COMMUNITY SPONSORED PLAN FOR ARTIFICIAL INTELLIGENCE

• Document Type: CREST
• Collection: General CIA Records
• Document Number (FOIA) /ESDN (CREST): CIA-RDP95-00972R000100030005-3
• Release Decision: RIPPUB
• Original Classification: S
• Document Page Count: 9
• Document Creation Date: December 23, 2016
• Document Release Date: September 4, 2012
• Sequence Number: 5
• Content Type: MEMO

Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
MEMORANDUM FOR:f irector of Data Processing, DDA
THROUGH Deputy Director for Administration
FROM
Philip K. Eckman
Director of Research and Development, DDS&T
SUBJECT Community Sponsored Plan for Artificial
Intelligence
1. Let me take this opportunity to congratulate you on
your assumption of duties as Director of Data Processing.
Sometime soon, we might get together so that I can brief you
on research we have in progress on behalf of ODP.
2. More currently, however, let me pass on the attached
recommendations from the IR&DC to John McMahon. We drafted
these for Dick DeLauer to send to John at his request. These
follow up on a DDCI request for specifics in support of a
previous IR&DC recommendation that the introduction into the
community of Artificial Intelligence (AI) be speeded up.
3. Some of the recommendations pertain to ODP, and you
may find others of them interesting. If you have any questions
don't hesitate to call me or whom, I am assured,
you know.
4. I look forward to getting together with you in your
new role.
Attachment:
as stated
This document is downgraded
to UNCLASSIFIED upon removal
of attachment
25X1
25X1
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
 Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
SEc
MEMORANDUM FOR: Deputy Director of Central Intelligence
FROM Hon. Dr. Richard D. DeLauer
Undersecretary of Defense for Research & Engineering
(as) Chairman,
Intelligence Research & Development Council
SUBJECT Artificial Intelligence Program Recommendations )
REFERENCE (a) DDCI Memorandum
(b) IR&DC Repor
Executive Summary
This memorandum responds to your request for recommendations
concerning the application of artificial intelligence (AT) technology in
the Intelligence Community (1C). Developed more fully below, the
Intelligence Research & Development Council (IR&DC) recommends that:
(1) Guidance be given to all IC Program Managers to include plans for
applying artificial intelligence within their recommended FY-85
programs. A sample of suggested wording for such guidance is
provided as attachment-A. 1
(2) A subcommittee of the IR&DC be established to provide needed
information exchange on Al matters within the Community. The scope
of this subcommittee would include technical information exchange,
interfacing between basic Al research and IC applications, and
coordinating IC AT activities. )
(3) The IC continue to rely on the long-term financing of basic Al
research by the Defense Advanced Research Projects Agency (DARPA),
coordinating its requirements through the IR&DC Al subcommittee.
(4) The IC, itself, continue to concentrate on applying Al and prepare a
list of applications which would solidify the commitment and alert
researchers and vendors to Community needs. An initial set of
candidate applications is included as attachment-B.
(5) Community ADP managers be advised that in all likelihood a more
diverse set of computing hardware and languages will be required to
support AT systems. Plans will need to be developed for maintaining
these tools and integrating them into existing computer environments.
(6) Several nascent Al applications laboratories in the Community be
expanded, and their timetables advanced, so as to form focal points
for Al in the Community. Two such laboratories are those in the
V
25X1
25X1
25X1
25X1
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
 Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
Engineering Topographic Laboratories (ETL) and the Office of Research
and Development (ORD).
(7) An unclassified Al computer facility he established for the Community
and networked with the existing Al research centers via the ARPAnet.
This facility would provide a cost-savings link between the
Intelligence Community and the Al research community.
(8) End-users of Al systems begin to allocate resources for technology /
transfer, recognizing their commitment to provide domain-specific V
expertise to be built into the systems.
(9) The component agencies work together to plan for the internal
development of Government personnel in Al.
V
As we indicated in our report, we are encouraged by recent breakthroughs in
Al and believe that there are many areas in the Community which could
benefit from applied Al technology. The recommendations contained in this
memorandum are aimed at accelerating current Community efforts in AT to
speed up the productive use of this technology throughout the Community.
Key among these recommendations are numbers 8 and 9. The projected
shortage of trained personnel, added to the normal security issues,
necessitates that the IC be able to supply a large proportion of its own AT
specialists. Therefore, training is paramount (Rec. 9). The importance of
Rec. 8, "End-User Planning", results from the lack of exposure of Community
components to the technology and their understandable general reluctance to
exchange current resources for future benefits. Although certain of these
recommendations may need to wait for FY-85, every effort should be made to
take preparatory steps in FY-83 84 and Program Managers should be asked to
address this specifically.
1. DCI Guidance
Given the strong potential for Al applications in the Community, the
IR&DC recommends that guidance be given to all IC Program Managers to
aggressively pursue the technology and to plan for Al in their FY-85
programs. Each individual Program Manager will, of course, decide where Al
technology can best support his or her program. It is therefore
recommended that each Program Manager prepare a list of candidate projects
for early attention. A sample list is provided in attachment-B. Because
of the burgeoning technology and the Community's relative inexperience in
applying Al to intelligence problems, we recommend that the DCI resist at
this time the temptation to name a single Al focal point or lead office for
the Community. We suggest, instead, that the Community rely on developing
a program of information exchange concerning applied AT. Such a program
can and should be sponsored through the Al subcommittee of the IR&DC.
2. IR&DC Subcommittee on Al
To coordinate the exchange of information in this field, we recommend
that a new subcommittee be established within the iR&DC to exchange
technical information, track basic research, evaluate Community
requirements from a technical perspective, and review technically the
development of Community Al systems. This subcommittee would serve as the
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
 Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
major coordinating body for Al activities within the Community as well as
coordinate Al research efforts on the outside. Specifically, as indicated
below, this subcommittee will interface with DARPA (a member of the IR&DC
but not of the IC). As another example, a series of specific
recommendations have been developed by the EXRAND, including several
involving Al technology and these are endorsed by the IR&DC. These are
summarized in attachment-C. As a part of its charter the new IR&DC Al
subcommittee would monitor the progress of Al activities such as those
proposed by EXRAND. 1.
3. Basic Al Research
While artificial intelligence has been under development for more than
25 years, there remain many fundamental problems to be solved. We
recommend that continued encouragement be given to DARPA, NSF, ONR, and
other funding agencies to support basic research in artificial
intelligence. Historically, these agencies have been providing more than
$10 N per year for Al research during this period. We hope that similar
levels of support would be continued in the future. The IR&DC would then
serve a coordinating role in expressing the needs of the Community to these
funding agencies.
4. IC implications of Al
There are many areas where AI technology could be applied to
intelligence problems. Several initial efforts are already underway to
investigate the application of expert systems, natural language processing,
image understanding, and logic programming technologies and methods. A
preliminary list of potential Community applications appears in attachment-
B. This initial list presumably will be expanded and continually refined
by the IR&DC AT subcommittee. As usual, identifying suitable applications
and expressing Intelligence Community requirements to the research
community will be a challengin task requiring a deep understanding of both
Al technology and IC needs.
5. Al Computing Environment
Systems involving artificial intelligence generally have special
requirements which necesitate a computer environment that is somewhat
different from the classical data processing centers of the Community.
Consequently, the development of AI applications in the Community will
necessitate the member agencies' acquiring a diversity of computer
equipment (e.g., machines specifically designed to run the LISP programming
language). The various Program Managers should begin planning now to
allocate resources for these facilities. Admittedly a diversity of
computing equipment (and vendors) will be a hardship on ADP managers.
However, such equipment is essential at least for in-house research and
development efforts. Of course, every effort should be made as systems are
developed to ensure that operational systems will be as compatible as
possible with individual agency ADP masterplans and traditional mainframe
equipment. F
6. Classified In-lfouse Al Laboratories
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
 Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
Current efforts to establish classified, in-house AI laboratories such
as those in CIA-ORD, and ETL should be encouraged and expanded with
additional funds. The special research-user relationships needed to
develop applied Al support to operations argues for several distributed
facilities as opposed to a centralized approach. Quality Government
laboratory facilities will not only provide the computing resources
necessary to conduct Al research and development but will also provide a
mechanism for training Government personnel and attracting competent
researchers from the universities and private industry. While recognizing
the special security and privacy needs of each of the agencies,
consideratiod should nonetheless he given to the electronic-Networking of
selected Community AI research facilities to further the exchange of
knowledge and information in the field. Recent advances in network,
security and compartmented dissemination substantially reduce the risks
associated with such interconnections. F
7. Unclassified AI Laboratory
In addition to in-house facilities for developing classified AI
applications, the Community could also benefit from an unclassified
facility which is connected electronically to other unclassified AI
research centers outside the Community. Such a connection most likely
would be via the existing ARPAnet which already connects most of the Al
research centers in the U.S. The IC unclassified Al facility could be
operated under contract following models established by DARPA such as the
Information Sciences Institute at the University of Southern California.
This facility could be used to cut down on Government commercial software
development costs by providing Government-furnished computer time to
contractors in an environment similar to the classified environment into
which the operational system must eventually be placed. F__1
8. End-User Commitment
The basic paradigm of an Al system is to develop a functionally
capable system in the laboratory and then to load it with specific
knowledge about the intended application. This knowledge is generally
obtained from the customers for whom the system is being built. We
recommmend that guidance be given to the various components to plan for a
substantial consumer commitment to identify experts and to make them
available to provide the required domain-specific knowledge. Further,
these internal customers for AI technology will also need to plan adequate
resources for operating and maintaining such systems once they are
developed. Active user involvement in the technology transfer process
should be ensured from the beginning.
9. AI Training
Perhaps the most important consideration for a long-term commitment to
artificial intelligence will be the establishment of an adequately trained
Government work force. Sufficient understanding of the tools and
techniques of Al will be required of Government personnel to identify
suitable applications, to monitor contracts, to select hardware and
software, and to apply commercially developed products to the Community's
needs. Because of strong competition for the relatively few skilled Al
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
 Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972ROO0100030005-3
systems people available, the Community will have to establish training
programs to develop its own .internal expertise. Opt. ions available for
acquiring this expertise include:
9. 1 Establishing close relationships with the major Al research centers
through summer institutes, post-doctoral fellowships, hosting
professors on sabbatical, and extended on-site apprenticeships with
the centers.
9.2 Encouraging and assisting employees to seek formal courses or
graduate school training in AT and knowledge engineering. F
9.3 Conducting an AT seminar or lecture series for members of the
Community.
9.4 Sponsoring periodic Al symposia similar to the one recently hosted
by CIA-ORD and CIA-OSWR at CIA Headquarters. F-1
9.5 Developing a series of internal training courses for Al system
developers and end-users. F-1
9.6 Providing internships at Community Al laboratories for academic or
industrial AT researchers.
Early consideration should be given to recruitment and training in FY-83
and 84 to form a cadre of knowledgeable personnel around which to build a
viable program of applications beginning in FY-85.
10. The Scale of Community Efforts
It is premature to recommend a consolidated resource level to be
applied to the program Community-wide. Experience has shown that
substantive Al implementations in other areas require 25-50 person-years to
achieve significant results on a single application. While this is not
extraordinary when compared to other system development efforts, it is
important to recognize that Al systems can be costly and may require
several years to implement. At current prices, a well-designed Al
laboratory typically might cost $2-4 M. In fact, as outlined in
attachment-C, the ERRAND subcommittee of COMIREY recently recommended that
Al laboratories of ORD and ETL be enhanced by an additional $1 M each in
FY-83. Each Program Manager will want to assess the facilities available
to him and weigh these against his agency's requirements before submitting
a budgetary request. There are, however, several actions which could be
developed on a Community-wide basis and which can benefit each of the
agencies while avoiding redundancy. These activities of common concern
could include setting up common AT training programs, establishing an Al
software clearinghouse, and developing selected Al system development
tools. Case-by-case investigation will be required to assess the trade-
offs of the economy of scale of common actions versus meeting program-
unique requirements. F--]
11. The Need for Artificial Intelligence
To reiterate, the Intelligence Community is faced with a continuing
SECRET
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972ROO0100030005-3
 I
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
set of technical problc+ms involving the collection, processing, and
reporting of intelligence. Advances in collection technology promise to
continue to provide more data than we can process with common data
processing technology. Valuable human expertise is lost every year as
knowledgeable analysts retire or otherwise move on from their positions.
Confusion mounts over just what it is that a particular quantitative
technique really models or how to set the parameters to make it work.
Information systems continue to be developed which are inflexible to
changes in requirements and costly to modify. In short, much is being
spent on developing automated systems, but the practical utility of such
systems is not all it could be.
In conclusion, we believe that artificial intelligence is the only
technology today with the promise of yielding the kinds of information
systems needed to process the projected volumes of data and present it in a
way that is meaningful to and efficient of our human analysts. The time is
now to begin to build the kinds of systems which will be required by the
mid- to late-Eighties, systems which employ the techniques of artificial
intelligence. F-7
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
 I
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
,NF NTIAL
CO
ATTACHMENT-A
SAMPLE DCI GUIDANCE
As in previous years, I wish to emphasize the importance of developing
innovative technical and operational solutions to the intelligence problems
that face us. Creative planning and application of appropriate technology
should allow us to achieve the significant progress of which we are
capable.
One can scarcely be unaware of breathtaking advances in computers
which touch every corner of our lives. We must redouble our efforts to
take fullest advantage of this technology. I am particularly persuaded
that the most advanced area of computer science, known (to my dismay) as
"Artificial Intelligence", has much to offer the Community. To this end, I
have reaffirmed and strengthened last year's DCI GOAL 12: "To Enhance
Manpower Talent and Productivity J'. It has been modified to recognize
explicitly the potential contribution of advanced computer science. F-1
GOAL 12: To Enhance Manpower Talent and Productivity
Intelligence analysis and production remain a manpower intensive
effort and we continue to suffer from a critical shortage of scarce skills.
Methods must be found to ensure intelligence access to the necessary skills
and to leverage through computer technologies our resident talents. The
shortage of linguists and regional analysts will be particularly critical,
but engineers and technical specialists will also be in short supply. Ways
must be found both to increase the skilled manpower available and to
improve the productivity of available personnel.
The capability of the Intelligence Community to recruit, retain, and
ensure the productivity of necessary manpower remains limited. Colleges
and universities continue to produce insufficient graduates with the skills
required by the Community, nor is the shortage likely to be remedied
through the decade. And, while public perceptions are slowly changing, the
problem is still compounded by: an impression among undergraduates that
past personnel cuts and hiring freezes reflect the long-term trend; a
residual negative public attitude toward intelligence; and, competition
from the private sector for the same skills. Consequently, it remains
difficult to find qualified personnel, even when manpower positions are
available.)
You should endeavor to initiate programs that will improve the
Community's ability to recruit and retain the manpower necessary. Further,
you should endeavor to initiate programs, based upon the most promising
computer science advances, to multiply the productivity of available
manpower.
25X1
25X1
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
 Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3
Iq
Next 2 Page(s) In Document Denied
Declassified in Part - Sanitized Copy Approved for Release 2012/09/04: CIA-RDP95-00972R000100030005-3