Vincent P. Roske, Jr, Deputy Director, J8 (Wargaming, Simulation &
Analysis), The Joint Staff
New
concepts and techniques are taking form in our defense analysis
methodology. These new concepts have their
motivation in the defense debate. of
today. That debate ponders
issues involving the reaction of adaptive threats, the consequences of effects
based operations, the modes and value of information operations, the structure
and performance of command and control, and a host of other difficult to
analyze subjects. Our traditional
analysis methods do not seem to be very helpful when applied to these new
problems. Complexity science, open system dynamics, agent-based simulation and
genetic algorithms are just some of the ingredients that form the basis for
promising new, and very much needed, defense analysis methodologies.
Why don't our
old tools work on theses new problems?
A growing number of
subjects
in the defense debate belong to a class of problems often referred to as
"open" systems. They differ
from the systems we normally analyze with traditional operations research
methods that are closed system problems.
Closed systems are those around which we can define a boundary and
assume that energy (in any form) does not cross that system boundary. That this practice may not always be justified
may be evident
in the critique, "ignoring those data that don't fit the model."
The model, of course, is just our defined
representation of the real world phenomenon. Traditional systems modeling
approaches are "closed" representations of reality in the sense that
we prescribe the system and process — it’s our creation. This technique works very well for a large
number of problems we deal with, such as building an airplane, an autopilot, or
designing a military formation. These
are problems in which we are in control of the "system" and can
prescribe its structure and responses and prescribe the environment in which it
is to operate.
It
is as much convenience as perhaps unfortunate fate for us analysts that the
industrial revolution bequeathed us an arrogant sense that the world can be
thought of as assemblies of closed systems and that we can discern, define,
predict behavior, and control those systems.
From steam engines to assembly lines to electric power grids we have
come to believe that we are in control of our world; that if we push in
"here" we can predict what will bulge out over
"there." To the extent that we
are in control of a system's configuration and behavior, closed system analysis
methodologies work well. But what
happens when the problems we are analyzing are processes and environments we do
not control?
The
terms "adaptive" threat and "effects based" operations hint
strongly at a growing realization that in increasingly important ways, we are
NOT in control of many of the processes that are important to the success of
our military operations. For example, the flip side of the effects based
operations concerns unintended consequences.
Analyzing unintended consequences is a little like making a list of
things you haven't thought of. Yet, the
unintended consequences can ultimately determine the success or failure of a
military operation.
The
DoD is concerned today with how to "transition" the military
force. A more precise notion might be
that given the emergent behavior and adaptiveness of our adversaries and the
environments we face, our concern might better be on
becoming a "transitional" force.
"Transitional" may be
the characteristic that enables and informs a military force to adapt as
necessary in order to prevail. More important than being a new thing may be the
ability to be many things.
Command
and control process analysis is another subject that has always been difficult
to analyze. The difficulty comes when
trying to account for the creativity, initiative and perception of the human
factors. These are the factors that
ultimately drive the consequences, the effects, of the technology network that
we usually think of as
C2ISR. What
is a pound of C2ISR worth? The output of C2ISR systems seems to depend
on the perception, imagination of the people using the network and the
information in it.
Why
has the analysis of C2ISR been so unsatisfying?
The reason may be that energy, in the form of ideas, initiative and
imagination, expressed as perception and intent, are flowing across the
boundary of the technical systems we define and have been trying to analyze.
The result is that we do not control the cause-and-effect relationships in
those “systems" because the systems are open; the responses are basically unbounded
and unpredictable. Perhaps we
have been applying closed system analysis methodology to what is actually an
open system problem. The presence of the human being introduces energy across
the systems boundary and produces emergent and adaptive behaviors from the system. This is characteristic of open
systems, particularly of complex adaptive systems. This is the language of open systems
analysis. It's no wonder that we haven't
been very successful assessing what a pound of C2ISR is worth.
What about so many other Defense issues today?
Have we been using wrong analysis methods on these sets of problems, addressing
effects base operations, information operations, C2ISR, adaptive threats and
other defense issues? How would we begin to know if we are?
We might get a hint by asking, "Are the
answers we are getting solving our problems?" If the answer to that
question is "No," then maybe we are asking the wrong questions; maybe
the methodology is wrong for the problem.
Another
insight might come from asking, “Are we in control of the involved
processes?" Do we have control of
how the system might configure and how the system might react to stimulation,
or how the environment might in turn react to the system? If the answers to those questions are
"Yes," then it might be analytically appropriate to go ahead and
define a system representation, a model, and analyze its responses and claim
that response represents reality. If the
answer is “No,” then we might be misleading ourselves to prescribe the system
and thinking that we are learning
something about “reality” by analyzing it.
It
might be interesting, too, to wonder how much our ability to calculate, our
collection of algorithms, has conditioned our ability to perceive and build
models of the real world in the first place.
Does our model represent what is really happening, or is it only a view
filtered by our pre-developed sense of how things "should" work and
that already filtered by our sense of what we are able to calculate? Is there evidence to suggest that analysts
have suspected all along that closed system methods are not effective for many
issues we are asked to analyze? I think so.
We have
had an open system analysis method in use in the DoD for a long
time. It’s called Wargaming and it seems
to be being used more frequently and on a broader group of issues than
ever. In a classic command post exercise
we inject human decision making into a structured system, a simulated combat environment,
to generate open systems behaviors.
Human decision making represents energy
crossing the structured system boundary.
The analysis problem, of course, is that wargames are run in real-time
and are much too slow and expensive to gather more than one point on the
performance curve.
We
need more than wargaming in our kit bag of open systems analysis tools. We need sound understandings of how open
systems behave. We need to know how to represent open systems in our
computers. We need a calculus
(interesting to draw here on that Newtonian tainted term) that allows these
systems to exhibit emergent behavior and to adapt (a clearly non-Newtonian
process). We need to know what to
measure from them. And analysts will
need to take on new roles in problem solving different from their traditional
role which
was to observe the real world, define it as a system, calculate that system's
behavior, and then tell decision makers about that behavior in ways useful to
them.
Open
systems can adapt their design and an entirely new process can emerge quite
suddenly to satisfy the needs and goals of the components. One of the analyst's
new jobs is going to involve recognizing when a collection of components is
actually presenting a successful process, understanding how effective, robust
and enduring that process might be and how much we can expect to influence and
manage its evolution.
An
open systems analyst job may change to one involving: 1) Characterizing
components involved in a phenomenon (what they are, what they individually need
and produce); or, 2) Defining the goal that an assembly of them must
successfully accomplish. Said another
way, an analyst's new functions in open systems analysis concerns defining the
"Agents" and the "Test." The open system analyst predicts
nothing about the processes that might emerge among the Agents to succeed at
the Test. He defines a universe of
components, sets goals for them, and observes what emerges. Following that, the
analyst’s next big challenge in open systems analysis is to decide what can be
said about the emergent processes and how can practical
decisions be better informed from the
insights gained.
Evolving an open system
adaptive methodology:
Recently
we've seen the introduction of agent-based simulation to defense issues. Sometimes the agents are used as part of a
search algorithm to search databases and response surfaces. Mostly agent-based simulations represents
individual actors ¾ soldiers in small unit operations.
These are coming to be referred to as, "Dot Wars" models; red
and blue dots contesting with each other.
Dot Wars simulations demonstrate emergent behavior in the groups of dots
resulting from the simple rules provided for each dot's abilities and behavior.
Usually, each dot in the group is given the same set of rules. No rules or algorithms are provided for the
behavior of the group of dots. Yet, structured group behavior emerges, to be
discovered, from the rules guiding each of the dot actors. ISSAC, MANA, TRANSIM
appear to be examples of the agent-based simulation that exhibit
"emergent" behavior, but I suspect they may not yet be
"open" and "adaptive."
Although
they exhibit emergent behavior, these models are still closed systems in the
sense that no adaptation is going on within the processes. No energy is crossing the boundary of the
system of rules that prescribe the physical characteristics and behavior of the
Dot War agents. (Unless one stops the model, changes the rules, and runs it
again).
Dot Wars may be an important part of the
methodology needed to analyze open systems.
They may provide the "Test" for the agent components. But, to become an "open” system model,
change and adaptation must become a characteristic of the Red and Blue Dots.
One
technique that seems to demonstrate emergent behavior and adaptation involves
embedding genetic algorithms to breed increasingly effective changes in the
physical characteristics and decision rules of the Red and Blue agents. It also
includes increasing the variety of Red and Blue agents performing various
tasks. This technique may be seen in the "Counter Drug" model
development demonstration. The CD Model consists of Blue system agents,
representing search and interdiction air and maritime systems, and a Blue C2
agent to manage their mode of operation. These Blue agents "compete"
against drug running boats and aircraft agents
controlled by a Cartel C2 function that is learning from experience and making
similar decisions on how to adapt and better employ the drug running assets.
Red has economic factors to consider and tries to make money. Assemblies of Blue agents are trying to catch
the Red drug runners.
It
appears that the CD Model's combination of agent-based simulation and genetic
algorithms can develop tradeoffs between the quality of information, its flow,
and the decision rules for the employment of forces, verses investments in
physical numbers and performance characteristics of systems against an adaptive
adversary. The combined techniques of
agent-based simulation and genetic algorithms appear to enable the development
of configurations of ISR and operations assets (architectures?) as well as the
strategies for managing and evolving these to dominate an adaptive threat. Such open system methodologies may enable
measurable assessments of what a pound of C2ISR may be worth in various
contexts.
It
may be appropriate to consider that models like the CD Model may always
be at best just, "pseudo-open" models themselves.
"Pseudo-open" in the sense that the system boundary of the model may
have been expanded to an extremely larger, but still finite universe of
possible excursions and variations of the agents, their characteristics and
operating modes. One of the
questions that will confront future "Open Systems” analysts concerns the
equivalent of VV&A. How
"open" must the model be to be "representative?" "What does representative
mean?" It may be enough simply not
to be able to tell the difference between the unpredictable behavior of the
model and the unpredictable behavior of the real world ¾ as long as they are unpredictable in similar ways
and explainable after the fact. I'm
speculating here; just introducing the concept that matching detail in the
representation of the agents to the details of real process actors may not be
where Open System VV&A will be found.
The rules for closed systems modeling may not apply to modeling open
systems.
The CD
Model is a research test-bed. In coming
months research is planned to "calibrate" its agents to represent the
components of a real military operational process. This process may be the specifics of real
drug interdiction operations, Precision Strike or Time Sensitive Target
operations. The Test in the CD Model
will also be redesigned to examine assembly of effective, robust architectures
by the model. The goal is to use the
model to develop effective C2 linking and physical and decision management
characteristics for the sensors and platforms involved in the operation and
develop the rules for evolving and employing those components when confronting
an adaptive adversary. Another interesting
direction of research might involve developing strategies and means when Red
and Blue do not have purely opposed objectives.
The
advent of Open Systems analysis methodology does not mean that traditional
Operations Research tools will become obsolete.
It may, however suggest that the first task of analysts in the future
may be to determine if the problem is open or closed, and then to select
appropriate methods. If the processes in
question are open systems, then classic, closed system analysis methods may not
be very informative.
The
next few years are going to open new frontiers for analysts. They are already
on our doorstep. These frontiers may
provide dramatically new vistas on what is knowable and what are
"good" questions to ask about how the world works. They may radically change not only how we
analysts think about problems and processes, but also change how we use
computers to enlighten us. They may even
change the kinds of decisions that are made to influence our world. Clearly,
though, the demand on the analyst will be on new perspectives on the problems
and new methods for problem solving. The
seemingly vast and vaguely bounded field of Complexity Science offers a
foundation for understanding of open system process while techniques such as
agent-based simulation, genetic algorithms and others offer
potentially effective analysis methodology.