Software Agents: Characteristics and Classification

Qusay H. Mahmoud

School of Computing Science, Simon Fraser University

Burnaby, BC. Canada V5A 1S6
[email protected]

Abstract. Since the arrival of the World-Wide Web, and the explosive growth of
information on the Web, researchers in the field of Software Agents got interested in
information agents (also known as Internet agents). However, for the past couple of
years, agent technology has been a hot topic, and most likely, this is mainly due to
the popularity of the Java programming language, which represents an ideal
language for implementing software agents as it is the “Write Once Run Anywhere”
language. This is an important feature for software agents as it allows them to run on
all platforms of the Internet. The word “agent” however, has been misused. People
involved in agent research have a variety of definitions for that word. This paper
presents a brief overview of software agents, with emphasis on mobile agents. We
classify agents in different types along several primary characteristics that agents
should exhibit.

1 Introduction

Before we try to define what is a Software Agent, let us first try to understand
the meaning and characteristics of an agent. We are all, in one sense or another,
familiar with the concept of an agent. Probably most of us have dealt with travel
agents and we know the role undertaken by them. The main representative role of
a travel agent, for example, is that it acts on behalf of others. This characteristic
can be considered as the first fundamental property of agency. A travel agent acts
on behalf of a traveler in a variable degree of autonomy. In other words, when a
travel agent calls an airline to reserve a seat, they do so autonomously; that is,
they do not tell the airline to whom do they need a seat, they just say we need a
seat. This is the second characteristic of agency – autonomy. A third characteristic
of an agent is the degree of proactivity and reactivity in their behavior. For
example, once an agent receives the details of its tasks, it tries proactively to
attain the goals defined by the assigned tasks. And it reacts to the changes in the
available data by modifying its plan. Agents may also exhibit other attributes,
including: learning, co-operation, and mobility.

2 What is an Agent?

Based on the above discussion, an agent can be defined, along with its
characteristics as follow:

“An agent is an entity that:

• acts on behalf of others in an autonomous fashion
• performs its actions in some level of proactivity and reactivity
• exhibits some levels of the key attributes of learning, co-operation, and
mobility.”

This definition, which is based on the definition given in [5], is equally true for
software agents. But what exactly is a software agent and how does it differ from
a software object? You may think of a software agent as one (or more) software
object(s) that conforms to the above characteristics of agents and can be described
as inhibiting computers and networks, assisting users with computer-based tasks.
It is the responsibility of the programmer, however, to determine what an agent
can do, as well as the information required from the user or software to have an
agent performs its actions in a reactive manner. The behavior of the agent can be
set by another software, which you can think of as a sort of a super agent, that
forks (or clones) new agents when a task requires extra help.

3 Classification of Software Agents

Classification refers to the study of types and entities. There are several
dimensions to classify existing software agents. They can be classified according
to: the tasks they perform; their control architecture; the range and effectiveness
of their actions; the range of sensitivity of their senses; or how much internal state
they posses [6]. In this paper we classify agents, along several ideal and primary
characteristics that agents should exhibit. We identify three characteristics:
autonomy, learning, and cooperation, which we have discussed in Section 1. Our
classification here is based on [4]. Let us start by elaborating a bit more on these
characteristics. Autonomy refers to the characteristic that an agent can operate on
its own without the need for human guidance. In other words, an agent has a set
of internal states and goals, it acts in such a manner to meet its goals on behalf of
the user. In order to do so, an agent has to be proactive in the sense that it has the
ability to take the initiative rather than acting simply in response to its
environment [15]. Cooperation with other agents is necessary to accomplish a
complicated task. In order to cooperate however, an agent must posses a social
ability that allows it to interact with other agents. Finally, an agent is intelligent if
it is able to learn and sense as it acts and reacts to its external environment. These
three characteristics of agents are used to derive some types of agents to include
in our classification as shown in Figure 1.

Collaborative Agents
Cooperate Learn Interface Agents

Autonomy

Figure 1: A partial view of agent classification

3.1 Interface Agents

Interface agents perform tasks for their owners by emphasizing autonomy and
learning [16]. They support and provide assistance to a user learning to use a
particular application such as a spreadsheet. The agent here observes the actions
being carried out by the user and tries to learn new short cuts, then it will try to
suggest better ways of doing the same task. As P. Maes points out, the key
metaphor underlying interface agents is that of a personal assistant who is
collaborating with the user in the same work environment. Interface agents learn
to better assist its users in four ways [8]:

• By observing and imitating the user

• Through receiving positive and negative feedback from the user
• By receiving explicit instructions from the user
• By asking other agents for advice

While interface agents ask other agents for advice (learning from peers), their
cooperation with other agents however, is limited.

The objective of P. Maes work on Interface Agents was initially to migrate from
the direct manipulation metaphor [9] to one that delegates some of the tasks to
software interface agents in order to accommodate novice users. With the
th
explosive growth of the Web, and as we come close to the 20 century, we are
having as many untrained computer users as we have microprocessors. Therefore,
there is a gap that is coming more apparent as we go along with computers.
Current computer interfaces, for example, do nothing unless you give commands
from the keyboard, mouse, or touch screen [5]. Therefore, the computer is merely
a passive entity waiting for your commands to execute them. It does not provide
us with any help whenever we wish to perform some complex tasks. In the future,
interface agents as well as information agents (as discussed later) may know
users’ interests and act autonomously on their behalf. That is what researchers in
the area of agent technology have high hopes for.

3.2 Collaborative Agents

As the proliferation of computer communication networks was a big step toward

the development of “virtual societies”. Collaboration between individuals (in the
virtual society) requires that communication links be established and used
effectively. Distributed Artificial Intelligence, which is a subfield of Artificial
Intelligence, is concerned with a virtual society of problem solvers (agents)
interacting to solve a common problem [5].

The goal of collaborative agents is to interconnect separately developed

collaborative agents, thus enabling the ensemble to function beyond the
capabilities of any of its members. Implementing efficient ways of cooperation
among agents is actually one of the central issues for Multi-Agent Systems
development [19].

One of the motivations for having collaborative agents is to provide solutions to

inherently distributed problems, such as distributed sensor network [20], or air-
traffic control.
3.3 Information Agents

The explosive growth of information on the Word-Wide Web has given a rise to
information agents (also known as Internet agents) in the hope that these agents
will be able to help us manage, manipulate, or collate information from many
distributed resources [7]. One may notice however, that information agents seem
a bit similar to interface agents [16]. However, it is important to note that not all
types of agents discussed here started at the same time. So, with the explosive
growth of information, and the need for tools to manage such information, one
would expect a degree of overlap between the goals of some agents. One
distinction between interface and information agents, however, is that information
agents are defined by what they do, in contrast to interface agents which are
defined by what they are.

There are some interface agents, developed at MIT Media Lab, that have been
deployed for Web-based roles [8], and hence they can be considered as
information agents. One important characteristic of Information Agents is that
they may be static or mobile.

Information agents are most useful on the Web where they can help us with
mundane tasks. For example, we carry out actions that may consume long time
(e.g. searching the Web for information). Why does not the computer (e.g. an
information agent) carries out such tasks for us and later on present us with the
results?

3.4 Reactive Agents

Reactive Agents act and respond in a stimulus-response [18] manner to the

present state of the environment in which they are embedded. P. Maes highlights
the following three key ideas which underpin reactive agents [13].

• Emergent functionality: the dynamics of the interaction leads to the emergent

complexity.
• Task decomposition: a reactive agent is viewed as a collection of modules
which operate autonomously and responsible for specific tasks (e.g. sensing,
computation, etc.).
• They tend to operate on representations that are close to raw sensor data.
Up till now, there is a relatively few number of reactive software agent-based
applications. A good application area for them seems to be the entertainment
industry. Researchers at Philips are already working on digital video and 3-D
graphics-based reactive agent animation [4].

3.5 Hybrid Agents

Hybrid Agents refer to those agents whose constitution is a combination of two or

more agent philosophies within a singular agent [4]. These philosophies may be
mobile, interface, information, collaborative, … etc. The goal of having hybrid
agents is the notion that the benefits accrued from having the combination of
philosophies within a single agent is greater than the gains obtained from the
same agent based on a singular philosophy. An example of this is collaborative
interface agents [17].

NOTE: The author believes that mobile agents (discussed next) can actually be
considered hybrid agents, because in order for a mobile agent to be useful, it may have to
reactive, proactive, and collaborative, for example.

3.6 Mobile Agents

A software agent is a mobile software agent if it is able to migrate from host to

host to work in a heterogeneous network environment. This means we must also
consider the software environment in which mobile agents exist. This is called the
mobile agent environment, which is a software system distributed over a network
of heterogeneous computers and its primary task is to provide an environment in
which mobile agents can run. Note that not only an agent transports itself, but also
its state. When it reaches the new host, the agent should be able to perform
appropriately in the new environment.

3.6.1 Anew Paradigm for Distributed Computing

The central principle of today’s distributed programming is remote procedure

calling (RPC). The RPC approach, which was conceived in the 1970s, views
computer-to-computer communication as enabling one computer to call a
procedure in another. In RPC, all messages go through the network, each either
requests or acknowledges a procedure’s actions. This approach, however, has its
own limitations. Most notably, all interactions between the client and server must
go through the network as shown in Figure 2.
 Figure 2: RCP-based Client/Server Computing Paradigm

Another approach that is forming a new paradigm for distributed computing is

one that employs mobile agents. Initially this approach was known as Remote
Programming [1]. The Remote Programming approach views computer-to-
computer communication as one computer not only to call procedures in another,
but also to supply the procedures to be performed. Each message that goes
through the network comprises a procedure that the receiving computer is to
perform and data that are its arguments. The procedure and its state are termed a
mobile agent as they represent the sending computer even while they are in the
receiving computer as shown in Figure 3.

Figure 3: mobile agents-based computing paradigm

This approach is attractive since the reliability of the network is not crucial for the
following reasons:

• Mobile agents do not consume much network bandwidth. They only consume
bandwidth when they move.
• They continue to execute after they move, even if they loose network
connectivity with their creators.

Therefore, if a client requires extensive communications with a particular server

somewhere on the network, then implementing such a system using mobile agents
is attractive. This is due to the fact that an agent can move closer to the remote
server, reducing the network traffic, performs all tasks and comes back. During
that period the client machine doe not have to be switched on. It will have to be
switched on only when it is time to welcome back the agent. Figure 3, shown
above, illustrates the idea of this paradigm.

At this point, someone may say that this is exactly what process migration is all
about and this has been done in the 60’s. That someone would be absolutely right.
However, mobile agents are different in the sense that they exhibit the
characteristics of an agent as discussed in Section 1.

3.6.2 Applications of Mobile Agents

An attractive area to use mobile agents is in processing data over unreliable

networks. In such networks, the low-reliability network can be used to transfer
agents, rather than a chunk of data, from place to place. In this paradigm, the
agent travels to the nodes on the network, process the information on those nodes
(without the risk of network disconnection) and then return home. Another
interesting area that is attracting lots of attention is electronic commerce. For
example, instead of spending a huge amount of time going through on-line
bookstores to find the best deal on a book, firing up an agent to do this task would
save us a considerable amount of time. The agent would be programmed to visit a
number of bookstores and find the best deals on books we need.

Another interesting area of applications for mobile agents is in network

management. In today’s heterogeneous network environments, network operators
are required to have extensive knowledge of the diverse networks in order to
manage them. Such management requires the collection of large amount of data
from nodes on the network. Mobile agents represent an ideal tool for collecting
the data and analyzing it.

3.6.3 Advantages of Mobile Agents

Mobile agents seem to be useful for many different applications. One may
however, claim that virtually any task that can be performed with mobile agents
can be performed with other technologies (e.g. remote method invocation).
Despite the fact that there are not many distributed computing problems that
cannot be solved without mobile agents, nevertheless mobile agents make certain
applications easier to develop and may improve reliability and efficiency. Table 1,
which is based on [5] contains some of the claimed advantages of mobile agents
over conventional approaches.
Advantage Justification
Efficiency Mobile agents consume fewer network resources since they
move the computation to the data instead of the data to the
computation.
Less Bandwidth Most communication protocols involve several interactions,
which cause a lot of network traffic. Mobile agents consume
bandwidth only when they move.
Robustness and Fault The ability of mobile agents to react dynamically to adverse
Tolerance situations makes it easier to build fault tolerance behavior in
complex distributed systems.
Support for Hetero- Mobile agent systems are computer and network independent.
geneous Environments Therefore, a Java mobile agent can target any system that has a
Java Virtual Machine.
Support for Electronic Mobile agents are being used to build electronic markets since
Commerce they embody the intentions, desires, and resources of the
participants in the market.
Easier Development The construction of distributed systems can be made easier with
Paradigm mobile agents. Mobile agents are inherently distributed in
nature. Therefore, they are a natural view of a distributed
system.

Table 1: possible advantages of mobile software agents

3.6.4 Security in Mobile Agents

Despite the fact that mobile agents represent a new useful paradigm for
distributed computing, they have, however, seldom been used in practice. The
main reason for this is due to the fact that there are a number of related technical
as well as social challenges to implementing mobile agent-based applications.
Technical issues include, among others, difficulties with portability and security.
The security issue may also be considered as a social challenge – convincing
people to use mobile software agents to buy items over the Internet. As B. Laurel
[10] pointed out “few of us would hire an agent to push the buttons on our
calculator; most of us would hire an agent to scan 5,000 pieces of junk mail”.

The security challenge, however, is a complicated one. Mobile agents raise issues
similar to Java Applets. There are several security issues to be considered in
mobile agent-based computing. Some people in the computing community think
of mobile agents as viruses sine they may exhibit similar behavior.
Mobile agent security can be split into two areas [5]:

• Protection of host nodes from destructive agents.

• Protection of mobile agents from malicious nodes.

One approach to protect the host nodes from destructive agents it through
authentication and agent signatures where all unknown agents are rejected.
However, this does not really seem like a good solution, because how would we
know that a mobile agent is whom it claims to be?

The other area of security deals with the issue of protecting mobile agents from
hosts which may want to scan the agent for information; alter the agent’s state; or
even kill the agent. The crucial issue here is that the agent will have to expose its
data and information to the host in order to run on it [11]. Current research has
shown it is computationally impossible, however, to protect a mobile agent from a
malicious host [12]. Some researchers are looking at sociological (instead of the
hard computational model) means of enforcing good host behavior [12].

4 Conclusion

Agent technology is a hot topic that is keeping many people both, in academia
and industry quite busy. In this paper we have given a brief overview of software
agents and their applications. We explained the characteristics of an agent and
based on that we classified agents into six types: interface, collaborative,
information, reactive, hybrid, and mobile. The mobile agent class was discussed
in more details than any other class. This because in the author’s view a useful
mobile agent is inherently an information agent and should be collaborative and
reactive. In other words, a mobile agent is different from a mobile object since the
mobile agent has to exhibit some attributes of agency (e.g. autonomy, learning,
co-operation, proactivity and reactivity).

5 References

[1] J. White. “Mobile Agents”. In J. M. Bradshaw (editor), Software Agents.

AAAI Press/MIT Press, 1997, pp. 437-472.

[2] P. Morreale. “Agents on the move”. In IEEE Spectrum, April 1998, pp. 34-41.
[3] V. A. Pham, and A. Karmouch. “Mobile Software Agents: An Overview”. In
IEEE Communications Magazine, July 1998, pp. 26-37.

[4] H. S. Nwana. “Software Agents: An Overview”. In Knowledge Engineering

Review, Vol. 11, No 3, Oct/Nov 1996, pp. 205-244.

[5] S. Green, L. Hurst, B. Nangle, P. Cunningham, F. Somers, R. Evans.

“Software Agents: A review”. Technical Report, Trinty College Dublin, Ireland,
May 1996.

[6] S. Franklin, and A. Graesser. “Is it an agent, or just a Program?” A Taxonomy

for Autonomous Agents”. In Proceedings of the Third International Workshop On
Agents Theories, Architecture, and Languages, Springer-Verlag, 1996.

[7] P. Maes. “Intelligent Software”. In Scientific American, Vol. 273, No. 3, Sept.
1995, pp. 84-86.

[8] P. Maes. “Agents that Reduce Work and Information Overload”. In J. M.

Bradshaw (editor), Software Agents. AAAI Press/MIT Press, 1997, pp. 145-164.

[9] P. Maes, and Schneiderman, B., "Direct Manipulation vs. Interface Agents: a
Debate", In Interactions, Vol. IV Number 6, ACM Press, 1997

[10] B. Laurel. “Interface Agents: Metaphors with Character”. In J. M. Bradshaw

(editor), Software Agents. AAAI Press/MIT Press, 1997, pp. 67-77.

[11] D. Chess, C. Harrison, and A. Kreshenbaum. “Mobile Agents: Are they a

good idea”. In N. Vitek, and C. Tschudin (editors), Mobile Objects Systems,
Springer Verlag Lecture Notes in Computer Science, July 1996, pp. 25-47.

[12] A. Rasmusson, and S. Janson. “Simulated Social Control for secure Internet
commerce”. In New Security Paradigms ’96, ACM Press, Sept 1996.

[13] P. Maes (editor). “Designing Autonomous Agents: Theory and Practice from
Biology to Engineering and back”. MIT Press.
[14] M. Graham, and P.R. Wavish. “Simulating and Implementing Agents and
Multiple Agent Systems”. In Proceedings of the European Simulation Multi-
Conference, Copenhagen, Denmark, June 1991.

[15] M. Wooldridge, and N. R. Jennings. “Intelligent Agents: Theory and

Practice”. In Knowledge Engineering Review, Vol. 10, No 2, 1995, pp. 115-152.

[16] T. Koda and P. Maes, "Agents with Faces: The Effects of Personification of
Agents”, In Proceedings of HCI '96, London, UK, August, 1996.

[17] Y. Lashkari, M. Metral, P. Maes. "Collaborative Interface Agents". In

Proceedings of the Twelfth National Conference on Artificial Intelligence, Vol. 1,
AAAI Press, Seattle, WA, August 1994.

[18] Jacques Ferber. “Reactive Distributed Artificial Intelligence: Principles and

Applications”. In G.M.P. O’Hare, and N.R. Jennings (editors), Foundations of
Distributed Artificial Intelligence, John Wiley & Sons, Inc., 1996, pp. 287-317.

[19] B. Moulin, and B. Chaib-Draa, “An Overview of Distributed Artificial

Intelligence”. In G.M.P. O’Hare, and N.R. Jennings (editors), Foundations of
Distributed Artificial Intelligence, John Wiley & Sons, Inc., 1996, pp. 3-55.

[20] E.H. Durfee, and T.A. Montogomery. “Coherent Cooperation among

Communicating Problem Solvers”. In IEEE Transactions on Computers, Nov
1987, pp. 1275-1291.