MULTI-AGENT PROPOSITIONS TO MANAGE

ORGANIZATIONAL KNOWLEDGE

Position paper concerning a three-dimensional research project

Jorge Louçã, César Rosa, Francisco Guimarães, Valmir Meneses

ISCTE - DCTI, c.296, Av.das Forças Armadas, 1649-026 Lisboa, Portugal

ywords: Intelligent agents, Knowledge management, HCI on Enterprise Information Systems

Abstract: This paper presents the work in progress in a three-dimensional project, including the theoretical

foundations and main goals of the lines of research incorporating our project: user modeling in a distributed

cooperative system, interactive cooperation in a multi-agent structure, and knowledge representation in a

cognitive agent architecture. These lines of research are complementary and share a main goal, to make

propositions regarding the use of multi-agent systems in organizations, namely in what concerns support to

decision making processes and, in a general way, knowledge management within organizations.

1 INTRODUCTION

This paper presents the work in progress in a three-

dimensional project, including the theoretical

foundations and main goals of the lines of research

incorporating our project: user modeling in a

distributed cooperative system, interactive

cooperation in a multi-agent structure, and

knowledge representation in a cognitive agent

architecture. These lines of research are

complementary and share a main goal, to make

propositions regarding the use of multi-agent

systems in organizations, namely in what concerns

support to decision making processes and, in a

general way, knowledge management within

organizations.

What makes knowledge management hard is the

ill-structured and subjective nature of different types

of knowledge present in the organization. Moreover,

distinct organizational actors interact in ill-structured

domains. Knowledge management should be

distributed, as the organization is multi-dimensional,

should characterize individuality of organizational

actors, representing their beliefs and supporting their

reasoning processes, should support interactions

between organizational actors, as working processes

are for the most part based on interactions, and

finally should propose an adequate working

environment, helping people in their tasks.

Aiming to support knowledge management

processes with the characteristics above, we propose

to found our research on cognitive mapping,

allowing the simplification of complex ill-structured

domains of knowledge, and on multi-agent systems,

allowing to use artificial cognitive agents to

characterize organizational actors and to represent

agents interactions. This way, our project has an

interdisciplinary nature, putting together domains

like distributed artificial intelligence, user modeling,

knowledge representation and reasoning techniques.

We also take our inspiration from social science

theories, mainly from psychology and sociology.

Previous work concerns the use of cognitive

maps to represent artificial agents beliefs

(Louçã,2003a), a distributed architecture to support

organizational actors interactions (Louçã,2003b),

and a conceptual framework to represent emergent

social phenomena in organizations (Louçã,2003c).

436

Louçã J., Rosa C., Guimarães F. and Meneses V. (2004).

MULTI-AGENT PROPOSITIONS TO MANAGE ORGANIZATIONAL KNOWLEDGE - Position paper concerning a three-dimensional research project.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 436-441

DOI: 10.5220/0002657504360441

 SciTePress

This paper is organized as follows: section 2

describes the theoretical foundations and main goals

of the research concerning user modeling in a

distributed cooperative system; section 3 regards

interaction in a multi-agent structure; section 4 is

related to our propositions about knowledge

representation in a cognitive agent architecture;

finally section 5 presents a brief conclusion.

2 USER MODELING IN A

DISTRIBUTED COOPERATIVE

SYSTEM

In a competitive context, where the environment

continually changes, organizations are compelled to

dynamically adapt themselves. Changes are political,

demographical, legal, technological, concerning

consumers and competition. Organizations tend to

consider strategy a flexible and continuous process,

facing dynamic environmental factors. Strategic

processes should allow evaluation and control,

proposing corrective measures as result of

organizational learning.

Evaluation and control depend on continuously

obtained metrics, supporting decision-making at

each activity level and, by aggregation, at

organizational level. That’s why systems such as

EIS – Executive Information Systems and DSS –

Distributed Support Systems are important software

tools to support decision-making processes. Their

main feature is to facilitate communication and

dialogue at all organizational levels. To do this, they

should monitor different activities inside the

organization, be used to communicate and compare

data, and finally support discussion and negotiation

around strategic issues.

Organizational structure, working processes and

interactions inside the organization should be

tailored to EIS-DSS tools. The way technology is

used and its connections to structure, tasks and

people are essential, like suggested by the Leavitt

diagram, expressing the cohesion between key

concepts in an organization (Wilson,1995).

Let’s consider an organization where cognitive

agents support actors. Figure 2 depicts this general

idea. Specific software tools and knowledge-based

systems compose artificial agents. In this

environment, interactions between actors can be

made through artificial agents in a multiagent

system. Information systems design is determinant to

model such distributed systems, and has become

more than a technique issue – in particular, the user

modeling perspective is, in this context, a way of

Artificial

agent

Organizational

actor

Message

between

artificial

agents

B e l i e f s

Micro

representation

of ind. beliefs

Beliefs

I n t e r a c t i o n s

Figure 1: Generic architecture for agent supported interactions in an organization (Louçã,2003c)

Figure 2: Leavit

t diagram (Wilson,1995)

MULTI-AGENT PROPOSITIONS TO MANAGE ORGANIZATIONAL KNOWLEDGE: POSITION PAPER

CONCERNING A THREE-DIMENSIONAL RESEARCH PROJECT

437

modeling IS according to the cognitive character of

users, modeling their reciprocal interactions and

with sub-systems in the organization. This context-

awareness allows capturing, analyzing and

manipulating user’s contextual data.

Other research projects use the notion of user

context to model an IS. (Bauer et al.,2000) observes

user’s activity and recognizes his goals, considering

the context of the user, his behavior and educational

level. (van Elst & Abecker,2002) describes a

process-oriented knowledge management

architecture, where users are coupled with roles in

particular contexts, looking to support the

integration of individuals in the organization.

Generally, those propositions are based on the

observation of users behavior, recognizing context

and roles in workflow systems, supporting tasks and

providing on-line help. Nevertheless, they don’t

address the fundamental issue of decision support

based on a panel of ratios and data concerning

strategic orientations. Also, they don’t deal with

user’s collective goals neither with intuitive

interfaces according to individual and/or collective

goals.

This line of research has its theoretical

foundations on cognitive mapping, cognitive

artificial agents, and also on interaction models

inspired on user modeling. Its goal is to bring

together disparate notions such as strategy,

organizational processes, people and technologies.

To do this, the notion of workflow will be used,

depicting departments, groups or individuals,

implementing negotiation rules, emotional

intelligence and processes orientation according to

organizational strategic purposes. This way, each

task will be embedded with some organization high-

level goal, improving overall consistency of

individuals, behaviors and roles.

An important issue about knowledge

management software concerns the kind of

relationships and communication protocols used by

artificial agents to support organizational actors

interactions.

3 INTERACTION IN A MULTI-

AGENT STRUCTURE

This line of research concerns interactions between

artificial agents to support communication inside the

organization. The main proposition is to adopt a Web

Service (or Service-Oriented Architecture)-like

architecture for agents to communicate (W3C,2003).

The advantage of a Service-Oriented Architecture is

that web services provide standard means of

interacting between different software applications,

allowing agents to communicate between themselves

as well as to interact with the external environment.

This way, same communication protocols could be

used both between agents and between agents and

external entities, improving flexibility and

autonomy.

Service-oriented-like architectures are

characterized by communication between agents

(senders and receivers) and by the definition of

services as “sets of functionalities” provided by

agents. Provider entities are persons, departments or

organizations proposing services through their

artificial agents. On another hand, requester entities

are persons, departments or organizations wishing to

make use of those services. According with this

basic architecture, requester agents will exchange

messages with provider agents, in order to requester

entities attain services proposed by provider entities

(W3C,2003).

A goal of this line of research is to define

WSD – Web Services Descriptions, that is, message

formats, datatypes, protocols and transport

serialization formats to be used by agents. Another

point is to define the location at which a provider

agent can be invoked, and where it may provide

some information about a given service.

Message semantics (service description) represent

a kind of “contract” between requester and provider

concerning a given service, on how and why agents

will interact, regarding also the meaning and

purpose of interactions (W3C,2003).

Web Services and Service-Oriented Architectures

have been recently presented as the subject of

several research projects (Arabnia & Mun,2002).

Specifically, our goal is to propose a multi-agent

architecture where cooperative agents announce

their services to the multi-agent community. Agents

will also announce the availability of services to the

Figure 3: Basic architectural roles in

ervice

oriented

like architectures [W3C,2003]

ICEIS 2004 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS

438

exterior, through Internet. This would be useful in e-

commerce environments, where organizations

provide web based on-line services.

According to several litterature sources (Robins

et al.,,2003) (Li et al.,2003) (Luck et al.,2003), the

advantages of SOA architectures are the

implementation of open, generic and modular

architectures with loosely coupled infrastructure,

reduced costs and simpler agent integration in the

multi-agent system, interoperability between Web

services, scalability, extensibility and manageability

of Web services, integration with the World Wide

Web, all resulting on competitive advantage and

organizational innovation.

Thes SOA architecture we propose can be used to

allow interactions between multi-agent systems.

According to Figure 4, the open cube is used to

symbolize a department or company site. Each actor

(person) interacts with the system through an

artificial agent. Within these MAS, a capabilities

directory is maintained. Such directory registers its

entries in a UDDI compatible format as described in

(Sycara et al, 2002). Each agent publishes its beliefs

and capabilities in this directory upon entrance in the

MAS. When needed, an agent can query the

directory for desired capabilities and receives a list

of agents that fulfill the query. It is then free to

interact with the agents without further query to the

directory. Such directory service is provided only

inside the company site boundaries. This service

also provides the Broker address in case an agent

needs to interact with other MAS outside its own

domain.

A Broker service (Sycara et al, 2000) is used to

allow interoperability between remote MAS. The

Broker encapsulates the translation and formatting

of the message interchanged through the MAS.

This distributed cooperative architecture will

allow interactions between artificial cognitive

agents. The next line of research explores conceptual

tools to represent organizational actors knowledge

and reasoning procedures in cognitive agents.

4 KNOWLEDGE

REPRESENTATION IN A

COGNITIVE ARTIFICIAL

AGENT ARCHITECTURE

Multiagent systems are completely distributed – the

reasoning process goes on internally to each

artificial agent. This feature allows the

representation of heterogeneous agents, using

complementary technologies and representing

different cognitive models. The knowledge

representation and reasoning technologies that have

been used to this purpose are chosen according to

their specific features, each attending to some things

and ignoring others (Luck et al.,2003). When

choosing a given technology, we are in fact selecting

a point of view about knowledge representation and

reasoning. Each technology is an approach to the

task of determining how well it approximates to the

reality we mean to represent (Sowa, 2001). For

instance, logic concerns a point of view of individual

entities and relations between them, rule-based

systems consider rules of inference, frames represent

prototypical objects and semantic nets are graphical

representations of different kinds of entities through

a network topology. Each of these approaches has

both benefits and drawbacks. In fact, the choice of a

given technology is motivated by the characteristics

of a given domain, as well as by some insight

indicating how people reason intelligently. On

another hand, formal technologies are problematic in

practice. Recent research in multiagent systems has

searched for new technologies. These should be

simple and operational enough to be used in

organizations, and quite powerful and adapted to

hill-structured domains. According to this idea,

cognitive maps have been proposed to model beliefs

of cognitive agents in a multiagent environment, as

reported by (Chaib-draa,2002) and (Louçã, 2003a).

A cognitive map is a graphical representation of

the behavior of an individual or a group of

individuals, concerning a particular domain.

Cognitive maps can be employed at a micro level, to

represent individual cognitive models, and at an

institutional level through the use of collective

cognitive maps. Psychologists mainly use cognitive

maps as data structures to represent knowledge.

Generally, this kind of cognitive model facilitates

communication inside a group, supporting

discussion and negotiation between the elements

having different points of view. This way, cognitive

maps can be used to detect conflicts. Several

software systems are proposed to represent

organizational discourse into cognitive maps,

Figure 3: SOA

based Open Cube Architecture

MULTI-AGENT PROPOSITIONS TO MANAGE ORGANIZATIONAL KNOWLEDGE: POSITION PAPER

CONCERNING A THREE-DIMENSIONAL RESEARCH PROJECT

439

describing mental models in artificial agents (Chaib-

draa,2002) (Louçã, 2002a and b).

A cognitive map is composed by concepts

(representing things, attitudes, actions or ideas) and

links between concepts. Those links can represent

different kinds of connections between concepts,

such as causality or influence links. Figure 4

exemplifies a cognitive map, where links can

represent very positive influence (++), positive

influence (+), negative influence (-), and very

negative influence (--). This particular type of

cognitive map is used to represent strategic thought

in organizations, as reported by (Louçã, 200a).

Figure 4: Example of a cognitive map (Louçã, 2003a).

The main interest of cognitive maps is their reflexive

character, allowing people to became conscious of

implicit knowledge, through the visualization of

direct and indirect links between concepts. We each

construct our private versions of reality and deal

only with those constructions, which may or may not

correspond to some real world (Louçã, 2003a).

Cognitive mapping can be compared with other

knowledge representation technologies used in

artificial intelligence, such as conceptual graphs

(Sowa,2001). Conceptual graphs propose a

symbolic representation, where the meaning of a

concept is drawn from its position in the concepts

hierarchy. Concepts semantic can be understood

through the nature of links with other concepts. This

way, a concept can have different meanings,

depending on the specific context where it’s used.

Knowledge representation is done through a direct

graph composed by nodes and arcs. Nodes represent

physical or abstract objects, with properties and

values. Arcs are links between nodes, meaning

particular types of relationships between two given

nodes (Sowa,2002). We think that cognitive

mapping concerns a less constraint knowledge

representation technique – in cognitive maps,

concepts are not integrated in some particular

hierarchy, allowing a non-restrictive representation

of ill-structured domains.

Starting from the work already discussed in

several conferences (Louçã,2003 a, b and c), this

line of research will propose a methodology of

reasoning based on an agent cognitive model. This

methodology will use cognitive maps as instruments

to represent and operate complex cognitive

structures known as schemes and schemas. Schemes

and schemas are basically composed by concepts.

Agent’s reasoning composes schemes, used to

transmit complex ideas and plans in multi-agent

communication.

From the cognitive mapping point of view, a

scheme can be considered an ensemble of related

concepts, used in a particular context to represent a

complex thought or to solve a given problem. The

notion of scheme has been deeply discussed in

several scientific domains. The origin of the

argumentation that has been used can be found in

Descartes and Berkeley, with the notion of mental

image, but it was since the writings of the

philosopher Kant (18th Century), where he refers

“innate structures which organize our world”, that

schemes became an important subject of research in

cognitive science (Estivals,2002). In the beginning

of last century, the psychologist Bartlett studied

human memory and introduced the notion of

schema. To Bartlett a schema is an organized

structure of experience, containing general abstract

concepts of individual experiences (Estivals,2002).

Therefore, the notion of schema concerns a structure

– a schema is a data structure containing generic

concepts in memory. The notions of schema and

scheme are frequently coupled in cognitive sciences,

such as to Piaget, who used schemes to explain the

evolution of knowledge throughout stages of

development in young child (Piaget,1967). On

another hand, psychological theories concerning

deductive reasoning are based on the idea that

contents and context are essential to reasoning, that

deduction is supported by the mechanisms of mental

representation, and by proceedings concerning those

representations (Quelhas,2000). We stand that an

operational representation of schemes and schemas

can be useful to study human deductive reasoning -

this is the main reason why we will put together the

notions of

scheme, schema and cognitive mapping in

a multi-agent environment.

This line of research identifies several

complementary goals:

• to adapt cognitive mapping to represent

agent’s beliefs;

• to recognize complex cognitive structures in

agent’s cognitive maps, such as schemes and

schemas;

- -

- Adapt

employees to

changes

- Employees

thrust

- Employees

notion of

"group"

- Dialogue

- Interact

- Accept

suggestions

- Periodical

reunions

- Age of

employees :

30 / 35

- Internal

debates

- Resistance to

change working

habitudes

- -

- Inovate

working

processes

- Create

propositions

- Research to

improve working

processes

- Invest in R & D

- Professionnal

learning

ICEIS 2004 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS

440

• to represent the context of a given concept

through the notion of scheme;

• to conceive a reasoning methodology based

on schemes;

• to use schemes to communicate between

agents.

5 CONCLUSION

This paper has presented the work in progress in a

three-dimensional project, including three

complementary lines of research incorporating our

project: user modeling in a distributed cooperative

system, interactive cooperation in a multi-agent

structure, and knowledge representation in cognitive

agent architecture. These lines of research have

diversified theoretical foundations: cognitive

mapping; multi-agent interaction; cognitive agents;

knowledge representation; interaction protocols in

open systems, such as service-oriented

architectures; cognitive science foundations and

contemporary psychological theorists. We intend to

make propositions regarding the use of multi-agent

systems in organizations, namely in what concerns

support to decision making processes and, in a

general way, knowledge management within

organizations. Finally, the result of our project will

be provided to the research community, throughout a

software library allowing the implementation of

these ideas in organizations.

REFERENCES

Arabnia, Hamid R., Mun, Youngsong (Eds.), 2002,

Proceedings of the International Conference on

Internet Computing, IC'2002, Las Vegas, Nevada,

USA, CSREA Press, 2001, ISBN 1-892512-38-6 -

Volume 3.

Bauer, M., Dengler, D., Meyer, M., and G. Paul, 2000,

Using Task Models for Efficient User-Agent

Interaction, IUI2000 Workshop on Using Plans in

Intelligent User Interfaces.

Chaib-draa, Brahim, 2002, Causal Maps: Theory,

Implementation, and Practical Applications in

Multiagent Environments, IEEE Transactions on

Knowledge and Data Engineering,

November/December, 14(6).

Estivals, Robert, 2002, Théorie Générale de la

Schématisation 1 : épistémologie des sciences

cognitives, (in french), Paris : L’Harmattan.

Li, Yinsheng, Ghenniwa, Hamada and Shen, Weiming,

2003, Integrated description for Web service-oriented

agents in e-Marketplaces, The 16th Canadian

Conference on Artificial Intelligence AI 2003 -

Business Agents and the Semantic Web

(BASeWEB’03).

Louçã, Jorge, March 2003, Modeling context-aware

distributed knowledge, AAAI Spring Symposium

2003 on Agent-Mediated Knowledge Management

(AMKM 2003), Stanford University, San Francisco,

USA.

Louçã, Jorge, June 2003, Representing emergent

sociocultural phenomena: propositions to support

current emergentist trends in psychology and

sociology, NAACSOS (North American Association

for Computational Social and Organizational Science)

Conference, Carnegie Mellon University, Pittsburg,

USA.

Louçã, Jorge, October 2003, A Conceptual Framework to

Represent Emergent Social Phenomena, The Agent

2003 Conference on: Challenges in Social Simulation,

University of Chicago, Chicago, USA.

Luck, M., McBurney, P., Preist, C., (Eds.), 2003, Agent

Technology: Enabling Next Generation Computing,

Agent Link Network, available at

http://www.agentlink.org/

Piaget, Jean, 1967, La psychologie de l’inteligence, (in

french), Paris : Armand Colin.

Quelhas, Ana Cristina, 2000, Raciocínio Condicional:

Modelos Mentais e Esquemas Pragmáticos, (in

portuguese), PhD dissertation presented at the

Université de Provence en Aix-en-Provence (France),

ed. ISPA - Instituto Superior de Psicologia Aplicada,

Lisbon, 2

edition.

Robins, Bill, Sleeper, Brent, McTiernan, Chris, 2003, Web

ServicesRules: Real-World Lessons from Early

Adopters, Stencil Group Report, available at

http://www.stencilgroup.com/ideas/reports/2003/wsrul

es/wsrules.pdf

(Sycara et al, 2002) M. Paolucci, T. Kawamura, T. R.

Payne, and K. Sycara.. Importing the semantic web in

uddi. In Proceedings of E-Services and the Semantic

Web Workshop, 2002.

(Sycara et al, 2000) H.C. Wong and Katia Sycara. A

taxonomy of middle-agents for the Internet. In

Proceedings of the Fourth International Conference

on MultiAgent Systems, July, 2000, pp. 465 - 466.,

2002

Sowa, John, 2001, Processes and Causality, available at

http://users.best.net/~sowa/ontology/causal.htm

Sowa, John, 2002, Architectures for Intelligent Systems, in

a Special Issue on Artificial Intelligence of the IBM

Systems Journal, vol. 41, no.3, pp. 331-349.

van Elst, Ludger and Abecker, Andreas, 2002, Domain

Ontology Agents for Distributed Organizational

Memories, in Rose Dieng-Kuntz and Nada Matta

(eds.): Knowledge Management and Organizational

Memories. Kluwer Academic Publishers.

W3C, 2003, Web Services Architecture, W3C Working

Draft – version of 8 August 2003, available at

http://www.w3.org/TR/ws-arch/

MULTI-AGENT PROPOSITIONS TO MANAGE ORGANIZATIONAL KNOWLEDGE: POSITION PAPER

CONCERNING A THREE-DIMENSIONAL RESEARCH PROJECT

441