|
|
This research has been motivated by the lack of models and languages in the visual design field that are able to address design solutions, which continuously adapt in response to the dynamic changes both in the information itself and in the goals or intentions of the information recipient. This paper postulates a multiagent model of ssdynamic design - a theoretical framework of design that provides a model with which the visual designer can think during the course of designing. The model employs a decentralized model of design as a premise, and borrows its conceptual model from the improvisational performance, such as dance and music, and bases its theoretical and technical framework on the field of multiagent systems. A design solution is considered an emergent behavior of a collection of active design agents, or performers, each of which is responsible for presenting a particular segment of information. The graphical behaviors of design agents are described by their dynamic activities - rather than by the traditional method of fixed attributes. The model is illustrated with two design projects, Dynamic News Display System and E-Mail Reader, both of which were implemented using a multiagent design simulation system, perForm, along with an agent description language, persona.
Keywords: visual design,
visualization, dynamic information, automatic design, multiagent
systems
1.1. Problems and Opportunities
Traditional visual design encapsulates information into fixed forms,
such as print or film, so that the message can be distributed or
stored. In the design of computer-based communication, design problems
become more dynamic as the media become more interactive and include
more temporal information. It is therefore important to create a
design that continuously adapts in response to the dynamic changes of
context, that is, information and the reader's intention. I argue that
visual design as a field limits its own contributions to traditional
methodologies, despite increasing efforts from within the field to
improve the visual design of digital media. This lack of models and
languages prevents designers from exploring those solutions that are
unique to computer-based communication.
The problem on which I have focused in this research is identified by
two unique characteristics of computer-based communication: dynamic
change in context; and the capability of temporal presentation
(Fig.1).
First, dynamic change in context (i.e., information and the reader�s intention) raises a new problem in the design of computer-based media. For example, on-line information systems, such as news databases and traffic information systems, are updated as information changes. The design in such a medium must reflect the dynamic changes in information over time. Interactive media provide readers with personalized access to information. The pace and order of reading and the amount and selection of information change over time based on the intention of the individual user. Consequently, in the design of computer-based media, a designer often finds it impossible to design for solving a particular problem, and instead must represent a way of designing in a form of a computer program that can generate design solutions at run-time.
But, what do I mean by a way of designing, or process of designing? What kind of form does this process take? In this paper, "a way of designing" is an explicit description of how individual design elements, unknown at the time of designing, change over time according to changes in the immediate context. A goal of this research has been to propose a conceptual model and a descriptive language that designers can use in the course of creating such dynamic design solutions.
The second characteristic of computer-based media is the capability of temporal presentation. It introduces new challenges and opportunities in design. Designers are no longer limited to fixed forms in order to communicate information. For example, a gradual color shift can be used to convey complex emotional quality; or a rapid repetitive movement can be used to indicate a quantity.
However, although traditional visual design has dealt with temporal presentation media, such as television and film, there have been only a few studies on temporal forms [4]. As a consequence, the visual design field lacks the rich models and languages that would allow designers to discuss visual forms presented over time. A classic example of such a model is a color space, e.g., the Munsell model [6], which has provided designers with a conceptual model of color, along with a vocabulary set, allowing the use of rich color harmony as well as the analysis of complex interaction of color. This research proposes such a model for the visual forms that change over time.
These characteristics have raised various questions that need to be answered: How can a designer conceptualize and describe a solution that can be interrupted by some immediate changes in information and in the reader's intention? How can a designer describe a design that contains an undetermined number of design elements constantly changing? How can a designer contemplate design problems as continuous problems, rather than a series of discrete problems?
I suggest that it is valuable to develop a model of design that would provide designers with a conceptual framework during the course of solving design problems, in order to extend the fixed nature of the traditional design-object so that it will include more continuous and responsive characteristics. The model is also intended to contribute to the development of computer systems that represent and generate continuous and responsive design solutions. In this paper, I shall call this framework a multiagent model of dynamic design, or maDes.
1.2. Background
There have been a number of researchers who proposed generative
design systems (e.g., [3][5][8][11]). However, most of their
representation schemes do not consider the fluid nature of
computer-based communication. Rather, those representations are
modeled after traditional static design (e.g., a layout, or a
presentation) where a design solution is described as a collection of
declarative statements that are intended to fix formal attributes. On
the other hand, maDes is intended to encourage designers to perceive
(1) a design problem in computer-based communication as a continuous
stream, instead of a collection of discrete problems; and (2) a design
solution as a fluid emergent entity generated by dynamic activities of
design elements, instead of a set of design elements with fixed
attributes.
maDes has been influenced by three distinct areas of studies. First is
the cognitive models of design that suggest the multiplicity of
knowledge involved in designing (e.g., [10][12]). These theories have
encouraged a distributed representation as a natural means for
representing a generative design process. Second, in order to develop
a distributed model, I have drawn an analogy from improvisational
performance, such as dance and music [1][7]. The intrinsic nature of
improvisation that addresses performers' responses to the spontaneous
changes in context during a performance has provided an insight for a
design solution that must respond to dynamic changes in its context.
Finally, the research in multiagent systems - a branch of distributed
AI - has provided a theoretical and technical foundation to this
research [2]. In particular, maDes has adopted cooperative reactive
(or situated) agents [9] as constituents of multiagent design systems.
Imagine that you are a designer. Think of a news display system that
allows users to access a news database that is constantly updated.
News stories arrive at the display system as they are issued, and a
user can view them as they are published. The purpose, or more
precisely, intention, of a user may also vary over time. For example,
a user may want to read news articles according to locations; or a
user may need to look at a particular category. Your task as a
designer is to create a visual design solution for the display that
reflects the dynamic changes in both reader's intention and the
database. Since information is only available at run-time, you are not
allowed to directly adjust attributes of potential design elements,
such as typeface and color, unlike the newspaper design. Thus, you
must design a way of designing, or design process, in order to
dynamically solve this continuous design problem. In addition, since
the display is capable of presenting temporal expressions (e.g.,
animation), you are not limited to the traditional notion of static
design solutions. Rather, you are free to use temporal forms. For
example, you may use repetitive motion (e.g., up and down) as a
"label" for an important news story, as you would use a hue (e.g., red)
for the same purpose in the design of a news magazine.
I have just described that your task as a designer is to create a design solution that is responsive. And the solution is a way of designing that you have to describe. Now, how would you want to describe such a design solution?
maDes provides a designer with a unique framework for describing such a design solution. For example, Fig.2 is a scene from an experimental on-line news display designed based on the proposed model. Here, the design solution is an emergent expression created by the situated activities of design agents that are responsible for presenting headlines, news stories, placenames, a clock, a date text, and a map, each of which has its behavior specified by a designer. A headline agent left-aligns its text to its placename first when it is born; and when a new headline is issued at the same location it aligns the text to the new headline. Headline agents also change the translucency of its text based on its age. Placename agents change their text color when there is a headline associated with their location. When a news story is being read by a user, other design agents make their visual representations highly translucent so that they do not distract the user's reading. And so on... The solution here is created by a team of collaborating agents.
A designer's task when working with maDes is to anticipate
potential changes in the context, and specify kinds of expressive
forms for design agents to perform according to their immediate
situations. An analogy to the designer's role here can be made with
that of a director of an improvisational dance performance - selecting
dancers with the desired expressive skill (body forms) and coaching
their situated acting through a thorough rehearsal. In the course of
creating a final dynamic solution, a designer would have to rehearse
behaviors of agents iteratively by simulating real dynamic design
situations.
3.1. Overview
maDes models a design solution as a system consisting of a collection
of smaller design systems. Each system, or design agent, is
responsible for presenting a particular segment of information. A
design agent is a system that can modify its expressive behavior as
the context changes and can cooperate with other design agents. In
traditional visual design, a design element is usually described by a
set of fixed attributes; while in maDes, a design element is described
by a set of dynamic activities.
The term context here is used from the perspective of an agent, and it is comprised of the information, the reader's intention, and the immediate presentation environment (i.e., surrounding visual elements). Any of the three constituents is potentially dynamic.
3.2. Agent's ability
I have described that the design agent presents information according
to a given context. But, how can a designer describe agents' activities?
The proposed multiagent model of design adopts reactive agents as its
constituents [9]. As described in Scenario, design agents are like
performers in an improvisational performance, in that they perform
reactively using their skills, without deliberately planning their
actions. Performers seem to simply act on immediate situations, but it
does not imply their performance is not intelligent. Rather,
deliberation is considered embedded in their reactive pattern, or
skill, through previous training. In dynamic design, the behavior of
the design agent is described as a set of situation-action patterns.
If a design agent is capable of communicating a particular segment of
information at any given situation in such a manner that a designer
desires, the agent's ability is said to satisfy the designer's
intention.
Agent's ability is specified either for a particular agent, or for a class of agents. When it is specified for a class of agents, agents that belong to the same class share their ability. For example, there may be a set of design agents belonging to one class that are all responsible for presenting a headline in a dynamic electronic news system. In dynamic design, a designer often has to focus on behaviors for a class of design agents, rather than a particular design agent, since a particular information set is usually unknown at the time of designing. However, there is nothing to prevent the designers from creating a single unique agent. A clock agent in Dynamic News System is an example of such a unique agent.
In maDes, the ability of the agent is determined as well as described by the following characteristics:
Physical Realization
Physical realization is a perceivable representation of the agent. It
is important to recognize that an agent is not bound to a particular
type of expression. For example, an agent that is responsible for
presenting a headline in Dynamic News Display may express itself using
text or through voice. Since the agent is an active system, it
chooses an appropriate realization according to a given context.
Physical realization(s) for an agent is determined based on the
available technology and a particular design problem at hand.
State and Sensor
State is a unit of what the agent "knows." State includes: physical
properties (e.g., color) that are determined by the agent's physical
realization and changed by actions described next; information about
other related agents; and information about user's intention, and
changes in information for which the agent is responsible.
The design agent is assumed capable of sensing information from the external world (external to itself), which include: information to be presented (e.g., news data), the reader's intention, and other agents' activities. A set of sensors is defined for an agent by a designer, and the value observed by a sensor is a special type of state.
Action
Action is the abstraction for the agent's basic ability - just as
body-form is the abstraction for describing performer's skill in
dance. There are three basic categories of actions that the agents can
perform: formal, communicative, and external. Formal action is an
expressive action that influences the agent's form, such as typography
and color. For example, a flash shown in Fig.3, which can be used to
attract a reader's attention, is a formal action. Communicative action
is the act of sending a message to other agents, which is used when
explicit coordination is necessary. An example of communicative action
is a message sending action taken by a placename agent to inform its
selection to the associated headline agents. Finally, external action
is a type of action used to influence outside the multiagent system,
such as an application program. For example, there can be a quit
button agent that terminates the application program.
Each formal action - or temporal form - must be carefully described by ae designer for a particular design problem. An action can be performed instantaneously (e.g., change font), or it may take a certain duration (e.g., gradually glow size). maDes provides an abstraction that allows structural and analytical descriptions of forms expressed over time.
Any meaningful formal expression can be considered an action. However, what is considered a meaningful action is determined by how a designer views a design problem and its solution. For example, consider the flash action shown in Fig.3. A designer may consider it as a meaningful action to attract a reader's attention. Alternatively, the first half of the same action (tn~tn+2), which changes the color of text from dark to bright, can be considered another action that serves a different purpose. The abstraction of action provides a framework to structure design as a set of meaningful temporal forms.
Strategy
Strategy is the abstraction over actions, which describes the agent's
ability to achieve a particular goal. Strategy is like a performer's
ability to express a certain theme in a musical improvisation. A player
achieves a particular theme by selecting an appropriate action
sequence at the right timing. A set of strategies determines the ability
of a design agent.
To select an appropriate action as a response to its immediate situation, the agent must be able to recognize various contexts by which it is potentially situated. In other words, the designer of a dynamic design solution must identify possible situations for individual agents. A situation can depend on an agent's own state, another agent's state, information it's presenting, a reader's intention, or a combination of these. For example, the placename agent in the news display can recognize when a cursor is above its text, when its text is clicked, when there are headline agents associated with its place, and when some news story is being presented.
Situations are structured in a hierarchical manner. Suppose a user is
interested in reading news articles based on locations (sit-1). Then
if a reader selects a placename with which a headline agent is
associated, it generates a sub-situation for the headline: a reader is
potentially interested in reading its story (sit-2). Notice that,
sit-2 is situated within a larger situation sit-1. You can also find
another substitution within sit-2 when a reader selects the headline
agent: a reader is interested in reading its story (sit-3). In other
words, when the agent is situated in one situation, it does not have
to watch out for all the situations it can recognize; rather it only
has to consider sub-situations that potentially happen within the
immediate high level situation. Fig.4 presents a schematic diagram of
situations that are hierarchically defined. The abstraction of
strategy and situation provides designers with a means of analyzing
and structuring dynamic design problems in a systematic manner.
Given abilities to recognize necessary situations, a strategy is specified as a simple procedure and is composed of a set of actions and a set of reactive rules, based on the reactive agent model proposed by Singh [9]. The simplest strategy consists of a sequence of one or more design action. For example, a headline's strategy to attract a viewer's attention may consist of an action to gradually become red and increase its size. From a perspective of a headline agent, this strategy can be written as:
S-1 attract-viewer's-attention-strategy:
perform flash-action
More complex strategies provide the agent an ability to determine an appropriate action to perform according to its immediate context. An example of this type of strategy, for the headline agent, is to attract a viewer when the news item is important. This strategy is achieved by adding a conditional statement to S-1:
S-2 attract-viewer's-attention-strategy:
If my news article is important to my reader
perform flash-action
otherwise
do-nothing
A strategy can also consist of a set of other strategies. For
instance, a simple strategy for a headline agent to use
attract-viewer's-attention-strategy until it is deleted by a
user can be defined as follows:
S-3 basic-strategy:
while I am not deleted,
use attract-viewer's-attention-strategy
Also, a strategy can consist of a sequence of multiple strategies.
Notice that these two strategies, basic-strategy and
attract-viewer's-attention-strategy, involve three situations as shown
in Fig.5. First the top level situation is when the headline agent is
not deleted. Then, there are two sub-situations: one when the agent's
news article is important, another when it is not.
Strategy is a useful abstraction which a designer can use in the course of exploring design solutions, as well as in describing the final solution. It provides designers with a framework for identifying changes in context, and for determining appropriate dynamic and responsive design solutions.
3.3. External view
Fig.6 summarizes the agent's ability. The design agent acts on the
immediate situation using a strategy that is designed to achieve its
current goal. The information observed from the external world and
messages sent from other agents are used by the strategy to determine
appropriate actions to perform.
It may seem complex to understand how an agent behaves, with a given
set of strategies and actions. One method that helps one understand an
agent's behavior is to consider its history. Fig.7 shows a history of
a headline agent. The agent repeated performing flash-action through
ti since its news story is important. Then,
its situation changed at tn, where the news
story became no longer important. At tn,
the agent stopped performing flash-action and started perform
do-nothing and kept doing nothing until now through tj.
Also, an agent's future can be understood in terms of histories. Given a particular point in time, there are many possible histories. As time progresses, an agent possesses one history in its past. Fig.8 is a schematic diagram showing the concept of possible histories found in the future of an agent. The notion of history is an external description of the agent behavior, but it is not explicitly represented. History is simply an analytical device that can be used to understand the agent's behaviors.
3.4. Organization of Agents
maDes emphasizes a decentralized and lateral interaction among design
agents, as opposed to a centralized and hierarchical one (Fig.9). Like
dancers or players on a football team, who do not have strict
hierarchical control, the design agents collaboratively act in order
to achieve a global goal. However, an agent can become a local leader
with some authority over other agents, like the lead dancer or the
quarterback. The leaders can orchestrate other agents. For example, a
headline agent in the dynamic news design can be a leader that
oversees a story agent and a photograph agent.
I have described that a design solution as a whole is an emergent behavior generated by both implicit and explicit collaborations among design agents. This emergent solution is considered a group strategy, and its communicative quality is considered the ability of a group. However, there is no explicit description about group strategies. Group strategy and group ability are like those of a football team, and composed of a collection of strategies used by individual agents.
3.5. Design Method
In addition to the theoretical framework, this research suggests a
methodical process which a designer can follow through to solve a
design problem. Fig.10 shows a schematic flow of a design process with
maDes. This method fundamentally resembles a typical prescriptive
design process paradigm - analysis, synthesis, and evaluation.
The first phase is the analysis of a design problem. The designer must understand the nature of information, the goal of communication, and types of intended readers. The second phase is the decomposition of the problem. A design problem is described as a set of design agents. In this phase, types of design agents, their roles and organizational style are determined. The third phase is the specification of the agents' behaviors. Here, a designer specifies the behaviors for individual classes of design agents using actions and strategies. Then the solution, or partial solution, created in the third phase is evaluated by simulating agents' behaviors in various contexts.
Although the design method is described in a linear fashion, the
design is not completed by a simple four-stage process; rather it is
an iteration of exploring and examining agents' behaviors and
relations among them. The role of a designer is like that of a
director in the performing arts, in that, the design process consists
of a course of dialogue between a designer and the design agents. A
designer must carefully determine the behaviors of agents in such a
way that each agent can play its role according to the changes in the
immediate context, and can contribute to an emergent design solution.
perForm is an experimental software program that is implemented based on maDes to examine its theoretical framework with concrete design examples. perForm provides an agent description language called persona, along with a multiagent design simulation engine, which simulates parallel activities of design agents. persona is implemented in lisp and provides a set of macros that are used to define agents' abilities, including actions, strategies, sensors, and messages. perForm uses a special 3d graphics library for the realization of the design agent, enabling the use of high quality typography to examine maDes with design solutions that are not too simplistic. The software is written using sgi's Performer graphics library on sgi's onyx Workstation with the Reality Engine graphics.
The following sections illustrate how dynamic design can be created
with maDes, using perForm. Because of the space
limitation, the description of agents' behaviors are given in the
style of stories, instead of strategies written in persona.
5.1. Overview
The design problem is to create a visual interface to an on-line news
database, as introduced in Section 2. In this experimental project, I
intended to create a visual interface that can continuously provide
users with an overview of entire news articles issued within a certain
time period (e.g., past 12 hours). I also decided to present news
articles based on their physical locations where they are issued.
5.2. Decomposition and Roles of Agents
The information is decomposed into the following agent types: Headline
Agent, Story Agent, Map Agent, Placename Agent, Clock Agent, and Date
Agent. Notice that a news article is represented by a headline agent
and a story agent. The role of the Headline Agent is to quickly inform the
content, as well as to present the age of its associated news article.
The role of the Story Agent is simply to provide its message content. The
reason to decompose a news article in two different agents (headline
and story) is to create a visually less overwhelming interface. It is
also conceivable to create a class of design agent that is responsible
for presenting the entire news article. However, it was rather
meaningful to represent a news story using two agents.
5.3. Solution
This particular example used a set of news articles categorized as top
U.S. news from the Clarinet database on May 19, 1995. Fig.11.a shows
the display just before 9 o'clock in the morning. There are three
headlines at Los Angels, one at Kansas City, and one at Washington
D.C. The placename agents responsible for presenting these three
placenames have recognized that there are news stories associated with
them, and changed their color from light gray to bright orange. Also,
notice that three headlines at Los Angeles have larger type size than
the others. This scene is taken right after a reader moved a cursor
over to the text "Los Angeles." First, the Los Angeles agent notices that a
cursor is on top of it - a new situation; and then it informs headlines
associated with it that it is being focused upon. Having been informed by the
placename, those headline agents now "know" that their placename is
being focused upon. According to my instruction for this situation,
headline agents gradually grew their type size. It then kept that size
while a cursor was on its associated placename, in order to make their
text more readable.
When a headline agent is born, it first flashes its text close to a viewpoint (in z-axis) using white (Fig.11.d-f). Then, right after that, it gradually aligns the text to its placename. This introductory action is achieved by using a strategy with a flash action and a move action. After introducing its text, the headline agent uses a heigher level strategy until the age of the headline agent's news article reaches to the specified limit. This strategy provides the headline agent an ability to act based on how a user is interacting with its associated placename as well as with itself.
Each headline agent left-aligns its text to a headline that is issued next at the same place; or to its placename if it is the most recent one. The Headline Agent always tries to maintain this alignment using the same strategy. The Headline Agent also looks at the age of its news article and changes the translucency of its text proportional to its age. For example, the headline presented at Kansas City shows that its story is older than other news stories.
If a placename is selected, headline agents associated with that place use a different strategy to grow its text size, and to move to top left position. Fig.11.b-c presents a sequence of scenes after Los Angels is selected. Three headline agents gradually move their text to the upper left part of the display. In this design, selecting a placename means to further examine articles at that place. Thereafter, if a user selects a headline by clicking its text, it informs its news story agent of its selection. This changes the situation for a story agent, which has been using a strategy to hide. Then, it uses another strategy to make its text visible (Fig.2). Notice that there have been other news stories arrived to other cities, while a user has been interacting with news articles at Los Angeles.
Behaviors of the other agents are relatively simple. The Clock Agent finds current time based on whether the system is used in quick review mode or real-time mode, and displays it in the form of a clock. It also changes color of its clock every hour based on a previously chosen set of 24 colors around a hue circle. The Date Agent keeps checking current date and displays it in the from of a text. The Map Agent just keeps displaying a map. In an earlier design the Map Agent changed the brightness of the map based on the time of a day; however, I decided not to use that strategy since color coding schemes of other agents were difficult to maintain.
Finally, notice that, in Fig.2, texts representing placenames and headlines, other than the selected headlines and its news story, are highly translucent. When the headline and the placename agents notice (using their sensors) a situation in which a news story is being presented , they use a strategy which tries not to visually distract the user's reading.
5.4. Summary
Imagine the design of a newspaper, for instance. The design of
Dynamic News Display clearly has shown a different method in which
designers can approach a problem, when compared to a traditional
model. maDes was able to highlight the fluid nature of design problems
and the potential of dynamic and responsive design solutions. maDes also
provides a means to describe a rich and emergent solution out of
relatively simple distributed agents.
6.1. Overview
The design problem in the second case study is to create a visual
interface to an e-mail system, E-Mail Reader. The interface must
represent: an arrival of a new message, reply-replied message
relationships; whether or not a message is read by the user; and the
number of mail messages. In addition, I intended to make the interface
playful, avoiding conservative structured layout, using interesting
temporal forms and vivid colors.
6.2. Decomposition and Roles of Agents
The information is decomposed into the following design agents: Sender
Agent, Subject Agent, Message Agent, Clock Agent,
Number-of-Messages-Agent (MumMsg agent), Reading-Mode-Switch-Agent
(Switch agent), and Date Agent (Fig.12). Similar to a news article in
the Dynamic News Display, a mail message is decomposed into three
design agents: a sender, a subject, and a message. In addition to the
obvious roles that are implied by these three agents, additional
communicative roles are assigned to them. First, Sender Agent is
responsible for representing whether or not its message is read, the
temporal relationship to other messages, and reply-replied
relationships. Sender Agent is also responsible for informing its
message agent when it is selected by the user. Subject Agent must
visually represent its relationship to its sender. The message agent
must present its text when its sender is clicked. The roles of Clock
Agent, Date Agent, and NumMsg Agent are those of which their names
suggest. The role of Switch Agent is to indicate the current reading
mode of the system, and to switch a reading mode between single and
relational (discussed later) when it is clicked.
aFig.12. Interactive scenes from the E-Mail Reader, showing how the sender and subject agnets introduce their text by coordinating each other.b
c
d
6.3. Solution
Similar to the Dynamic News Display, design agents are realized using
a 3D space. However, unlike the News Display, the
view point is rotated 45 degrees clockwise around the y-axis, creating
a sense of perspective (Fig.12). The clock agent uses a lego-block
pattern to create a playful atmosphere. Unlike other graphical
elements, which use a single realization element such as text, the clock
agent uses multiple parts: numbers, lego background, in addition to
the clock itself. There is no need to decompose a clock into smaller
design agents since it is a meaningful unit.
When a new email message arrives in the mail system, the sender agent presents its name text at the far right in the x-axis and at about the height of the clock in the y-axis; and then flies the text towards the clock. The sender agent randomly chooses the position for its text somewhere between the center and right edge of the clock (Fig.12.a-c). This random positioning in the x-axis is intended to make the names of senders easier to distinguish from adjacent ones, and to create a playful emergent form. As the sender agent approaches the clock, it uses squashing and stretching action, and stops. The design of this action is influenced by a technique used in traditional animation in order to provide a viewer a sense of life. After arriving at the clock, the sender agent adjusts the y position of its text at the top of the clock if it is responsible for the most recent message, and otherwise places it just below the next sender text (Fig.12.d).
aFig.13. a is a close-up view showing how a sender and a subject coordinate. b-c are scenes while a message is presented by a message agent. In d, a leader agnet is waiting for a previous message to finish its presentation by moving back and force.b
c
d
As the sender agent finishes performing its introductory action, its associated subject agents introduces its text by gradually showing the text right next to the name of the sender, using its introductory strategy (Fig.13.a). Since the sender agent stops at a random location, its associated subject agent uses its sensor to adjust its final location.
The Sender Agent is sensitive to a reader's clicking, and if its text is clicked, it informs its associated message agent about the selection. After receiving a message from its associated sender agent, Message Agent recognizes a new situation, and uses a different strategy to present its text in front of other graphical element facing towards the view point (Fig.13.b-c). Message Agent keeps the text visible until a reader clicks the text. Simultaneously, the sender agent that is clicked faces its text to the view point, become translucent, and start "dancing" in the background. The subject agent associated with the message places its text just behind the message and rotates it 60 degrees counter clockwise in order to make it readable. While a message text is presented - a new situation - other sender agents make their text defocused, decreasing the contrast, and their associated subject agents make their text transparent, in order to make the message currently chosen easier to read.
Individual strategies, as well as group strategies, introduced by now are relatively short-sighted. In other words, although a group of agents presents a complex visual message with multiple contents, such as sender's name, message text, and subject, each of which uses its own strategies in parallel, each presentation remains relatively brief. What if there is more information than that can fit in a screen? What if a designer desires to present a chain of causal relationships in sequence? The next simple example will illustrate how a group of agents can collaborate to create a sequence of presentation.
Fig.13.d shows a scene after the agent whose sender name is "Ishantha
Lokuge" is selected under the relational reading mode, and becomes a
leader. Since Ishantha's message is a reply to the message sent by
David Small, the leader agent (responsible for Ishantha's name)
requests the agent responsible for David's name (replied agent) to
start a presentation. While David's message is being presented, the
leader agent waits and visually indicates the fact that it is
performing next, by shrinking and stretching in the x-axis
(waiting-action). Then, after a user finishes reading David's message,
the leader agent recognizes it and begins its own presentation.
Simultaneously, the leader agent informs reply agents (sender agent
whose messages are the reply to the message associated with its
message) so that they can present their messages after its own
presentation is over. This action changes the situation of the reply
agents; and the closest reply agent, which is presenting next, starts
performing waiting-action indicating its next presentation. Finally,
when a user finishes reading the leader agent's message (Ishantha's),
the leader agent stops its presentation and asks the closest reply
agent to start its performance. The first reply agent informs the next
reply agent when it is deselected. The remaining reply agents simply
relay their presentation, one by one. Fig.14 shows a schematic diagram
of this collaborative presentation.
6.4. Summary
The second study experimented with another type of collaboration
orchestrated by a temporary leader agent. It has demonstrated that a
collection of agents can generate a rather longer presentation
sequence, as opposed to a short-sighted reactive design solution. I
also tried to utilize temporal form as a more meaningful way to
communicate messages. For example, the squash/stretch action for the
sender agent is used to generate a sense of life; the dancing action
for the sender agent is used as an "icon" expressing a sender's name
for the message currently presented.
I have presented a multiagent model of dynamic design - a theoretical
framework which provide designers with an approach towards creating
design solutions that are responsive to dynamic changes in information
and the user's intention. I have also emphasized the opportunity of
utilizing temporal forms as a communicative means. Two case studies
have demonstrated that the model is a plausible framework for
highlighting the dynamic nature of computer-based communication, and for
creating a complex design solutions in systematic manner with
simpler descriptions of design agents.
I would like to thank the members of the Media Laboratory's Visible Language Workshop for their suggestions, and Dr. Munindar Singh for our conversation via email on his theory of multiagent systems.
This work was performed at the MIT Media Laboratory, and supported, in
part, by ARPA, NYNEX, and Alenia Corporation.
