perm filename SRI[D,LES] blob
sn#153833 filedate 1975-04-03 generic text, type T, neo UTF8
I INTRODUCTION
II CANDIDATE APPLICATIONS
A. Aircraft Records Analysis
1. Problem Statement
2. Problem Assessment
B. Economic/Political Models
1. Problem Statement
2. Problem Assessment
C. Interactive Scene Analysis for Cartography
1. Problem Statement
2. Problem Assessment
D. Maintenance Aids (Near Term)
1. Problem Statement
2. Problem Assessment
E. Maintenance Sensor-Requirements Analysis
1. Problem Statement
2. Problem Assessment
F. Manipulators for Dangerous Tasks
1. Problem Statement
2. Problem Assessment
G. Mapping and Modelling of Geographic Areas
1. Problem Statement
2. Problem Assessment
H. Missile Range Picture Analysis
1. Problem Statement
2. Problem Assessment
I. Political Modelling
1. Problem Statement
2. Problem Assessmnt
J. Remotely Piloted Vehicles
1. Problem Statement
2. Problem Assessment
K. Resource Exploration
1. Problem Statement
2. Problem Assessment
L. Seismic Analysis
1. Problem Statement
2. Problem Assessment
M. Ship Recognition
1. Problem Statement
2. Problem Assessment
N. Tactical Commander's Management Aide
1. Problem Statement
2. Problem Assessment
III CONCLUSIONS
I INTRODUCTION
Artificial Intellience has existed as a research enterprise
for approximately 15 years, during which time the scope and power of
its techniques have steadily increased. In harmony with this, there
is a growing conviction that artificial intelligence can be applied
to solve a large number of specific user-oriented tasks.
We consider here a number of such tasks that represent
potential applications of artificial intelligence. For each application,
we shall present a brief description of the task and offer a preliminary
assessment of its significance and difficulty. Where possible, we
also note the agency of the Federal Government that has an interest
in or provided us with information about a particular task. Because
we have not investigated all the tasks equally thoroughly, we also
indicate a "confidence factor" that reflects the depth to which we
have considered the task and, consequently, have confidence in the
statements and opinions set forth.
The results of our study are summarized in Table 1. The first
column lists, in alphabetical order, the various potential application
tasks. Columns 2 and 3 contain our assessment, on a scale of 1 to 10,
of the potential significance of the task to both the Department of
Defense and to our society at large. A task was judged to have high
significance (near 10) if it effected large savings or influenced
a broad spectrum of users. It was judged to have low significance
if it were narrow in scope and highly specialized.
Columns 4 and 5 contain our assessment, on a scale of 1 to 10,
of two aspects of the technical difficulty of performing the task.
(Great difficulty is rated at 10.) The two aspects reflect the degree
to which the task is ill-defined, and the degree of technical risk
that one assumes in attempting to reach a useful level of performance.
Columns 6 and 7 set forth our estimates of time and money
needed to achieve a laboratory prototype system. We have again used
a relative scale of 1 to 10 because we have not investigated each
task in sufficient depth to estimate the time and manpower requirements
accurately. Informally, we think of the maximum rating of 10 as
representing a (say) 5 - 7 year research program, and a minimum
rating of 1 as representing an immediate application for which a
prototype exists today.
The final column contains the confidence factor (on a scale
of 1 to 10) that represents a self-evaluation of our ability to make
statements about the application task.
Before proceeding with the spcific task descriptions, we should
emphasize that these applications are merely a small selection from the
very large class of potential applications of artificial intelligence.
They were selected on the basis of such factors as interest to the
defense community, availability of information, and interests of
individual SRI staff members.
II CANDIDATE APPLICATIONS
A. Aircraft Records Analysis
1. Problem Statement
The military maintains a vast number of records about
the history and status of its aircraft. These files include (1) a
configuration file describing the precise configuration of each
aircraft, together with ancillary information such as the manufacturers
of the various components, (2) a failure file that records every failed
part for every plane, (3) a maintenance file that records all repairs
and preventive maintenance on each plane, and (4) a "red book" that
contains past and forecast flight schedules. Similarly, a pilot file
is kept on each pilot to record his training, flights, pilot errors,
and so forth.
One use of this data is for the analysis of airplane
crashes. In general, the purpose of analysis is to discover any
common theme that links several crashes together. For example, a
common theme might be a failed 1/2 horsepower motor manufactured
by the same firm and installed on three different types of aircraft
that have been involved in crashes. A common theme might also involve
human error: for example, different maintenance mechanics, all of
whom were trained by the same instructor, may have made the same error
that has resulted in several crashes.
The artificial intelligence application would be to
devise a program capable of accessing the various files and analyzing
them to find links among a set of aircraft crashes.
2. Problem Assessment
This problem appears to be rather difficult because
it is only moderately well-defined, and because apparently no close
analogy to the required program already exists. In fieldable form, it
also requires an ability to manipulate very large data bases, a
problem only now beginning to receive serious attention. However, a
small version of the required system could probably be developed in
the medium-term future.
Interested agencies: all of the armed services, as
well as the civilian aviation industry and regulatory bodies.
B. Economic/Political Models
1. Problem Statement
Econometric models of a great many varieties are in
common use today. In contrast to this, useful political moels are
virtually non-existent. It would be desirable to overcome at least
part of the deficiency by merging an existing econometric model with
at least a rudimentary political model that dealt only with those
political consequences that affect the econometric model. In particular,
the political model would deal with those events that alter the assumptions
or data upon which the (typically analytic) econometric model is based.
2. Problem Assessment
The widespread interest in modelling "soft" processes
of a political, economic, or sociological nature make this problem as
important for its longer-term implications as for its immediate utility.
Although superficially a difficult problem, recent work
has indicated that useful levels of performance may be available in the
near term.
Interested agencies: The intelligence community would
be an immediate and direct beneficiary. However, if the technology
were developed it would be likely to diffuse to a much larger set of
users.
C. Interactive Scene Analysis for Cartography
1. Problem Statement
Both the military and civilian communities devote a
considerable amount of effort to processing aerial photographs; the
advent of satellites has increased this tendency. The purpose of the
processing may be purely cartographic--i.e., to make maps--or it may
be to obtain intelligence information about anything from troop movements
to the suitability of a site for locating a new factory.
One particularly time-consuming step in the cartographic
process is the tracing or outlining of cartographic features and land-use
categories on an orthographic photograph. For example, an operator
must today laboriously trace out roads, rivers, lakes, forested areas,
and so forth on an orthophoto displayed on a console. It would be far
more preferable if he needed to indicate a given feature only by
pointing at it with his cursor rather than by outlining it. A
computer system could then abstract the distinctive characteristics
of the particular feature, and would extrapolate from the single point
to the entire feature. Thus, for example, the computer could outline
a lake given a single point, thereby reducing operator time dramatically.
The computer would display its results before finalizing them, thereby
giving the operator an opportunity to correct any errors.
2. Problem Assessment
The enormous amount of pictorial data currently being
collected and analyzed make this problem an important one. Moreover,
initial success on the cartographic problem would be likely to lead
into more difficult areas of interactive photo-interpretation. The
technical difficulty of the particular problem we cited seems remarkably
low. The technical tools needed, insofar as we can determine, exist
today in several laboratories.
Interested agencies: Army Engineering Topographic
Agency, U.S. Geological Survey, other cartographic and photo-interpretation
agencies.
D. Maintenance Aids (Near Term)
1. Problem Statement
There is already a well-developed technology of applying
computer systems to troubleshooting mechanical equipment. Notable
examples include the ATE/ICE system for diagnosing malfunctions in
Jeep Engines [1] and the DEPOT/MAIDS system for depot-level
maintenance of diesel truck and tank engines [2]. These systems
use straightforward methods that do not involve any AI technology.
We think that the use of AI techniques could make
substantial contributions to the performance and efficiency of these
kinds of systems. In particular, achieving the following improvements
would seem worthy and could be added using existing AI technology:
a) More flexible man-machine communication involving
a greater use of natural language, spoken word
recognition, and voice output.
b) Ability for the technician to volunteer information
he believes relevant before the program asks for it.
c) Greater ease of modifying existing programs to
incorporate new troubleshooting procedures.
d) Addition of programs that can give advice about
repair procedures and advice about how to assemble
and disassemble the equipment.
2. Problem Assessment
The importance of the equipment maintenance problem
faced by DOD is sometimes underestimated merely because maintenance
is not a "glamorous" problem. Nevertheless, substantial savings as
well as improved performance will occur from increased maintenance
efficiency. While some progress has been made toward using computers
to help automate the process, much more can be done, and the AI
technology for doing so exists.
Interested agencies: U.S. Army Tank and Automative
Command (TACOM), U.S. Army Frankford Arsenal.
E. Maintenance Sensor-Requirements Analysis
1. Problem Statement
Skyrocketing maintenance costs throughout our society
are motivating an increasing interest in finding ways to automate the
maintenance process. One easily discernible trend has been to build
sensors into a particular device or system and to use the sensory
information they provide to direct the maintenance process. For
example, the aids mentioned in the previous section use sensors to assist
in vehicular maintenance. As these methods gain acceptance, there
will be an increasing need to identify, for a particular piece of
equipment or for a class of equipment, the combination of sensors
that are most cost-effective.
One approach to this problem would be to make use of
the emerging family of artificial intelligence systems for diagnosis
and trouble-shooting (for example, the SOPHIE system [3], MYCIN
system [4], and our own CBC system). Such programs could be used to
evaluate the utility of a candidate set of sensors by running in a
simulation mode. A large number of different faults could be
simulated, and the program would note the difficulty in pinpointing
each fault using the candidate sensor set. A large number of different
candidate sets could be evaluated in this manner.
2. Problem Assessment
The 60 billion dollar operations and material budget
of the Department of Defense is one measure of the importance of this
problem. The successes of the existing artificial intelligence diagnosis
systems suggest that the problem can be approached using current technology
as a base. As it stands, however, the problem is not very well defined,
leading to a corresponding difficulty in assessing technical risk
and resource requirements.
Interested agencies: Throughout the Department of
Defense; for example, the Navy Materiel Command. Successful technology
development would also have a broad impact in the automotive industry.
F. Manipulators for Dangerous Tasks
1. Problem Statement
Many modern processes involve handling radioactive or
toxic substances. In order to handle such substances safely,
remote handling devices, or teleoperators, have been developed. Although
the history of such devices goes back at least to the 1940's, the
current generation of in-service manipulators does not incorporate the
sophisticated control techniques that have been developed in several
of the leading artificial intelligence laboratories. One possible
application would be to merge a fully automatic manipulator, as developed
in these laboratories, with a supervisory level of human control in
order to produce a teleoperator with extended capabilities. A
laboratory prototype system of this sort has been demonstrated by
SRI; its purpose is to help dismantle highly toxic devices whose
use has been proscribed by Congress.
2. Problem Assessment
The near-term technical feasibility of extended
capability teleoperators does not appear to be in question. Similarly,
the ultimate need for such teleoperators of various agencies and
contractors of the Federal Government appears obvious. There is,
however, a question as to how widespread the need is for these devices
strictly within the Department of Defense.
Interested agencies: Edgewood Arsenal, Energy Research
and Development Agency.
G. Mapping and Modelling of Geographic Areas
1. Problem Statement
Computer modelling of geographic regions has been an
active enterprise for many years. In 1967, World Data Bank I was
assembled, consisting of approximately 100,000 points specifying basic
cartographic data (coastlines, international boundaries, and the
like). In 1973, World Data Bank II came into existence, comprising
over one million points specifying coastlines, international boundaries,
major drainages, and so forth. (World Data Bank I is available to
the public; II will be available in the not too distant future.)
Current software can access a world data bank in
several different ways under user control. For example, it can
"window" the data, and display it under any of several standard
cartographic projections. In addition, various other data (like
thematic overlays delineating, for example, land use) and software
(like conventional correlation algorithms) can in principle be used
in conjunction with the basic cartographic data. It would be of
further interest to add symbolic models and procedures to this
conventional software. For example, the economic/political models
discussed previously could be expanded to make explicit use of basic
cartographic and thematic data--to compute, say, the vulnerability
of certain supply routes to geopolitical changes.
2. Problem Assessment
A major feature of interest in this problem is the
demand it makes on combining several disparate source of data and
knowledge. It is a member of a class of such problems whose hallmark,
when approached "manually", is the segregation of knowledge into
specialties that often do not communicate freely with one another.
Consequently, an artificial intelligence program able to deal with
these disparate knowledge sources has the potential of
achieving better performance that a single human expert; moreover,
it is likely to result in increased communication among the human
experts in the various specialties. In the short term, the latter
property is likely to be the more important one.
As the situation currently stands, the problem is
ill-defined.
Interested agencies: the intelligence community would
be an initial user of a system aimed at the stated problem. If the
technology were developed, however, it would almost certainly diffuse
to a much wider set of users.
H. Missile Range Picture Analysis
[THIS SECTION WILL BE UPDATED WITH A CONSIDERABLY BROADER VIEW
OF THE TASK.]
1. Problem Statement
The White Sands Missile Range employs several high-speed
motion picture cameras to photograph the flight of each missile tested.
From these photographs, the position and attitude of the missile, as a
function of time, can be calculated. Two years ago, an automatic
film scanning system was purchased by White Sands. An algorithm for
tracking the missile was developed by an external consultant. This
algorithm is simple and computationally fast and, accoding to the
consultant, reasonably accurate. However, it does not make use
of all the information poentially available in each picture frame. The
challenge is to improve the algorithm, in accuracy, speed, or both.
2. Problem Assessment
The current algorithm is sufficiently simple to make
it likely that a modest improvement in accuracy is possible. The
simplicity of the current algorithm makes it unlikely that a more
accurate algorithm would also be faster. Accordingly, we rate the
technical risk as high, in that it appears unlikely that additional
development will produce a dramatically improved algorithm.
The problem is a very specialized one, having interest
and implications only for White Sands.
Interested agency: White Sands Missile Range.
I. Political Modelling
1. Problem Statement
Computer-based political models of national scope do
not, to our knowledge, currently exist. The intelligence community
currently employs political analysts who "model", in poorly understood
ways, a geographical area. It would be of interest first to understand
the form of model(s) used by the analyst, and then to encode the
model(s) for computer implementation. Models of this sort could
serve as an interactive aide for an analyst, could provide backup,
and could serve as a training aide for new analysts.
2. Problem Assessment
This problem represents an extremely interesting
research project for artificial intelligence, but is both ill-defined
and likely to involve considerable technical risk. If successful,
however, the technology developed would be likely to have broad
implications for a variety of applications areas.
Interested agencies: Initially, the intelligence
community.
J. Remotely Piloted Vehicles
1. Problem Statement
There are a number of existing or potential uses
for remotely piloted vehicles (RPV's) within both the defense and
non-defense spheres of interest. The most prominent existing application
of RPV's is for airborne vehicles of several varieties. Perhaps
less obvious is the potential use of RPV's in such underwater applications
as undersea recovery or deep sea drilling. The hallmark of these
applications is the requirement to maneuver and perform tasks under
the local control of a combination of visual, sonic, or other sensors.
Because artificial intelligence has been concerned with problems of
this sort, it appears plausible that it may have techniques to contribute
to the control of RPV's.
2. Problem Assessment
The problem as it stands is ill-defined. Most
airborne applications are highly classified, while undersea applications
have not, to our knowledge, been seriously discussed. Accordingly, it
is difficult to predict the extent to which relatively straightforward
extensions and applications of artificial intelligence techniques can
make a contribution.
K. Resource Exploration
1. Problem Statement
The process of exploring for mineral resources is
a lengthy and involved one whose success depends critically on the
knowledge of the field geologist actually surveying a geological
province. If he has close familiarity with all the varieties of
ore bodies that may be encountered in the province, he is far more
likely both to observe relevant characteristics and to interpret
correctly the characteristics that are noted. This, in turn,
dramatically increases the likelihood of discovering significant ore
bodies.
Unfortunately, it is unrealistic to expect one
geologist, or even a small group of geologists, to possess this breadth
and depth of knowledge. Instead, as in other professions, a geologist
specializes much more narrowly. As a result, there are many documented
cases of ore bodies that remained undiscovered after "thorough"
exploration, only to be discovered subsequently by a geologist whose
specialization happily matched the physical situation. Of course,
there is no way of accurately estimating the number of ore
bodies that have been missed by exploration programs and that remain
undiscovered, but it is commonly agreed that the number of such ore
bodies is significant.
The methodology for discovering ore bodies is closely
analogous to the process of diagnosing diseases. In each case, there
is an underlying state of physical reality whose identity the
practitioner seeks to establish through a series of indirect observations
and tests. Typically, none of the tests are definitive, so the
practitioner must in principle consider several alternative models
for the underlying state and then seek to confirm at least one of them.
The success of the MYCIN [4] system for medical diagnosis, and the
emergence of roughly analogous newer programs for diagnosing equipment
malfunctions, suggests that a geological consultant to aid in the
exploration process is feasible.
2. Problem Assessment
The fundamental significance of the problem, measured
against the background of a world in which many mineral resources are
in short supply, is unquestionably very high. The preoccupation
of the Department of Defense with this situation is reflected in the
number of current studies dealing with present and projected shortages
of critical materials.
The problem appears to be readily approachable from the
current base of artificial intelligence technology.
Interested agencies: Department of Defense, Bureau
of Mines, U.S. Geological Survey.
L. Seismic Analysis
1. Problem Statement
A problem of long-standing interest has been the
detection of underground nuclear explosions through the analysis of
seismic data, and a number of statistically oriented waveform
analysis programs have been developed. One approach to improving
the performance of such programs is to augment the traditional signal
processing programs with artificial intellience programs capable of
dealing with symbolic data. For example, intelligence reports indicating
a high level of activity at a known test site might be used to tune
the signal processing programs for increased sensitivity to that
particular geographic location. Current developments in sonar processing,
notably the HASP program, provide a model for an application program
aimed at this problem.
2. Problem Assessment
The problem as it stands is ill-defined; an initial
effort would be needed to define the current state of seismic signal
processing programs in order to assess the likelihood that artificial
intelligence techniques would contribute to their performance.
The problem is a specialized one but, if successful,
could have implications for radar signal processing. One important
feature of the seismic problem is the absence of a requirement for
real-time response.
Interested agency: U.S. Arms Control and Disarmament
Agency, U.S. Geological Survey.
M. Ship Recognition
1. Problem Statement
A recurring task for crew members on submarines is to
identify the class of a surface ship from its image in the periscope.
Remarkably, the variety of ship classes is so great that even experienced
sailors can identify by sight only a small fraction of them. Current
procedure calls for two cooperating sailors to do the job. One views
the surface ship through the periscope and calls out features (e.g.,
"two stacks"), while the other searches through a mass of pictorial
data (perhaps silhouettes) in an effort to match the description with
a particular picture. It may be possible to automate at least part
of this procedure through the application of artificial intelligence
techniques.
2. Problem Assessment
The problem is a very specialized one whose technical
success will depend heavily on a judicious task specification. For
example, a fully automatic procedure, working directly from (say) a
video image of the periscope scene, is almost certainly too difficult
for a near-term application project.
Interested agency: U.S. Navy
N. Tactical Commander's Management Aide
1. Problem Statement
The tactical commander of, say, an aircraft squadron,
performs many typical management functions in the course of his daily
operations. He is concerned with such items as degree of combat
readiness of his aircraft, supply of fuel, spares, and other material,
the state of readiness of his personnel, past and projected flight
schedules, and so forth. The artificial intelligence community has,
within the past year or so, shown an increasing interest in developing
methods for helping management personnel deal with issues of this sort.
The suggested methods typically deal with functions like file management
aids, automatic alerting mechanisms, and message routing mechanisms,
all driven by models of both the individual and his organizational
environment. In this climate, it should be possible to develop
management aids specifically aimed at the needs of the tactical
commander.
2. Problem Assessment
The problem is of broad significance, with potential
impact on many levels of management throughout the Department of Defense.
It seems very likely that some useful functional aids could be supplied
(on a laboratory demonstration basis) in the near term. Indeed, technical
uncertainties center more on how much can be achieved how fast, rather
than on the question of whether anything useful can be achieved.
Of much greater uncertainty is the degree of organizational
problems that would attend any attempt to introduce this technology on
a wide scale.
Interested agencies: Initially, agencies and services
throughout the Department of Defense. Ultimately, agencies and organizations
throughout both the public and private sector.
III CONCLUSIONS
We have undertaken here to provide some initial suggestions
and evaluations of potential applications projects for artificial
intelligence. Although, to reiterate, the applications we have
considered are only a small sampling from the set of all possible
applications, they probably represent a fair cross-section nonetheless.
Some of the systems considered emphasize what has been called
"pure thinking." They accept input only from a conventional computer
terminal, perform some deductions or other "intellectual" activity,
and output the results to a conventional terminal. Other systems
involve "doing" as much as "thinking." They may accept input from a
variety of sensors as well as from a conventional terminal, and their
output may control real-time devices. Some of the systems are aimed
at providing highly interactive aids to a user; others may be envisioned
as operating in a more nearly stand-alone mode.
If we assume that our assessments, as summarized in Table 1,
are accurate enough to form a basis for discussion, then several of
the tasks stand out as being ripe for early development. We can say
with some confidence that Interactive Scene Analysis for Cartography
(Row C) has considerable significance for the Department of Defense,
has low technical difficulty, and can be pursued with a quite modest
program. An application of artificial intelligence to maintenance
problems (as exemplified by Rows D and E) also offers the possibility
of significant payoff for moderate levels of risk and resource.
Finally, we can say with considerable confidence that a program aimed
at Resource Exploration (Row K) has the potential of being of very
great relevance, but involves somewhat greater difficulty and would incur
a higher cost.
Certain other program areas appear to be clearly unsuitable
for current exploitation, either because the potential payoff is too
small or because the risk is too great. Missile range instrumentation
(Row H) and Ship Recognition (Row M) both have only a small potential
payoff, especially when measured against their difficulty. Remotely
Piloted Vehicle Control (Row J) has a moderately high potential payoff,
but appears to involve high levels of both difficulty and cost.
The Tactical Commander's Management Aid (Row N) is an
interesting possibility presenting the strongest contrasts. It has
very great potential significance, is of medium technical difficulty,
but is likely to be costly to undertake and may have significant
organizational problems associated with it.
An analysis of the set of tasks suggests several interesting
taxonomies for application tasks generally and for these applications
in particular. One taxonomy would be based on the underlying technology
needed to support a class of applications. To illustrate this, let us
consider four general areas of current artificial intelligence research
or interest: Symbolic Models, Large Files, Perception, and Common-Sense
(or Plausible) Reasoning. Techniques emerging from these areas of research
are needed to support the various tasks as follows:
SYMBOLIC MODELS
Aircraft Records Analysis
Economic/Political Models
Maintenance Aids (Near Term)
Mapping and Modelling of Geographic Areas
Political Modelling
Remotely Piloted Vehicles
Resource Exploration
Seismic Analysis
Tactical Commander's Management Aide
LARGE FILES
Aircraft Records Analysis
Economic/Political Models
Mapping and Modelling of Geographic Areas
Political Modelling
Ship Recognition
Tactical Commander's Management Aide
COMMON-SENSE REASONING
Aircraft Records Analysis
Economic/Political Models
Mapping and Modelling of Geographic Areas
Political Modelling
Resource Exploration
Seismic Analysis
Ship Recognition
Tactical Commander's Management Aide
PERCEPTION
Interactive Scene Analysis for Cartography
Maintenance Sensor-Requirement Analysis
Resource Exploration
Ship Recognition
A second useful organization for the set of tasks groups tasks
together according to their "synergistic possibilities;" that is,
according to the likelihood that a successful implementation of one
application task could be incorporated in another application task to
improve performance and/or capability. Here we have many possibilities,
and cite only a few for illustrative purposes.
MAINTENANCE
Near term maintenance aide, maintenance sensor requirement study.
INTERACTIVE COMPUTER GEOGRAPHY
Interactive Scene Analysis for Cartography
Mapping and Modelling of Geographic Areas
If we consider interactive computer geography as a new "application
task," then we would include this task itself in a new group:
PLAUSIBLE DEDUCTION FROM (PARTIALLY) GEOGRAPHIC DATA
Interactive Computer Geography
Resource Exploration
Seismic Analysis
Program
Significance Difficulty Requirements
Level
Non- Ill- Technical Time of
TASK DoD DoD Defined Risk Scale Effort Confidence
--------------------------------------------------------------------------------
A. Aircraft Records
Analysis 5 6 7 7 7 5 4
B. Economic/Political
Models 4 8 2 4 3 4 10
C. Interactive Scene
Analysis for
Cartography 7 3 2 2 2 3 8
D. Maintenance Aids
(Near Term) 8 7 4 3 4 4 8
E. Maintenance
Sensor-Requirements
Analysis 7 5 7 5 6 5 4
F. Manipulators for
Dangerous Tasks 3 6 3 2 2 4 8
G. Mapping and
Modelling of
Geographic Areas 7 5 5 6 7 4 7
H. Missile Range
Picture Analysis 2 1 1 7 3 3 7
I. Political Modelling 7 5 7 9 9 6 8
J. Remotely Piloted
Vehicles 6 5 7 8 9 9 5
K. Resource
Exploration 7 9 2 3 5 5 9
L. Seismic Analysis 3 5 6 6 5 6 1
M. Ship Recognition 3 1 4 7 6 3 3
N. Tactical Commander's
Management Aide 9 9 5 4 5 8 3