from an NPC in the scene (First Sergeant Jones) but from a tutor built into the system itself. We decided not to give this feature a personality or a face but merely to let it appear as text on the screen. This seemed to be the least obtrusive way to introduce tutorial information into the experience. We tested various uses of the tutor and decided to bring in the intelligent tutor after the participant made two non sequitur responses in a row. We also made the statement more direct and included a third person reference to the entire event. In our example, the intelligent tutor statement read as follows:

Note that the final sentence is a complete restatement of the choices.

The intelligent tutoring system within Leaders served two purposes. The first was clarification, as illustrated here. A more sophisticated use of the intelligent tutor was designed to come into play if the participant went through the simulation for a second time and made the same mistake again. In this case the system would remember the participant and his or her response and would explain the teaching point behind the decision node. The participant would still have the free will to choose the wrong answer, but would be much better informed about the point and the intention of the lesson.

In the case of this example, a participant who has gone through the simulation for a second time and encountered this decision point, then asked a series of questions and eventually gave the wrong answer for the second time, would get this statement from the intelligent tutor:

The Leaders team created the full conversation question and answer system for the simulation, and in the process gave the simulation participants a greater sense of free play. We also designed an even more sophisticated solution to the problem of free play, which, though not implemented, was detailed in a report to the ICT and the Army. It is described here.

SPIDER WEBS

One classic way to add a sense of freedom to the story is to create story structures that look more like spider webs than branching trees. In these structures the outcomes of one set of decision elements (decision nodes) can lead to many different nodes. This greater flexibility in the development of the story allows a finer degree of differentiation between possible directives by the participants. If there are five paths out of a decision point instead of only two, the participant has a greater sense of control of the environment because the nuances of their choices matter more. Moreover, the dangers of the ever-branching tree are limited because the same number of decision nodes exist within the exercise, it is just that they interconnect because there are just more paths in and out of each node.

In a branching story structure where there are four levels of decisions and two choices coming out of each decision node, beginning with one decision then going to two possible outcomes, then four, eight, and sixteen nodes, any one path will allow the participant to experience no more than four nodes. That means that, of the 31 nodes that are created for the exercise, 27 nodes are not used. Allowing the participants access to those nodes through the device of spider webbing provides a greater use of the nodes. The result is that the participant can replay the simulation many times and have many different experiences at each playing, even if they do repeat some of their early choices. If AI is employed to keep track of the participants’ paths through the story, the system can also direct the player onto those paths that they had not previously traversed and as a result expand the participant’s options.

The trick to creating this kind of simulation story is all in the telling. As we will see, it is important that nodes lead one to the next in such a way that the tension in the story—and the logic of the story—builds. The problem with Shared Outcomes (that they disrupt the clusters of content) can be overcome by creative writing and the AI capabilities of the computer. Not every node can lead to another. But it is possible to identify certain characteristics about certain decision nodes so that an AI system can know whether or not one node can be the outcome of another node. Of the 31 nodes in chapter 2 of the Leaders simulation only four of them had conditions that made it necessary that another node precede them. That means that 27 of the nodes could follow any of the other nodes and the consequences would seem logical.

To create a spider web-based branching storyline in which most nodes can lead to most other nodes, an analysis of each node must be made, certain characteristics must be identified and coded, and then an AI system must be created which recognizes these characteristics and blocks the use of the nodes in those situations in which their use would be inappropriate. In addition, the story structure itself must be coded (as metadata), along with information about the previous path that particular participants have taken. If these elements can be taken into consideration, it is possible to have a much more flexible version of the simulation chapter which could be replayed over and over again with the participants taking a completely different path each time.

Here is an example. Soldier X has been through the Leaders simulation once. He has had made all the possible decisions (four per chapter). The system can tell by his identification number who he is, which decisions he had made, and in what order. The first question the system then asks itself (the next time the participant goes through the system) is: Which other possible start nodes can the participant get this time? Since there are a limited number of nodes that have the qualities needed to start the simulation, that decision is made rather quickly. From that point on the possible path is chosen by the system during the real-time play of the exercise. So the participant goes into a node, makes his decision, and then the system chooses a new target decision point (node) based on several factors. Figure 20.1 shows a sample form that indicates information and ratings that the authors of the Leaders simulation gave to the decision node that we have been using as an example in this chapter. That node is called Hand in the Plan.

We can see a number of elements are identified so that they can be considered in choosing the next node:

• Is there a required preceding node or a required follow-up node to this node?
• Which characters are present in the scene?
• What is the background of the scene?
• What is the amount of dramatic tension in the scene?
• What is the degree of pedagogical importance of the decision?

Figure 20.1 Molecule Identification Form.

In addition, this form is used to record other pertinent information about the node:

• Its name and number.
• The text of the question itself.
• Summary positive and negative responses.
• Specific wording of the outcome scenes.
• The number of question and response pairs available to the participant.

In the case of our example we can see that, while this is an important decision, it requires certain prerequisite information not available in the node itself, so that another node must precede it. So while Hand in the Plan could follow a good many nodes, it cannot be the first node in the chapter.

Note the ratings for dramatic tension and pedagogical importance. This is part of the metadata for the node, and is done to make sure that the simulation follows the rules of storytelling and pedagogy. Ideally the nodes get more dramatic as the story progresses. On the other hand, pedagogically it is important to put the nodes with the highest degree of instructional importance early in the chapter so that there is a higher probability that they be encountered.

In the case of the decision in our example, the concept is not especially diffi-cult nor does it carry especially dramatic consequences. Combine that with the fact that it requires a prerequisite node and it looks like Hand in the Plan is an ideal node for the middle of the simulation exercise, not the first nor the final node of chapter 2.

HANDCRAFTED SPIDER WEBS

Though it would be possible to create an AI system that assembled the simulation as it were in progress, sort of like an engineer standing on the front of a locomotive and laying out sections of track as the train roars ahead, it is clear that there are only so many combinations of nodes that follow all the rules for drama and challenge and the presence of characters and available prerequisites, etc. This means that with these limitations it would be possible to construct something like 25 different handcrafted, hard-coded branching storylines that would provide a great deal of participant reusability and represent almost all the possible paths that a player could take through the simulation. The high-powered functionality of having an AI engine that would build the story structure as the simulation progressed might not be achieved in the creation of such an exercise, but the learning experience would be just as good.

All the foregoing, however, assumes a simulation structure like Leaders, where the participant stands still at the command center and his or her subordinates ask all the direct questions that need to be asked. If the participant is able to move around the terrain, encounter different characters, and initiate conversations, we have a whole different kind of simulation, one that was not addressed in the limited scope of the Leaders project, but certainly one worth exploring. Participant-driven movement over a large terrain populated with numbers of soldiers, all of whom can respond to questions, adds new variables to the simulation and would certainly be better served by something like the auto-assemble spider webs we have just described. In fact, there are even more sophisticated kinds of simulation strategies designed to give participants a greater sense of free will, which address many of these problems.

SUMMARY

Giving participants a sense of free will is one of the core needs of effective story-driven simulations. Even a simulation based on branching story lines such as Leaders is capable of employing a number of techniques that improve the participants’ sense of free will. One such technique involves giving the participants the ability to ask questions, creating a wide range of answers to these questions, and developing strategies for handling non sequitur statements. Intelligent tutoring operates against the player’s sense of free will but can still be used effectively if it is employed as a tool that operates outside the confines of simulation. Spider webs allow branching storylines to give the participants greater range throughout the simulation and, if well crafted, can add greatly to the sense of freedom, but to do this they must follow a rigorous set of rules that assure that they are used effectively. Handcrafted storylines may be just as effective as those driven by artificial intelligence, but become more difficult to manage in simulations where the participants can move around the terrain and have discussions with anyone they meet.