Metapattern for converging knowledge management with artificial intelligence

Pieter Wisse

Conceptual modeling is an aspect of knowledge management, artificial intelligence, etc. As knowledge management (KM) and artificial intelligence (AI) are far from unequivocal concepts — they vary with ‘situations,’ as this paper argues — modeling methods have varied accordingly.

Metapattern (Wisse 2001) is designed to encompass variety in agreement with perspectival inquiry (Haynes 2000). It essentially shifts atomicity to particular behaviors (Wisse 2002b). Metapattern is sketched here with an emphasis on assumptions, followed by remarks on related directions in knowledge management and/or artificial intelligence.

An exchange of redundancy

Social psychology instructs about the situational nature of behavior. An actor, or agent, is assumed to reside in various situations. Hence variety exists in the actor’s behavior. In fact, a particular behavior completely corresponds with the actor as far as a particular situation goes. Adding situation and behavior therefore turns inside-out the treatment of an actor as entity/object. Only an actor y’s barest identity remains necessary and sufficient for relating a situation x and a behavior z (Figure 1.a). The whole of actor y is now reflected by his particular behaviors across relevant situations (Figure 1.b).

Figure 1: The conceptual triad of situation, (actor’s) identity and behavior.

Please note that juxtaposition of behaviors rests on repeating the — reference to the — actor’s identity. An instance of identity exists for every relevant situation, i.e. where the actor has a particular behavior. Through repeated identities, differences (particular behaviors) are reconciled with unity (one actor).

In traditional modeling (entity-attribute-relationship, subject-predicate, object orientation, object-role, etc.), repetition of identity shows up as redundancy and is therefore ruled out. Metapattern, however, accepts repeating identities for situations as the deliberate price for avoiding hierarchical decomposition strictly within the boundaries of one actor itself. It is a small price compared to — the disadvantage of — a hierarchical modeling approach to behavior which leads to redundancy when behaviors in different situations are similar. For some behaviors inevitably are, to some extent, anyway, and all the more so as the modeling horizon broadens. Aspect orientation, for example, is an attempt to overcome redundancy of behavioral modeling. In software engineering, it is more generally known as separation of concerns. Every approach ultimately leaving the wholeness of the actor intact is bound to fail, though. The issue simply dissolves with metapattern.

Flexibility through relative concepts

Metapattern’s compactness and flexibility comes from the assumption that situation, (actor’s) identity and behavior are relative concepts. A rigorous set of modeling constructs applies throughout. Following the spatial orientation of Figure 1, decomposition can proceed both up- and downward.

Upward, for example situation x₁ can itself be considered as constituted by several actors’ identities — presumably all different from I_y — appearing in a correspondingly less determined situation. Introducing levels, the original situation x₁ may be designated as situation x_{m, 1}. The result of one step of upward decomposition is shown in Figure 2. The limit to upward decomposition lies in ambiguity. It can proceed as long as identities at the next lower level, such as I_y for the step illustrated here, can connect with only a single identity in what previously was held as an undifferentiated situation.

Figure 2: Upward decomposition.

Upward decomposition has a paradoxical ring to it. It can be extremely productive, though, to attempt it from any instance of an identity as a starting point. Is a person’s particular behavior correctly located in a city? Or is specific reference to his home — within the city — more apt? Or doesn’t it matter in which city his home happens to be (and was the original situational designation irrelevant, even)? In Figure 3, the right hand side of Figure 2 has been adapted to indicate what after one step of upward decomposition now count as situation, identity and behavior.

Figure 3: A shifted configuration of situation, identity and behavior.

Reading these figures in reverse order gives an impression of downward decomposition; the original identity is encapsulated within a more determined situation while the original behavior is decomposed into identities, each with situational behavior at a lower level of the conceptual model.

There are no limits to upward and downward decomposition. A upper boundary condition must be formally set, however. It reflects the model’s horizon, corresponding to the most generally accounted for, least determined situation. In metapattern’s visual language (Wisse 2001) it is a thick horizontal line.

A conceptual model takes on the shape of a lattice of nodes. Some nodes are connected to the ‘horizon.’ Different instances of equal identity may be connected laterally to indicate that an actor’s behavior in one situation is invoked from — his behavior in – another situation.

Temporal variety

How metapattern accommodates dynamics also goes beyond the fixed, singular instance of identity as implied by traditional modeling methods. On the assumption that each instance of an identity is supplied with a unique … identifier, such instances can be moved about in a controlled manner. For example, the home adress of person A was x₁ up to time t₁, and from then on it has been x₂. The more radical decomposition has moved toward a lattice consisting mainly of identities, only leaving relatively little behavior specified otherwise, the more finely grained management of temporal variety is.

Knowledge management across universes of discourse

The introduction to metapattern is admittedly sketchy. It should suffice to claim for metapattern that it invites modeling beyond given universes of discourse, language games etc. Another language game, for example, can initially be installed as a roughly outlined situation. Upon closer inspection, some relevant behaviors may be seen to coincide with behaviors encountered in other ‘language games,’ all captured from a common horizon. The overall conceptual model can be tuned accordingly, gradually supplanting the original classification of such language games with a more tightly knitted, integrated representation of situational behaviors. With so-called legacy systems as embodiments of limited universes of discourse and language games, metapattern shows how improvements can be practically achieved, i.e. goal-oriented, step-by-step, manageable, cost-effective, etc.

Open interconnection, now katalyzed into reality by the Internet, is rapidly making obsolete the conceptual confinement implied by traditional database technology. One particular database’s data dictionary is no longer the limiting measure for a universe of discourse and its control. Information from disjunct, often even varying, physical sources must increasingly contribute to conceptually significant assemblies. This explains the interest in the so-called semantic web. Ontology is hot. Understandable from the largely technological bias, solutions are still mainly proposed with reference to resources. Metapattern approaches conceptual problems … conceptually, resulting in an unambiguous design as possible for application of information and communication technology.

Accounting for (individual) differences in knowledge management

Metapattern readily supports management-of-knowledge where knowledge is considered the result of objective analysis. Indeed, in daily practice much of human activity conforms to objectivity. The train leaves at a certain time for all who care to travel with it. The price of a loaf of bread is set regardless of the individual customers, etcetera. But how about individual differences? To what extent should knowledge management allow for human differences in daily practice, too?

It is especially when account must be taken of essential differences that metapattern supports necessary and sufficient elaboration while maintaining cohesion. Does the experience of one employee provide the appropriate model of behavior for another employee? Is a report of an experience rightly called knowledge in the sense of useful? Should conditions of usefulness be made explicit?

It might be that the careless assumption of equality, similarity, and so on, between situations is risky. In such cases, it is recommended that situations are explicitly classified with regard for individual involvement. This requires the — identity of the — reporter to be explicitly stated, too, as part of the situation of the experience-turned-report for knowledge management. In addition, the reporter’s motives might be relevant, suggesting a perspectival turn (see below).

What metapattern first of all can help to establish is whether formal knowledge management is at all worth substantial efforts. Suppose, when documenting an experience, a reporter is asked to include more information than he can realistically be expected to provide. That would immediately make using what was nonetheless recorded as knowledge extremely risky. Wouldn’t a superior alternative simply be a personal reference, to be called upon only when a particular need arises? Conceptual models should assist making practical judgements before incurring the bulk of costs.

Again, metapattern doesn’t impose any conceptual limits upon modeling knowledge even when individual differences must be included. The relative nature of its key concepts (situation, identity and behavior) guarantee open-ended opportunities for decomposition both upward and downward. Precisely because conditions for rigor are secure, the question of relevance deserves priority.

Perspectival turn in artificial intelligence

A pragmatic view of artificial intelligence emphasizes independency of individual agency. A whole actor is considered — and actually designed, constructed, used, etc. — as the artifact. Then, artificial intelligence emerges as an irreducible aspect of the encompassing artifact for controlling its behavior.

Metapattern suggests an actor invokes a particular behavior as befitting a particular situation. The patterns an actor is primarily recognizing are therefore: situations. Recognition is not merely passive, though. An actor, artificial or not, would be overwhelmed when it lacks a pattern of its own to establish situations-as-patterns. Traditionally, an actor’s self-directed activity is attributed to goals. A goal is seen as different in kind from, for example, a behavioral specification that should be executed to accomplish it.

Figure 4: Semiotic ennead.

An extension of the semiotic triad into an ennead (Wisse 2002a) argues for a relative status of goal, or motive. In Figure 4, the term ‘motive’ is preferred. In semiosis, an actor constructs a sign from a configuration of — again, the relative concepts of — motive, focus and concept. In short, he brings a perspective to his own observational behavior. The resulting sign is transformed in cycles of semiosis into another configuration of motive, focus and concept. That is, semiosis changes the actor’s perspective, resulting in one particular behavior or another.

The ennead is a sophisticated metamodel for connectionism. On the basis of the formal articulation of perspective, a classification of intelligence, mind, etc. may be erected. At the one extreme is an actor whose motive, focus and concept remain fixed. Only values of the concept(s) can change. This is the proverbial thermostat. At the other end of the spectrum is an actor who is able to change perspective completely, i.e. including his uppermost motives. Findings in evolutionary psychology are straightforward: man’s perspective is nowhere near perfectly plastic.

From such a classification, artificial intelligence can aim for results both more relevant and rigorous. For — artificial — actors whose perspectives may change, metapattern now leads to conceptual models for also recording the effects of perspectival plasticity. That is, semiosis both extracts from and feeds into a comprehensive repository.

Convergence of KM and AI

As knowledge management advances to cover ever more complex human endeavors it inevitably needs to acknowledge the relevant distribution of perspectives among stakeholders. The nature of the challenge is similar to empowering an artificial actor with behavioral variety. Knowledge management and artificial intelligence can converge through the conceptual technology of metapattern.

References

Haynes, J.D. Perspectival Thinking: for Inquiring Organizations, Palmerston North (New Zealand), ThisOne and Company, 2000.
Wisse, P.E. Metapattern: context and time in information models, Boston (USA), Addison-Wesley, 2001.
———— Semiosis & Sign Exchange: design for a subjective situationism, including conceptual grounds for business information modeling, Voorburg (Netherlands), Information Dynamics, 2002a.
———— The ontological atom of behavior, in: PrimaVera, working paper nr 2002-06, Amsterdam University (Netherlands), http://primavera.fee.uva.nl/, see publications/working papers, 2002b.

note

The major part of this text has subsequently been included in ‘The Relationship between Metapattern in Knowledge Management as a Conceptual Model and Contragrammar as Conceptual Meaning,’ co-authored with J.D. Haynes (in: Proceedings of the First Workshop on Philosophy and Informatics, Deutsches Forschungszentrum für Künstliche Intelligenz, research report 04-02, 2004).