A Merged Future for Knowledge Management and Enterprise
Modeling
Draft
Editor:
H T Goranson
Advanced
Enterprise Research Office, Old Dominion University
Abstract
Both knowledge management and enterprise
modeling have strong interest communities; each has a sustainable market
in the enterprise supporting practitioners and theorists. But both have
apparent structural barriers to fulfilling early promise. This paper reports
on speculations toward marrying the two.
ICEIMT Provenance
The International Conference on Enterprise
Integration Modeling Technology (ICEIMT) was initiated in the early 1990’s
as a major planning activity to support an intended, large joint research
effort between the United States and European research establishments.
The “conference” included a substantial study and workshop prologue to
aggressively examine high value futures and strategies.
The research program did not mature
under combined administration as anticipated, but the ICEIMT product was
enormously influential in defining the discipline and industry of enterprise
integration and enterprise modeling. ICEIMT was repeated five years, and
again ten years later. The first workshop of the third ICEIMT was held
in Paris, December 2001, and addressed the topic of the relationship between
knowledge management and enterprise modeling. One workgroup took the charter
of examining possible combined futures and the research roadmap these futures
require. This paper is the result of that study.
Author Information
Participants in the work group were
Guillermina Tormo Carbó — Universidad Politecnica De Valencia, Spain;
Ted Goranson — Old Dominion University, USA; Michael Huhns — University
of South Carolina, USA; Jim Nell — National Institute of Standards and
Technology, USA; Hervé Panetto — Research Center For Automatic Control,
France; and Michael Wunram — Universität Bremen, Germany. Additional
contributions and insights came after the physical meeting from Bernard
Katzy — CeTIM. Ted Goranson was the editor for this report.
General Background: Knowledge Management and Enterprise Modeling
Both knowledge management and enterprise
modeling are well established in enterprises today. In both cases, one
can find:
-
A robust vendor and consulting community.
-
Well established university research groups, funded in different
ways.
-
An active press, targeting both managers and technicians.
-
Enough promise — supported by case studies — to fuel continued
investment and implementation.
but also:
-
Some spectacular failures.
-
Some vexing limits in successful implementations.
The workgroup examined those barriers
and found that a combining of techniques from knowledge management and
enterprise integration shows promise in addressing them in each.
Enterprise Modeling and Integration: Background
Enterprise integration addresses the
need to optimize operations, usually of large and/or complex enterprises.
This is a fundamental business need with direct and measurable benefit.
Though many optimizations can be done at a local or subsystem level, the
most effective strategy either starts with the system level or includes
it as an important component. For some time, there have been many techniques
to “model” processes and other elements of the enterprise. Modeling in
this context means creating an explicit representation, usually computable,
for the purposes of understanding the basic mechanics involved. One often
uses that understanding to measure, manage and improve the process or element.
A basic problem is that there are
many types of elements to be modeled in an enterprise, and many perspectives
and contexts in which those models would be “viewed.” Enterprise integration
in this context combines models and their uses in such a way that the whole
system can be seen in various coherent ways and for multiple purposes.
enterprise integration provides a sort of model of models or model framework
in which these components can be interrelated. All enterprise modeling
approaches share the fundamental strategy of integrating at the model level,
rather than at the application and software level, but there are wide differences
between enterprise modeling strategies otherwise.
Some enterprise modeling systems are
wholly computable. Enterprise resource planning (ERP) is a widely employed
enterprise integration subset that focuses on specific tasks, delivering
planning and control functions. It generally requires a constrained modeling
approach and heavy use of generic models, thus restricting the processes
for better or worse. The more general enterprise integration philosophy
is framework based, such frameworks supporting:
-
Levels of model genericity (to enable model — and best practice
— reuse).
-
Relationships among different views (for instance views needed
to see organizational linkages versus information flows).
-
Relationships among different types of basic entities in
the enterprise; for instance, activities need to be modeled in a fundamentally
different manner from roles or resources.
CIMOSA [citation] is a strong example
of such an integrating framework, major elements of which are standardized
internationally as ISO 15704 Requirements for enterprise reference architectures
and methodologies. enterprise integration frameworks are widely used, especially
in the subset of ERP noted above and a similarly focused subset of product
data management (PDM), which supports activities centered on the evolution
of product features as they are transformed by processes in the enterprise.
Enterprises which use computer aided design heavily implement enterprise
modeling in this fashion.
Enterprise Integration: Problems and Limits
We have enough experience with different
editions of enterprise integration and the enabling enterprise modeling
to understand general limits. The major problems are of two types:
-
enterprise modeling — at least traditionally — makes certain
assumptions: that one knows what they want to make (or do), who will do
it, and a precise notion (perhaps to change later) about how each element
of work will be done. Because the primary leverage from the approach is
the system view, some substantial part of the system must be included in
the model. But those enterprises desiring a system view at some cost, additionally
wish to include additional elements of the system: strategic planning,
marketing and product design (if applicable). These processes aren’t as
easily captured as process models however: they have “soft” elements like
unknown futures, tacit knowledge and poorly understood cultural and collaborative
dynamics. [ICEIMT2 citation on the challenges of softness.]
-
enterprise modeling usually deals with the normative, stable,
deterministic case. In other words, managers expect their world to remain
as it is because they are going to great lengths to engineer an operational
enterprise. Dynamic environments, evolving processes, shifting partnerships
and changing products are a way of life for many enterprises. So if enterprise
integration is employed, it must be more of the federated type than the
unified (ICEIMT1 and 2 citations on federated.) That means the enterprise
integration system must ideally be cheap to assemble, must change the source
models and process in little or no way, be responsive to change — even
indicate change, and be to some extent self-organizing and adapting. Current
systems do not impress in this regard.
Very likely, adding knowledge management
techniques to the mix can mitigate these problems, possibly in a revolutionary
manner.
Knowledge Management: Background
Knowledge management solutions address
several needs that all share the underlying notion that enterprises depend
heavily on individual and institutional knowledge — and the recognition
that it be better understood and managed. knowledge management is a set
of philosophies, tools and techniques to support various functions within
this need. It is strongly needs-driven, with both needs and basic tools
originating from the management science sector. It should be noted that
knowledge management and enterprise modeling differ in this fundamental
way. While both address pressing business needs, enterprise modeling originates
from the industrial engineering and operational perspective and is technique-centric;
knowledge management originates from the management perspective and is
needs-centric. These two communities have a poor track record of deep collaboration,
which may explain why such an apparent synergy has been hitherto unexploited.
The discrete problems addressed by
the knowledge management community are:
-
A need to capture individual knowledge to make it “institutional”
knowledge so that it can be reused in the enterprise, and persist when
an expert leaves.
-
There is often a second- order intent to use standardized
knowledge elements and communication methods to develop and support corporate
culture for competitive benefit.
-
Many wish knowledge management to improve support for the
“learning organization:” education at the individual, team and enterprise
level.
-
The development of knowledge “metrics.” The notion behind
this is that significant investment is wrapped up in knowledge, and there
is currently no good way to quantify the value of the result. Metrics needed
by two functions: financial accountants (to evaluate capital knowledge
assets); and planners using simple cost-benefit analyses in decisionmaking.
-
Auditability of intellectual property. Where the bullet above
concerns value, this one concerns ownership: tracking the initiation of
an idea and the various inputs in order to parse who contributed what and
when and prove it in court.
-
Selfawareness. The notion here is that the better you “know”
yourself and your relationship to the world, the better you can change
and manage yourself. Notably, this notion is the very same driver as in
enterprise integration. But in the knowledge management world, it is more
focused on strategic planning, where enterprise integration focuses more
on operations.
-
knowledge management is often invoked as the backbone around
which diverse corporate cultures will be combined after a merger or acquisition.
Because the needs of knowledge management
are more diffuse, the tools and implementations are too. Many tools are
simply ways of aiding collaboration by structuring the way information
is stored, indexed and shared. Also, many of the techniques are “soft”
and merely philosophical, motivational, or concerned with awareness-building.
There is a high level of puffery in the industry, and it is heavily consultant-driven.
enterprise integration and knowledge management are comparable in the rough
order of expectations, the intuitive recognition of value, the formation
of promising successes stories, and the often-repeated costly mistakes.
Knowledge Management: Problems and Limits
The general problems with knowledge
management systems are of two types:
-
knowledge management systems are “soft,” almost by definition.
They deal with intellectual property for which no good value metrics exist;
they deal with collaborative contexts which are not well modeled; and they
implicitly address the slippery reality of “tacit” knowledge. Many knowledge
management systems deal with strategic planning which means they address
uncertain futures, but without leveraging extrapolations from the “as is”
situation. “As is” in this context means the insight about who and what
a company is that is often tied up in the enterprise integration (or other
operational) system, whether or not formalized and automated.
-
knowledge management systems deal with “know-how,” but with
little emphasis on the “how.” In other words, the knowledge is not sufficiently
bound to the work of the enterprise, or what that work might become. One
part of this problem is the age old lack of linkages between strategic
planning and operational management — it is not just an impedance mismatch
between functions, but between methods and basic representations as well.
This mismatch frequently produces strategic decisions that make little
sense.
Just from this brief overview, the
reader may already be anticipating suggestions from the working group on
how the strengths of one approach could correct the weaknesses of the other.
knowledge management needs formalisms (which might help with metrics) and
anchoring in the enterprise’s actual work; enterprise integration needs
ways of dealing with knowledge about context and other soft elements, specifically
including tactic knowledge.
Near Term Future: Deductive Trust and Process Situating
The workgroup recognized a few near
term synergies between enterprise modeling and knowledge management.
“Knowledge” in the knowledge management
context is “justified true belief.” Each of those three words convey different
dimensions of trust in the information. Usually that trust is “inductive;”
the trust is based on (in ascending order of “closeness” to your own judgement:
-
Authority: Someone in the enterprise represents [do you mean
asserts or certifies?] that the knowledge is to be trusted. This person
might be trusted by you, in which case you are trusting that person as
a certifier of sorts; but usually, you are delegating trust.
-
Votes: The second case above involves a certifying agent
that has the authority of the enterprise, which can be seen as a case of
enough votes of the right kind. This type involves votes directly on the
information itself. You might not have cause yourself to trust the information,
but some group dynamic provides additional confidence, by aggregated authority
or broadened depth. (There are likely several group mechanisms involved
here, but the workgroup did not exhaustively explore them.)
-
Experience: You have seen this case before with enough similarity
and enough times to have confidence that it will turn out the same way
the next time.
But there is a different basis on
which one might base trust, a “deductive” basis which involves understanding
the cause and effect mechanics behind the situation in sufficient detail
to determine the outcome. For example, one may have experienced many sunrises
so have “inductive” confidence that the sun will rise again tomorrow. Or
that person may have deductive trust based on a knowledge of the planetary
mechanics that produce sunrises. That type of trust produces a better foundation
for justified true belief.
In the business enterprise, deductive
trust is much preferred, because it is auditable: decisionmakers can —
if so inclined — “audit” the trust behind the knowledge by zooming in on
the underlying physics. Most knowledge in an enterprise is of the inductive
type and this is reflected in current knowledge management systems, whereas
managers want most to be of the deductive type. Enterprise models capture
cause and effect dynamics within the enterprise, so a marriage would seem
manifest destiny. In this case, each element of knowledge in the knowledge
management system is linked to modeled processes (representing activities)
in the enterprise modeling.
Such a linkage can be made during
the (already costly) modeling and knowledge capture processes without unduly
extending the difficulty of either. The benefit to knowledge management
would be rather profound: some significant portion of the knowledge will
be (or be expected to be) deductively auditable by linkage to actual processes.
Another way of putting this is that knowledge in a knowledge management
system is know-how; current knowledge management approaches focus on the
“know,” but not the “how.” Linkage of knowledge management to enterprise
modeling provides the how. And that “how” linkage provides a significant
benefits: maintaining knowledge costs money — maintaining vitality in that
knowledge base costs more.
Knowledge managers need to know which
knowledge to “forget.” If there is a robust linkage to processes (current
and future), the knowledge has no apparent relevance to the business. That
should prompt an examination with one of the following results:
-
The enterprise modeling is incomplete and needs to be extended.
In this case, the existing knowledge indicates what processes need to be
better modeled or added to the process model inventory. Experience indicates
that this can be a powerful technique for modeling processes that have
“soft” mechanics (like many marketing processes for instance). The net
effect is that important processes that might otherwise be omitted, can
be included, some “holes” being filled by reference to knowledge in the
knowledge management system.
-
The knowledge is determined to be not relevant, and can therefore
be deliberately forgotten. The ability to know what is not relevant is
an important step in a system’s knowledge of itself, which in turn is likely
a necessary condition for being a “learning organization.” Knowledge should
be deliberately “forgotten” because it is out of date; because of machine
constraints (storage limits or search time problems); because it can be
more robustly handled by a collaborating agent; or because it is the kind
of basic knowledge that an enterprise needs to reinvent from time to time
to sharpen its learning skills. An example of the latter is a schoolchild’s
hand calculator deliberately forgetting how to multiply in order to force
the child to learn for herself.
-
The knowledge is determined to be relevant, but not well
supported by processes in the existing enterprise. This would indicate
modifying the enterprise. Often the solution in this instance is to develop
business partnerships with entities that can support the knowledge process
linkage either by supplementing the source enterprise, or maintaining that
knowledge itself.
-
The situation is the complement of the first case noted,
where an enterprise modeling is more complete than the knowledge base.
This can be used as an indicator of knowing what you don’t know within
the universe of interest. (This rounds out the four likely conditions for
full knowledge management: knowing what you know in a trusted way; knowing
what you can forget; knowing what you do not know; knowing where else you
can get the knowledge.)
Philosophically, one can understand
the synergy in another way. Knowledge resides in the individual, but has
value in the context of the enterprise. knowledge management can be seen
as the management of pieces of knowledge, while enterprise modeling can
be seen as the compositional framework for those pieces.
Another way of understanding the problem
is to consider a breakdown of knowledge management into four elements:
revealing of information; formation and management of facts; formation
and management of relationships and contexts among facts; and applying
that knowledge to effect. Today’s knowledge management systems do the first
two well enough, but need help with the other two.
Enterprise modeling may help with
understanding contexts. The basic idea behind enterprise modeling is just
this notion: taking fragments of information within the enterprise and
placing them in a larger context. enterprise modeling provides a registration
framework for the parts that relates one to another. But this framework
relies on artifacts of the modeling process that capture local interdependencies.
knowledge management systems incorporate a wider array of representations,
but even the most formal likely will lack these local features that allow
global registration. The news is not all bad, though, because of the appearance
of “ontology” based methods. Ontologies are formal descriptions of elements
and behaviors, originally devised to help share knowledge between systems
employing different representations.
A focus on ontologies should provide
a bridge between enterprise modeling and knowledge management as both disciplines
are becoming more ontology-focused. But the leverage is likely to come
more from the enterprise modeling side, because enterprise models are based
on the notion of activities and outcomes which automatically captures a
notion of local dependencies among information elements. This notion is
what — at root — allows compositions into larger context and systems. The
state of the art in process ontologies is the Process Specification Language,
developed at the U. S. National Institute of Standards and Technology [PSL
citation] and proposed as an international standard.
To provide a bridge between knowledge
management and enterprise modeling, PSL is the likely starting point. In
particular, the combination of PSL-like ontology structure and CIMOSA-like
composition strategies can be overlain on existing knowledge management
tools and theories to provide for systems behavior and business context.
Both PSL and CIMOSA (or substitutes) will have to be examined carefully
for needed extensions. Neither was designed for this larger, more ambitious
role.
Enterprise modeling can certainly
help with the “effect” problem in knowledge management. This is the problem
of linking each piece of justified knowledge to a business role. The trick
here (the workgroup believes) is the slight shift of emphasis from the
normative notion of “task” in enterprise modeling. enterprise modeling
is concerned with doing work, and processes that perform tasks are the
logical currency. But knowledge is more naturally seen as being applied
to solve problems. So a “problem” centric notion of the basic unit is proposed
as a bridging strategy. A problem is seen as a combination of a task (or
set of tasks) together with an element (or elements) of knowledge.
At first glance, this seems an immediately
implementable strategy to take short term advantage of synergies between
existing enterprise modeling and knowledge management tools and techniques.
The workgroup proposes serious research focused on this likely “low hanging
fruit.”
There is a precedent for the sort
of merger suggested here, and an example of how quickly the result can
spread and become the normal way of doing things. Financial management
is a matter of collecting many pieces of information and managing them
in much the same way that knowledge management intends to manage knowledge.
In fact, financial knowledge is a simple case — the qualitative case —
of general knowledge; so knowledge management is a generalization of financial
management.
About two decades ago, accounting reached
a crisis very similar to the knowledge management crisis today: (financial)
knowledge was collected but not relevantly “situated.” All of the problems
noted above existed in some form. The response was Activity Based Costing
(ABC), which simply uses a reduced form of enterprise model to ground the
individual costs and provide a way of intelligently assembling and relating
them. ABC went from a proposal like this to standard practice in less than
a decade; substantial benefits resulted. The nearterm enterprise modeling/knowledge
management proposal simply extends this logical evolution. As with the
ABC revolution, a key strategy is to continue the same basic tools already
in place; in this case, that means to continue using the operational and
business process modeling methods that are already part of the management
toolkit.
In the knowledge management context,
most knowledge management is non-formalized and non-managed so of course
it is non-computable. Informal knowledge management is a human-to-human
phenomenon based on personal networks. So this end of the merged knowledge
management/enterprise modeling system must leverage and ride on top of
the human infrastructure. (The better knowledge management systems already
do.)
Medium Term Futures: Fact Based Decision Making
Enterprise modeling is generally focused
on tactical optimization and similar types of self examination. But many
enterprises have their most pressing needs in strategic planning in the
context of uncertain futures. The more uncertain the future, the more significant
the threats and opportunities, but the less valid are simple extrapolations
from the past.
The importance of thinking about the
future is paramount for many enterprises, and for these real resources
must be committed for designing processes to be able to respond in an agile
way. Decisions are weighty and should be as much based deductively as possible.
Often this is termed “fact based decision making,” and often it is supported
by iterative simulations of what-if situations.
The connection with both enterprise
modeling and knowledge management is straightforward and obvious. “Traditional”
enterprise modeling structures processes in a way that systems can be optimized.
enterprise modeling for simulation (though not recognized as such) does
precisely this with the twist that the models are executable representatives
of the processes. Models in most conventional enterprise integration systems
don’t have this character, they are representatives used to understand,
not control processes. But the extension to control is not so great in
many cases, and indeed modern enterprise integration systems perform substantial
but limited control. The further extension to simulatable elements is also
not so great, generally involving substituting synthetic stimuli for real
ones. So it seems quite logical and cost effective to speak of enterprise
modeling in the context of strategic simulation, especially when the basic
unit is the problem as suggested above.
(It should be noted that the advantage
does not flow the other way. Most built-from-scratch simulation systems
use “models” that cheaply emulate the behavior of processes. This cheapness
is usually achieved by not modeling the underlying “physics” of the system;
also the granularity is not determined by the unit of work as seen at the
level of the work, but at some coarser subsystem granularity. As a result,
simulation-derived models cannot easily be adapted for wider purpose.)
The workgroup has three recommendations
to make at this medium term horizon and in the context of strategic, fact-based
simulation.
-
That the merger of enterprise modeling and knowledge management
be extended (and justified by) the use of the combined, structured knowledge/process
base for simulation. The advantages are potentially profound, because of
the reuse of information, the running start in well-founded infrastructure
that works, and the hard-won existing, practical binding to the way things
are really done. The technical challenges seem to be in “packetizing” knowledge
elements from the knowledge management side and adding a few new expressions
to modeling methods on the enterprise modeling side.
-
That notions of reuse be better exploited. The advantages
of this are seen as similarly profound. The basic problem is that knowledge
management systems are generally case-based, meaning that the knowledge
(and representation of the knowledge) is bound in specific cases containing
details that are irrelevant, artifacts of how the information appeared.
It is hard work to wade through cases to find relevant insights, extrapolate
what you need and apply it in a specific new context. The preferred alternative
is to build analogy-based knowledge management systems, which index and
manage information at the more generic and reusable abstract level.
Analogy-based systems are hard to
build, and certainly not expected in the near term. But the first step
toward such systems may not be so far away. It concerns clear guidelines
about what is generic and what specific to a task, problem or application.
As it happens, enterprise integration frameworks are nearly universal in
dealing with this problem in some way. Unfortunately, usually the solution
is a matter of art specific to the expert who is the source for the knowledge
being modeled. It probably is the case that every practical determination
of what is generic must be captured in this manner. In other words, it
is a type of metaknowledge that is captured at the same time and using
the same methods as the “base” knowledge. The format comes from the integrating
framework.
The bottom line is that knowledge
management systems can take a large step toward identifying generic analogies
by adopting enterprise modeling methods when collecting knowledge from
experts.
-
The final medium term recommendation concerns knowledge feedback,
or self reinforcing truths. An example of self reinforcing truth is when
a prominent stock analyst predicts a stock will rise. It does in part because
of her recommendation, which further reinforces confidence in her “analytical”
ability. It turns out that many dynamics in an enterprise may be of this
type to some extent. For example a quality metric may indicate quality
because second order dynamics may have adjusted or grown up around it to
promote quality results. For instance, a quality metric may be related
to number of inspections, and the precision of those inspections adjusted
to the fact that the system drives to many. In fact, the same quality could
be achieved with fewer inspections, but only by breaking the cycle of driving
toward many, promoted by the “truth” feedback.
Both enterprise modeling and knowledge
management systems have this problem. Usually it is concealed in so-called
“tacit” knowledge, which is the concern of many knowledge management systems.
But tacit knowledge is a famously black hole, not exhaustible. Good knowledge
management practices will help identify which tacit knowledge needs to
be captured and why, and (sometimes) at what cost. But these truth feedback
loops are best identified when they are deliberately broken as experiments,
for instance actually trying to reduce the number of inspections while
taking concurrent action elsewhere. One can practically do this only in
simulated enterprises, which brings us back to the merger of enterprise
modeling, knowledge management and strategic simulation.
The workgroup did not have time to
make specific recommendations of steps and research issues toward solving
this problem. But there is a general feel that low hanging fruit is available
when the problem is well stated and the more near term steps noted above
are taken.
Longer Term Futures: Self Organizing Enterprises
The workgroup considered the next
generation of enterprise integration systems. These are likely to exhibit
federating behavior and to do so using an agent system. They are also likely
to cover more — perhaps much more — of the enterprise. The new scope will
include at minimum some strategic planning and product definition as one
dimension of expansion, and some human, knowledge and collaboration dynamics
in the other dimension.
Agents in this context would likely
be the result of evolution from first generation models that represent
the superficial behavior of a process, and the second generation noted
above where the models capture underlying physics and can be exercised
in a simulation environment. Third generation models will be agents, small
pieces of software code that include the model and have the ability to
negotiate among themselves to optimize the system.
The result will be federated enterprise
integration where the system self-integrate. Note that the integration
is at the model level, not the enterprise proper. But since these models
have the ability to control, the effect is much the same.
This vision of enterprise integration
was already identified in the second ICEIMT when creating a capability
model for integrated systems [cite ICEIMT chapter]. A high level of integration
was when a process had the ability to see itself, see its context in the
system, and change itself to optimize the system — perhaps in collaboration
with others — even when it would apparently “harm” the agent. Presumably,
the risk-reward environment would be structured to reward this behavior,
even — it was speculated — reward an earnest search for such optimization
even when fruitless.
A higher level of integration is achieved
when an agent has the ability to see into the system, following a relationship
chain of some sort, discern a change in the system that would optimize
the system, and effect that change. In this scenario, all of the agents
involved would be rewarded in some way. For example, you may have a set
of processes that do nothing but search and optimize for agility against
a likely general change. If the enterprise were a virtual enterprise, this
agent would be looking at processes involved in the work and others not
currently engaged. All processes are in different formats, use only partially
integrated applications, and cross business and cultural boundaries. Agents
in these companies would be expected to enthusiastically support simulations
that could eliminate them from the partnership. In fact, each company is
expected to devise novel notions to support this process. This is possibly
an achievable goal, with at least one project underway [citation].
In this case, distinctions among knowledge
bases, operational process models, business processes, financial metrics
and simulation agents will have all but disappeared. But there clearly
are barriers. Perhaps the key barrier concerns realities of agent mechanics.
As noted above, these agents need to know themselves and what they know,
know what they do not, know where to get trusted information remotely,
know what to forget, and know the system’s goals and associated metrics.
Perhaps it will collaboratively determine those goals.
Knowledge managed by these agents
will include soft elements such as unknown futures, tacit knowledge and
collaborative (cultural) dynamics. The system will integrate (in addition
to factors currently handled by enterprise integration frameworks) product
features, process features, and system features. Product features are things
like holes and handles; process features are things like drilling and assemly;
system features are things like profitability and stakeholder satisfaction.
System features incorporate the system optimization metrics. The managing
context will be through bounding constructs (for instance discretely supervised
profit centers), practical constraints (such as physical transport of subassemblies)
and financial and implementation motivations.
The good news is that lots of work
by bright people is going into the general case. The business case provides
a much more simple universe than “real life” because businesses (but not
necessarily their employees) can be assumed to be motivated by financial
rewards that are quantifiable. There are only complications about deferred
rewards (market share, stock price, increased capability, new markets and
the like). Moreover, the business application can justify significant investments
in research and products — a repeatable improvement of only a few percent
means hundreds of billions a year. Moreover, agent systems seem inevitable
because it is the only scalable strategy for either knowledge or model
management.
The bad news is that agent mechanics
add another layer of complexity, so agent systems will be engineered for
simplicity. One almost certain strategy will be to devise agents that all
behave the same. The reason is that each agent has to know how the others
will behave; if they are all the same and the agent “knows itself” (or
has recourse to examine itself), it can predict how others will behave.
There are likely to be a whole swamp
of thorny research issues, but the workgroup focused on two that are likely
to be key. They are related.
The first involves harmonizing the notion
of uniform agents with the wild variety of models likely to be involved.
Recall that at this level of federated integration, diversity of methods
is expected, even encouraged. Obviously, some sort of agent wrapper must
be devised. The work indicated in the near and midterm agenda sketched
above indicates that this wrapper structure will almost certainly be designed
at the ontology level, built on extensions to PSL. This work will begin
on a firm basis because the first extensions to the existing PSL base will
be known agent needs. The most prevalent approach would be to use “speech
acts” which have several formal advantages and the elegant property of
being intuitively related to processes as they are currently modeled.
The second challenge indicated for attention
by the workgroup is the so-called multilevel agent problem. This problem
has an analog in the real world: not all processes need or want the same
level of freedom. Some collection of processes or organizational elements
will be bound more tightly within the enterprise. For instance, several
processes will typically be collected in a partner company. The processes
act as agents, but the company does too, and one is not a simple sum of
its constituents. Similar aggregations may occur by functions and many
aggregations may overlap.
The research challenge is to design
the wrapper so that it can both support the aggregation process and accommodate
the agency of these higher-level agents. Clearly, this strategy will be
framework-based, by methods extended from today’s enterprise integration
frameworks.
Extra Considerations
In addition to the ambitious agenda
noted above, the workgroup raised three issues to be considered by the
enterprise integration and knowledge management communities.
The first is a common suggestion that
needs to be underscored. enterprise integration and knowledge management
are generally thought of as something that large firms do to preserve their
way of doing things: maintaining centralized control. The agenda as outlined
above recognizes that reality but adds the clear alternative of smaller
companies or profit centers opportunistically aggregating to act as large
enterprises. That means that a future merged strategy must be devised with
sensitivities to small and medium enterprises. Flexibility and tailorability
must increase and complexity and cost decrease from current practice.
The second is the complement. Implementing
new infrastructure of the level of cleverness outlined will change some
fundamentals of how business is done. Some optimization must be considered
at a higher level than the larger enterprise, beyond to national and societal
interest. This is especially cogent as the initiating research will likely
be funded by government agencies.
The final concern extends that notion
in a structural way. Some technologies seem inherently abusable, while
others seem self-correcting by design. For example, the internet will likely
be an inherently democratizing force despite the best efforts of large
companies to “own” it or repressive governments to coopt it. The workgroup
recommends a project to study how to ensure that this new direction for
merged enterprise integration/knowledge management is inherently “good”
and designed in a way that prevents capturing by inevitable corporate attempts
to bend it one way or another for selfish purposes that compromise other
elements of society.
Return to previous page