Communication Of Intent In Military

Command and Control Systems

Lawrence G. Shattuck
Department of Behavioral Sciences and Leadership
United States Military Academy
West Point, New York 10996

David D. Woods
Cognitive Systems Engineering Laboratory
The Ohio State University
1971 Neil Avenue
Columbus, Ohio 43210

In C. McCann and R. Pigeau (editors),

The Human in Command: Exploring the Modern Military Experience.
New York : Kluwer Academic/Plenum Publishers, 279-292, 2000.

Abstract
Military command and control (C2) is a type of distributed supervisory control system.
These systems are characterized by remote supervisors who work through multiple
local actors to control a dynamic process. The agents can be separated by both space
and time but still must coordinate their activities to achieve the goals of the system.
Coordination normally occurs through the use of predetermined plans and procedures.
However, these plans and procedures can be underspecified and brittle when a local
actor is confronted with an unanticipated situation. In these instances, the local actor
must adapt the plan in a manner consistent with the intent of the remote supervisor.
Remote supervisors guide the adaptation by imparting their presence to local actors
prior to controlling the process.

This research investigated the communication of intent in military C2 systems by using a

mixed-fidelity simulation. U.S. Army battalion commanders and subordinate company
commanders developed operations orders using their standard planning procedures.
Anomalies were developed that forced them to rely on the senior commander’s intent
rather than the written plan. Battalion and company commanders were asked to
determine what action they would take in response to the anomaly. The responses of
the battalion commanders were used to evaluate the responses of the company
commanders. Results of the simulation indicated that even though military C2 systems
have a specified method to communicate intent, company commanders were able to
match their battalion commander’s intent in only 34% of the thirty-two episodes.
Explanations for the results are discussed and methods to improve performance are
suggested.

1
1. Introduction

Military command and control (C2), air traffic control, and nuclear power plant
control rooms are examples of distributed supervisory control systems.
Distributed supervisory control systems are hierarchical and cooperative. They
include remote supervisors who work through intelligent local actors to control
some process. A remote supervisor typically provides plans and procedures to
multiple local actors. These plans and procedures often are inadequate to
cope with the unanticipated variability of local situations (Woods and Roth,
1988). As a result, local actors must adapt the plans and procedures to the
situation based on their understanding of the remote supervisor’s intent. The
research reported here investigated how remote supervisors (senior
commanders), separated by both time and space from local actors
(subordinate commanders), impart their presence by communicating intent to
coordinate and adapt underspecified plans and procedures.

Normally, plans and procedures are developed by designers before the system
is implemented or by remote supervisors who have an incomplete (or
inaccurate) view of the local situation. These plans and procedures often do
not aid local actors in their process control tasks if they follow them rotely. In a
field study of a human-intelligent machine system, Roth, Bennett, and Woods
(1987) learned that correct problem-solving paths deviated from the
predetermined plans and procedures in 78% of the cases. Deviations were
found to be the norm because domain experts and designers are unable to
anticipate all possible local conditions. The plans and procedures were
underspecified, requiring technicians to “supply knowledge and act outside of
the scope and direction” of the expert system (Roth, et. al., 1987).

Woods, O’Brien, and Hanes (1987) also found in their studies of nuclear power
plants that “good operations require more than rote rule following.” They
identified two types of errors that occurred when “events demanded a relatively
variable sequence of component actions and extensive feedback from the
environment in order to adapt to unpredictable constraints or disturbances.”

• “Type A problems where rote rule following persisted in the face of

changing circumstances that demanded adaptable responses.”

• “Type B problems where adaptation to unanticipated conditions was

attempted without the complete knowledge or guidance needed to
manage resources successfully to meet recovery goals.”

Local actors either failed to adapt plans and procedures to local conditions or
they adapted plans and procedures without considering the intent of the
remote supervisor.

2
Suchman (1987) supports the concept of local actors adapting plans and
procedures. She states that “instructions must be interpreted with respect to a
collection of actions and circumstances that they never fully specify.” Local
actors cannot blindly follow predetermined plans and procedures issued by
remote supervisors or by system designers without regard for local conditions.

Local actors need a framework or a context for adapting their plans and
procedures when responding to novel situations. They must understand the
supervisor’s intent underlying the plans and procedures. The dictionary
describes intent as “a purpose; object; aim” (Webster’s New World Dictionary,
1970). Another dictionary (Webster’s II New Riverside University Dictionary,
1994) describes intention as “an aim that guides an action.” In
communicating intent, the remote supervisor explains the goals of the system
(object or aim) and the reason for pursuing the goals (purpose). This
description provides a framework for adapting the existing plans and
procedures to respond to novel situations. Intent is the means by which the
remote supervisor imparts his or her presence to local actors. This sense of
presence assists local actors in responding the same way the supervisor
would if the supervisor were able to view the situation through the eyes of the
local actors.

There is an inherent tradeoff for the remote supervisor in establishing the

framework for adaptation. Supervisors must determine the latitude or flexibility
they will give actors to adapt plans and procedures to local situations.
Supervisors who establish centralized control inhibit local actors. These
supervisors show little regard for the impact of local situational factors. At the
other extreme are supervisors who give local actors complete autonomy. In the
latter case, remote supervisors are ‘out of the loop’ and the response across
multiple local actors may be coordinated and synchronized only by
coincidence. In fact, the system may actually move away from its goals or end
state. Somewhere between these extremes are systems in which local actors
are able to modify and implement procedures to achieve the supervisor’s intent
based on an analysis of local conditions and a sense of the supervisor’s
presence.

2. A Model Of Distributed Supervisory Control Systems

Military C2 systems, nuclear power plants, air traffic control networks, and
many industries exhibit some form of distributed supervisory control.
Although these domains are very different in their specific objectives, there are
distinct similarities in the underlying structures and functions used to achieve
their objectives or end states. In spite of their ubiquitous nature, little has been
written about the characteristics of these systems. Any model of distributed
supervisory control that could be applied across diverse domains must
address the following scheme:

3
 A remote supervisor uses a communications process to
provide local actors with plans and procedures and to impart
his/her presence. The degree of control established by the
remote supervisor influences the ability of the local actors to
adapt to unanticipated conditions based on the actors’
assessments of their local environments.

2.1 Remote Supervisor

Advances in communication and automation no longer require supervisors

and actors to be at the same location. Supervisors can be separated by both
space and time from their subordinates. Such technology affords greater span
of control but has its pitfalls. Supervisors who attempt to function as local
actors will quickly lose sight of high-level organizational goals. They cannot
continuously monitor all processes, filter the information, and determine the
appropriate course of action for each local actor. They must remain detached
from the details of the local actors’ environments so they can evaluate the
system’s progress relative to the high-level goals.

Prior to controlling a process, the remote supervisors and local actors may
engage in cooperative activities, including planning, training, and rehearsals.
During periods of low activity in controlling a process, the need to exchange
information among the supervisor and actors is minimal. When an
unanticipated, time-sensitive event occurs, a remote supervisor must rely on
local actors to provide him/her the information needed to assess the situation
and to formulate a plan at the very time that the actors are most busy. Woods
(1994) describes this ebb and flow of activity as characteristic of systems in
which cognitive activity is distributed across multiple agents. If the
supervisor’s intent is effectively communicated prior to an unanticipated event,
actors are more likely to make proper decisions with respect to overall system
goals. Supervisors will not become mired in one local situation and lose sight
of the system’s goals. Thus, during peak periods, actors are able to devote
more cognitive resources to controlling the local situation and less to the task
of updating the supervisor.

2.2 Local Actors

A remote supervisor and local actors have different perspectives of the

environment. Local actors operate at the sharp end (Reason, 1990) where they
interact directly with the process being controlled. They have privileged access
to a narrow portion of the system. Remote supervisors operate at the blunt
end and can only influence the control process indirectly through policies,
plans and procedures, and the allocation of resources.

4
Local actors have specific goals and are able to continuously monitor local
changes in system status. These local conditions, in turn, affect how the
actors respond to the plans and procedures of their remote supervisors. For
example, if the local actors cannot communicate with the remote supervisor,
actors must rely on the plans and procedures provided them. If local
situations are such that the plans are no longer valid, actors must modify and
implement the plans on their own. Even then, however, local actors cannot act
independently. They must reconcile their actions with their supervisor’s intent
and coordinate their activities with other local actors. Failure to coordinate
their activities with other agents could result in system failure.

2.3 Communications Process

The methods used to communicate between the remote supervisor and the
local actors include voice, text, and graphics. Exchanges can occur face to face
or electronically. Functions of the communications process include:

• Distributing plans and procedures;

• Specifying the degree of control;
• Imparting presence;
• Updating agents on status of system.

A remote supervisor will communicate plans and procedures to local actors

but the actor is not expected to implement them until some future point in time
when the system is most likely in a different state, one which might not have
been anticipated by the supervisor. (For example, at Time T, when the System
is in State S1, implement Procedure P. However, in Time T, the system is in
State S2 and Procedure P may not apply). The content of the communication
may consist of nothing more than the procedure itself, or it may contain
additional information, including system goals, rationale, constraints, and
other considerations (Klein, 1993). It is vital that the local actors understand
the intent so when local conditions are different than those envisioned by the
remote supervisor, the actors (if given the latitude) can take action that will
move the system closer to its goal state.

2.4 Plans and Procedures

Distributed supervisory control system functions are based on plans and

procedures (or guidelines). Plans and procedures are often developed by
designers in conjunction with domain experts long before a system is
operational. By their very nature, they are brittle. The plans and procedures
specify what actors should do and how they should do it for each state of the
system. However, designers and experts are not able to predict every situation,

5
system state, or set of interactions for which plans and procedures should be
developed.

Plans and procedures are also developed just prior to beginning (or even
during) the control process. These situation-specific plans and procedures
require supervisors to have an accurate view of the control process. Since
supervisors cannot directly observe the process, local actors must devote both
time and cognitive resources to updating the supervisors’ knowledge of the
local situation when time is critical and cognitive resources may be committed
to near-capacity.

2.5 Degree of Control

There is an inherent tradeoff for the remote supervisor in establishing the

framework for adaptation. Supervisors must determine the latitude or flexibility
they will give actors to adapt plans and procedures to local situations. Ashby
(1956) wrote that “only variety can destroy variety.” If in some game, D can
make 10 moves, but R can only make one move, then the variety in the
outcomes will be as large as the variety in D’s possible moves. If, however, R
can make two moves rather than one, then the variety of outcomes is reduced
by as much as one-half. The remote supervisor determines the number of
“moves” available to the local actors vis-à-vis the environment by the degree of
flexibility afforded to the actors.

Remote supervisors establish the degree of control they exert on local actors
through plans and procedures. Specific, detailed plans and procedures
establish centralized control and reduce the flexibility of actors to adapt to
unanticipated variety in the local environment. Generalized plans and
procedures increase the flexibility of local actors but decrease the likelihood
that there will be a coordinated response across the system.

3. Military C2 as a Type of Distributed Supervisory Control

In the domain of military command and control, senior commanders (remote

supervisors) and their staffs engage in a deliberate, self-paced decision-
making process to formulate a plan for future operations. The plan is then
communicated to subordinate commanders (local actors) for implementation.
In this domain, the system includes an intelligent, uncooperative, unpredictable
enemy that attempts to exploit weaknesses in the plans of the friendly forces.
Detailed battle plans do not remain viable much past the onset of hostilities.
When confronted with unanticipated events, subordinate commanders must
initiate an event-driven decision-making process. Their task is to identify and
achieve the senior commander’s higher-order objective even if the specified
mission can no longer be accomplished. Subordinate commanders must rely

6
on the senior commander’s statement of intent to bound the space of
acceptable solutions.

In most distributed supervisory control systems, intent is communicated

passively. It is implied in the procedures and difficult to discern. However, in
the domain of military command and control, senior commanders work
diligently to develop a statement of intent that is communicated to subordinate
commanders prior to starting a tactical operation. Such explicit statements of
intent ought to increase the likelihood that local actors will make decisions in
accordance with the goals of the remote supervisor. Favorable research
results in a military C2 system could be extended to improve local actor
compliance with remote supervisors in other domains.

The present study used active duty, infantry and armor battalions to
investigate how intent guides the event-driven decision-making process so
that the senior commander’s higher-order objectives are achieved. In a mixed-
fidelity simulation (see Figure 1), four army battalion commanders and their
staffs developed tactical operation orders based on

2 BN CDR’s
BN CDR Response
Anomaly
Presented
CO CDR’s
Response BN CDR
Plan & Intent Compared With Decides If
1 Communicated 4 Responses
BN CDR’s Match
Response And Intent

3
CO CDR’s
CO CDR Response
Anomaly
Presented

Figure 1. Sequence of events for tactical simulation.

a brigade (next higher echelon) operation order they were provided. The
battalion commanders and their staffs disseminated the orders, which
included statements of intent, to subordinate company commanders. These
company commanders (four per battalion) developed their own operation
orders and briefed them back to the battalion commanders.

An investigator then reviewed copies of the battalion and company operation

orders. Two types of anomalies were created for each battalion. In the first
anomaly, the unit was blocked from completing its specific mission but could
still achieve the higher-order objectives of the battalion commander. In the
second anomaly, the unit had unexpected success relative to the specific
mission and had to decide what to do next. In both cases, the intent statement

7
of the battalion commanders provided sufficient information to aid the
company commanders in responding to the anomalies.

The battalion commanders were presented with both anomalies. They were
asked how they expected the subordinate company commanders to respond to
the anomalies. Their answer became the basis for evaluating the responses of
their subordinate company commanders. The anomalies were then presented
to the company commanders. While being videotaped, the company
commanders used a think-aloud protocol as they considered how to respond
to the anomalies. The battalion commanders watched the video tapes and
judged the responses of their subordinates relative to their own responses.
Finally, former battalion commanders, serving as neutral observers, watched
the video tapes. The neutral observers provided valuable commentary on the
intent process within each of the battalions.

4. Results of Simulation

Four battalions, each with four company commanders were given two
anomalies, generating thirty-two problem-solving episodes. General
information extracted from the verbal protocols included the following:

• a summary of the company commander’s response.

• the judgment of the battalion commander on whether or not the
company commander’s response matched his intent.
• the basis for the company commander’s response.
• references by the company commander to coordinate his actions.
• references by the company commander to any element of the plan.
• length of the protocol.

Battalion commanders judged that the company commander’s responses

matched their intent in only seventeen of the of thirty-two episodes (53%). In
three episodes, however, the responses matched only by coincidence. The
company commanders chose to remain in place not because they understood
the battalion commander’s intent but because they misinterpreted the
information available to them. In three other episodes, although the battalion
commanders judged the decision of the company commanders to match their
own, they were, in fact, substantially different. Battalion commanders
considered them a match because the company commanders were “thinking
along the right lines.” If these six episodes are considered mismatches, then
the responses matched in only eleven of thirty-two episodes (34%).

During the detailed analysis of the think-aloud protocols, six distinct

categories of utterances were identified. These categories were used to parse
the protocols and to identify the type of information used by the company
commanders to respond to the anomalies. The six categories are:

8
 • Need for more information
• System status
• Reference to procedures
• Reference to intent
• Course of action
• Coordination activities

Company commanders who were successful in matching their battalion

commander’s intent initially determined the system status (i.e., the disposition
of friendly and enemy forces). They made specific references to procedures and
the intent statement in the battalion operation order. They also acknowledged
that they had to coordinate their activities with commanders of adjacent units
prior to taking any action.

Unsuccessful company commanders generally did not refer to the battalion

commander’s statement of intent. In addition, unsuccessful commanders
also exhibited several other behaviors. Some commanders exhibited flawed
domain knowledge. For example, one commander’s incorrect knowledge of
tactics and time/distance factors caused him to initiate a system response
that could not have been completed in the time available. A few commanders
had a low tolerance for situational uncertainty. They decided not to act until
they were given more information to reduce their uncertainty. In some
instances, commanders misassessed the information available to them. Even
though they were given information on the status of the environment (i.e.,
enemy units), they did not incorporate it into their mental model of the system.
Some commanders also exhibited a rigid adherence to procedures in spite of
situational information that indicated they were facing a novel, unanticipated
situation. When a major, unanticipated event occurred on an adjacent part of
the battlefield, these commanders would not deviate from their assigned
mission, even though the event jeopardized the higher-order goals of the
system.

5. Implications of Results for Distributed Supervisory Control Systems

Results of this study suggest several areas that can be explored to improve
communication of intent, and, therefore, the performance of distributed
supervisory control systems confronted with unanticipated variability.

5.1 Path vs. State Solutions

Rich and Knight (1991) believe that solutions to problems can be either paths
or states. Military doctrine suggests that intent should focus on state
solutions yet, in practice, solutions are often paths. The doctrine states that

9
intent is designed to “focus subordinates on what has to be accomplished in
order to achieve success, even when the plan and concept of the operation no
longer apply…” (FM 100-5, 1993). In implementing the doctrine, commanders
are taught that their intent statements ought to include these components:
purpose, method, and end state. In this study, the method sections were often
very specific. In a sense, they established a canonical path through a Problem
Space (see Figure 2). The method tells Local Actor 1 to proceed from Start
State to State 2, State 3, and, finally, the End State. An actor may progress
unimpeded through the Problem Space. Or, in the case of Local Actor 1, there
may be an obstacle that prevents the system from reaching the End State as
planned. However, the event (or breakdown) that blocked mission
accomplishment was not unexpected so there is a contingency plan that can
be implemented. So even though a breakdown occurs, Local Actor 1 follows a
pre-planned path through the Problem Space.

Canonical Path
Remote
Start End
Supervisor 2 3 4
State State
Level

Problem Space

t
ten
In
Expected
Breakdown

Local Implement
Start Plan End Unexpected
Actor State
2 3
State
Breakdown
Level 1. Recognize breakdown. Implement
2. Identify correct contingency Start Plan Solution End
2 3 Space
in plan. State State
3. Implement contingency. 1. Recognize breakdown.
2. Situation goes beyond the
Local Actor 1 literal plan.
3. Local actor cannot achieve
endstate without intent.

Local Actor 2

Figure 2. Path vs. State Solutions.

For Local Actor 2, however, the event that occurs between State 2 and State 3
is unanticipated. Very specific intent statements offer no help. There is no
alternative, pre-planned path through the Problem Space to the End State. In
order to move through the Problem Space to the End State, the intent
statement should illuminate and bound a Solution Space. Actors should be
free to move through the Solution Space based on their analysis of the local
situation. The intent, then, should guide the local actor’s adaptation process.

10
5.2 Ambiguity of Natural Language

The ambiguity of natural language also presents an obstacle to distributed

supervisory control systems when confronted with unanticipated variability.
Misunderstandings arise when remote supervisors and local actors do not
share a common understanding because they are separated by both space
and time. Suchman (1987) states that many expressions “rely upon their
situation for significance.” In describing distributed decision making systems,
Fischoff & Johnson (1990) state that “people need to translate their thoughts
into some language before those can be shared with others. Their ability to
use the language sets an upper limit on the system’s coordination of decision
making, as does the system’s procedures for information sharing.” They also
state “that terms will unwittingly be used differently” by agents in the system.

These difficulties are compounded in distributed supervisory control systems

because intent is communicated from remote supervisor to local actor at Time
1 and in Context A, but often is not interpreted and implemented until Time 2
and in Context B. Remote supervisors should make every attempt to ensure
that local actors clearly understand the intent. Embedding the intent in
hypothetical contexts is one method that can be used to check the
understanding of the local actors.

5.3 Flexibility vs. Synchronization

Highly-coupled, technologically-laden systems call for flexible yet

synchronized responses to anomalous, unanticipated situations. In
distributed supervisory control systems, a remote supervisor oversees multiple
actors, each controlling a local process. It is not enough for local actors to
implement the correct procedure. The procedure must be initiated at the correct
time and place relative to the actions of the other agents in the system.
Decisions made and implemented in isolation can interact in unexpected ways
and actually prevent the system from reaching its goals.

5.4 Imparting Presence

Remote supervisors have no other means to control a process except through

the actions of local actors. Plans and procedures are too brittle to effectively
guide the decision making processes of these actors. The plans and
procedures must be tailored to the local situation. A remote supervisor with
high-level system knowledge and a clear understanding of system goals would
effectively adapt the plans and procedures based on local conditions. Remote
supervisors, however, must remain in a position where they can retain a
comprehensive view of the entire system. Since they cannot be physically

11
present with the local actor, they need to impart their presence to guide the
adaptation of plans and procedures.

When remote supervisors impart their presence, they are equipping local actors
with the strategic-level goals, constraints, and tradeoffs of the system.
Presence empowers a local actor to make decisions similar to those that the
supervisor would make if the supervisor were at the actor’s location.
Communicating intent is the primary means by which remote supervisors
impart their presence.

6. Conclusion

Van Creveld (1985) states “Confronted with a task and having less information
available than is needed to perform that task, an organization may react in
either of two ways. One is to increase the information-processing capacity, the
other to design the organization, and indeed the task itself, in such a way as to
enable it to operate on the basis of less information. These approaches are
exhaustive; no others are conceivable.”

Those who design distributed supervisory control systems must choose one
of the two alternatives proposed by Van Creveld. Advocates of technology-
based solutions will undoubtedly seek to increase the information-processing
capacity of an organization. The logic is that if they increase the ability to
process information, they will reduce the amount of uncertainty. Less
uncertainty will result in better (and probably more centralized) control of the
system.

Uncertainty, however, is not so much a function of information-processing

capacity as it is the ability of the designer to predict and identify all conceivable
states of the system. But as stated earlier, highly-coupled, complex,
distributed systems will often confront remote supervisors and local actors
with unanticipated states. Designing distributed supervisory control systems
with the ability to function with less information requires that local actors have
a variety of responses at their disposal and the authority to implement them at
their discretion. The process of imparting supervisor presence through the
communication of intent is essential and worthy of further investigation.

7. References

Ashby, W. Ross (1956). An Introduction to Cybernetics. New York: John Wiley

& Sons.

Department of the Army (1993). FM 100-5 (Operations). Washington, DC:

Department of the Army.

12
Fischoff, B. and Johnson, S. (1990). The possibility of distributed decision
making. In Committee on Human Factors Commission on Behavioral &
Social Sciences & Education & Education National Research Council
Distributed Decision Making Report of a Workshop. Washington, DC:
National Academy Press

Klein, Gary A. (1993). Characteristics of commander’s intent statements. In

the Proceedings of the 1993 BRG Symposium on Command and
Control Research.

Reason, James. (1990). Human Error. New York: Cambridge University Press.

Rich, Elaine, and Knight, Kevin (1991). Artificial Intelligence. New York:
McGraw-Hill, Inc.

Roth, E.M., Bennett, K.B., and Woods, D.D. (1987). Human interaction with an
“intelligent” machine. International Journal of Man-Machine Studies, 27,
479-525.

Suchman, L. (1987). Plans and Situated Actions: The Problem of Human-

Machine Communication. Cambridge, England: Cambridge University
Press.

Van Creveld, Martin (1985). Command in War. Cambridge, MA: Harvard

University Press.

Webster’s New World Dictionary of the English Language (2nd Ed.). New York:
The World Publishing Company, 1970.

Webster’s II New Riverside University Dictionary. Boston, MA: Houghton Mifflin

Company, 1994.

Woods, D.D. (1994). Cognitive systems in context. In the Proceedings of the

1 6th Annual Conference of the Cognitive Science Society.

Woods, D.D. and Roth, E.M. (1988). Cognitive systems engineering. In M.

Helander (Ed.) Handbook of Human-Computer Interaction. North-
Holland: Elsevier Science Publishers B. V

13