CIA/ORD INTELLIGENT SYSTEMS PROGRAM
Document Type:
Collection:
Document Number (FOIA) /ESDN (CREST):
CIA-RDP95-00972R000100020003-6
Release Decision:
RIPPUB
Original Classification:
S
Document Page Count:
17
Document Creation Date:
December 27, 2016
Document Release Date:
December 14, 2012
Sequence Number:
3
Case Number:
Publication Date:
February 4, 1982
Content Type:
MISC
File:
Attachment | Size |
---|---|
CIA-RDP95-00972R000100020003-6.pdf | 1.1 MB |
Body:
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
I MEMORANDUM FOR. I
(AJJZ
)k 0 A,4, Idx
~RQ 1/ ll.?~ RJR- 1-` fq' ' QF- b .f
14 I (t:3) &tat
-b -)/ 0 bP 04--f- 5jo e
pie c,c. , t~1~ ,Ae.t ac ud,~ 1~,,a frr
u b y (piti) p tee
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
STAT
STAT
STAT
STAT
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
CIA/ORD INTELLIGENT SYSTEMS PROGRAM
ORD is planning to establish a major new program of research
and development in the field of intelligent systems. Intelligent
systems are "smart" computer programs which can perform logical
deductions, solve complex problems, and explain their actions to
the systems' users. ORD has been tracking this technology for
several years. Recent significant successful applications in
universities, government, and industry indicate that this tech-
nology has now matured to a level sufficient to warrant a sub-
stantial Agency investment. We believe that the techniques of
intelligent systems will benefit the Agency through
? Improving the productivity of Agency personnel by performing
certain functions which previously required human intervention
? Making computer systems easier to use and understand
? Improving our ability to write and maintain computer programs
to perform complex tasks
? Institutionalizing and making more widely available the
specific knowledge and expertise of Agency experts in many
fields.
In general, intelligent systems technology can help us in the
Agency to control the cost of meeting our customers' projected
need for more intelligence and better information.
The technology of intelligent systems involves the creation
of computerized knowledge bases of factual data, problem-solving
heuristics, and logical inferencing mechanisms which are specific
to a particular problem domain. Potential Agency applications of
this technology include directed search through massive data
files, diagnosis of system malfunctions, user-friendly interfaces
to complex computer models or networks, image understanding,
computer-aided design of micro-electronic circuitry, understanding
of natural language text, and many more. The ORD Intelligent
Systems Program will investigate both the technology and its
application. An important aspect of this program will be the
development of an understanding of which aspects of the technology
are appropriate for which types of Agency applications.
The primary purpose of the proposed program is to transfer
intelligent systems technology to the Agency. In seeking this
goal the program has these general objectives:
? To provide continued technical support for initial Agency
experimental efforts in developing intelligent systems
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
o To introduce and demonstrate the value of intelligent
systems technology to a wide range of Agency analytical,
scientific, and system development functions
o To acquire and develop tools and skills within the Agency to
apply intelligent systems to the Agency's information problems.
Hence, the Intelligent Systems Program will concurrently address
tool development, actual applications, and understanding the
theory behind the intelligent systems/knowledge-based approach.
We believe that widespread use of intelligent systems
technology in the Agency is inevitable. However, the cost-
effective introduction of this technology to the Agency will
require a program of careful experimentation, analysis, and
organizational learning regarding the capabilities, character-
istics, and use of this type of computer system. The Agency
needs to begin now to expand its capacity to absorb this tech-
nology and establish in-house expertise and experience in intel-
ligent systems design, development, and use. The proposed
program is designed to provide the Agency with the experience we
will need as we face the information (and knowledge) processing
environment of the Eighties.
Our DDI and DDA counterparts have expressed their interest
in the Agency pursuing this technology. We are continuing to work
closely with them in the development of this program.
FUNDS: FY84 FY85 FY86 FY87 FY88
$1.20M $1.75M $2.OOM $1.50M $1.50M
Additional Staffing Required: +3 positions
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
ba~l
1. PROGRAM TITLE: Intelligent Systems
Submitting Agency: CIA
2. FUNDS: FY84 FY85 FY86 FY87 FY88
$1.20M $1.75M $2.OOM $1.50M $1.50M
3. PROGRAM SUMMARY:
The ORD Intelligent Systems Program is designed to provide
the research, development, and technology transfer required to
acquire, implement, and effectively use intelligent systems
throughout the Agency. Intelligent systems are advanced computer-
based information processing systems (hardware and software)
which can solve complex problems and perform logical deductions
from data and rules provided to them by human experts. These
capabilities are typically beyond the current capabilities of
today's data processing-oriented systems. The multi-year Intel-
ligent Systems Program will involve efforts in the areas of
application development, product evaluation, tools acquisition,
and the general process of learning how to build and employ
intelligent systems.
4.1 Program Overview
The intelligent systems approach appears to be a major step
forward in making the computer easier to use. By raising the
level of the interface through which the end-user communicates
with the computer we can allow analysts and other personnel to
take advantage of the power of the computer without requiring
them to become a computer "expert." The technology of intelligent
systems is expected to provide the Agency with some new techniques
by which we can
Improve personal productivity
Make computer systems easier to use
Perform procedures which were previously difficult or costly
to express in traditional computer languages
Solve some analytical problems that were previously untractable
Improve our ability to build and maintain computer software
Increase the general accessibility of the knowledge accumulated
by the Agency's substantive experts in particular subject
areas.
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
A principle goal of the development and use of intelligent systems
in the Agency will be to increase the use of computers while
holding down the costs of system implementation and maintenance.
Information and the knowledge of how to use it are an
important and valuable resource of the Central Intelligence
Agency. Growing requirements for intelligence production demand
that our analysts be able to process, sometimes quite rapidly,
considerable amounts of information in a variety of forms. To
meet these requirements the Agency is making a substantial invest-
ment in automated tools for information handling. Such tools,
however, are not always easy to use. Learning vast amounts of
computer jargon and protocols can be a confusing and frustrating
experience for the analyst and may result in lower productivity.
Improved techniques are needed for developing "smarter" system
interfaces which will make our information systems easier to use
by shifting the burden of communication as much as possible from
the end-user to the system itself.
In addition to support for handling vast quantities of
information, Agency personnel also need assistance in actually
processing and analyzing this information. The process whereby
raw data is converted into finished intelligence is often complex
and usually involves extensive knowledge and sophisticated skills
and expertise on the part of our analysts. The development of
this knowledge and expertise is a costly, time-consuming, and
labor-intensive process. Our analysts' knowledge requires con-
tinual update and revision as world events change, as new intel-
ligence requirements develop, and as new information systems are
constructed to provide automated support for handling information.
Techniques are needed to enhance the flexibility of our informa-
tion processing systems, to reduce the overall cost of retraining
Agency analysts as the environment changes, to help retain'specific
subject-area expertise which may otherwise be lost when Agency
personnel retire or transfer, and to provide more sophisticated
aids for processing increasing amounts of complex data and fusing
that information into meaningful and accurate intelligence. One
approach to solving these problems is the development of auto-
mated systems which can be programmed and modified easily to
assist humans in evaluating data and making decisions in a changing
environment. We call such automated systems "intelligent systems."
Intelligent systems are computer-based systems which answer
questions, perform relatively complex reasoning, manipulate
symbols as well as numbers, apply formal and heuristic logic,
explain their behavior, and in general solve procedurally-oriented
problems with an ability comparable to that of a human expert.
The technology is an outgrowth of a branch of Computer Science
called Artificial Intelligence which has been studying the
L
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
problem of developing more intelligent behavior in computer
systems for more than 25 years. Over the past decade the Federal
Government (largely through the Department of Defense) has spent
hundreds of millions of dollars in support of research and develop-
ment to provide techniques for building more intelligent systems.
More recently, private industry has become interested in the
intelligent systems approach and several commercial applications
have been developed.
ORD has been tracking this technology for many years. After
some early feasibility studies and prototype system development
efforts in 1980-81, ORD has concluded that intelligent systems
technology offers significant improvements in the application of
computer technology to the Agency's problems. The time appears
to be right to invest significantly in transferring this tech-
nology to the Agency. The multi-year ORD Intelligent Systems
Program will follow a three-part strategy of general tool and
product assessment, specific Agency application development, and
the training of Agency personnel in the techniques of "knowledge
engineering." The Agency is expected to improve its intelligence
production through this program by providing its analysts with
the ability to cope more effectively with the increasing volumes
of information and processing requirements which are projected
for the Eighties.
4.2 Program Objectives
The proposed program is intended to provide the planning
mechanism for the orderly acquisition and use of intelligent
systems technology within the Agency. The products of this
technology program are expected to have a significant impact on
many of the Agency's information processing activities from
collection to analysis and reporting as well as the way future
information systems are designed and implemented. In addition,
we expect to see intelligent systems technology become more
prevalent in the tools and systems which the Agency buys. As the
capabilities and commercial availability of intelligent system
products grow, the internal demand for such services will also
grow. The goal of ORD's initiative in this area is to prepare
the Agency (system developers as well as end users) to absorb and
apply these technologies effectively.
The objectives of this program are:
To provide continued technical support for initial Agency
experimental efforts in developing intelligent systems
To introduce and demonstrate the value of intelligent
systems technology to a wide range of Agency analytical,
scientific, and system development functions
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
To acquire and develop tools and skills within the Agency to
apply intelligent systems to the Agency's information problems.
To accommodate these objectives the program has been designed to
encompass continued support of ongoing applications, identification
and initiation of new systems employing these techniques, generali-
zation of tools and approaches developed for specific applications,
and the training of Agency personnel in how to build and use such
systems.
4.3 Intelligent Systems Technology
The aim of intelligent systems technology is to develop
"smarter" computer programs, i.e., programs which can communicate
on a higher level with end-users and which possess considerable
relevant knowledge about specific subject areas. These programs
perform functions which previously required human expertise to
accomplish and generally were considered beyond the capabilities
of ordinary computer programs. Whereas more typical applications
of computers might involve information storage, retrieval,
processing, and communication, intelligent systems are programmed
to employ additional knowledge about the information being
processed and use this knowledge to reduce the search time
required to find relevant information, to make logical inferences
using uncertain data, to plan strategies for achieving some goal,
or to converse with end-users in "natural" language rich with
their own domain dependent jargon and meaningful operations. The
domain specific knowledge and the skills to apply
it
in intelligent
systems comprise what is typically referred to as
the
knowledge
base, and intelligent systems which are structured
in
this manner
are called knowledge-based systems.
The knowledge-based systems approach should be considered
when a computer program is needed but there is not enough in-
formation about the problem and how to solve it to write a con-
ventional software system specification. In general, conven-
tional computer programs require that an algorithm be known for
solving a given problem (e.g., a mathematical formula, a closed
procedure that can always arrive at a yes/no decision, or a well-
defined series of bit manipulations that always results in the
right things being done to the right data.) Knowledge-based
systems, however, allow computers to go beyond algorithmic
solutions and to address problems where solutions (right or
wrong) are not always guaranteed. Such systems typically employ
heuristics or rules-of-thumb to select strategies for determining
what to do next in trying to solve a particular problem. These
heuristics are most often obtained by system designers from human
(domain) experts (e.g., image analysts, agronomists, engineers)
in a series of interview sessions that function much the way
on-the-job training of new professionals might be conducted. The
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Ed~.!aL'~aimd
knowledge of these heuristics and the skill to apply them
effectively (usually acquired through years of experience) are
what make the individual an expert in his or her particular field
and what give the intelligent system its power.
Knowledge-based systems offer promise in problem situations
where there are people who, given enough time, can solve the
problem. Knowledge-based systems do not possess the ability to
solve problems that humans cannot. Neither can they solve
problems for which humans cannot articulate how the problem
should be solved. However, they can be used to imitate the
behavior of scarce human experts in trying various approaches to
a problem, offering advice on various aspects to consider or
experiments to conduct, or recognizing blind alleys and suggesting
alternative strategies. Typically, the body of knowledge compiled
in a knowledge-based system represents the combined expertise of
several experienced people in a field and hence can serve as a
useful consultant to the professional by considering multiple
points of view.
Knowledge-based systems have been applied successfully outside
the Agency.in a number of areas including medical diagnosis,
mineral exploration, design and configuration of computers,
analysis of chemical structures, and molecular genetics. Current
areas of investigation include subjects closer to intelligence
production such as automatic image analysis, troubleshooting
computer systems, and the management of SIGINT collection. The
development of a new substantial knowledge-based system is a
medium- to large-scale undertaking involving, perhaps, two to
ten person-years divided between the system developers (knowledge
engineers) and the domain experts (scientists and analysts) who
contribute their knowledge. Smaller applications (involving
tasks a human can learn in a month or so) could be operational in
a year or less.
Each new application of knowledge-based systems technology
does not have to start from scratch. New applications can some-
times be developed from existing knowledge-based systems by
supplying new knowledge (facts and rules of behavior). Further-
more, general purpose knowledge base management systems are being
developed which provide basic inferencing tools. New applications,
then, require the development of the new knowledge base and
problem-solving strategies appropriate to the particular subject
domain. In addition, tools and aids for knowledge engineers are
also being developed which will facilitate the elicitation of
knowledge from the experts and the design of the knowledge base.
Knowledge-based system development techniques differ signi-
ficantly from conventional software methods. The development of
intelligent systems is not preceded by writing monolithic
, YR
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
specification and design documents. Instead, knowledge engineers
start with extensive interviews with human experts. The knowledge
gleaned from these interviews is transcribed into a partial
prototype of the final system and presented back to the expert
for review and suggestions for modification. Because intelligent
systems are typically designed and coded in high-level languages
with sophisticated automated development tools, the original
functioning prototype can be developed rather quickly and sub-
stantial alterations to the design and the code can be made in a
relatively short period of time. In this manner, the expert and
other end-users become intimately involved with the system as it
develops, understand how it works, and insure that the system
finally produced addresses the problem as it has evolved (not as
it was when the project was originated).
4.4 Potential Agency Applications
There are a number of potential application areas where
intelligent systems may be useful in the Agency. These areas
include
Large-scale simulation and modeling aids
Intelligent assistance for routine processing
Focusing searches through massive data files
Diagnosis of system failures
Automated aids for image understanding
Computer-aided design of micro-electronic circuitry
Improved techniques for understanding natural language
Enhanced interfaces to data base management systems
Computer-aided instruction
We expect that as the Agency gains experience with intelligent
systems technology additional applications will readily be
identified.'
4.4.1 Simulation and Modeling
One of the most important Agency application areas for
intelligent systems is in the development and use of complex
analysis models. Initially, we believe that models such as those
dealing with military assessment (e.g., M'IVS, PRIME, Air Defense)
or resource assessment (e.g., Soviet Economy, Chinese Agricultural
Assessment, Energy Resources Estimation) have the highest potential
payoff. Such models are often large and complex and may require
5-8 people to operate and maintain and as much as one year to
learn to use. Intelligent system approaches can be used to
construct more user-friendly, self-tutorial systems which assist
the user in specifying parameters and interpreting results.
These systems are designed to be more easily modified with their
knowledge explicitly represented in a readily comprehensible form
rr. P.,) T
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
to accommodate changes in model behavior. The Rule-Oriented
Simulation System (ROSS) at Rand Corporation is an example of a
large-scale modeling effort. ROSS is a revised (intelligent
system) version of a large Advanced Penetration Model previously
built for the Air Force to simulate strategic air defenses.
Within the Agency, an intelligent front-end to the Hydrology
model has been constructed based on the PROSPECTOR system at SRI.
4.4.2 Intelligent Assistants
Another application of the technology is to develop intelligent
assistants which can automatically perform routine tasks which
involve more detail than the human user cares to remember but
more flexibility than current non-intelligent systems exhibit.
Examples of such activities include logging in to various computer
systems (e.g., via the COINS network), scheduling appointments
and maintaining calendars, setting up parameters and complex run
streams for computer models, or storing and retrieving data from
local files as in an analyst's personal workstation. The Rule-
based Intelligent Terminal Agent (RITA) of the Rand Corporation
is an example of an intelligent assistant which has successfully
performed these types of functions.
4.4.3 Focused Search
In problems requiring extensive searching through massive
amounts of data such as SIGINT files, annotated imagery, or text,
heuristics can be developed to focus the search to the most
likely areas first. Heuristic search techniques have been
successfully demonstrated in the DENDRAL system at Stanford
University which generates plausible structures for organic
molecules from mass spectral data. Systems based on DENDRAL
techniques might be developed, for example, to assist in crypto-
graphic analysis where plausible machine settings or encipherment
procedures are being sought.
4.4.4 Failure Diagnosis
A fourth potential application area involves the general
diagnosis problem of determining the source or cause of system
failure by analyzing its symptomatic behavior. In approaching
these types of problems, systems can be built which formulate a
hypothesis (or diagnosis) and then proceed to organize the
available information (which may be incomplete or uncertain) to
either confirm or deny that hypothesis. Systems using these
techniques may be developed in the Agency to diagnose complex
system malfunctions (such as large computer systems or networks),
remote system failures (as in orbiting spacecraft), computer
security breaches, or even medical diagnosis of selected foreign
individuals.
ECR T
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
YVal~
Several diagnosis systems have been successfully built such
as MYCIN at Stanford which diagnoses bacteriological infections
in human blood, CADUCEUS at the University of Pittsburgh which
can diagnose some 500 different disease types of internal medi-
cine, and PROSPECTOR which evaluates geologic site suitability
for mineral exploration (developed for the U.S. Geologic Survey
by SRI International). In some instances (e.g., the SPILLS
system at the Rand Corporation), systems have been built which
can recognize that more information is required and ask the end-
user a specific question. Such systems may be appropriate to
support the intelligence collection process.
4.4.5 Image Understanding
Another high-payoff application of intelligent systems
technology is ORD's Image Understanding Program in support of
NPIC. This multi-year R&D effort is seeking to develop the
capability of automatically recognizing target images in a scene
and interpreting the meaning of the activities detected in a
series of images. It employs, in part, the ACRONYM intelligent
system from Stanford University. In addition to its expert image
interpretation task, the Image Understanding Program will also be
investigating the integration of multi-sensor information, the
assessment of image quality, and the development of an intelligent
assistant to track newly developing target activities. The
Intelligent Systems Program will complement the Image Understanding
Program by conducting research and development in the general
tools and techniques of knowledge-based systems. Some of these
tools may be of use in the Image Understanding Program. Con-
versely, the Intelligent System Program will seek to generalize
upon the knowledge-based techniques developed for the image
understanding work and determine how they might be applied in
other areas. Close collaboration is expected among the personnel
involved in these two programs.
4.4.6 Computer-Aided Design
An application area currently receiving considerable interest
in industry today is that of computer-aided design (CAD) and, in
particular, the design of micro-electronic circuitry. An intel-
ligent circuit design tool could be of significant help to OTS in
the development of special-purpose integrated circuits where only
a few of a kind may ever be produced and hence the entire cost of
the design must be borne by the Agency. The EL system of Stallman
and Sussman may serve as a basis for building such a tool.
4.4.7 Natural Language Understanding
Considerable work has been done to improve the computer's
capability to understand natural language. Many of these tech-
niques are following a knowledge-based approach. Correctly
I
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
interpreting natural language sentences would be of tremendous
benefit in the Agency in improving the user interfaces of our
computer systems and the automatic exploitation of text. Projects
involving text obtained from open literature (e.g., Projects
APOSTLE and DISCIPLE), transcribed by FBIS (e.g., Project MIDAS),
or contained in cables and messages to be routed to the appropriate
analyst (e.g., Project SAFE) could all be enhanced by knowledge-
based techniques of natural language understanding.
Development projects such as SRI International's LADDER,
System Development Corporation's EUFID, and Artificial Intelligence
Corporation's INTELLECT have focused on natural language front-
ends to data base management systems. Research at Yale, MIT,
Maryland, Brown, and Illinois Universities (among others) have
investigated the need for knowledge bases which model real world
events and relationships as an adjunct to understanding natural
language sentences. Understandably, most of this work has focused
on understanding English, although some research is being conducted
at the University of Texas in understanding other languages as well.
In the area of understanding spoken natural language, the
DARPA Speech Understanding Program has developed several techniques
of interest which may potentially have longer-term application in
the Agency. The knowledge-based approach of Carnegie-Mellon
University in the HEARSAY pro ect is noteworthy for its ability
to control the interaction ofdseveral knowledge sources con-
currently to solve problems cooperatively. Successful speech
understanding capability might ultimately have application in
monitoring foreign broadcasts, transcribing interviews, in covert
collection, or in maintaining the physical security of our
facilities.
4.4.8 Data Base Management
Large, on-line data bases can be confusing to end-users in
terms of knowing the right questions to ask, formulating. syn-
tactically correct queries, searching multiple data bases, or
drawing inferences from incomplete data. Intelligent systems
offer a way to improve the capabilities of our existing data base
management systems by providing mechanisms to permit deductive
inferencing from existing data using rules of inference compiled
by an expert in a particular subject area.
Augmented Data Manager ..( ADM is. one such to a can
be " u
"t6enlfahce the current capabilities of our existing data base
management systems (e.g., GIMS, RAMIS, or ORACLE).
4.4.9 Computer-Aided Instruction
The explanation feature of most knowledge-based systems ac
be used to explain to the user the logic which the system is
ILLEGIB
allt 'ALA
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
following at any point (i.e., the goal it is trying to reach and
the facts that it has already established). This explanation
approach has been extended in some projects to develop a system
which can instruct end-users in the knowledge which has been
built into the system. While obvious application areas for
computer-aided instruction techniques might exist in the Office
of Training and Education or the office of Communications, the
technology might also be applied with individual systems to make
them self-tutorial. The SOPHIE system at Bolt, Beranek, and
Newman was developed to teach electronic circuit debugging. The
GUIDON system at Stanford University was developed to tutor
medical students about the contents of its knowledge base. The
GUIDON system constructs'a model of each student's performance,
can lead the student as well as respond to the student's initiative,
and can present and explain medical diagnostic strategies.
The goals of the ORD intelligent systems program will be
accomplished along several fronts at once. ORD personnel will
serve as program planners, technical contract monitors, and in-
house intelligent system experts. In most instances contractors
will be employed for the actual design and implementation of any
intelligent system application or the refinement of pertinent
knowledge engineering techniques. However, to meet the goals of
this program, three additional positions will be required in ORD.
In addition, it is important to point out that because of the
crucial role which application domain experts play in the genera-
tion of knowledge for intelligent systems, heavy participation
will be required from the principal beneficiaries (experts and
end-users) as individual applications are developed.
4.6 Research Issues and Products
A principal product of this program will be the development
of actual working intelligent systems for specific Agency appli-
cations. The selection of which applications to address and in
what order will be an early focus of the overall effort.
Subsequently, the evaluation of completed applications and the
measurement of success will be another aspect of the research.
Concurrent with the development of specific applications
there is the need to build a body of expertise in the Agency on
the design issues involved with developing intelligent systems.
Such issues include techniques for eliciting and representing
expert domain knowledge, identifying and evaluating appropriate
intelligent system applications, designing and processing know-
ledge bases, and implementing and operating intelligent systems
(including computer resources, programming languages, and inte-
gration with the Agency's computer environment). Issues such as
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
ta.'S. is 1 .y ~.
these will generally be investigated in conjunction with the
development of actual applications although occasional study or
tool development projects may be performed to obtain a general
capability which may be applied to several applications. At
times, in-house experimental or prototype applications may be
developed to demonstrate a concept or the feasibility of an
approach.
A potential spinoff of the proposed program is the development
of intelligent systems tools and techniques having general appli-
cability to other computer science concerns such as data base
design, information system architecture, or mathematical model
development. The transfer of these tools to other system develop-
ment efforts and, indeed, any potential alteration in the way in
which the Agency designs and develops systems will be pursued
closely in this program.
4.7 Program Strategy and Schedule
The proposed program encompasses both support to ongoing
knowledge-based,,,,,ORD programs and broadening the range of experi-
mental applications. As a result, part of the proposed budget
will support the acquisition of special hardware and software
needed to implement and evaluate the application of intelligent
systems. The budget will also support the acquisition of an in-
house knowledge engineering capability which can be directly
useful in applying intelligent systems technology within the
Agency. The rest of the funds will be used for contractual
support in the development of knowledge-based applications.
This program will build upon and generalize other ongoing
ORD and Agency projects; hence, much of the groundwork and early
experience with intelligent systems will be prepared in FY82 and
83. These limited ongoing projects will likely continue in FY84
as maturing applications. General tools and techniques will be
extracted and any lessons learned will be consolidated into a
body of Agency expertise relevant to knowledge engineering.
Additional, more substantial applications will be initiated in
1984, as appropriate, and these can be expected to extend for
several years of development thereafter. In outlying years it is
anticipated that the tools and techniques of knowledge engineering
and intelligent systems development will be successfully trans-
ferred to ODP, ODE, DDI and other system development and user
organizations.
The general operating philosophy of the ORD Intelligent
Systems Program will be to adapt existing tools, techniques, and
systems to Agency applications and the Agency environment. There
appears to be both sufficient R&D funding of the field (by DARPA
and others) and intense industrial interest in applying this
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
technology (such as oil and mineral exploitation, design of
integrated circuits, or computer sales hardware configuration) to
ensure continued progress. ORD's investment strategy under this
program will be to build upon existing intelligent system proto-
types in early applications and to learn as much as possible
about appropriate techniques of knowledge engineering.
5. EXPECTED PROGRAM BENEFITS
The goal of intelligent systems is not to replace people by
machines but rather to assist humans with the more mundane aspects
of information processing. By using its preprogrammed knowledge
base the computer can perform repetitive functions and routine
processing with less human intervention. In this way humans can
have more time for creative thinking and analysis. The producti-
vity of the user can be enhanced (rather than reduced) by the
computer through the development of better human/machine inter-
faces, more independent problem-solving on the part of the computer
(such as error-correcting or trying alternative approaches),
having the capability to explain its behavior to the end-user,
and possessing the ability to be easily modified through explicit
representation of its knowledge. Ultimately, the payoff of this
technology is expected to be the ability to take better advantage
of the power of the computer to produce more intelligence with
our existing resources.
The primary benefit of intelligent systems to the intelligence
consumer will, of course, be an improved finished intelligence
product. By consolidating group expertise into one knowledge
base which is available as a consultant, an applications office
should be able to improve the quality and consistency of its
products. Further, because such systems are designed to accom-
modate change, the Agency should be better able to respond
quickly to changing intelligence requirements or "what if"
questions from our consumers.
The ORD Intelligent Systems Program will serve the broader
Intelligence Community by permitting the sharing of relevant
technology and by bringing into focus explicitly the difference
in assumptions and inferencing techniques employed by various
agencies as such systems are built. Indeed, one significant
benefit of constructing any intelligent system is the codifica-
tion of knowledge in a previously uncodified field. Workers in
the ORD program will remain in contact with knowledge engineers
from other Community agencies (such as NSA) to share experiences,
trends, and new techniques. It is quite possible that intelligent
systems developed for CIA applications may have broader applica-
tion throughout the Community either as a central resource
information system, as a distributed copy for other similar
applications, or as a general-purpose tool for the construction
of other intelligent systems.
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6
ORD has high expectations for the successful application of
intelligent systems technology to Agency problems. However, the
development of large-scale expert system applications at this
point must be viewed as having some risk. The technology is
still developing, and the Agency does not yet have a full under-
standing of the capabilities and limitations of intelligent
systems. Of much lower risk is the acquisition of and experi-
mentation with selected tools and techniques and the development
of knowledge-based intelligent assistants and enhanced user
interfaces to existing computer models and systems. Recognizing
this risk, the ORD approach to intelligent systems will be to
develop applications initially from selected well-bounded problem
domains and allow those applications to evolve as our under-
standing of the technology grows. An early task of the program
will be to develop criteria for selecting suitable applications,
assessing the chances of successfully completing the system, and
measuring the degree of success of the results.
Because intelligent systems technology offers a significant
increase in the'ausability of computers, wide use of this tech-
nology in the Agency is inevitable. ORD is using these technologies
in an increasing number of its projects, at least some of which
will lead to operational uses. In addition, intelligent systems
technology is receiving wide interest in industry, and we can
expect to find aspects of this technology in an increasing
portion of the devices and systems which the Agency will acquire.
Finally, users of data processing services within the Agency are
learning about intelligent systems capabilities, and fairly soon
we can expect ADP users to start demanding these types of services
and capabilities of the systems they use.
Declassified in Part - Sanitized Copy Approved for Release 2012/12/14: CIA-RDP95-00972R000100020003-6