School of Mathematics and Systems Engineering

Reports from MSI - Rapporter från MSI

Agent-based Interface Approach with

Activity Theory
- Human-Computer interaction in diabetic
health care system

Oct MSI Report 06149

2006 VäxjöUniversity ISSN 1650-2647
SE-351 95 VÄXJÖ ISRN VXU/MSI/DA/E/--06149/--SE
Acknowledgement
This Master’ s thesis in Computer Science has been written in Sweden at the School of
Mathematics and Systems Engineering (MSI) at VäxjöUniversity. The assignment and tasks
is carried out under the research of IMIS (Integrated Mobile Information System for Diabetic
Homecare) in Blekinge Tekniska Högskola (BTH). My supervisors at BTH have been Dr.
Guohua Bai as the project leader and Peng Zhang as the technician.
I would like to express sincere gratitude to my supervisors, both for the opportunity of the
interesting research subject and for the kindness help from all aspects during the study. Bth
located in Ronneby which is a very nice town in south coast of Sweden which is about 120km
far way from Växjö. It is a little bit hard to meet and discuss with my supervisors often but I
was always looking forward to it because I could always learn many things from them, not
only about the subject but also the important things in life.
Thanks for my supervisors leading me on the right track in the research, otherwise may be
this paper would not be successfully finished. I also would like to say thanks to the nice
faculty in MSI in VäxjöUniversity, thanks for the opportunity to study and graduate here. I
had a very nice time here in Sweden.

Växjö, Sweden, Sep 2006

-2-
Abstract
IMIS (Integrated Mobile Information System for Diabetic Healthcare) aims at providing
healthcare on both stationary and mobile platform, which is based on Engström’s triangle
model in Activity Theory. It focuses on the need for communication and information
accessibility between care-providers and their shared patients. Based on the identified need in
the target area, IMIS has decided to construct a network-based communication system to
support communication and accessibility to patients’journal. Since the system integrates
various roles from the heath care organization, it is a challenge to provide a useful software
program to the group members. In order to facilitate the application and enhance the
Human-Computer interaction of the system, agent technology is applied to increase the
flexibility factor so that the system could be self-adapted to a wider range group of users.
Besides, this thesis also introduces the approach of using social-psychology — Activity
theory in HCI, and discuss the integration of these different disciplines. The Multi-agents
System is applied with Gaia methodology from micro perspectives. From the macro
perspective Activity theory constructs the coordination mechanism of the different agents. A
prototype is applied based on the different model of our research.

Keywords:
HCI, agent, Activity theory, usability, flexibility

-3-
Contents

Acknowledgement ...................................................................................................................... - 2 -
Abstract....................................................................................................................................... - 3 -
Contents ...................................................................................................................................... - 4 -
Table of figures ........................................................................................................................... - 5 -
1 Introduction........................................................................................................................ - 6 -
1.1 Background –IMIS................................................................................................ - 6 -
1.2 Research questions ................................................................................................. - 6 -
1.3 Contribution of thesis ............................................................................................ - 8 -
1.4 Outline of the thesis................................................................................................ - 9 -
2 Theoretical background of HCI...................................................................................... - 10 -
2.1 Usability factors.................................................................................................... - 10 -
2.2 Limitation of direct manipulation in usability................................................... - 11 -
2.3 Agent technology .................................................................................................. - 13 -
2.4 Human Agent Interaction.................................................................................... - 15 -
2.4.1 Usability .................................................................................................... - 15 -
2.5 User initiative vs. agent initiative (mix initiative).............................................. - 16 -
2.6 Interface agent...................................................................................................... - 17 -
2.7 Character (personalization) ................................................................................ - 19 -
2.7.1 Tasks differences....................................................................................... - 20 -
2.7.2 Different preferences................................................................................ - 20 -
2.7.3 Approach to build personalization. (Learn ability)............................... - 21 -
2.8 User profiling........................................................................................................ - 23 -
3 Framework of agent-based interface.............................................................................. - 25 -
3.1 Methodology ......................................................................................................... - 25 -
3.1.1 Gaia— agent-oriented analysis and design methodology ...................... - 25 -
3.1.2 Activity theory.......................................................................................... - 26 -
3.2 IMIS system case study with Gaia ...................................................................... - 28 -
3.3 IMIS agent coordination mechanism based on activity theory ........................ - 29 -
4 IMIS prototype................................................................................................................. - 32 -
4.1 Task analyses .............................................................................................................. - 32 -
4.2 IMIS activity structure .............................................................................................. - 32 -
4.3 Architecture of IMIS interface .................................................................................. - 33 -
4.4 IMIS agent Communication...................................................................................... - 34 -
4.5 Scenarios..................................................................................................................... - 35 -
5 Discussion and future work............................................................................................. - 37 -
6 Conclusion ........................................................................................................................ - 38 -
7 Reference........................................................................................................................... - 39 -
Appendix A ............................................................................................................................... - 41 -
Appendix B ............................................................................................................................... - 47 -

-4-
Table of figures
Figure 1.1: Involved theoretical fields .............................................................................. - 8 -
Figure 2. 1: Agent Classifications from Stan Franklin and Art Graesser ........................... - 14 -
Figure 2. 2: Four ways for interface agent to learn ............................................................ - 22 -
Figure 2. 3 different components of user profile................................................................ - 23 -
Figure 2. 4: IMIS assistant agent........................................................................................ - 24 -
Figure 3. 1: Relationship between different models in agent-based system ...................... - 25 -
Figure 3. 2: basic structure of an activity........................................................................... - 27 -
Figure 3. 3: hierarchies of activities................................................................................... - 28 -
Figure 3. 4: stages of the agent-oriented analysis in the MAS system............................... - 29 -
Figure 3. 5: Agent coordination mechanism and agent work delegation ........................... - 30 -
Figure 4. 1: A diabetic health care structure with activity theory....................................... - 32 -
Figure 4. 2: Architecture of IMIS interface........................................................................ - 33 -
Figure A 1: Schema for role BG Handler........................................................................... - 41 -
Figure A 2: Schema for role Diagnosis Advisor................................................................. - 42 -
Figure A 3: Schema for role Medication Manager............................................................. - 43 -
Figure A 4: Schema for role Profile Vetter......................................................................... - 43 -
Figure A 5: Schema for role Side Effect Checker .............................................................. - 44 -
Figure A 6: Schema for role Patient Assistant.................................................................... - 44 -
Figure A 7: Agent model of IMIS ...................................................................................... - 45 -
Figure A 8: Service model of IMIS.................................................................................... - 45 -
Figure A 9: IMIS acquaintance model................................................................................ - 46 -
Figure B 1: IMIS Prototype for a care taker....................................................................... - 47 -
Figure B 2: Report of the patient’
s records ........................................................................ - 48 -
Figure B 3: Chart of average glucose by hour ................................................................... - 48 -

-5-
1 Introduction
Innovation of computer science has been paced fast. From the birth of the first mainframe to
the ubiquitous computers nowadays it only takes a few decades. Thanks to the development
of the Internet, the distance of human communication is shortened by the invisible virtual
connection. Probably many people still remember the skeptical and curious attitude towards
the Internet when it began to emerge in our life 10 years ago. At that time, people could not
believe it, when they were told it was possible to chat with another at a different location with
a computer. It sounds like magic.
Talking with people from the other side of the world instantly is not a dream any more. I
am doing it everyday. With the support of this kind of communication technology, computer
science researchers believe that human being will benefit from the new channel of
communication in various fields and are making effort to develop new products for this
purpose. Numerous new applications based on the Internet are under development. But not
many of them are really useful. Too many commercial issues dominate computer production
domain: airline or hotel reservations are available around the world but medical records are
inaccessible even when they might help save people’ s life. [1]
This is one of the reasons that I am with great pleasure to work with the health care
system, not only for its subject in computer science, but also because I strongly believe that
such kind of system will dramatically improve life of patients.

1.1 Background –IMIS

Diabetes is a very common disease that affects many people. It usually accompanies with
many complications. Diabetic health care is a very complex process which involves many
kinds of situations. In a traditional diabetic care process patient need to contact with their care
providers regularly in various ways in order to get the medical support, training and
supervision to take care of themselves [2]. Research shows the formal way to store patient’ s
record will increase the efficiency of care provider giving diagnosis about patient health
condition. Meanwhile, better communication among care providers will also improve the
treatment if cooperation and coordination are well organized.
There are several problems in the current process of diabetic patient health care [3].
Communication is inefficient and sometimes information is even not reachable in the most
vital time. This led to the birth of the research project –Integrate Mobile Information System
for Diabetic Healthcare (IMIS). IMIS provides a platform to store patient’s information in a
formal standard and integrates two themes: the first is to focus on patients and the second is to
focus on care providers. The two themes share the same communication platform and process
the same information at a proper time for a seamless collaborative work among every member
in the organization [2].

1.2 Research questions

Human-computer interaction (HCI) is one of the most significant subjects in software design.
Usability is usually emphasized as a criterion for system implement. Many programs have
been failed due to the lack of usability. Programmers only implement system from scratch,
they do not refer to any results, methods or recommendations from HCI research [4]. Some of

-6-
them even are not aware of the existence of others. Thus many programs are too complex for
users, have “unfriendly”interface or have no proper functions to fulfill user’ s requirements.
The reason is that most of software programmers do not consider the interface part as their
major concern and they only add it after the “more important”parts of the system have been
made [5]. On the contrary, HCI specialists insist on designing interfaces before programming
because attitude of users towards the interaction with target system is very important. Their
roles should be particularly cared since software systems are essentially tools for them to
perform everyday activity. Low usability will definitely impairs the value of the system.
It is our aim to development a both usable and useful system for people involved in
diabetic healthcare care. As it is said earlier, the IMIS system is to integrate two themes which
include both care providers and care takers, who have different roles in the system and their
individual goals through the system are different. To design a system interface which is fit for
every user in a complex organization like health care is very difficult if not impossible. The
interface has to embrace the necessary functions that are needed by different users in a health
care activity. Like most other health care systems, IMIS stores patient’ s medical record as a
care reference for both providers and patients. Users have individual tasks based on the
history record (stored in a term of database). They keep their tasks explicit all the time. On the
other hand, they should get relevant information easily and quickly when necessary due to the
crucial nature of our context. However, managing information and retrieving the relevant data
at proper time is not an easy job for users. It could be either too complex or time-consuming.
In this case, it has been decided to provide some help for users to ease the manipulation of the
system. Agent technology is adopted to achieve this goal.

With the facts known, the thesis will find answers to the questions below:
1. Can computer agent enhance the interaction between users and the IMIS system?
2. How to model and implement agent-based interface to provide a shared interface to
multi-users to enhance the communication between them in IMIS system?

It is my intention to find answer to these questions after I decide to work with IMIS
system. The potential answer of the first question can be either negative or positive, since
there have been some arguments about this for a long time. The IMIS research group believes
agent will facilitate health care process to some extent and the value of the system will be
increased when agent technology is adopted. Based on this fact I have collected some
materials in order to compare the different theories and examples to come to answers to my
research questions. The first question would be answered according analyses of previous
relevant researches about HCI and agent technology, and the second will be answered with
the practical research of IMIS system with the integration of theoretical analyze result.
Activity theory as a branch of social-psychology theory will be introduced and Engström’s
triangle model will be applied with IMIS in practical as well.
Activity Theory was developed by the Russian psychologists Vygotsky, Rubinshtein,
Leont’ev and others, with work beginning in the 1920’s. It is a novel approach to implement
as an underlying framework in HCI design field to build the gap between HCI research and
practical design [4]. In a multi-users system like IMIS, activity theory is very meaningful. In
the later chapter of this thesis I will apply it as a framework of agent-based interface

-7-
application.
We assume that applying agent Technology will increase the interaction between IMIS
system and the users. Having this belief as the hypothesis of my thesis, I will endeavor to
investigate this subject in the following chapter. In the experiment part, activity theory is
elaborated to decide the function of agent-based system. HCI, Agent technology and activity
theory as input of my thesis, will be integrated together and contribute to construct IMIS
system. The relation between these relevant theories is illustrated in figure 1.1.

Figure 1.1 : Involved theoretical fields

1.3 Contribution of thesis

The contributions of this thesis are primarily represented in three aspects. First, I will give a
complementation to usability principles with the flexible agent technology. And the second
part of the thesis is about designing the component of agent-based interface with activity
theory. The third contribution is to provide a prototype of the IMIS system as an experiment
result of the discussion form the previous chapter.

Human agent interaction

The tradition interface approach has drawback when interacting with users, especially when
users are distributed and belong to a complex organization (like healthcare). Agent’ s
properties are introduced and contributed as a flexible approach to complement the
disadvantage of direct manipulation when a more general usability criterion is required. The
benefit of HAI will be discussed along the different usability aspects proposed by Soren
Lauesen [5]. Some other relevant properties of agent will also elaborate as a reference of
future discussion. Researcher question one will be answered after the comparison of the
different two interaction approach and the mix-initiated system is proposed to contribute later
research and also other researches in similar field.

Model the different component of IMIS system with activity theory

The thesis adopts the theoretical analysis result into the practice case study based on the
context of health care activity. Gaia [6] will be approached as a software engineer
methodology in order to model the different agent component of the organization. Activity
theory as a social-psychology methodology will contribute the holistic framework of agent
communication mechanism when carry out a proper service (activity) in our context. Agent’ s
roles and different services will be defined from both micro (individual) and macro (societal)
prospective [7].

-8-
Prototype design of IMIS
An incomplete prototype design is also include in the thesis to illustrate the function of IMIS
system, question two will be clearer with this experimental implementation part.

1.4 Outline of the thesis

Theoretical background will be introduced in chapter 2. Usability aspects and features of
agent technology will be discussed and a comparison between agent and direct manipulation
will also be given to explain how agent would enhance the interaction. The methodology of
the thesis is elaborate in chapter 3 and the models and framework of the agent-based IMIS
interface are designed based on the methodology. Prototype of IMIS interface designed as
part of this thesis is introduced chapter 4. A simple walkthrough evaluation from scenario will
also be discussed in this chapter. A summary of the contribution and future work about the
thesis is given in chapter 5. Conclusion of our research questions is summarized in chapter 6.

-9-
2 Theoretical background of HCI
This chapter will list usability aspects in HCI research, introduce agent technology and
compares it with direct manipulation. Then discusses features of agent technology to analyze
how it could enhance the human computer interaction.

Definition: Human-computer interaction is a discipline concerned with the design, evaluation

and implementation of interactive computing systems for human use and with the study of
major phenomena surrounding them. [25]

HCI is an interdisciplinary area that touches many fields, the design of software program
has a high level goal from HCI prospective— the usability.

Definition: Usability focuses on the extent to which a product can be used by specified users
to achieve specified goals with effectiveness, efficiency and satisfaction in a specified context
of use.
[ISO 9241-11]

There are many discussions about usability principles in HCI design. As a general and
abstract criterion, usability is usually emphasized during design, evaluation, and
experimenting process (most HCI guidebook has a special chapter highlighted in the
beginning of the book to introduce usability principles). An interactive system with low
usability will likely be of very limited value to be adopted. Imagine frustrated users who
cannot carry out their tasks successfully, and in need of help when they use the system and
can not use it properly due to poor usability, their motivation and trust for using the system
will decrease. It is important that the usability should be comprehensive and embrace the
aspects concerning with human user. On the other hand as pointed out in the definition,
product are used by specified users, the usability should match different users because users’
criterion is various from one to another. But the traditional HCI approaches and principles are
sometimes constrained to achieve these usability aspects. Before trying to solve the problem,
it is important to know these criterions.

2.1 Usability factors

In this thesis, we consider the following aspects of usability, inspired by Soren Lauesen [5],
include broad issues in an interactive system.

 Fit for use (or functionality). The system can support the tasks that the user has in
real life. It concerns what functions the interface/system offers to user. Are these
functions satisfying with the needs of users? Or they are redundant and bring
unnecessary burden to users? In order to fulfill the expect task, interface should
include the functions that help to perform the specified tasks in real life. At the same
time, superfluous function may make the interface more complex. HCI designer
needs to know the users’work task in order to provide required functions.

- 10 -
  Easy to learn. How easy is the system to learn for various groups of users? Does
user easily get accustomed to use the system? Do they need special training? Well
designed interface doses not take a long time for user to learn its manipulations, even
though the interface should be used in various people who have different level of
knowledge (both knowledge about system manipulation and knowledge about the
context task). Try to ease the learning process make the interface more acceptable.

 Task efficiency. How efficient is it for the frequent user? The purpose to develop
software, as we mentioned, is not just to create a tool for people to finish task. What
is more important, it should facilitate the task otherwise it is worthless or even
harmful with occupying the limited resources of computers, especially for a user who
use the system frequently. System designer should consider efficiency to save both
the system resources and user’s time.

 Ease of remembering. How easy is it to remember for the occasional user?

 Subjective satisfaction. How satisfied is the user with the system?

 Understandability. How easy is it to understand what the system does? This factor
is particularly important in unusual situations, for instance error situations or system
failures. Only an understanding of what the system does can help the user out.

These usability aspects proposed by Soren Lauesen are general issues, which are parts of
Nielsen’ s “system acceptability” model [26], and overlapped with other HCI specialists’
usability theories. These factors are ideal situation but hard to implement all of them in one
system, feature of different interaction approaches will help to demonstrate the reason.

2.2 Limitation of direct manipulation in usability

Direct manipulation was original defined by Ben Shneiderman. “Direct manipulation systems
have rapid, incremental, and reversible actions. It reduces errors and encourages exploration”
[8]. The direct manipulation has many benefits and it applied in vast range of computer
programs, from operating system to internet application, from computer games to word
processors. In these systems, interface is a pure tool that mediates the activities between
human and computer, i.e., a mouse, monitor, and printer. Direct manipulating such interface is
a one way interaction between people and computer, it response only if human give it an
order/command, e.g., pushing a button, selecting a menu. The response of the system occurs
in a very short time (perhaps about 100 milliseconds). The user gets reply as soon as s/he
sends a manipulation. The strong point of the direct manipulation is that users have full
control over the system, they are aware of the situation. They do not feel confused or lost
when manipulation the system if something unexpected happen.
But researchers notice that there are also some limitations for users who direct manipulate
the system from interface:
Firstly, while manipulating the interface, human sometimes has to face the virtual

- 11 -
computing environment which they are unfamiliar with, e.g. errors or bugs that can be only
fixed by programmer or system designer, sometimes they have to guess the implicit function
of the system. However, normally users are not computer experts. In most cases interface is
designed for various groups of users. Users only concern about their tasks, and do not need to
know about the virtual environment of the system. A well designed interface should be able to
let most of users manipulate the system as easy as using an automobile— most user would not
care about the infrastructure of a car as long as it could be achieve our goal to send them to
the right destination.
Secondly, software becomes more complex and harder to control nowadays. Information
based on the Internet is rapidly increasing around the world. Consequently computer
programs become more and more complex today. Computers and internet bring us new
technology, which changes our life dramatically with various emerging software programs. At
the same time users are facing heavier and more complex work environment. Information in
the web browsers is not static as it was ten years ago. It is dynamically changing based on
different manipulations from the users. In a diabetic health process, the patient’
s glucose level
are not the same value every day, it even changes every hour or every meal. Diabetic patients’
body conditions are related with their diet, medicine and other unforeseen factors. Nurses and
doctors do not need to visit patients’home as often as before but they face the monitor with
more time and manage those irregular information. The normal way we cope with heavy and
complex workload, according to the other domains, is work delegation, for example,
professor delegates some research to their students. Of course it does not mean the professor
do not know how to finish the work, probably it is because they have no time to work for it.
Regarding to the complex work, there is no better way than to delegate as well. In this case,
people usually delegate the complex task to experts who are competent with it. Also take the
automobile as an example, when our car is broken, the best way is to send it to garage. The
construction of a car is too complex and to fix it is far beyond our ability, even though we
have knowledge about how to fix a car, perhaps most people prefer not to fix it themselves
because it is not their job, they have other things to do.
In a diabetic health care community, doctors and nurses work together with certain
diabetic patients. Doctors may delegate some tasks to nurses when they have too much work.
But the nurses also have a lot of job to do, and they do not only care the diabetic patients. and
the most important thing is when they use the new system, they may have the same problems
as every member in the system community— the complexity of the manipulation toward the
system, then what should they do? Delegate the tasks to others like fixing a car? But who is
competent to this work, the question will be answered in the following section.
Thirdly, the system interface has to be adapted by various users nowadays. In the
mainframe era, most system manipulators are experts in computer domain, because the
computer resources are very limited but now computing devices are ubiquitous and more
people are concerned into the system application process. From old to young, from computer
expert to naïve, they are from totally different domains. Normally, a program are designed for
a certain group of people, they have similar motivation to use the system. Like library system
for students to borrow a book, movie information system for movie funs to retrieve
information and share their favorite movies and e-business system for customers who want to
shop through internet. Ben Shneidermen [9] suggests that accommodation individual

- 12 -
difference is important when designing an interface. The individual difference of users here
has two aspects.
Most HCI researchers have noticed the first aspects and divide users into expert and
novice according to their ability to manipulate the system. When the system is adopted by
users, they will face a new system. Users are “novice”at the beginning and they may need
instructions or user training to learn how to manipulate the system and make the process
easier. After a long term of employment of the system, some people become “experts”, some
functions will be unnecessary for them any more but some new users still need help to get
acquaintance with the system. Doctors and patients become the system experts after they visit
the interface regularly, although direct manipulation researchers advised that designer should
consider this aspect, the changes in interface are quite inappreciable because the direct
manipulation approach towards interface is not flexible. The second aspect to classify
different users is labor division in collaborative support system such as IMIS. Most systems
have one main theme, and are applied for certain goals. Normally users have common ground
to use the system, not only the similar motivations, but also similar tasks and manipulations
towards the system. Integrated Collaborative support system has different situation than just
simply offering a platform to a specified group of people for single task. IMIS combine two
themes into one system and user’s tasks are different due to the focus of their jobs. Since
different groups of users will share a same interface in IMIS system. Immutable interface is so
limited to match so many usability principles mentioned before.
For all the reasons above, agent researchers proposed a more flexible approach to
actualize versatile interface which could tailor itself to adaptive those members in the system.
This research direction integrates artificial intelligence with HCI researches. Even though it
has gained much criticism, agent technology is developing fast and accepted by more and
more HCI researchers.

2.3 Agent technology

The term agent is widely used in computing, AI and other fields. There is not a universal
definition for it. People use different definition to describe their own intelligent system.
During the theoretical study period when I read and search for information about agent
technology, I found many different ways to describe the concept agent [10]. The definition
from Wooldridge’s research is quite incisive and popular.

Definition: An agent is a computer system that is situated in some environment and that is
capable of independent (autonomous) action in this environment in order to meet its design
objectives. [7]

As Wooldridge described, agent is capable of independent action in the environment in

order to meet its “design objectives”. From HCI prospective, our high level “design
objective”is to achieve universal usability for a broader range of users in the system. The
direct manipulation approach has limitations while interacting with users because the system
interface is inflexible, it is hard to design an interface that match for every user’
s demand.
Nevertheless the independent agent provides a flexible way to build a dynamic user interface
for the need of wild range of users. Before investigating how agent could provide flexible

- 13 -
issues to interaction we may need to know agent’ s properties.
In fact, from Wooldridge’s definition, an agent does not need to have advanced ability. A
pure reactive agent could also perform simple task and fulfill the design objective
competently. A good example is the thermostat which could retain the temperature in a
constant value. Obviously, intelligent agent is more useful than this kind of trivial agent.

Definition: An intelligent agent is a computer system capable of flexible autonomous action

in some environment. [7]

For the intelligent agent, the level of agent’s intelligence is not identical. The agent who is
working for the repeated and time-consuming task could have much less intelligence than one
that work for the complex task. An agent could possess different properties in different
systems according to their users working requirements. In Figure 2.1 we list the taxonomy of
intelligent agent’ s properties based on Stan Franklin and Art Graesser:

(sensing and responds in a timely fashion to changes in the

reactive
acting) environment
autonomous exercises control over its own actions
pro-active
goal-oriented does not simply act in response to the environment
purposeful
temporally
is a continuously running process
continuous
communicates with other agents, perhaps including
communicative socially able
people
changes its behavior based on its previous
learning adaptive
experience
mobile able to transport itself from one machine to another
flexible actions are not scripted
character Believable “personality”and emotional state.
Figure 2. 1: Agent Classifications from Stan Franklin and Art Graesser [11]

These properties are interweaved, in other words, an agent need to possess some
properties in order to get others. Wooldridge believes that three aspects below are essential for
an intelligent agent (which performs flexible autonomous actions):

 Pro-active /Goal oriented: generating and attempting to achieve goals; not driven
solely by events; taking the initiative.

 Reactive: maintains an ongoing interaction with its environment, and responds to

changes that occur in it (in time for the response to be useful).

 Communicative/collaborative (social able): It is the ability to interact with other

- 14 -
 agents (and possibly humans) via some kind of agent-communication language, and
perhaps cooperate with others (include people).

Agent should be intelligent and flexible to make up the limitation of direct manipulation
interface’
s drawback; hence these three properties are essential for an agent-based system.

2.4 Human Agent Interaction

After we apply agent technology in target system, the study of human computer interaction
becomes the study of how does human interact with agent based system. As a branch of HCI
research, we go back to see how agent satisfies the usability factors.

2.4.1 Usability
 Fit for use. The system can support the tasks that the user has in real life. The
functions from different groups in a integrate system are different. Some functions
are necessary for a certain group of people but useless to the others. System
developers have to apply functions in such kind that they could be hidden in stead of
disturbing irrelevant people. Agents manage the process and focus on the specified
need of individual user. In the holistic level, agent’
s functionality are considered from
three levels: communication, coordination and collaboration [12]
Communication means agent should enable the information sharing among the
users (diabetic workers in our context). Care providers should be able to
communicate with care takers (or their agent) and read or write information to them
at any time.

Coordination is based on the communication. It means the agent should be able to

mediate users’activities to avoid repeat or inefficient work in real life.

Collaboration is another general functionality that is usually considered in agent

based system. It means agents are also involved in task delegation. Thus we have
found a good solution to the second disadvantage of direct manipulation. Repeated
and complex work in direct manipulation interface would be possible to be
delegated to agent for autonomous action.

 Easy of learning. How easy may the system be taught to various groups of users? The
nature of direct manipulation system is stable; on the other hand, the natures of
human beings are various. People involved in integrated system have different
knowledge about expertise. A universal interface which is easy for every user to learn
is very hard to achieve in this case. Agent’s flexible properties will help different
users to solve this problem.
 Task efficiency. How efficient is it for the frequent user? Direct manipulation
designers stated to provide shortcut for frequent users, but most manipulation still
repeated and time consuming. In health care system like IMIS, efficiency is one of
the most critical aspects when patient’s record is need, especial in vital time. The

- 15 -
 flexible agent provides some autonomous functions such as managing data, warning
or reminding. These functions increase the usability of the system.
 Easy of remembering. How easy is it to remember indispensable details for the
occasional user? One way of designing interface that easy for occasional user is to
design referring to the tradition. It is easy for new users to do what they have known
and they probably prefer to use a system like this: we design our system based on the
old health care systems in the market. And the old paper-based work system is also a
good reference for us to apply the functions. In addition, the agent provides a
proactive property will remind user’ s task in case it is forgotten or conflict with
others.
 Subject satisfaction. How satisfied is the user with the system? Similar as the former
aspects, it is hard to satisfy every user in one system, especially with direct
manipulation interface. Agent is a possible and more flexible approach to achieve a
more satisfaction with more universal user. But how to achieve this is the question.
This will be discussed when we talk about agent’ s learning ability and other
properties later.
 Understandability. How easy is it to understand what the system does? If ask
Shneiderman: “Is it easy to understand what a system does with a direct manipulation
interface?”The answer would probably be positive and he will list many reasons to
convince us, e.g. take the automobile as example again, direct manipulation is
explicit, you change gear box and give gasoline when you want to accelerate and you
step brake when you want to stop. But regarding to the limitations of direct
manipulation we discussed before, error or implicit design sometimes exists in the
system: when people want to drive backward they have to manipulate the gearbox
with different or even opposite ways with different cars. Wrong manipulation leads to
damage of the car so gearbox layout has been changed and some effort are made to
avoid this. Some car gear box become automatic because many people don not need
it and it is more convenient and easy to use it. To use agent is more risky, yet could
be very beneficial. The understandability should be well concerned otherwise will
lead to failure of the system. Understandability will be discussed with character in the
later section.

2.5 User initiative vs. agent initiative (mix initiative)

Agent’s autonomous feature changes user computer interactive from explicit to a more
implicit manner. Direct manipulation researchers persist in importance of user control and
decision support. Unfortunately these technological developments are not in line with a
change in the way people interact with computers. The current dominant interaction metaphor
of direct manipulation requires the user to initiate tasks explicitly and to monitor many events.
“This metaphor will have to change if untrained users are to make effective use of the
computer and networks of tomorrow [13]”. This represents demand of software users who
suffer form heavy workload.
Both direct manipulation researchers and their opponent’ s contestations are quite
reasonable. Those researchers have noticed the different directions where opportunities lay for

- 16 -
innovating in the realm of HCI, and innovation from both fronts have been fast paced. A lot
of conferences and journals pitch agents against direct manipulation. However, Maes states
that they are actually complementary metaphors of direct manipulation [8]. The emergence of
agent changed the traditional interaction of user initiative (direct manipulation) to a
complementary approach of agent initiative, which is also referred to as indirect manipulation.
Agents are not an alternative approach for direct manipulation, their task is to improve
(enhance) the user ability to direct manipulation interface. The integration of agent with
traditional interactive interface will ease the manipulation of the interface. The agent behind
interface as an assistant also takes part of responsibility to manipulate the system. “Instead of
user-initiated interaction via commands and/or direct manipulation, the user is engaged in a
cooperative process in which human and computer agents both initiate communication,
monitor events and perform tasks”[13]. Lieberman stated that “Having an agent operating
directly in the user interface rather than as a background process increases the extent to which
the user perceives the software as acting like an assistant”[8]. Therefore we consider agent as
an intelligent human being assistant behind user interface to execute the task for their users in
the context.

2.6 Interface agent

Definition: Interface agent is a dynamic profiling act as a personal assistant, it is defined as

a character enacted by a computer that interacts with a user in a socially engaging manner.
Some interface agents also attempt to assist users in accomplishing computer-based tasks. As
agents, they can act autonomously to perform operations without explicit directions from the
user, and, potentially, they can collaborate with other types of software agents. [14]

Bernda Laurel [15] has listed several fields that interface agents are appropriate:
 Navigation and browsing: interface take initiative to observe the data, it also take part
of the manipulation of interface

 Information retrieval: interface agent not only observe the data, it also retrieval
valuable data automatically and analyze those data in order to give proper suggestion
to its user. Doctors and nurses work with many patients, they have to observe
patients’glucose records in order to monitor patient’s body condition. It is not
realistic that they stay in front of computer too long time to read different patients’
records all the time. Sometimes they need pay more attention to serious patient
frequently, or they may need some particular information in some emergence
situation (patient feel not good suddenly), interface could do this job to retrieve the
proper information in right time and offer those data to care provider to make
diagnose and to patient for some instruction.

 Sorting and organizing: there are many information stored in the database of system,
the data from different users have different contribution to the members in the system.
The data connected to other aspects influence the final decision from the user. Data
stored in the system are uncertain and irregularly. They are various and might be

- 17 -
 changed at anytime from their owner. Some of the data are very critical for patients’
health like the glucose level record, others such as diet and medicine taking record,
contact address, time schedule are very meritorious information that lead the diagnose
or heal care process more effective. Interface agent is a qualified data manager who
can sort and organize the data for corresponding users from different category of the
health care system. They can even create some data for users after getting their
authorization. There are some security and legal issues involved here when the
system takes some responsibility to manage users’ private data. Because the
limitation of this thesis, we will not discuss it here. However, other paper will discuss
about these issues in more details. .

 Filtering: the information is stored in the system. Although the information can be
well organized and sorted clearly so that it is easy to be found by users or interface
agents when necessary, there are so much information stored that the valuable piece
of information for the diagnosis should be picked out accurately. Different data have
different values for different doctors. Diabetes leads to other of complications,
different views on the patient’ s records from different aspects. And their focuses are
not the same. In other words, not all the information is useful for doctors to cure the
patients. Similar as information retrieval, filtering data requires the interface agents to
be more intelligent. A qualified assistant (interface agent) not only just get the
relevant information to their boss, they should also pick up the significant information
that their boss need for the task.

 Reminding: when people get a lot of work to do, they get easily stressed and the
probability to commit mistakes arises. Our research tries to reduce users’workload
but mistakes or neglects cannot be hundred percent avoided due to human’s nature.
Even though we assume that the users are retentive and the mistake are avoidable, if
not impossible, there are always some issues which is beyond users’abilities and lead
them to a wrong decision under consciousness. Interface agents can remind users
when necessary. And they also combine suggestions from different groups (colleague
or subordinate) to avoid inconsistency or conflicts, e.g. diabetic health care is a social
problem, different specialized doctor should give consistent diagnoses and medicine.
Interface agents should be able to remind doctor if the medicine that is prohibited by
other doctors is given to the user. The user should notice it as well because conflicts
may threaten their health or even life.

 Programming: the programming here is different from the term programming a

computer system. It has a broader meaning. As a procedural software program, the
creativity of agent is confined by its knowledge. Agent can accommodate some of its
function or abilities when it is necessary. Actually agents are usually more familiar
with the system than the “naïve”users in this field. But in the other fields (related
with context), they need the knowledge to assist its user, so interface agent could
“program itself”in some condition [13], this function concern with adaptive property
of interface agent.

- 18 -
  Scheduling: In Sweden, diabetic patients usually take care of themselves at home.
They administrate their diets and medicine. The best way to cure diabetes is to eat,
sleep, and exercise regularly everyday. Thus, personal schedule is important for their
body condition. Besides, Patients visit diabetic doctors sometimes. Some of the visits
are not really necessary and even not well prepared because the visits are not in a
proper time. Doctors are usually occupied with inpatients. Many patients share one
diabetic doctor. The patient visiting a doctor takes a very short time (perhaps between
seven to fifteen minutes). Thus the diagnosis time is precious and doctor and patient
should make use of the visiting time effectively. Interface agent can arrange time for
users. They instruct patients the proper eating, sleeping and insulin rejecting time and
also guarantee that the consultation with doctor is worthy and efficient.

 Advising: interface agent is an expert system. It combines many aspects and

considers the information that may affect the user’
s decision and gives advice based
on users’ intention. The interface agent’s advices should be constructive and
reasonable. Agent should know users’goals and intentions so to give pertinent advice.
This requires good communications with users. Next chapter (how to construct an
interface agent) will discuss this.

Besides, interface agents are also competent assistant for providing help, coaching,
tutoring, performing, the role of interface is versatile, but in the context system, specialized
capability is more appealing and helpful users.
How could the interface agent be a qualified assistant to our users to gain trust and
satisfaction, we will discuss this latter in interface’s personalization.

2.7 Character (personalization)

Agent’s initiative and user’s initiative are both important in agent-based system. Agents need
authorization or confirmation from user for the work that they are doing. Sometimes this is
out of control. Agents give suggestions to users in most cases. Users make decisions if the
suggestions are adopted or not. Like direct manipulations researchers always emphasize and
cite to debate their opponents. “User understanding is central, and user control is vital for
people to be successful.”[8] It is our believe that users should understand agents’activities
and we also insist that agents as intelligent units, remains highly autonomous, which is one of
the most critical aspects for a successful helping interactive interface [16].
We have discussed pro-activity by presenting the agent initiative capability in section 2.3.
But some questions are still open: how does interface agent know the users personal
information (e.g. preference)? When should interface agent take initiative and how could they
perform initiative to satisfy users need in a perfect way? Maes gave us one answer in her
research: “it (agent) takes initiative because it knows what your interests are, it can, for
example, tell you about something that you may want to know about based on the fact that
you have particular interests.”[6] This instructed the relationship between proactive and other
properties that they should cooperate together to answer our research questions. And character
is one of the most important because people in different positions have different “interests”.

- 19 -
2.7.1 Tasks differences
It is common to consider interface agent as an assistant who possess the same character as the
real human assistant in real world environments (perhaps a much more submissive assistant
who will never maintain different opinion to its user). Thus it makes sense to construct the
interface agent that simulates the real human assistant. As we assumed that interface agent
could be a versatile assistant that could qualify many kinds of jobs based on the context. But a
versatile assistant and competent assistant are not the same term in the real life. A diligent
secretary who is familiar with his or her work probably get dismissed if he or she does not
follow his boss’s will or does not know his boss’s intention, which means people will never
tolerate a assistant with such kind of behavior. It reveals a truth as a primary principle while
constitute the characters of interface agent that interface agents should focus on the intention
of users in order to tailor itself to satisfied the user’s requirement. For example, doctor’s
interface agent should focus on the patients records and give warning or suggestion in right
time, nurse’s interface agent pay more attention on medicine plan and relevant information.

2.7.2 Different preferences

Horvitz [17] has pointed out some problems with the use of interface agents: “poor guessing
about the goals and needs of users, inadequate consideration of the costs and benefits of each
agent action, poor timing of agent actions and inadequate attention to opportunities that allow
a user to guide the invocation of agent services and to refine potentially suboptimal results
provided by the agent”. Thus it is a remarkable question for interface agent developers while
designing a interface that we should consider how interface agents overcome the problems
above and provide best interactions to each user and what is more important, is how interface
agents “provide their users proper assistance at the most appropriate time”[18].
The reason we adopt personalization in diabetic care system is obvious. We can not find
exactly two same people in the world. Patient’records and safety level are different and, they
also prefer different behaviors according to the same record. Personalization of this kind
ascribes to the desires of different interaction way. An example from the doctor’ s perspectives
in diabetic health care process can explain why doctor what to interact with patient in
different ways: two diabetic doctors may have different habits to view a patient’ s record
because they work in different ways and also because they have different experience. It is also
quite normal that they would like to view the data in different ways for different patients. The
doctor who works with a new patient may need to turn to more comprehensive information to
get to know the general body condition of the patient. On the contrast, if one knows the
patient well and is familiar with his/her diabetes situations, s/he could give a quick diagnosis
from experience and knowledge accumulating from previous clinic records. Much
unnecessary information can only slow down the diagnosis process and make the work
inefficient. As the same reason, we could make the conclusion that “in the same way as each
person works and interacts in a certain way with his/her personal assistant, each user interacts
in a personal way with an interface agent”. [18]
Some experimental studies from [18] have proved the hypothesis that when people work
in different environments, their requirements are different, and their preferences, habits and
interests are various from one to another as well. There is one thing we need to take into

- 20 -
account that users tolerate toward agent’s suggestion is important for agent’ s survival. Agent
should always focus on gaining trust and competence from certain user. They should be
qualified to their work and actions from them should be understood by the user, since user is
more passive in an agent initiative system. They will abandon or use the agent, or what is the
worse case, they may refuse to use the system if its action is too brusque and impact on user’ s
ongoing work. A famous negative example is from Microsoft, the Microsoft office agent has
gained vast of criticism from designers for its disrupting users work, even if Microsoft
Company made office agent easier to hide in the update version, complains keep on coming.
So now the agent does not appear in the default in the newest Office software.
In our context, the requirements from users are not very different. Care providers focus on
users’history records and the preferences. Meanwhile it is notable that agents give
suggestions to doctors after analyzing the data from the patients, but they could also make
decision according to the analyzing result. We propose agents have more autonomous
functions to provide contribution for competent tasks, but the errors they made should never
beyond users’tolerance. Whether should interrupt the user or not, should agents give
suggestions or just do it without informing the user? Those factors should be deliberated by
agent before carry out.
Besides personalization in the context system, customization is also adopted to ease the
interaction between user and the system. Compared with personalization, customization is
easier to implement and more explicit. Many systems have noticed the significant of
customization, and they provide many ways to users to customize the system environments,
such as changing the background color, highlighting the most commonly used text, etc. A
good example is Microsoft windows XP operating system. The difference between
customization and personalization is that the former is generally initiated by user and the
latter is basically generated by agent.

2.7.3 Approach to build personalization. (Learn ability)

“Personalization involves a process of gathering user-information during the interactions with
the user, which is then used to provide appropriate assistance or services, tailor-made to the
users needs” [19]. The process of gathering user-information is also the term as learning
process of interface agent.
There are three approaches to building agents according to Maes [13]:
The first approach consists in making the end-user program the interface agent. It is more
like a “semi-autonomous agent” because its autonomous ability is restricted. It collects
user-programmed rules for processing information related to a particular task. The second
approach which is called “knowledge-based approach”is more advanced and autonomous. It
consists in endowing an interface agent with extensive domain-specific background
knowledge about the application and the user (called a domain model and user model,
respectively). At run-time, the interface agent uses its knowledge to recognize the user’
s plans
and to find opportunities for contributing to them.
Both of the two approaches have drawbacks while applied. The first semi-autonomous
approach relies on user providing information to them, despite agent from this approach is
more trustable. But it is not competent because the agent is not independent and users have to
do much work with it. User will get impatient if he/she has to teach the agent all the time.

- 21 -
Knowledge-based approach has adopted more widely in intelligent user interface, but it brings
on problems both from competence and trust issue: the agent is customized by programmer;
the extensive knowledge is from someone else even it embraces a broad range and designed
to fit for all. It has conflicts with our belief that personalization is important.
So an alternative approach is demanded to gain more competent and trust from user. The
learning approach proposed by Maes enhances the communication between agent and user
thus enable designer to build user profiles based on machine learning techniques.
In the learning approach, agent is given minimum of the background knowledge from
designer. During this early period, the interface agent is not very familiar with the habits and
preferences of its employer so that assistance is confined and may be not very helpful. But it
can learn appropriate behavior from its employers and even from other agents for the sake of
action on behalf of user for their work.

Figure 2. 2: Four ways for interface agent to learn: 1) it observes and imitates the user’
s
behavior. 2) It adapts based on user feedback. 3) It can be trained by the user on the basis of
example. And 4) it can ask for advice from other agents assisting other users. [13]

A learning agent acquires its competence from four different sources (see figure 2.2). The
dashed line in the figure indicate the four information flow for agent to gain knowledge
(1) It observes and imitates the user’ s behaviors. Agent observes the activities of users,
picks out user’ s interests and summarizes his/her habits or preferences while
performing actions in the tasks.
(2) It adapts based on user feedback. Communication between user and agent is vital.

- 22 -
 User remains control over and agent gains trust when the user gives direct or
instruction after agent’s activities. For example, user may shut down some functions
or refuse some suggestions from agent if they are not satisfied with them.
(3) It can be trained by the user on the basis of example. Users can train the agent based
on their needs. If they need agents to do something particular, they can customize
the function by giving some hypothetical examples to tell them what to do in those
cases.
(4) It can ask for advice from other agents assisting other users. The communication
between agents is beneficial and appealing. In an information system which cover a
big range of users, it is no denying that users are various from one to another. On
the other hand, some of them share the information in the system and they have
many similarities to cope with the information. As a result, they may perform the
same actions in the environment. That is to say, the agents who work with different
users may also share the same information and knowledge. Taking advantage of this
feature and to learn that knowledge from other agent will ease the work from user’ s
prospective.

2.8 User profiling

Figure 2. 3 different components of user profile [18]

In [18], information was sorted into three different categories based on the personalization
issue we discussed above: 1) Application independent information include the static
information of basic description of user such as name, age, gender, telephone number, email
addresses, etc. 2) application dependent information include issues related to user’s work
requirement, needs, interests, goals, etc. 3) user-agent Interaction information contains
information describe the appropriate action agent should carry out based on user’s preference
to communicate with the system, for example, reminding and suggestion are more acceptable
than warning and rudely interruption.

- 23 -
 The information described in Figure 2.3 is from the four sources showed in figure 2.2 and
could be described as general issues about user when interface agent manipulate the system
and try to finish a task. The IMIS interface agent with the information listed in Figure 2.3 is
illustrated in Figure 2.4.

Figure 2. 4: IMIS assistant agent

In order to constitute personalization for different users, agent constructs user profile for
each user. After building a user profile, the information is stored in the profile. Like human
being need brain to thinking and make decision. When agent takes initiative to finish a task
for its user it needs to make use of the knowledge stored in the profile.

Conclusion
This chapter investigates how HCI can benefit with the agent technology. After a short review
of the Human computer interaction approach and a detailed comparison we can draw a
conclusion that old direct manipulation approach is limited and intelligent assistant (interface
agent) for manage and manipulate information from web information system is a flexible and
remarkable approach to achieve more usability and flexibility issues. The knowledge we have
gained and which is presented in this chapter contributes to our understanding of how human
assistants (interface agent) can be integrated into our system to facilitate users and what the
effects of this effort might be. We also contribute to the understanding of how such interface
agent can and should be designed. Our pragmatic approach to intelligent help will enable
designer to construct and deploy an agent-based working help system at an existing diabetic
health care prototype system, lending support to our hypothesis that interface agent renders a
solution to many of the complex and heavy workloads formed by the employing of new
system with fully automatic approaches. This in turn allows us to study the usage of the
system and complement the usability aspects for further work.

- 24 -
3 Framework of agent-based interface
This chapter will give a description of agent-based interface design methodology and
social-psychology theory— activity theory, and then combine these methodology to model
IMIS agent system and construct agent coordination mechanism.

3.1 Methodology
3.1.1 Gaia— agent-oriented analysis and design methodology
The existing technology in software engineering, i.e., object-oriented analysis and design, is
not competent enough to conceptualize notions in agent-based system. So many previous
approaches fail to capture and realize flexible, autonomous and other properties of agent [6].
Wooldridge, Jennings & Kinny inspired an agent-orient analysis and design approach to solve
this software engineering problem. Figure 3.1 shows the relationship between different
models of an agent system in a general case.

Figure 3. 1: Relationship between different models in agent-based system [6]

The figure above also illustrates the stage of agent-orient analysis and design. There are
two stages in the engineering process: agent-orient analysis and agent-orient design. Gaia [6]
summarized these two processes:

 The analysis process

1. Identify the roles in the system. Roles in a system will typically correspond to:
 individuals, either within an organization or acting independently
 departments within an organization
 Organizations themselves.
Output: A prototypical roles model— a list of the key roles that occur in the system,
each with an informal, unelaborated description.
2. For each role, identify and document the associated protocols. Protocols are the
patterns of interaction that occur in the system between the various roles.

- 25 -
 Output: An interaction model, which captures the recurring patterns of inter-role
interaction.
3. Using the protocol model as a basis, elaborate the roles model.
Output: A fully elaborated roles model, which documents the key roles occurring in
the system, their permissions and responsibilities, together with the protocols and
activities in which they participate.
4. Iterate stages (1)-(3).

 The design process

The Gaia design stage can now be summarized:
1. Create an agent model:
 aggregate roles into agent types, and refine to form an agent type hierarchy;
 document the instances of each agent type using instance annotations.
2. Develop a services model, by examining activities, protocols, and safety and liveness
properties of roles.
3. Develop an acquaintance model from the interaction model and agent model.

These two processes model the notions of organization (diabetic health care for in our
case) from abstract concepts to concrete concepts respectively. The central and basic notion of
Gaia is role model. It corresponds to the roles of organization in real life. As an example in
diabetic health care, there are individuals that take the role of doctor, diabetic nurse, district
nurse, stuff nurse, personal assistant and so on. The instantiated roles from different members
are not static all the time. In another word, many individuals could be qualified to serve for a
patient as long as they have the required knowledge. On the other hand, these instantiated
roles have to interact with each other in order to take a proper service in most case.
To define a computer system as a set of roles is a novel approach. It works well in our
agent-based system and also helps me to fine the point where activity theory could contribute
to the model process.

3.1.2 Activity theory

The activity theory was original emerged and elaborated by Russian psychologists in the early
of 20 century. It is a philosophical, psychology and cross-disciplinary framework for the
study of human practices as development progresses. Recently Bonnie and Kuutti figured out
that the traditional cognitive approach in HCI is limited because it is unable to penetrate the
human side of the interface [4]. And their alternative approach— activity theory in HCI field
is noticed and accepted by more and more researchers. Especially in Scandinavia, there is a
thriving tradition to integrate this cross-disciplinary theory in HCI studies.
In activity theory, there are many activities carried out in the organization. Our analysis
objects are those individual activities as basic unit. Structure of activity theory is shown in
figure 3.2. An activity is undertaken by a subject, the subject uses tools (also term as artifacts)
to manipulate or achieve an object, this process leads to transforming object into outcome
(illustrated in the upper triangle). The object is also regard as the motivation for an activity
and could be either concrete or abstract like plan or idea. The relationship between subject

- 26 -
and object of an activity is mediated by tool (artifacts). A tool could be anything used during
the transformation process, the form of tools could also be concrete or abstract (such as some
theories, way of thinking or form in the organization).
Since the structure with these three notions is too simple to explain the relations between
an individual and its environment in an activity, Engeström extended the basic structure to a
more complete version (see figure 3.2). In this updated structure, a subject exists in a
community. At the same time the community shares the same objects. The community and
subject relationship is mediated by rules, and the community and object relationship is
mediated by “division of labor”. According to Kuutti [4], there may exist a set of rules for the
subject to decide how it can be a member in the community, these rules include implicit and
explicit norms, conventions and social relations within a community. Division of labor under
the organization mediates the relations of community and object. It refers to the explicit and
implicit organization of a community as related to the transformation process of the object
into the outcome.

Figure 3. 2: basic structure of an activity [20]

Uden and Willis [27] Stated several benefits that activity theory can offer to HCI design:
•It offers an approach to conceptualize relationships between individuals, communities,
technologies and activities.
•It models expertise as an active, collective phenomenon, and in the importance it ascribes to
collective learning; it provides the understanding of context in which computer-supported
activities take place during design and evaluation.
•It considers the computer as a special kind of tool mediating human interaction with the
world. Meaningful goal-directed activities constitute the context for both mental processes
and external actions.
It validate that well designed interface should be transparent [9] because user should not
consider interface as a tool to achieve their goals in the organization, they are actually acting
though the interface [2].
In addition, Kuutti [4] also presented the three level hierarchies in activities: activity,
action and operation. These hierarchies are illustrated in Figure 3.3. They are corresponding
to motivation, goals and conditions (see Figure 3.3). The lowest level of this hierarchy is
operation. Operation is something that we have known so well that we do not need to
deliberate before we do it. As long as the prerequisite or the condition for this operation is

- 27 -
fulfilled, it can be realized. The middle level is action which consists of several operations. It
is driven by goal, that is to say, before we are going to do something new to us, we must have
a plan or a goal and always keep it in our minds. However, when we get used to this action
and do not need to deliberate or make a plan for it, this action collapses into operation. The
highest level is activity, which in turn consists of many actions. The activity is driven by the
motivation. The motivation is need or desire that the subjects of the activity should answer
together. An activity is considered collective, since it always focuses on complex
interrelations between the individual subject and his or her community [24].

Figure 3. 3: hierarchies of activities

It is notable that in this hierarchy, one operation may belong to different actions and
similarly the same action could be used as contribution to various activities. This feature is
coherent with the agent-orient’ s role model. That is, one role could participate in different
services. A service could be considered as a collective activity and it consists of many
subjects with different roles that interact with each other to achieve their individual goals.

3.2 IMIS system case study with Gaia

According to the Gaia agent-orient methodology described above, the stages of our target
agent-based system could be described in the figure below: this figure depicts the stages of
Gaia analysis process. The multi-agents system consists of agents with various roles, some
roles could contribute to help patient like doctor, and others could act like other care providers
in the organization. Permissions indicate the authority (or “rights”) of corresponding roles
which could be read a piece of information from other roles, modify this information or
generate new information. Responsibilities include two properties, which are liveness
properties and safety properties. Liveness properties state that “something good happens”,
and safety properties state that “nothing bad happens”[Gaia]. The former properties shows
something actions will be done to achieve the goals with a certain permissions and the latter
one represent the precondition when should agent initiative those actions. Protocols enable
roles to communicate each other and the interaction model composes of roles interacting with
those protocols.

- 28 -
Figure 3. 4: stages of the agent-oriented analysis in the MAS system

Figure 3.4 illustrate the stage of the Gaia analysis in IMIS system. Because the limitation
of this thesis, I only list some roles’schema to illustrate how Gaia could be applied in our
system. For more information of different IMIS models, see Appendix A. As we mentioned
before, there are actually several groups form different department share the IMIS system.
During the Gaia agent-orient analysis and design we define these roles: BloodGlucose
Handler (BGH), Diagnosis Advisor (DA), Medication Manager (MM), Profile Vetter (PV),
SideEffect Checker (SEC) and Patient Assistant (PA). IMIS multi-agents system consists of
many roles models like these.

3.3 IMIS agent coordination mechanism based on activity theory

Besides the model from Gaia method, a more general Human-agents coordination mechanism
in IMIS system has also been constructed based on activity hierarchy. As we have introduced
the three hierarchies in activity theory, the work group of health provider is divided into three
levels: co-construction, co-operation and co-ordination [3]. Co-construction represents the
understanding and translation of an activity among the whole work communication.
Co-operation represents the action level where actions are performed, which is driven by a
certain group of goals. And co-ordination represents some automatic operations. The three
levels hierarchy based on the activity theory is distinguished in the agent coordination
mechanism showed in Figure 3.5. This figure has the same form with the IMIS group member
(human) coordination mechanism which has been elaborated in [3]. And figure 3.5 illustrate
the agents coordination mechanism in our context— IMIS system, it also shows the agent

- 29 -
delegation in health care system. The members in diabetic health care are connected with
agent in this mechanism. The dash-dot lines stand for the internet based communication with
either stationary PC or pocket PC. The double way arrows between user and agent stand for
the Human agent interaction. The arrows between different agent shows the agent task
delegation flow, and the dash line classify the work provider into three distinct levels, their
activities are triggered by patient, because every care service are initiated by the patient health
records.

Figure 3. 5 : Agent coordination mechanism and agent work delegation

In real health process, doctors on the top co-construction level are in charge of the whole
process. And they interpret and analyze the information from patient. To decrease doctor’ s
work load, and increase coordination to avoid conflict, the doctor’s autonomous agent
initiative some delegation according to the analyze result to high-level nurse’s agent. The high
level nurses include diabetic nurse and distinct nurse. They are on the second level. As I
mentioned, people on this level emphasize to perform the actions driven by certain set of
goals. In IMIS, these health care actions are divided and delivered by doctor (or probably, his
interface agent) from the higher level. Nurses on this level are usually responsible for medical
cure. They have explicit goal to perform their tasks. Different nurses focus on different
aspects. Their actions cooperate together to cure the patient. After agent receives instruction
from higher level agent, they will interact with nurse to inform them about the task, or they
can autonomous react the task delegation. Like their master, interface agent in this level work
for their individual goals to avoid conflict between each other. The high level nurse in turn
will delegate task to low-level care providers. Staff nurse and personal assistant are on the
third level. And it corresponds to the operation level in activity: they only do what they are
told. Unlike higher providers, they only work in a certain condition: staff nurse inject insulin
according the medication plan and personal assistant do some home service rather than home
care to improve patient health indirectly. Their interface agent on this level is driven by
conditions, e.g. update records about the patient after health care. The information can be
helpful for care provider (or their agents) on higher level for future use.
In this agent coordination mechanism, each member in real word could have an interface

- 30 -
agent accompany with them. These competent agents will work on behalf of each member.
They work on the same level as their user and also work in the same way according to the
user profiles which are built from a long term communication. Tasks are delegated and
finished by communication with different agents on different levels.

Conclusion
As an experimental case study based on IMIS system, we have modelled the MAS framework
with Gaia and activity theory. This chapter have discussed the models by following Gaia’s
analysis and design procedure. After defining roles model, interaction model in analysis
process and the corresponding agent model, service model and acquaintance model in design
process, agent’ s functions are available in IMIS system. These definitions also enable us to
combine activity theory when we introduce the agent coordination mechanism. Based on this
mechanism, different agents collaborate to achieve the same goal (cure the patient). With the
help of agent, the complex health care process is decomposed in terms of many sub-tasks and
they are in turn delegated to other agents. Of course agent will take the responsibility to fulfil
all the time consuming work and to help user as we discussed in the theoretical analysis.

- 31 -
4 IMIS prototype
This chapter introduces the experimental implementation process of IMIS system (screenshot
and functionality detail of the prototype, see Appendix B). According to our study, there are
two layouts of interface in agent-based system: the human-computer interface which enable
direct manipulation and the agent-environment interface which enable autonomous services.
These two layouts are integrated as a tool to help users in health care organization. This
chapter provides a prototype of agent-based IMIS system. Before we start to define the
function of our system we need to understand the context. Hence we will start this chapter
with task analyses.

4.1 Task analyses

Task analyses or also sometimes called requirement gathering process is the first step in
prototype design. It decides the functionalities of the system. The task analysis of our
agent-based system includes these issues:
Identify different users from different association (both care providers and patients)
Get their relevant information from user profile, which includes history communication
inform for each user, related work group’ s information or patient’s history records, etc.
Observe the given information from patient in runtime and give reaction.

4.2 IMIS activity structure

In IMIS system we propose activity theory as an underlying framework to analyze the context
of our research. In a diabetic system, many activities could be carried out in order to take care
of a patient. Relations of component in the system are shown in Figure 4.1.

Figure 4. 1: A diabetic health care structure with activity theory

This structure is an application of activity theory in our case. The subjects of the care
activity are care providers (doctor, nurse, personal assistant), and the objects are patent and
their daily body records. Care providers use artefacts (information from Health care system)
to mediate the activity in order to transform the object into outcome. They will give some
diagnosis according to patient’ s health situation. In the whole organization, a patient is usually
cared by a certain group of providers. Therefore the communication consists of many work

- 32 -
groups which have a certain responsibility to work with different patients. Norm (rules) which
mediates the community list some laws to specify the work for the community. The labour
division specifies the detail division of care provider members in the organization.
In the human-computer interface which interface agent is not adopted, the subjects and
objects are the human beings who should handle the activity by themselves. We have also
added agent as a part in this activity structure. Agents as a part of the health care system
perform some activities on behalf of users. Thus some agents work in the position of subjects
and some works for objects as it is illustrated in Figure 4.1. Their work load will be reduced
and it is much more efficient to care a diabetic patient with this structure.

4.3 Architecture of IMIS interface

Unlike other web based information system application which faces to specific customer,
IMIS system is a multifunctional system that could be used for people in different group. And
it is cover the whole heal care process, from primarily home care to hospital clinic. The
system interface should represent many functions needed to carry out the activity based on the
context. To provide a more flexible and comprehensive interface to strength interaction
between user and system, we have extended activity theory to analysis the task requirement.
The architecture of IMIS interface is described in Figure 4.2.

Figure 4. 2: Architecture of IMIS interface

The IMIS interface contracture includes six main components based on activity and support
the function required in a diabetic care process.

 Tools: shared agenda and paper reports of patient’

s records

- 33 -
  Subject: (Work group\health care provider): it includes several groups of care
providers. Because care providers usually work cooperatively against certain patient,
category of those different groups facilitate to find information about them when
necessary. Conflict can be avoided and work becomes more efficient by listing their
schema. It is clearer for the group supervisor to delegate task and arrange the activity
if the detail information are available.

 Community (Network): network list the information of the community of the

organization. It embraces related information of health care in a more broad scope,
such as information of bank, society, supermarket, etc. these organizations are
involved in the health care process. For example, diet is important for patient so that
they need to check the food information in supermarket. They also might need to find
information about medicine if they need to buy some from shop.

 Rules: this section lists some rules or existed laws existed in the organization or the
society as a reference in the care process.

 Objects (Care taker): it includes the information about the patients (care takers). The
information of care takers is important to make the diagnosis. We list patient’s
address, agenda, routine, requirement assessment, economy, personal network,
message and records. These are relevant to the diagnosis making from subjects.

 Work division: this part includes the information about work division’ s information,
for example, the area leader, physiotherapist, psychologist and so on.

IMIS system is an integrated system with a big range of community. The user groups are
divided into two main part— care taker and care provider. They use the same portal to visit the
system. When user login the system, interface agent will be created and it will automatically
identify different group of people. The purpose of this process is to identify different users
and get their profile, and bring them to the proper service page. This could also skip some
unnecessary process for reliable and frequently user so that the interface became more
user-friendly for them.

4.4 IMIS agent Communication

The functionality of agent contributes on three levels: communication, coordination and
collaboration. Coordination and collaboration requires communication, it is the precondition
of the former two [3].
By using interface agent in the IMIS system, the work stress in communication from
care providers and care takers transform to the communication of their corresponding agents.
Since interface is autonomous assistant, it takes initiative to request and to respond to other
users’agents. Instead of the old ways of communication by phoning or sending information
by fax, agent communication is much more efficient and we provide a good platform to
record the communication message, interface agent will manage the communication

- 34 -
information and save them into user profile. History message stored in the platform could be a
good reference for future diagnosis.
Agents communicate with each other in various ways in the system. They not only
communicate with the other agents (which may have different responsibility) but also
communicate with human. Thus agents use some languages to understand each other. These
languages are called Agent Communication Language (ACL), such as KIF[21], FIPA[22], and
KQML[23]. In order to understand human user, agent also need to be able to translate human
nature language to ACL.
A common approach to implement human— agent communication is to offer a
consultation dialogues. It is a visual way that user could give agent direct clue to agent about
their interests. But more clever and autonomous agent should be able to observe user’ s
behaviour and transform the invisible information into database with ACL. These data could
be reused next time when the user initiates the same activity. This way is very appealing when
user frequently repeats the same action, but it is important to remember that the information
piece agent captured by guessing could lead to misunderstanding without confirm by user,
one motivation of communication between agent and user is to keep on correcting agent’ s
goals to take initiative. Agent’s goals to take action and the user’s intention should always
keep coherent. In the early learning stage the information about the user in agent’ s profile is
limited and errors or misconceptions are easy to occur. Knowledge accumulation is a
long-term process, luckily, the learning ability make agent more and more mature and
competent. Thus, after communication and correction, agent will have explicit goals for their
actions. E.g. diabetic doctor may frequently query a patient’ s data after clinical cure. Agent
guesses this patient need special observation, and stores his/her data to doctor’ s profile. After
being verified by doctor and then setting the goal to observe this special patient, agent will
initiate a couple of actions to help doctor to get the information before doctor make the
queries every time repeatedly. Sometimes doctors frequently fetch the data in a short time for
some other reasons. They do not want to keep track of the information for a long time. Wrong
actions which ascribe to lack of communication may annoy doctor.

4.5 Scenarios
Before we start the scenario of system design, I believe important to introduce a true story
from [3]. This story is about a diabetic patient and it reveals the collaboration problem in
health care: Maria (not real name) has several health problems of which one is diabetes of
type one. There is a staff nurse from the home care center of the county council, who visits
her home twice a day to give her insulin injections. In addition, a personal assistant from the
municipal social service helps her with some housework. Maria seems not very concerned
about her diabetes, and she often eats food and candy that is not good for her. The
concentration of blood glucose often goes up steeply. One day the personal assistant called the
home care center and reported to the district nurse that Maria did not feel well. The district
nurse suspected that Maria had eaten something sweet. The district nurse checked Maria’s
record of her recent medical history. Then the district nurse went to Maria’s home with a
device to measure the blood glucose. After asking several questions to Maria, the district
nurse took out the device and tried to measure the blood glucose for further decision.
However, the district nurse was not familiar with the device and could not understand what it

- 35 -
showed. After a phone call to the diabetic nurse (primary care nurse) who works at hospital,
the nurse drew a conclusion that Maria was in an acute situation. Immediately Maria was sent
to the doctor.
In this story, coordination between every member in a health care process is not well
arranged and organized. One problem in the story is there is not a formal way to store the
information: the staff nurse knows the patients body condition very well, but his information
may not available when diabetic nurse or doctor need to make the diagnosis. IMIS platform
solve the missing information problem, but the information should be interpreted by authority
(e.g. doctor) to make the finial diagnosis, and relevant task will be delegated to lower level
care providers, with this way, diabetic health care is more efficient and effective.

 Scenario

The same patient Maria is now offered a computer and the IMIS system. She measures her
blood glucose every day with some medical device and input the data into computer. When
the device gets the blood glucose information and reports it to the IMIS service, her interface
agent will analyze the data and tell her if the body condition is ok or not. Besides, she also
arranges and records her daily activity which may influence the blood glucose. Of course
agent as an assistant will remind her about those activities and records the result to reduce her
work load. When the agent found blood glucose level is in dangerous level, it will give her a
warning and the proper activity to suppress the blood glucose. After adopting agent’ s
suggestion, she feels better. But in the following days, the same symptom appears again and
she does not feel well. Her personal assistant also finds this. But personal assistant’ s agent
does not have permission to retrieve patient’ s history record for some security issues.
Fortunately, patient’s interface agent can easily get the history records. Therefore an alarm is
send by patient’ s agent to her high level care providers with those relevant records. With the
valuable records, high level care providers make a conclusion that patient Maria need to be
sent to hospital immediately so that a ambulance is sent to Maria’ s home. This decision will
be saved by agent and delegation will be delivered back to the low level care providers to
inform that special care a important before patient are sent to hospital. Their schema is also
automatically changed by agents since patient will stay in hospital for several days.

- 36 -
5 Discussion and future work
In this thesis, we have designed the IMIS system functionality for both agents and direct
manipulation interface. A prototype is provided to illustrate these functions. In the previous
discussion (see section 3.2), we have proposed the model of individual agents and agent’ s
services from micro and macro prospective respectively. We emphasize the macro level of
agents when designing their functions in this paper. An agent coordination mechanism
integrated with activity theory is therefore introduced based on the previous research to be
applied into the prototype of IMIS system.
With the agent-based approach, the IMIS system is more user-friendly because it could
learn from different users and tailor itself to fit for different users’requirement. This is a very
important and basic issue for an integrated system. In the practical case study, we adopt the
Gaia to define the roles of the organization. It is a very useful method since define roles of an
organization when developing a system is still a big challenge for most designers. Roles in the
organization is more flexible and it conceptualize the notions from obstruct to concrete
services. It is much easier for programmers to implement the agent-based system.
There is no denying that some new problems come about with our new solution in the
diabetic health care system, and also some more work need to be done in future research.
Security: In IMIS system, patient’ s personal health records are available on internet, most
of these record are private and should not be open by irrelevant person with out permission. In
addition, other important private information like bank account and pension information is
also available through the internet. Thus security issue is a significant problem which should
be taken into account when design the system. We have noticed this problem and endeavor to
avoid unnecessary trouble cased by security issue (e.g. we identified the authority of different
care providers when they login and retrieve the data, low level care providers can not get the
personal records without any permission). However, more security problem should be
concerned when agents has the authority to handle and transfer the data in the future work.
Evaluation: Evaluation is a long and iterative process that should cover the whole design.
We have done some heuristic work though evaluation during the previous design, but a more
comprehensive and systematic evaluation should be done in future work to complete and
improve the functionality of IMIS system.
Implementation: In the current IMIS system research, a prototype has been developed.
The prototype includes Hand-drawn mock-up, tool-drawn mock-up, screen prototype and
functional prototype. But the final version of agent-based system is still under developing.
The implementation will use agent design tools such as JBuilder, VisiBroker, JADE (Java
Agent Development Framework) and so on.

- 37 -
6 Conclusion
Now we return to answer our research questions (see section 1.2). By approaching agent
technology, the traditional human-computer interaction becomes human-agent interaction.
Compared with the direct manipulation in HCI, an agent is a multi-way communication
between users and the system. It satisfies the usability criterions better and also makes the
system more flexible and more user-friendly to a broader range of users. Agent will enhance
the interaction between human users and computer system.
The second question is a bigger challenge not only in our context but also in other
application in similar domain. Users are able to share an interface even though their tasks are
different. We provide customize and personality to build user profile for different users. And
what is more important is: we use activity theory to analyze user’ s tasks and model them with
agent-orient method. This experiment also implicit why task-centered design (with activity
theory) start to become a tendency in HCI and researchers realize “the activity surrounding
users could be more emphasized than users themselves”[24]. With the activity theory based
coordination mechanism, delegation is very explicit when activity is motivated by patient, and
each care provider in the system has a very explicit goal to work for the patient. The
relationship between every member in the system is also very clear and information flow is
formal and stable. Once information is created or modified, other related members will be
informed. These services are carried out by IMIS interface agent who is working on behalf of
users like an assistant. Their services are with many protocols that defined in the case study
process. With the help of the agent-based system, the work in the organization could become
more efficient and effective. However, due to the limitation of my thesis and the current work
progress of IMIS system, the prototype designed in the paper is not completely evaluated.
One the other hand, security is also a very significant issue in IMIS system. A more
comprehensive evaluation will be carried out after more functions of IMIS are implemented
and security issues will be particularly discussed in other thesis.

- 38 -
7 Reference
[1] Shneiderman, B. 1997 Between Hope and Fear. Communications of the ACM Volume
40, Issue 2 ISSN: 0001-0782, p 59 - 62

[2] IMIS: www.IMISCare.org, (Jun 2006)

[3] Zhang, P. 2005 Multi-agent Systems in Diabetic Health Care–from the perspectives of
Activity Theory and Systems Science. Blekinge Institute of Technology Licentiate Dissertation
Series No 2005:05 ISSN 1650-2140 ISBN 91-7295-060-9, p 3-56.

[4] Kuutti, K. 1995 Activity theory as a potential framework for human-computer interaction
research Context and consciousness: activity theory and human-computer interaction ISBN:
0-262-14058-6, p 17-44

[5] Lauesen, S. 2004 User Interface Design: A Software Engineering Perspective. ISBN:
0321181433, p 3-38.

[6] Wooldridge, M. Jennings, N. Kinny, D. 2000 The Gaia Methodology for Agent-Oriented
Analysis and Design Autonomous Agents and Multi-Agent Systems Volume 3, Issue 3 Year
of Publication: 2000 ISSN: 1387-2532, p 285 - 312

[7] Wooldridge, M. An Introduction to Multi-agent Systems, JOHN Wiley & Sons Ltd, 2001

[8] Shneiderman, B and Maes,P. 1997 Direct manipulation vs. interface agents, Interactions,
vol. 4, p 643-661

[9] Schneiderman, B. 1998 Designing the User Interface: Strategies for effective
Human-Computer Interaction. Addison Wesley, p 18-90.

[10] Honavar, V. July 1999 Intelligent Agents and Multi Agent Systems. A Tutorial presented
at IEEE CEC 99. Washington, D.C.,
www.cs.iastate.edu/~honavar/agent99.pdf (Jun 2006)

[11] Franklin, Stan and Graesser, Art; 1997. Is it an agent or just a program?: A taxonomy for
autonomous agents. In Intelligent Agents III: Agent Theories, Architectures, and Languages.
Springer-Verlag, p 21-35.

[12] Zhang, P et el. 2006 Using Agent to Coordinate Diabetic Health Care. School of
engineering, Bleking Institute of Technology. (Unpublished yet)

[13] P. Maes. 1994 Agents that Reduce Work and Information Overload. Communications of
the ACM, 37(7), p 30-40

- 39 -
[14] Lincicum, S. Introduction to Interface Agents.
http://www.ous.edu/onlinenw/2003/executive/lincicumExecSumm.pdf (Jun 2006)

[15] Laurel, B. (ed.) 1999 The Art of Human-Computer Interface Design. Addison-Wesley, p
355-367

[16] Lieberman. 1997 Autonomous Interface Agents. CHI 97, p 22-27.

[17] Horvitz, E. 1999 Principles of Mixed-Initiative User Interface. CHI 99, p 15-20.

[18] Silvia Schiaffinoa,b & Analia Amandia. 2004 User–interface agent interaction:
personalization issues Int. J. Human-Computer Studies 60, p 129–148

[19] Bonett, M. 2001. Personalization of web services: opportunities and challenges. In:
Ariadne Issue 28,
http://www.ariadne.ac.uk/issue28/personalization/intro.html (Jun 2006)

[20] HASU M. & ENGESTRÖM Y. July 2000 Measurement in action: an activity-theoretical

perspective on producer–user interaction, Volume 53, Number 1, p 61-89(29)

[21] Genesereth, M.R. and S.P. Ketchpel, 1992.Knowledge Interchange Format, Version 3.0
Reference Manual, Computer Science Department, Stanford University.

[22] FIPA, The Foundation for Intelligent Physical Agents, 2001.

http://www.fipa.org. (Jun 2006)

[23] Finin, T., 1993 Specification of the KQML agent communication language. DARPA
knowledge sharing initiative external interfaces working group.

[24] Norman, D. 2005 Human-centered design considered harmful .interactions Volume

12, Issue 4 ISSN: 1072-5520, p 14 - 19

[25] Hewett, Baecker et el. ACM SIGCHI Curricula for Human-Computer Interaction.
http://sigchi.org/cdg/cdg2.html. (Jun 2006), p 5

[26] Nielsen, J. 1994 Enhancing the explanatory power of usability heuristics. Proceedings of
CHI 94, New York, NY: ACM, p 152-158

[27] Uden, L. & Willis, N. 2001 Design user interface using Activity theory. Proceedings of the
34th Annual Hawaii International Conference on System Sciences (HICSS-34)-Volume 5 -
Volume 5 ISBN: 0-7695-0981-9, p 5031

- 40 -
Appendix A
This Appendix includes the detail of the Gaia analysis and design process in IMIS case study.

Gaia analysis of IMIS

In health care organization, a normal care process is based on the information of patient as a
journal, which includes factors that may influence patient body condition, e.g. Blood Glucose
(BG) record. Diabetic patient has regular life style in case their glucose level out of tolerant
scope and cause other acute disease. Patients measure their Blood Glucose many times a day
(more than 4 times) to observe whether the body condition is fine or not and consequently
decide if daily routine need to be changed or specially health care is required. In the health
care organization, every member has specified work roles. Doctors observe the patient’ s
records and give a general diagnosis and then delegate the work to sublevel providers such as
nurses who are in charge of certain medicine to suppress the BG level. Other care providers in
different levels such as personal assistants also have specified responsibilities to help the
patient. Due to the limitation of the thesis, it is not necessary to list roles of all the work
division. Part of a general diabetic health care process is illustrated by the model showed in
this appendix.
The diabetic health process is initiated when the BG level are found unusual during the
regularly measurement and report. BG Handler checks patient’ s Blood Glucose records and
set alarm when the BG level is out the save level, and return suggested actions, inform will
also be sent to patient.

Figure A 1: Schema for role BG Handler

The BGH expression in Liveness properties of role BG Handler (figure A1) shows the
protocol “AwaitAlarm” and “GenerateDiagnosis” will repeat infinitely. Once the safety
properties are true, Diagnosis will be read from DA and sent to patient.

- 41 -
 Figure A 2: Schema for role Diagnosis Advisor

When the alarm is set by BGH, DA will read the supplied information about the patient,
which includes the general information about patient’ s name, address, telephone, history
record and also detail information in patient’s profiles, because patient’
s tolerant BG level are
various from one to another even though the ordinary safety BG has a general scope. Further
more, BG record dynamically changes because it is influenced by other aspects such as food
and exercises. So it makes no sense if just simply make a conclusion from a suspect BG
record without consider relevant factors. DA roles interact with PA to get history record and
interact with PV to evaluate the BG level under personalized information. Besides, the
activity “FindWorkDivision”will find work group of patient and delegate relevant work task
to them after diagnosis has been made.

- 42 -
 Figure A 3: Schema for role Medication Manager

MM is a role that act like nurse who is responsible for the medicine management. It will
give a medication plan after get the delegation from DA. Meanwhile, it interacts with SEC to
make sure the medicine is safe for patient. The result will return to patient when the medicine
is safe and feedback is also available to change the patient profile.

Figure A 4 : Schema for role Profile Vetter

- 43 -
Figure A 5 : Schema for role Side Effect Checker

Figure A 6: Schema for role Patient Assistant

- 44 -
Gaia design of IMIS
Having finished the agent-orient analysis process of our application system, the next stage is
design process with Gaia. According to the Gaia design procedure, the first model is agent
model which are indicate in figure A7

Figure A 7: Agent model of IMIS

This agent model shows the agent instances that will realize these agent types at run-time.
We can consider this instance process as to build relationship with a set of different roles.
Gaia specifies that there may in fact be a one-to-one correspondence between roles. In other
words, each role corresponds to an agent, and some agents can consist of several role models
and one role model could be component of many different agent types.
Therefore, as we described the interface agent in chapter 2, an interface agent who is on
behalf of patient could consist of PA role, Doctor’s interface agent consist of GBH and DA
roles due to their high degree of interdependence, and nurse’s interface agent may compose of
MM and SEC roles. It is notable that patient may also use MM role to set their medicine taken
plan and other roles could also be components as their assistant. There is no borderline and
limitation among these roles. They are very flexible so that they can contribute to users with
different goals and tasks at run-time.

Figure A 8: Service model of IMIS

The second model of Gaia design is the service model. Service model of agents can be
regarded as activities that are carried out at run-time. Here we take medicine plan agent as an
example. Briefly there are five services (activities) that can be identified in the IMIS system.
The first one is “get delegation”which is derived from services of other agent (DA). It gets
delegation request as input and get no output. The pre-condition of “get delegation”is the
delegation that is sent and the post-condition is the type of delegated care service.
The service corresponds with “VetProfile”protocol is “vetprofile”. Its input is supplied

- 45 -
patient personal information and particularly, the medicine information about the user. The
output is the allergic medicine list which is stored in user profile. The pre-condition of this
service is user profile about allergic medicine that is available. And the post-condition should
be that the chosen medicine does not conflict with other given medicine and it does not have
bad influence to the supplied patient. Otherwise agent will choose other medicine and make a
new plan for patient.
The “GenerateMedicationPlan”protocol is associated with service “make medicine plan”.
The input of this service is patient detail information and output is the medication plan. At the
same time, SEC will send request to patient in order to ask if the new given medicine is fine
or not. This service is carried out under other protocols which belong to SEC, if the new plan
is fine to patient. Information will be sent to notify the patient. Otherwise, patient profile will
be modified and next time, the same medicine will not be chosen.
In service model, agents behave both reactively and proactively. They also collaborate
(social-able) to achieve the same goal. Other properties discussed during the theoretical
analysis are also embraced in this model. For example, when patient’ s BG level is above the
safe level, an alarm will be set to initiate to ask for the appropriated actions. After getting the
delegation, the interface agent chooses Metformin as a medicine to suppress the BG level
under the risk line, after check the profile description about the allergic medicine. There is no
history record about the bad effect of this medicine. Thus the recommended medicine should
be fine for the target patient. However, after taking the medicine patent feels bad and the
feedback of the phenomena to agent, agent will modify user profile and next time when
“ProfileVetter”retrieve the allergic medicine list. This medicine will not be chosen and other
alternative medicine will be recommended instead.
The last model is the acquaintance model (figure A9). It elaborates the communication
(social-able property) ways among agents. Figure only shows a part of agents’communication
relationship in the whole IMIS system.

Figure A 9 : IMIS acquaintance model

- 46 -
Appendix B

IMIS Prototypes
According to the scenario described in chapter 4, a prototype has been developed after
alternative design to display the functions of IMIS system. As show in figure B1, a patient
login the system, agent will greeting him and bring him to his service page which has his
activity list and records. These records are categorized and it will facilitate the process to
retrieve these records for different providers. To provide more detail information, patient
could write some notes for the records manually. Agents observe the updated date and give
instruction, e.g., the tolerant blood glucose value stored in patient’
s profile is between 85-231.
Usually after eating the blood glucose become very high but it should reduce to the normal
level 1 or 2 hours after taking some food. Otherwise it could be dangerous for patient. When
the measured record is above the upper level in the profile, a notification is given to patient
towards the record. Besides a warning will be displayed patient’ s body status to provide a
more explicit clue of the body condition. Below the patient information is the agent window
where agent initiative to find the solution and give a suggestion to patient. Patient could
choose either accept the suggestion or refuse to adopt is. When suggestion is accepted, the
record will be stored automatically by agent. Of course patient could also choose report to the
agent if given medicine is not appropriate. Preference of medicine will be changed either by
agent or by patient him/her self. In the bottom of the interface user could manually find other
relevant information which is sorted by the activity theory. (Care taker, network, tools, work
division, work group).

Figure B 1: IMIS Prototype for a care taker

- 47 -
 Figure B 2 : Report of the patient’
s records

In IMIS system, patients could set the alarm to care provider any time when they don not
feel well. And they can provide the corresponding records in form of a report to relevant care
provider as a tool to make the diagnosis. Care providers who have the permission to view
these records also could get the report before diagnosis or consultation. The report is showed
either by text table or chart. When a patient send an alarm to doctor, the doctor will probable
retrieve the glucose records from a short time to find the reason. If the information is not
sufficient to make a diagnosis, glucose from a long period will showed in the table. Records
showed in chart are more explicit and intuitive for users, especially for care takers who do not
have much knowledge to interpret the records.

Figure B 3 : Chart of average glucose by hour

- 48 -
Matematiska och systemtekniska institutionen
SE-351 95 Växjö

Tel. +46 (0)470 70 80 00, fax +46 (0)470 840 04

http://www.vxu.se/msi/

- 49 -