ursuantto E O 13526 MDR Case # 5477 WGBPDUV0WGBPD11 LrUJrnu Jf VU GBPD W $lIJrn $ W $UJIIl $ w jJrn l1 jJUJIIl MAR - APR 1986 E-X-P-E-RT-S-y-S-TE-M-S-----FO-R-N S-A - -U- - - - - - - ------1 1 BULLETIN BOARD U o o o o o o o o o o o o o o ooo o o o 6 AN INTRODUCTION TO AI U o o o o o o o o' ' o l 1 70 LEVELS OF CLASSIFICATION U o o o THE 39 STEPS U o o o o o o o o o o o o o o 11 PUZZLE U o o o o o oo o HANBlE VIA eOMINT CHANNELS ONLY TOP-SEERH- o o o o o o o o o o o o o o o o o o 13 NOT RELEASABLE TO CONTRACTORS CLA55IFI Q 8Y JSAtcS5M 1 a Qi b 551 YO i Originating Agen'y' OeterR liniiltign RQqwir@g CIO LL -------------u 4009980 86-36 j PubU d by PI Teo'n q and Stand I ldh irl NOS 3-4 - ' ' ' March-April 1986 ' ' ' ' u ' Collection Computer See Qrity ' ' ' 1 1 963-5877 968-8141 f y r t o ' i J Oli i S CS i 1 l l Information Science ' ' --' 1 963-1145 963-3095 Intelligence Research ' - 1 language _ ' 963-3057 ' J 963-5566 Mathematics 963-3648 Puzzles _ _ 1 963-4191 Science and Technology 1 Special Research _ Vera R Filby 968-8014 Traffic Analysis _ 'l obert J Hanyok 963-5734 Illustrator TIGHTENING OUR BELT _ 1 '--- 1963-3057 That's a lot for a spartan periodical with nary a frill ruffle or furbelow to do without We don't use color the paper is minimum quality usable on a press and the format is the most cost-effective Even so in the past year we've done what we could to bring costs down o For word processing we've moved to the Xerox Star from the Unix Editor on the BARDOLPH TSS resulting in impressive savings in labor costs especially in making printouts and in doing layouts o To avoid retyping we're using the Data Conversion Center for converting floppies made on other word processors o We've gone over to the high-speed Solna press which is more economical but prints only multiples of 16 pages To submit articles or letters by mail send to Editor CRYPTOlOG P1 If you used a word processor please include the mag card floppy or diskette along with your hard copy with a notation as to what equipment operating system and software you used via PLATFORM mail send to cryptolg at bar1c05 bar-one-c-zero-five note no '0' in 'log' I Always include your full name organization and secure phone number I I o We are trimming the distribution list We can use your help on this last item o If you are going to the field or into reti rement let us know so that we can remove your name from distribution Organizational not personal copies are sent to the field You can have a personal copy again when you return o If your organization is now smaller than it was request fewer copies o As the number of personal copies in your element increases eliminate an appropriate number of organizational copies I I o We're publishing half as many issues o We are down to a one-person staff the editor who has other duties as well For Change of Address send name and old and new organizations to Editor CRYPTOLOG P1 Thanks Contents of CRYPTOlOG should not be reproduced or disseminated outside the National Security Agency without the permission of the Publisher Inquiries regarding reproduction and dissemination should be directed to the Editor FOR OFFWIAL USB ONLJPY ern 400998P L 86-36 SBORB'I' EO 1 4 c P L 86 36 I P L 86-36 EO 1 4 c Mar-Apr 1986 CRYPTOLOG page 1 HAI'WLB ViA COItifft r p CUA fNBLS OnLY SfilCRB aCID P L 86-36 EO 1 4 c 4009980 Mar-Apr 1986 o CRYPTOLOG o page 2 tIAIfBL'E Vi-A COr HPf'f cnftP i ELS Oi Tb Y S86R 'F ero 4009980 P L 86-36 EQ 1 4 c SEGRE'F Mar-Apr 1986 o CRYPTOLOG o page 3 HA Hn V A COAQ -WT Clv rWE b OW bJPY S OR8T ero 4009980 -'I P L 86-36 I 1 I EXPERT SYSTEMS--FOR NSA u One of the more profitable advertising gimmicks in com uter software today is the use of the word expert ' Numbers of products claim to be expert systems or to have expert capabilities After we cut through the hype do we find that expert systems actually have any value When are they useful Can we use them at NSA WHAT CAN AN EXPERT SYSTEM DO Putting rules into a knowledge base can help to organize knowledge that was not well understood or not centrally located Books and papers can also be used for these purposes but the user interaction with the expert system gives it an advantage Also the expert system actually tries to apply the knowledge it does not just serve as a repository for that knowledge An expert system could be used to explain rules and regulations particularly those that have In general an expert system starts with a accumulated over time and are not logically set of facts obtains more facts from the user consistent Expert systems have been used most through interaction and using a database of successfully to diagnose problems in expensive knowledge about the application produces a complex mechanisms reasoned conclusion The system can in most cases explain its conclusion by displaying the logic WHAT IS AN EXPERT SYSTEM chain An expert system is very good at applying heuristic rules rules of thumb to a set ofloosely related data and deriving a conclusion -- much like an expert does It has been said that an expert is one who jumps to the right conclusions because the expert picks just the few facts that are important to a conclusion often bypassing complex computations and arrives at a conclusion while other people are still sifting facts An expert system on a computer can be used to record the heuristic rules that a particular expert uses -- to clone the expert This can be used to spread the expertise around or to substitute for an expert who is leaving It can also be used to train new experts -- to organize the expertise so that it can be taught The expert system leads the student through a correct logic chain with sample data an improvement over traditional computer-aided instruction systems that merely quiz a student repeatedly I use the term expert system narrowly to refer to software that separates the application domain knowledge from the program's control structure and that can explain its reasoning to a user The repository of domain knowledge is commonly called the knowledge base as in data base and the prow-am control structure an inference engine ' This is a specialized program to read the knowledge base and process it logically to derive some conclusion or conclusions The most common class of expert system building tools or shells on the market today there are dozens of them use a knowledge base of rules in the form of IF some condition or conditions THEN some condition or conditions The conditions usually are expressed as a parameter value combination and may refer to other rules which are determined to be true or false in a chain oflogic going back to data entered by the user or sensed by the system externalll' This chaining usually goes backward from a 'goal Mar-Apr 1986 o CRYPTOLOG o page 4 FOR OFFI6IAL USEl ONLJPY ero 4009980 rule to the original data but may go forward from a set of data to one or more conclusions see box Key to the usefulness of the expert system is the fact that the sequence of rule use is data-driven and may vary drastically in different runs The inference engine selects rules as needed and chains through the knowledge base in the most efficient manner Most expert systems have some ability to handle uncertain data A piece of data can be true not true or somewhere in between probably possibly not likely almost certainly almost certainly not etc Data that is probably true pushes the system in the direction of a positive conclusion and data that is probably false pushes the system in the opposite direction In addition to dealing with uncertain data most expert systems have some mechanism for handling unknown data by assigning a medium value or extrapolating from known data or by ignoring the rules involving unknown data Some expert systems have mechanisms for representing knowledge in some form other than simple rules e g complex data objects that can have different kinds of values or attributes and can inherit attributes from certain other objects PROBLEMS WITH EXPERT SYSTEMS Some knowledge simply doesn't need to be organized in a randomly arranged data base The steps follow each other in a purely sequential fashion and the solution can be programmed more simply and efficiently in a traditional programming language The structure of knowledge representation is very important Considerable thought needs to be given to this just as you would give considerable thought to the design of an important and complex computer data base Inexact reasoning is a black hole Although it is possible to combine values that are inexact no one has proven that one way of doing this is better in a general sense than any other way It appears that some methods are better for some applications some for others but the selection of a method is not well understood Conflict resolution can be tricky The inference engine matches patterns of rules to find the best rule to apply in any particular instance If two or more rules match the pattern some conflict resolution technique must be used There is not a good understanding of how to resolve these conflicts SAMPLE RULES KNOWLEDGE BASE 1 if SCREEN is BLANK then POWER is OFF 0 5 2 if KEYBOARD is not WORKING then PROGRAM is HUNG 0 4 3 if DISPLAY is ERROR then PROGRAM is HUNG 1 0 4 if PROGRAM is HUNG then FIX is CTL-ALT-DEL 0 5 5 if POWER is OFF then FIX is SWITCH POWER ON 1 0 6 if FIX is KNOWN then DIAGNOSIS is COMPLETE Forward Chaining The system tries rule 1 if the screen is blank then rule 1 concludes that the power is probably off confidence factor of 0 5 The system then looks for chains forward to a rule which is true if the power is off rule 5 Rule 5 then says that the fix is to switch the power on Once the fix is known rule 6 concludes the diagnosis If the screen is not blank rule 1 the system tries rule 2 and then rule 3 untilit either exhausts the knowledge base or finds a rule that works whose premise is true Any chain of rules which reaches the goal rule rule 6 ends the program Backward Chaining The system starts with the goal rule rule 6 If the fix were known it could stop It therefore looks for a rule which if true would make the fix known rule 4 or rule 5 The system picks rule 5 and checks its premise is the power om Rule 5 chains backwards to rule 1 if the screen is blank the power is probably off and the fix is to switch the power on rfrule 1 is false the chain fails and the system must try another chain it tries rule 4 which chains to either rule 2 or rule 3 The system continues backtracking through the knowledge base until it either completes a valid chain or exhausts the rules 8000 rules a system which runs on an IBM PC might be limited to as few as 100 rules EXAMPLES OF EXPERT SYSTEMS UntH recently all expert systems were built to run on mainframe or mini computers Only in the last couple ofyears have PC-based tools Expert systems tend to use up large amounts entered the market of computer resources - both time and memory One of the oldest expert systems is MYCIN Most of the smaller systems are severely limited in developed at Stanford University to dia ose the number of rules they can process one of the infectious diseases and to prescribe antibiotics most successful expert systems DEC's XCON uses Mar-Apr 1986 CRYPTOLOG page FOR OFFICIAL USE ONLY 5 4009980 Although MYCIN has never been used operationally its techniques have been copied by many commercial expert system packages One of the most successful expert systems is DEC's XCON which configures DEC computer systems Using 8000 heuristic rules XCON takes a customer's order and identifies all the parts that are needed to fill the order making sure that they all fit together DEC estimates that XCON saves the company at least $18 million a year Before BUILDING YOUR FIRST EXPERT SYSTEM Pick a problem Use the criteria described in Selecting Your First Application and add other criteria if they are important to you Pick a problem you know well one that looks too small to bother with because it probably will take more work than you thought but pick a problem that matters to someone -- preferably your boss Pick an expert You want someone who really is THE expert and who wants to help If you are the expert so much the better Develop a prototype Set some relatively short period of time and see what you can do in that period The schedule depends on your probl m ofcourse but three to six months should be enough to accomplish something that can be evaluated Evaluate the performance of your prototype system Does it make a good decision or provide satisfactory advice Evaluate the reasoning process do those right answers come from a valid line of reasoning Is the user interface acceptable for operational use Does the system run efficiently enough to be usable Will it still be usable when you get the full system more rules built Will the system be cost-effective when compared to the current manual way of doing it XCON was built the human experts found the task boring most of the time and therefore made numbers ofmistakes Traditional programs had been tried but didn't work the rules kept changing the knowledge was scattered and the totality of the logic was not well understood More recently and on a smaller scale the Campbell Soup company built a small expert system on a PC to diagnose breakdowns in their giant soup cooking machines A few hundred rules enabled a non-expert to diagnose most problems effectively The General Electric company built an expert system to diagnose GE locomotive problems The improvement in repair time was so dramatic that GE's market share increased by a large margin Autometric an NSA software contractor is developing an expert system to diagnose software problem reports They estimate that the prototype system with about 200 rules correctly identifies a problem module about fifty percent of the time saving about four hours of diagnostic time each week Eventually the system should enable a relatively inexperienced programmer to solve most software problems One of the chief reasons for developing this system was that the experts who had designed and built the software didn't want to maintain it and staffing became a problem Burroughs has a prototype expert system to diagnose personal computer problems for users calling a Burroughs hot line This hot line has been staffed by experts who are bored to tears by the mundane problems posed by most callers The expert system will enable an unskilled person to answer the hot line and effectively diagnose most user problems The experts will then work only on the more challenging problems I lin R8 has develo ed a PCbased expert system to identify the claSSIfication of cryptanalytic information according to a set of rules published by PI Proceed with incremental development of R53 researcher$ are working on a number of an operational system if your evaluation expert systems including one to support produces acceptable answers to your questions DEFSMAC Other expert system experiments are If the answers don't look good you need to fix underway in G4 Pl Wl A2 and T3 whatever is wrong or go back and try a different 1----------'--------------- application or possibly you need iI different expert system building tool Prepare the users for using the system This is the most imyortant item if you are building an operationa system DEC's XCON almost failed after three years of research because the people doing the work manually thought erroneously that the system would replace them and they would lose their jobs Mar-Apr 1986 o CRYPTOLOG o page 6 o FQK QfFICI Is i i O IsY P L 86-36 4009980 GONFIDSN't'IAL P L 86-36 EO 1 4 c SELECTING YOUR FIRST APPLICATION Your application doesn't have to match all of these criteria but the more items it matches the better your chances of success BULLETIN BOARD Traditional technology doesn't work Before DEC hired CMU to build RI they tried three times to write the program in FORTRAN I I The knowledge domain is narrowly defined and primarily involves l0t ical deduction One way oflooking at it IS to ask if a human expert could solve the problem over the telephone I Recognized experts know how to solve the problem You will have trouble teaching a machine how to solve it if you don't know how yourself I The task involves a few minutes to a few days of an expert's time I A traditional approach to solving the problem requires combining a large number of steps in a large number of complex sequences I Wrong answers will not result in loss of life or other unacceptable consequences Because of the heuristic rule approach P L 86 tiIIlalresults renot aranteed The resulting system willhaveahigh payoff You don't want to put a lot of eff rt into a system that doesn't really do anythmg Dave Fltzpatnck E42 968-8418 Bill Patterson B5093 963-3490 I n a r w ar ' e - o - r - s o 1' tr w a r e - s u - p p o rtj - a n - e - x p e rt -system might help diagnose problems in large complex computer systems or analyze network performance problems An expert system might assist with computer language translation when the two languages involved have different structures for example from FORTRAN to Ada GETTING STARTED If you are interested in developing an expert system look around there is probably someone not too far away who is experimenting with the technology and can heIP yon v dQ 1 find someone close at hand calll He is chairman of the SA Arti lCla nte rgence Working Group arid he can put you in touch with a numberofpeople to talk to 0 I ooooo I ooooooooooooooooooooooooooooooo ' ooooooooo SOLUTION TO NSA-CROSTICNo 62 Arthur J Salemme Joys of Plural Dropping CRYPTOLOG January 1978 When our daughter and a friend of hers were both five and were eating lunch at our house my wife gave them each a sandwich and a slice of American cheese cut in strips The little friend particularly enjoyed the latter and asked what thel were Cheese ' she was told 'Well then she asked 'can I please have another chee ' oooo oooooooooo 'p' 'L ' 6'-' O P L 86-36 Mar-Apr 1986 CRYPTOLOG page 7 eOI'fi'lfl N't'IAL IIAP'fDJ5B llI A GOMl t'J' OHAN fEL8 ONLY -------------- 4009980 I tasks traditionally thought to require human intelligence I in a manner approaching the working of the human mind This pursuit includes research in the cognitive sciences in an effort to understand more fully the human capabilities upon which intelligent machines are to be modeled H TR U T U 2 AI is distinguished from conventional computer science by its focus on symbolic rather than numeric qualitative rather than quantitative processes CTHJr i New capabilities for enabling machines to imitate man have been created as a result of fundamental developments in diverse fields including computer science the psychology of learning physiology and information science The work of cryptanalyst Alan Turing hypothesizing a machine that could perform mathematical operations in a binary format and developing a test to evaluate whether a disguised correspondent was machine or human Claude Shannon applying Boolean algebra to electronic circuitry and Norbert Wiener cybernetics are only a few examples ofcontributions that made AI possible The term artificial intelligence appeared concurrently with the first major event to bring together from a variety of academic backgrounds those who shared a common interest in enabling machines to perform intelligent tasks The event was the 1956 Dartmouth Conference organized by This is an abridgement ofan article that appeared in the February 1986 issue ofthe Cryptanalysis John McCarthy and attended by others who were to become prime movers in the field in subsequent Intern Bulletin years Although the Dartmouth Conference did Artificial intelligence AI comprises elements of not yield the results predicted it did set the atmosphere for interdisciplinary cooperation and many disciplines It was once explored only by challenge In addition two conference academics now it has become a topic of vital interest to both government and private industry participants Allen Newell and Herbert Simon both connected with Carnegie Tech now Carnegie It is worthwhile for the cryptanalytic community Mellon and the RAND Corporation came to the to familiarize itself with developments in AI conference with a working program demonstrating because ofpotential applications to cryptanalytic the intelligent processing of information that tasks conference participants had discussed and sought The history characteristics and applications of During the 1950s and 19605 expectations were artificial intelligence provide a fascinating high that machines could be programmed to exploration into the essence of the interaction imitate human thinking in the realm of common between man and machine that characterizes contemporary human existence sense that is everyday acts of reasoning that are performed by most people without conscious DEFINITION AND HISTORY recognition of the processes involved As the passing years revealed the unexpected complexity Although the discipline known as AI has been in of this task AI in the 19705 was marked by a existence for over thirty years no single definition philosophical and methodological shift toward is accepted by all the experts A survey of specialization Programs were written that would enable a particular machine to perform definitions given by leaders in the field however has yielded these concepts in common intelligently within a very limited domain In addition whereas once the goal was to equip a 1 AI seeks to enable a machine to perform computer with a general set of theories that would enable it to make a wide variety of decisions the page 8 FOR OFFICIAL USE ONLY Mar-Apr 1986 o CRYPTOLOG 4009980 focus shifted to providing an impressive array of facts about a given topic Other areas of artificial intelligence are also the objects ofintensive research efforts Natural language processing is the ability to program and receive feedback in a conversation-like context Computer vision and speech recognition are receiving attention from DARPA for their defense potential In addition robotics educational computer interactions and a variety ofoffice workstation uses for AI are currently under investigation and in some cases implementation Since the late 1970s a m ajor trend in AI has been commercialization What was once a topic that concerned academics and a limited number of research corporations has now become an industrial and a governmental fad of sorts especially within the US Department of Defense Moreover AI is an integral part of an emerging technical war between Japan and nations of the West In 1982 Japan announced a massive computer research initiative to take place during the next ten years Entitled the Fifth Generation of Computer Systems the project aims to put Japan in the forefront of world computer technology by the 1990s Many of the project's goals fall into the area defined as AI Response to the challenge was not slow in coming Britain Europe and the United States initiated efforts related to those of the Japanese AI is included in DARPA's Defense Advanced Research Projects Agency ten-year development program and in the plans ofMCC Microelectronics and Computer Technology Corporation a corporation of21 companies under the leadership offormer NSA Director Bobby Inman PROSPECTS FOR AI It soon becomes apparent that scientific opinion is divided regardin the ability of AI to substantiate its claIms The early AI pioneers were premature in declaring their ability to convert the mysteries of human intelligence to machine form Although the people currently at the forefront of AI development communicate a more cautious and realistic outlook the recent transformation of AI into a prime business venture has subjected the discipline to the exaggerations of commercial advertising While Ai is still in development significant contributions have already come from AI research and technology AI professionals are quick to AI PROCESSES AND APPLICATIONS recognize that the potential of machine capabilities has barely been tapped The next decade should AI software has developed in channels largely prove critical to the question of whether AI is able unknown to conventional programming languages to prove itself to the rest of the scientific Instead of using essentially linear reasoning AI community as the concerted efforts of government programs endeavor to imitate the flexibility of the universities and industry seek to overcome the human mind LISP developed in 1958 is the most current limits of computer technology popular programming language used by AI experts in the United States Short for List rrocessing ACKNOWLEDGMENTS LISP allows the organization ofdiita in networklike structures Items ofdata may be descriptive statements lists or many other data groupings Frames and scripts are two popular ways of storing information Frames contain slots into which data is inserted scripts have a similar structure including a series of scenarios into which the data might be expected to enter The hardware associated with AI falls into two categories computers designed especially for AI usage and AI adaptations of other computer systems Advertisements describing AI systems reveal AI-related innovations use of windows and a mouse pointer to allow greater flexibility in examining the relationships between data sets the adaptation of computer graphics to capture the essence of symbolic processing and a variety of developmental tools to assist the programmer in formulating programs that explOIt all of the networking potential of AI languages Many of these features are being used in areas other than those associated specifically with AI One AI application that is now receiving much attention is the development ofexpert systems This is described in another article in this issue Mar-Apr 1986 P L 86-36 Books and Articles Used I t'A rtificial Intelligence Expert Systems NSA briefing 1985 McCorduck Pamela Machines Who Think A Personal Inquiry into theHistory and Prospects of Artificial Intelligence San Francisco W H Freeman 1979 Mishkoff Henry C Understanding Artificial Intelligence Dallas Texas Texas Instruments 1985 Waldrop M Mitchell Artificial Intelligence Science reprint from 24 February 1984 23 March 1984 27 April 1984 15 June 1984 10 August 1984 and 14 September 1984 0 CRYPTOLOG page FOR OFFIOh L USB ONLY 9 ero 4009980 P L 86-36 'fOFSECRE'f LEVELS OF CLASSIFICATION U 1 r05 SAO U Reco ized classification levels and the factors involved In establishing them are as follows 'f'S 660 TOP SECRET shall be applied only to information or material when its unauthorized disclosure could reasonably be expected to cause excentionallv 2Tave damae e to the national securitv 1 8 eeO SECRET is used for information or material the unauthorized disclosure of which could reasonably be eXfected to cause serious damaQ'e to the nationa securitv 1 FOR OFFICIAL USE ONLY U The caveat FOR OFFICIAL USE ONLY FOUO is a unique markin in that although it can be used to preclude pubhc dissemination of certain information it is not in itselfa level of security classification The rules governing the use ofthis handling instruction can be quite confusing so we often decide or are told by others to mark an item FOUO without understanding its meaning the impact ofits use or whether it may be legally used in a given context U Certain information that has not been given a security classification under the criteria for assignment of CONFIDENTIAL SECRET or TOP SECRET maybe marked FOUO and withheld from the public domain under Freedom of Information FOI Exemptions Additionally at NSA such unclassified information is generally covered by Public Law 86-36 the National Security Act of1959 Section 6 of this act states nothing in this act or any other law shall be construed to require the disclosure of the organization or any function ofthe National Security Agency of any information with respect to the activities thereof or of names titles salaries or number of the persons employed by such Agency Thus this Act provides FOR OFFICIAL USE ONLY protection ofour organizational designators unclassified missions and functions budgetary information and personnel strength Only information that has not been previously available to the public may be marked FOUO U Included among the FOI exemptions permitting FOUO protection are D CONFIDENTIAL is the appropriate classification when the unauthorized disclosure of the information or material in question could reasonably be expected to cause identifiable damae e to the national securitv I U UNCLASSIFIED material will not normally be marked stamped UNCLASSIFIED unless i I is essential to advise the recipient ofthe material that it was examined in order to assign a classification and was then determined not to require one 1 Regulations orders manuals directives and instructions relating to the internal personnel rules or ractices of a DoD component if such publicatIOn or inspection would not affect the general public or if release to the public would hinder effective performance of DoD functions 2 Internal advice recommendations and subjective evaluations as contrasted with factual matters that are reflected in records pertaining to the decision-making process of an agency or DoD component 3 Information in personnel and medical files and personal information in other files that if disclosed to the public would result in clearly unwarranted invasion of personal privacy 4 Investigative records compiled for the purpose ofenforcing civil criminal or military law U Please remember that properly used the FOR OFFICIAL USE ONLY caveat can afford us a great deal of protection against the public disclosure of unclassified information that is not in the public interest 0 Mar-A p r 1986 CRY PTOLOG page 10 EO 1 4 c P L 86-36 T9P SI3 JRS UAlofBLE VI A eO JPYffif'T' OUItP'mB bS Q TkJPY - aCID 4009980 CONFIQEN'J'IAL THE 39 STEPS U P L 86-36 _ _ _ _ _lt13 This article was originally published in the December 1980 issue of the T15 Technical Bulletin very Few populated areas in the new building A limited amount of production was running on the system since all customers programmers etc U In 1957 NSA was scattered around the were located at Arlington Hall and Nebraska Washington Area at Arlington Hall Virginia at Avenue The 704 was one of the most powerful Nebraska Avenue in D C and in barracks at Ft computers on the market From an operational Meade Maryland The Agency had acquired a new view it was a simple machine It ran one program building today OPS 1 but it was not quite ready at a time The operator's function was to load the for occupancy The Agency had also acquired a program it was on cards into the card reader set new computer - a second mM 704 The decision any necessary sense switches there were 6 mount was made to install the new machine at Ft Meade and ready any required tapes check the printer rather than at ArlinJ rton Hall because that and press the program load button complex was schedufed to move Having installed U The computer room was a sterile place In the machine it seemed a prudent thing to run it addition to the computer its furnishings consisted One Friday in October 1957 at 11 30 p m of two government issue gray desks one of which was placed in front of the machine console a gray D The Baltimore-Washington Parkway was table a gray coat rack and a few gray chairs all of unusually dark that night The fog masked the which sat on a gray raised floor This exotic color lights of the few southbound cars and even the combination was offset by a pile of brown burn warm glow from BINDERS which at this time bags and a brown Frieden Calculator circa 1812 which sat on the gray table The atmosphere had yet to succumb to the arsonist was not to be always convinced the operator that it was raining seen The old Studebaker seemed especially reluctant to pursue its course to Route 32 The He was constantly amazed when at the end of the scene was something out of Rod Serling's THE shift he would emerge into sunshine TWILIGHT ZONE though in this case more aptly it was the MIDNIGHT ZONE Eventually the u It was 1 a m The only sound in the room was Studebaker groaned to a stop in front of Gatehouse the aching drone of the air handler punctuated by a periodic clacking of the printer Sitting in front of 3 Carrying his brown bag and thermos the the console it was hard for the operator not to be operator passed by the nodding Marine guard and mesmerized by the lights He had heard of an descended into the basement The eve shift delayed its departure only long enough to tell its operator who got some erotic pleasure from the replacement that his partner on the shift had console lights but hard has he tried he could receive only soporific sensations He was running called in sick The promise of a long lonely night KNOLL IXC Actually the machine was running was about to be fulfilled KNOLL IXC It was a production job and it had U The 704 occupied some portion of space which been running for weeks It was the only production job or any other kind ofjob available today belongs to the CAP area The area was a restricted area and in fact at the time was one of a Mar-Apr 1986 o CRYPTOLOG o page 11 CONfo'IBRN'f'IAL tI A MfiLE Vi2 e8 HU'P OIhUHfEJ S OUbY c aCID 40099'80 4 P L 86-36 GONFIBBN'f'IAL dump and had produced changes to put the printer output on tape also At 4 a m he was ready to implement his change He loaded the program It began It read the pattern cards from tape It cycled It Nroduced output ON THE CARD PUNCH U The operator recovered The original version of KNOLL IXC was back in production His version was in the burn bag Tomorrow would be better It was time to shut down u It was too early for 'lunch' yet the emptiness of the room induced a similar sensation in his stomach The bologna sandwich potato chips and the coffee consumption took 30 minutes It was now 2 a m only 6 hours to go There were three projects available One track down the cat that everyone agreed was somewhere under the false floor Previous efforts at this hunt had proven fruitless The operator decided against the project The cat he was sure had its own secret entrance to this secure area U The second project was the repair of the Frieden Calculator This electro-mechanical marvel had not worked since the day it came over from Arlington Hall The fall from the truck probably had something to do with its condition The operator and his partner had decided to repair it and in so doing had ended with 15 parts left over Its present condition however was no worse i e it still didn't work The operator vetoed this task since the hammer had been missing the last two nights and such a tool was needed for the delicate repair procedure U The third project was do-it-yourself programming U In retrospect the 50's were an extraordinary time for learning about computers The Agency was in the forefront in the acquisition of hardware ABNER ATLAS and the IBM 700 705 and 704 were on board But where were the people the mechanics who knew how to use them They did not exist As a result the Agency embarked on a course ofdo-it-yourself It provided its populace with a million-dollar machine a manufacturer's manual and said Learn It and they did U The operator's partner had spent some time looking at a dump ofthe KNOLL IXC program He had determined where the program read the cards He produced changes which read the pattern cards from tape rather than the card reader The effect was that after doing a load card to tape the operator never had to service the program for the entire shift except for the printer Not to be outdone this operator too had been looking at the U While there was a great deal of compute time available the number of applications programmed were not enough to justify weekend work on overtime The Agency was thus in the ideal posture for processing on demand a posture never to be seen again Since it was the weekend Saturday morning the operator began the shut down procedure He powered down the tape drives the printer the main frame He placed all tapes all cards all listings in the oversize shopping cart He secured the area and pushed the cart down the deserted roadway to the elevator U U At the time when the building had yet to be approved for beneficial occupancy there were very few secure areas In the North end first floor there was established a vault room for the computer operations activity It had a 3-way combination lock U The elevator refused the operator's call Try as he would it would not come What to do The operator was glad he had been lifting weights Pulling the loaded cart up the 39 steps to the first floor was not in hisjob description but he was not enough of a PMM lawyer to know it It was a thing that had to be done U It was now 8 o'clock The darn combination didn't work Here he was the fate of the free world in his shopping cart He couldn't open the vault Once again my age my father's age the playboy centerfold's age Would it work It worked U He left It was not raining Depositing his empty brown bag and his empty thermos beside him he urged the Studebaker to start It did As the car staggered down the B W Parkway the operator wondered if he had done any good U KNOLL IXC made a major contribution to what the Agency is all about It is a fact of Hfe then as now that our operators are prime movers in successes which they know not of 0 CRYPTOLOG invites readers to submit vignettes ofhistorical interest Mar-Apr 1986 o CRYPTOLOG o page 12 GONFIBBN'f'IAL IIANDfJEl ViA eOMIN't' OiIA UtElfJS ONLY 4009980 PUZZLE PAGE Solution to CRYPTARITHM MORAL DISQUISITION Jan-Feb 1986 Find the five-letter anagram to fill all blanks 1 234 5 6 7 to K R Y P T 0 S in Deo so the saying goes o 0 o 3 2 4 067 5 8 0 C I E T Y 5 7 9 8 1 6 4 One cannot ----- it though the sense one knows However after mon ----- one day When ----- and port had both been cleared away 9 0 388 3 9 Number the columns left to right 1-7 Note that no two letters refer to the same digit Column 1 shows that C is greater than S Hence column 7 shows that Y CoS Now column 1 shows that there must have been a carry from column 2 so that 1 K S C so Y K 1 I fell to musing would God ----- a Cain Who -- -- the heads of men instead of grain At jousts the knight with ----- to win a prize Since column 2 produces a carry P in that column cannot be 9 Therefore column 4 shows that P equals O ----- his opponents down to half their sizes Column 4 introduces no carry into column 3 since Y K 1 C must equal 9 There must be a carry into column 2 so R 0 9 And column 1 informs that S 8 K With him is damned the conqueror To murder also he who ----- and burns ooooooo Reprinted from the ASA Review Vol 1 No 4 July-August 1950 Hence K cannot equal 0 4 8 or 9 An assumption of any of the remaining six values for K gives tentative assumptions YandS Letter 0 cannot equal 0 9 K Y S or 9-S sO there remain four possible values after a K assumption Then R 0-9 column 6 gives T K O column 5 produces no carry so E is less than I and must be the smaller of the two digits still unaccounted for Solution to NSA-CROSTIC No 61 A 1 Salemme Guide to Russian Technical Translation NSA 1974 The translator must determine the precise type of rotary-wing aircraft being referred to in Russian as BEPTOJIET so that his rendition will conform to precise English usage He must not avoid the issue by translating it in formal context as chopper or eggbeater Jan-Feb 1986 P16-MAY86-Y12-105312 C P K I IKe The six K assumptions and the four 0 assumptions make for just 24 things to try far fewer than the 10 possible permutations often letters Continued analysis or a computer program leads to the unique solution PERKY STOIC Alternatively knowledge that the correct arrangement constitutes a phrase would quickly eliminate wrong assumptions e g ifK equaled 1 the phrase sould have the unpromising beginning PK CRYPTOLOG page 13 FOR OFFICIAL US E ONt f This document is from the holdings of The National Security Archive Suite 701 Gelman Library The George Washington University 2130 H Street NW Washington D C 20037 Phone 202 994-7000 Fax 202 994-7005 nsarchiv@gwu edu