Incorporating the presence of the viewer into a decentralized system is straightforward. The viewer may exert influence over the emergent functionality of the system the same way any other component of the system does, by altering an aspect of the environment or influencing the operation of other components.

In this way, the viewer is a "full-fledged" member of the system and consistently integrated into the experience. This contrasts with the model of hypermedia, where the consistency of viewer interactivity depends on the author's consistency of establishing links. When systems are designed by more than one author, or when the base of content grows, maintaining a consistency of experience becomes an increasing burden. A carefully designed decentralized system inherently limits this problem.

One key idea in these new architectures is that of "emergent functionality." The functionality of an agent is viewed as an emergent property of the intensive interaction of the system with its dynamic environment. The specification of the behavior of the agent alone does not explain the functionality that is displayed when the agent is operating. Instead the functionality to a large degree is founded on the properties of the environment.

An important implication of this view is that one cannot simply tell these agents how to achieve a goal. Instead one has to find an interaction loop involving the system and the environment which will converge [towards] the desired goal.

An agent is viewed as a collection of modules which each have their own specific competence. These modules operate autonomously and are solely responsible for [the computation] necessary to achieve their specific competence.

Communication among modules is reduced to a minimum and happens on an information-low level. There is no global internal model, nor is there a global planning activity with one hierarchical goal structure. [The] global behavior of the agent is not necessarily a linear composition of the behaviors of its modules, but instead more complex behavior may emerge by the interaction of the behaviors generated by the individual modules. [3]

Although the approach taken in the Automatist Storytelling System closely conforms to the ideas of Autonomous Agents, it is significantly different than previous applications of this methodology to the area of storytelling.

For instance, in Maes' own subsequent work, agents are applied in the following way:

Many forms of entertainment employ characters that act in some environment. This is the case for video games, simulation rides, movies, animation, animatronics, theater, puppetry, certain toys and even party lines. Each of these entertainment forms could potentially benefit from the casting of autonomous semi-intelligent agents as entertaining characters. [5]

Describing the operation of one such system designed around the story of The Three Little Pigs, Maes and Rhodes state that:

In our model, a story emerges from the interaction between discrete, autonomous characters, controlled either by humans or artificial systems. Each character has its own beliefs [and] motivations... [Characters] choose among their possible actions those that most fit their beliefs and motivations at the time. [6]

Much of the power of this visualization scheme stems from the notion of a normalized system. When a user requests information about Cambridge, activation is injected into that city and spreads to related graphical elements. The system translates the amount of activation into the amount of visual emphasis given to that element -- its size, brightness, and depth. Every activation value is normalized to the total amount of activation present in the system. As in a monetary system, as activation is added to one element, others implicitly devalue. Thus, by gradually increasing the value of one or more elements, the entire system responds to stay in a kind of visual equilibrium.

Thus, a key property of a closed or normalized system of values is that an individual value is only meaningful with respect to the larger "containing" system. This principle, which might be called a Relative Value System, forms the basis of the ConTour interface and is also used for the "Materials Listing" component of the Dexter interface.

Clip-based annotation, in contrast, provides a further level of control to authors by placing them in the position of structuring the story materials into functional units. Clip-based schemes seem to match the basic filmmaking process of refining "raw footage" by picking out useful or "good shots." Typically, these shots function as a unit; a description scheme that treats them as such is often sufficient. In addition, clip-based annotation may be employed for materials that aren't explicitly temporal such as still images and text documents.

The granularity of each unit of content is a key issue. Each piece must in some sense be self-contained and coherent on its own. On the other hand, due to the simplicity of the representation, a given chunk should only be about one particular set of things. In other words, each piece should form a kind of story phrase, complete enough to be coherent, yet not covering too large a range of ideas. Ultimately, the pieces must lend themselves to being dynamically edited together with other related pieces.

In the story applications described in this thesis, video clips tend to be between thirty seconds and two minutes in duration.

In ConTour and Dexter, discrete units of content (materials) are described by units of description (keywords).

By connecting a material to a keyword, the author forms a potential link to other materials that are described by that keyword.

Lacking explicit links, sequencing decisions are made during the viewing experience based on implicit connections via keywords. Deferring sequencing decisions in this way has two consequences: First, the base of content is truly extensible. Every new material is simply described by keywords, rather than hardwired to every other relevant material in the system. In this way, the potential exponentially-complex task of adding content is managed and made constant. Second, because sequencing decisions aren't pre-coded, viewers may play a more active role in the construction of the experience. Instead of using pre-determined links bound to a specific purpose or organizational scheme, the viewer may influence how they want to move from one material to the next.

Materials in these stories are described by a set of particular people, places, times, and themes. As far as the system is concerned, however, the categorizations are irrelevant. Materials may be connected to more than keyword within the same category, useful for instance when a character in the present recalls an event in the past, or when the material addresses several themes simultaneously. Likewise, materials need not be associated with a particular category's keywords at all, useful for "general" clips not necessarily tied to a specific time, place, or character's voice.

By using a keyword hierarchy as opposed to an explicit notion of slot (such as clips having "person slots" and "location slots") the representation is kept as simple as possible. A key point in both ConTour and Dexter is that it's not the representation, but what's done with the representation that's interesting and powerful. Furthermore, isolating elements of description in this way is in keeping with a decentralized approach; in this case, slots would needlessly bind components of the description together and disallow their potential usefulness as independent entities.