A distributed intelligence approach to Knowledge Management

1
A distributed intelligence approach to Knowledge
Management

• Matteo Bonifacio Paolo Bouquet
• University of Trento
• in collaboration with
• Istitituto per la Ricerca Scientifica e
  Tecnologica
• (ITC-IRST)

2
Outline of the talk

• Knowledge is intrinsically distributed
• However, most architectures of KM systems push
  toward centralization
• Consequences of this contradictory situation
• A distributed architecture for KM
• Re-locating AIs contributions in a distributed
  approach to KM
• Scenarios and applications

3
Knowledge is (intrinsically) distributed

• Physically knowledge is created by different
  individuals in their daily work and initially
  stored in their minds
• Socially individuals belong to communities that
  provide different interpretive schemas (different
  identities and languages, heterogeneous
  semantics)
• Pragmatically the same piece of knowledge can
  be used in many different ways

4
Consequences

• Advantages
• strong sense of identity
• multiple perspectives
• specialization
• Problems
• accessibility
• replicability
• generalization

5
(Implicit) assumptions

• Current KM systems and architectures embody a
  collection of
• epistemological, and
• managerial
• assumptions that are inconsistent with the
  distributed nature of knowledge

6
Epistemological assumptions

• From subjective knowledge to objective
  representation (accessibility)
• From context-dependent to fully general
  representations (replicability)
• From socialization to globalization
• From heterogeneity to homogeneity

7
Managerial assumptions

• After all KM is an instance of management
• knowledge is an asset (resource)
• whose dissemination (task)
• must be centrally driven (allocation)
• and controlled (monitoring)

8
Centralized architectures for KM
Organizational Contexts
Organizational Intelligence
Semantic Differentiation
Homogenization
Organizational representational structure
Categorization
Sources
Corporate memory
9
Why it cant work

• Theoretical reasons
• contextuality is irreducible (McCarthy,
  Giunchiglia Ghidini, Fauconnier, .)
• knowledge lives in communities (Lave Wenger,
  Boland Tenkasi, .)
• role of paradigms/frames/mental models (Kuhn,
  Nonaka Takeuchi, Orlikowski, .)

10
And indeed it does not work!

• Communitys perspectives are overridden in the
  organization of K
• People do not feel at home in the centralized
  organization of knowledge
• They prefer to duplicate information in local
  repositories
• Any information systems design that neglects use
  and user semantics is bound for trouble down the
  line - it will become either oppressive or
  irrelevant
• Susan Leigh Star, Sorting Things Out, 2000

11
The case of a worldwide consulting firm
Differentiation considered just as ambiguity
Organizational Contexts

Organizational Intelligence
Semantic Differentiation
Homogenization
Organizational representational structure

• Inefficient
• Ineffective

Categorization
Sources
12
A distributed intelligence approach

• Accessibility access does not make sense without
  taking context into account
• Replicability exchange across communities
  happens through a process of meaning negotiation
  (semantic interoperability) between different
  perspectives
• Generalization dynamic emergence of shared
  contexts

13
Two kinds of processes

• There is a qualitative difference between
• perspective making (building a shared perspective
  within a community)
• perspective taking (interacting with the
  perspective of a different community)
• Boland Tenkasi, 1995

14
Architectures for Distributed KM

• IT architectures must be coherent with these
  assumptions
• Corporate knowledge is the result of coordinating
  many autonomous knowledge sources
• A KM system must be designed as a system for
  managing many autonomous (local) knowledge
  systems (distributed architectures)

15
A distributed architecture for KM

Meaning negotiation
Group Agent 1
Group Agent 2
Wrapper
Wrapper
16
Main components

• Information sources each community produces and
  collects documents and data according to its
  objectives
• Context an explicit representation of a
  communitys perspective (local map)
• Intelligent Agents meaning negotiators across
  perspectives of different communities

17
Advantages

• Perspectives are explicitly - though partially
  -represented in the system
• People interact with the system using their
  perspective
• People can search the system through their own
  perspective as a filter
• People can learn from other peoples perspectives

18
Re-locating AI contributions

• Supporting perspective making
• Supporting perspective taking
• Knowledge level organizational analysis and
  designing

19
Perspective making

• Making organizational contexts explicit
• text and data mining, automatic categorization,
  linguistic analysis
• Representing contexts
• languages for representing local
  conceptualizations (e.g. ontology-based)
• Reasoning within a context
• Theorem Proving, Case-Based Reasoning
• Maintaining contexts
• intelligent context editors, personal and group
  assistants

20
Perspective taking

• Understanding the relationships between local
  knowledges
• formal models of the relationships between
  contexts, languages for knowledge sharing and
  integration
• Context matching
• algorithms for assessing semantic similarity,
  pattern matching, sub-symbolic models
• Meaning negotiation
• models of cooperation between autonomous agents,
  agent communication languages, planning
• Capitalizing on perspective taking
• models of belief revision, algorithms for belief
  updating

21
Analysis and design

• Organizational analysis
• agent oriented requirement engineering
• Design
• agent oriented software engineering
• Platforms
• multi-agent platforms, BDI implementations

22
Scenarios and applications

• Distributed semantic-based search engines
• Distributed workflow management systems
• Community-based content delivery
• Web-services (processes interoperability,
  collaborative planning)

23
The project EDAMOK

• Enabling Distributed and Autonomous Management Of
  Knowledge
• Funded by Provincia Autonoma di Trento for 3
  years (started October 2001)
• About 30 researchers and programmers mostly from
  IRST and University of Trento
• The role of scenarios and business cases