SURPRISE '97 - Intelligent Agent Technology in Travel Reservation Systems

Imperial College of Science Technology and Medicine Departments of Computing and Electronic Engineering

May 20 1996

CAPABILITIES AND APPLICATIONS OF SOFTWARE AGENTS
by George Georgallides ( gg5@doc.ic.ac.uk )

Abstract:
The use of the word agent in order to describe an object, either software or hardware, directly implies an attempt to implement human behaviour artificially. This is quite a daring statement and the fact that a plethora of software applications are labeled as agents does not mean that science has reached a point where it can substitute human personnel with machines. On the contrary the word agent has been misused mainly for commercial purposes. In order to deceive or to impress, companies have named their products AGENTS, thus undermining the original definition of the word.

The truth is, as far as software agents are concerned , we have a long way to go yet.







1 A Broad Definition of a Software Agent


[7] Over the years the general idea of how an agent should look like has remained unaltered, while the particular characteristics of individual agents have varied according to their specific tasks.

For example all software agents should be able to communicate on a high level basis with human users as well as with other agents, using e-mail, speech and other similar ways of communication. Many times, though, inter-agent communication can exist at lower levels than agent-human communication. For example human users need speech and graphical interfaces, while agents may communicate via a simpler language.

Ideally agents should act according to the given facts of a situation and their past experience, but this method of approach is still at very early stages, since it involves learning, judgment of situations and in general elements that make up the human consciousness. The following model was proposed by Deepak Kumar and Stuart C Shapiro [6] .



Figure 1 - Model of a Software Agent




The above model has the ability to decide whether a request involves judgement according to the agent's beliefs, or if it is just a matter of performing a logical operation. The part labeled Agent's Beliefs can either be a predetermined set of beliefs provided by the manufacturer, or it may have the ability to learn new things and update old beliefs.

Autonomy, or the ability to act independently, is a vital characteristic that can make an agent's behaviour be more human like and boost its productivity. Clearly there is a direct link between the quality of the technology implementing the agent's beliefs and the extend of its autonomy.

In this paper we mainly focus on distributed agents that can run on different applications and platforms. The advantage of this is that agents may wander around in networks undertaking several tasks. One or more Fascilitator agents may act as policemen within communities of agents, dealing with assigning agents to different tasks and reporting on the availability of agents. In such a system, otherwise called Distributed Artificial Intelligence , different agents may collaborate in different ways in order to generate optimum results.

According to SRI Open Agent Architecture these autonomous agents should communicate via an Interagent Communicating Language [2] , (ICL), varying from java, C++, Delphi etc.



2 Software Agent Standards Today


[1] Before we go on we should note that all Agent projects that we encountered are under development and no one claims to have created the ultimate agent. Also we note that an agent is not always indivisible and that it may consist of several smaller sub-agents.

Software agents today vary from Internet Information agents, Traffic Control agents to even more human like agents; see section 2.2 .




2.1 Autonomous Agents


Web Spiders [3] , an active example of information agents, can wander around in the world wide web gathering , indexing or classifying information according to their clients' requests. From the ethical point of view, Spiders may be developed in the future that will even be able to perform censorship within the web.

A good example of an autonomous software agent is SUMPY [5] , an agent which was built with the purpose to preserve the order and optimize disk utilization in a UNIX file system. SUMPY wanders around in the system and is able to know its position at any time by issuing commands and evaluating the system's responses, (i.e. response time etc.). The way in which the system responds will influence SUMPY's decision as to whether it should compress or backup files or simply to continue wandering around. It also has the ability to put itself into sleep when there is high CPU demand by the users, since the agent's operation is CPU consuming.

SUMPY is made up of four separate layers that work independently and can intercommunicate. These layers could be labeled as WANDER, COMPRESS, BACKUP and SLEEP. Each one issues commands to the system at random and irrespective of each other. The fact that they work in parallel reduces the overall system failure probability while it increases the overall efficiency and productivity of the system. But clearly this example of an agent does not fullfil all the requirements that we summarised and shown in figure 1 .


2.2 A Clerical Agent


A more human-like agent, Virtual Mattie [4] , was created in the Institute for Intelligent Systems of The University of Memphis. VMAttie's purpose is to perform the day to day tasks of a clerical worker autonomously and without any human intervention. Everyday duties include e-mail or other form of high-level communication with humans and/or other agents, acknowledge incoming messages, arrange weekly seminars, update mailing lists, compose and send reminders to speakers, etc.

The architecture of VMAttie allows reasoning, scheduling, problem solving and even learning and creativity. Research on this project still continues and it has been suggested to implement emotions such as guilt for not sending a notice on time and anxiety for not understanding a concept. It is obvious that this is a quite advanced software agent that attempts to cover most of the aspects that make up a, real-life, human agent. Aaron Sloman calls such agents complete !



3 Summary
With the aid of some examples we have seen how different agents can be created to perform different tasks. We have also seen how intelligence levels vary from agent to agent. For example SUMPY, section 2.1 , is an agent with limited cababilities as far as forming opinions or beliefs is concerned, whereas VMATTIE, section 2.2 , can be classified as a more intelligent agent.
It is important to note, though, that even if one day the perfect agent is created, several social implications will have to be considered. For example real people may use small lies in their every day work while they are negotiating and no one can accuse them, but a software agent built to lie may give rise to legal problems.



List of References:



---




1	Intelligent Agents: A Technology and Business Application Analysis
	Kathryn Heilmann, 1995
	Usefulness :                     (8)
	Readability :                     (8 )
	Content :
	On-Line : http://haas.berkeley.edu/~heilmann/agents/#exsum
	Obtained From :
	Notes : Information on Intelligent Travel Agents




---





2	The Open Agent Architecture
	SRI Artificial Intelligence Centre,
	Usefulness :                     (5)
	Readability :                     (8 )
	Content :
	On-Line : http://www.ai.sri.com/~oaa/main.html#2
	Obtained From :
	Notes :




---





3	Ethical Web Agents
	Nasa,
	Usefulness :                     (4)
	Readability :                     (8 )
	Content :
	On-Line : http://rbse.jsc.nasa.gov/eichmann/www-f94/ethics/ethics.html
	Obtained From :
	Notes :




---





4	Virtual Mattie - An Intelligent Clerical Agent
	Stan Franklin,
	Usefulness :                     (7)
	Readability :                     (8 )
	Content :
	On-Line : http://www.msci.memphis.edu/~franklin/vmattie.html
	Obtained From :
	Notes :




---





5	Sumpy: A Fuzzy Software Agent
	Hongjun Song,
	Usefulness :                     (6)
	Readability :                     (8 )
	Content :
	On-Line : http://www.msci.memphis.edu/~songh/publications/sumpy.html
	Obtained From :
	Notes :




---





6	Architecture of an Intelligent Agent in SNePS
	Deepak Kumar and Stuart C. Shapiro,
	SIGART Bulletin vol. 2, No.5
	Usefulness :                     (5)
	Readability :                     (7 )
	Content :
	On-Line :
	Obtained From :
	Notes :




---





7	Software Agents: An Overview
	Hyacinth S. Nwana, 1996
	The Knowledge Engineering Review vol. 11:3
	Usefulness :                     (8)
	Readability :                     (8 )
	Content :
	On-Line :
	Obtained From :
	Notes :




---