Effective Description of Enterprise Processes

Roger Burkhart
Deere & Company
John Deere Road
Moline, IL 61265
roger@ci.deere.com

Position Paper
Workshop on Specification of Behavioral Semantics in Object-Oriented
Information Modeling
Workshop on Object-Oriented Reflection and Metalevel Architectures
OOPSLA `93

DRAFT
July 31, 1993

1. Introduction

This paper proposes that a common set of primitive modeling concepts be
adopted to express and specify the behavior either of enterprise
objects (belonging to an application domain) or computational objects
(objects built in software and residing inside a computer, but
typically presumed to represent or correspond to some sort of external
domain objects). The applied world of enterprise or business process
modeling and the independent world of formal software specification
each have significant elements to contribute to the other. The
remainder of this paper identifies some of these elements, and outlines
a unified approach for describing behavioral phenomena in either
domain.

The world of business, including as major subdomains the organization
of people and tasks, the management of economic abstractions such as
money and contracts, and the technical analysis and realization of
physical processes and artifacts, is extremely complex. Existing
approaches to modeling a business capture only a fraction of the
complexity that exists in any real business domain. Rather than
describing or prescribing the real activities of the business,
information modeling has typically retreated to defining only static
structures of information that can be recorded as periodic snapshots
over some limited portion of the business state.

The object-oriented perspective contributes a key element to the
problem of business description: the principle of behavioral
encapsulation. The only relevant characterization of an object, as
viewed by agents external to that object, is one framed entirely in
terms of observable behavior. A formal description of business objects,
which assigns intended meanings to each of its terms and grammatical
constructions, must ground itself in primitive phenomena manifested at
the interfaces between objects. Software objects can be regarded as a
special case of business objects, since technical artifacts built
inside a computer are within the domain of objects a business deals
with.

2. Describing Business Complexity

Extending the scope of behavioral description to the unbounded
complexity of a real business domain is instructive for many reasons.
In particular, it forces recognition that many elements difficult or
intractable to deal with in existing formal frameworks are nevertheless
essential, unavoidable realities that must be addressed by a general
descriptive framework. Among these realities are the following:

 - Concurrency. In any complex domain, many things are happening at
  once. A general framework must deal with concurrency and the autonomy
  of concurrent processes at its outset. Objects in general are active;
  they encapsulate autonomous processes that accept and generate
  threads of interactions across an encapsulation boundary. Objects can
  also be internally concurrent; their encapsulation boundary can
  surround a collection of concurrently interacting subobjects.

 - Partial information. No part of the system can have information
  about any other part of the system, unless specific interactions have
  occurred to transfer that information (either directly or by
  inference out of information that has been transferred). Channels or
  mechanisms for information dissemination must themselves be described
  in the characterization of the total system. There can be no
  assumptions about global time or synchronization; synchronization
  must be mediated by mutual events that establish a common timeline.

 - Visibility and connection. For reasons of scalability, a process can
   directly affect or query only a finite set of other processes. The
   ability to reference or affect other processes must be inferred out
   of this network of direct connections. This basic network can be
   extended by processes that serve as mediators for others, creating a
   network of mediated interactions that is typically much more densely
   connected than direct interaction. Modes and layers of interaction,
   together with all processes that enable them, are fundamental to
   understanding the potential for action within the system.

 - Creation and destruction. A process not only has limited knowledge
   of other processes it can interact with, but the collection of these
   processes is not fixed. Processes can create or destroy other
   processes, and the ability of any process to communicate with others
   can change continually during its lifetime. A general framework for
   system description must account for the highly dynamic character of
   process configuration, and cannot assume any closure over the rest
   of the environment which any carved-out portion might eventually
   encounter.

 - History. The action of an encapsulated process can depend in general
   on any part of its history, including every interaction in which a
   process has participated since its creation. An external observer
   cannot assume any restriction on the ability of an object to
   remember any part of its history using whatever forms of internal
   representation the object might employ. The ability of the past to
   affect the future cannot in general be captured by any finite state
   summary. The history of an object is bounded on one end in that all
   interactions are successors of a single creation event, but is
   unbounded on the other in that an object may continue to participate
   in further interactions as long as it exists.

 - Reflection. The external behavior of an object can be a function not
   only of its history and an initial set of constitutive rules, but of
   internal processing that extends or evolves any initially
   predictable behavior. Objects can process representations not only
   of their history, but of their own capabilities, and they can
   compile and apply these representations to become effective over
   their own subsequent behavior. Monitoring and modification of
   behavior is a process that feeds on its own results, enabling a
   continual refinement of behavior into more complex forms.
   Indeterminism, in the most extremely surprising, creative, and novel
   ways, saturates the interior of a general object.

 - Ontology. A formal language or framework says nothing without a
   commitment to the meaning or interpretation of its terms and
   formulas. The word ontology has recently been adopted to refer to
   the conceptualization, or term definitions, expressed by a
   particular formal system. As domains expand and begin to overlap,
   particular ontologies must be integrated with each other.
   Progressively deeper and more comprehensive ontologies are required
   to accomplish this integration. Ultimately an ontology is needed in
   the true philosophical sense of this term: a systematic account of
   everything that can possibly exist, including an ordering of
   inherent dependencies. Object-oriented description begins a
   commitment to one such ontology: it proposes that the world be
   understood in terms of the phenomena or appearances generated by
   active elements that observe, affect, and control each other, either
   directly or through their appearances.

One way of thinking about the problem of business description is that
it must deal with the entire program of distributed artificial
intelligence (DAI), which studies problem solving and general behavior
in systems of intelligent agents. To cover a more complete domain, the
particular focus on problem solving and intelligence must be extended
to material worlds and apparently unintended phenomena. These too might
be explainable in terms of interacting elements possessing only
primitive behavior and yet leading to approximations of intelligence in
greater or lesser degree. Across such levels, an open challenge is to
develop a consistent account of intention or agency: to explain what it
means for an object to act so as to seek or cause some outcome.

Object-oriented programming sometimes suffers a vague disrepute arising
from a tendency to anthropomorphize objects. Perhaps this taint can
be dissipated by asking instead what it is that animates an object.
The ability of an object to act, and to participate in actions with
other actors, is deeply important to the core of what an object is. It
usually defines why we draw the boundaries of an object where we do and
label "these" phenomena as belonging to "this" object.

3. Using Formal Models

While the business domain points the way to a tar pit of inextricable
complexity, the world of formal systems, and formal software
specification in particular, offers a more comforting world of
carefully limited and mastered complexity. Both worlds are important,
because many phenomena of business naturally limit themselves to
special cases of the general model of complexity suggested above.

Mapping the features of a general model may well serve needs of both
domains. A commitment to a comprehensive scheme for describing all
possible phenomena, no matter how extreme or intractable the resulting
complexity, results in a simpler and more adaptable foundation than
approaches that just plunge into the middle of the scale of complexity
without concern for eventually meeting up with other important entry
points.

Formal systems, like almost any other artifact, are contained in the
domain of business objects (business exploits anything potentially
useful to its aims). Their effective application requires that they be
understood in the context of their intended use, operation, and
semantic interpretation. Deterministic machinery, and its possible
interpretation as formal language, has many potential roles in the
operation of a business. Business puts a premium on predictability and
control over many of its subprocesses, and therefore puts in place many
prescriptive structures that deliberately exclude any unplanned role
for complexity or contingency.

A special category of deterministic machinery is designed to carry a
weight of semantic interpretation, so that it can serve as formalized
language. Formalized language is important for at least two basic
reasons: to help people establish precision and avoid ambiguity (and
perhaps achieve better understanding as a result), and to allow
automated processing of language structures, either to perform
inference and simulation, or to establish causal interaction between
the formal system and some component of its environment it is intended
to track or control.

First-order logic is a particular formal language that serves as a
final arbiter of precision and avoided ambiguity. Its machinery is
cumbersome for direct use in most applications, but any other language
should be able to specify its mapping to the neutral form of
first-order logic to stake its own claim to formal status. The
selection of predicate logic as a basis for formalizing the conceptual
schema of a database [1] or for knowledge interchange [2] is entirely
appropriate, since no more complete or inherently capable method for
formalizing the intended meaning of a static communication has ever
been found. More esoteric forms of logic, such as modal or higher-order
logics, can be interpreted by expanding an initially first-order
universe to reflexively include components of the formal logical system
itself, such as predicates, expressions, and constructs devised
specially by an agent to organize and manage representations built
within the formal system.

To ensure sufficient extensibility, the semantic framework of
first-order logic needs to include not only the objects belonging to a
universe of discourse, and the formulas in a language, but also a
subject who constructs and interprets these formulas. Potentially this
subject-oriented semantics might have as much or more to say about the
process of knowledge acquisition by a subject (its purposes or
expectations, the historical or contextual situation, effect on
existing beliefs, etc.) as it has to say about the external objects
about which knowledge is being acquired. This semantics would also
address the process and mechanisms by which a knowledge-organizing
agent establishes its presumed relation to the environment where it
observes and acts.

4. Models for Specification

One salutary characteristic of first-order logic is its neutrality with
respect to ontology: any scheme for describing a universe of
discourse can be supported by suitable definition of terms. Once a
commitment to basic ontology is made, however, one of its challenges it
is to explain the subject-object interactions out of which logic and
linguistic description arise. These include the prescriptive use of
representations by subjects who impose plans and specifications on
their environment.

An ontology of behavior and process, by situating logic and its
semantic relations in a pervasive setting of interacting processes,
points to a more general way of thinking about formal specifications of
all kinds.  Each such specification can be regarded as a blueprint for
a machine, that if fed information about all interactions occurring in
some domain, would continually maintain a single result value signaling
that the interactions either do or do not conform to the specification.
All forms of precise behavior specification could then be interpreted
as indirect but equivalent ways of specifying such an observer machine.
The machine need not actually be implemented to be useful; it could
serve simply as a model by which to interpret statements of other
specification languages.

An indirect specification of behavior, in terms not of the behavior
itself but of a process designed to review and test behavior, allows
the statement of rules and conditions that can be stated in no other
way. Additionally, it can both specify and implement the interactions
by which a specification can be made causally effective over the
behavior it intends. Abstract models of computational processing can be
employed to avoid overspecifying the details of a particular
computational machine. Whether implemented or abstract, the process
that has the most powerful ability to review behavior is one that can
also simulate the kinds of behavior it expects to observe. The most
general model for enterprise simulation would consist at least of a
dynamically configuring community of interacting agents busily
processing their histories and replacing their capabilities in a fully
reflective way. Such machinery could establish running models of
desired enterprise behavior (for example, complex problem-solving
processes) to establish performance and quality criteria that could be
established in no other way.

A simulation of interacting agents that includes the reflective ability to
modify its own configuration and processing is the most expressive model by
which to monitor and control enterprise behavior. An initial and very
sketchy outline of such a model was offered in a previous paper [3].
That paper also briefly discussed and referenced some related work,
such as the Actors model of distributed computation [4]. Details and
major features of the proposed design continue to evolve, but the
overall program - an abstract and general model of concurrent
computation by which to specify enterprise behavior - continues to
appear feasible.  Whether it would ever succeed in mastering the real
complexity of a business (which includes the routine confrontation of
unprecedented circumstance) remains questionable.

References

[1]  _Concepts and Terminology for the Conceptual Schema_, Technical
     Report 9007, International Organization for Standardization,
     Geneva, Switzerland, 1987

[2]  Genesereth, M. R., Fikes, R. E., _Knowledge Interchange Format:
     Version 3 Reference Manual_, Logic Group Report Logic-92-1,
     Computer Science Department, Stanford    University, Stanford, CA,
     1992

[3]  Burkhart, R., "Process-based Definition of Enterprise Models," in
     _Enterprise Integration Modeling: Proceedings of the First
     International Conference_ (Charles Petrie, ed.), MIT Press,
     Cambridge, MA, October 1992.

[4]  Agha, G., Actors: A Model of Concurrent Computation in Distributed
     Systems, MIT Press, 1987