ARTIFICIAL INTELLIGENCE STEERING GROUP TRIP REPORT

Document Type: 
Collection: 
Document Number (FOIA) /ESDN (CREST): 
CIA-RDP86M00886R000500050006-9
Release Decision: 
RIPPUB
Original Classification: 
U
Document Page Count: 
62
Document Creation Date: 
December 21, 2016
Document Release Date: 
August 22, 2008
Sequence Number: 
6
Case Number: 
Publication Date: 
May 10, 1984
Content Type: 
REPORT
File: 
AttachmentSize
PDF icon CIA-RDP86M00886R000500050006-9.pdf3.12 MB
Body: 
Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 UNCLASSIFIED The Director of Central Intelligence Washington, D.C. 20505 Intelligence Research & Development Council ARTIFICIAL INTELLIGENCE STEERING GROUP TRIP REPORT DATE/PLACE 10 May 1984 MITRE Corporation McLean, Virginia AISG STAT In lieu of our May meeting, the AI Steering Group visited the Washington C3I Operations of MITRE Corporation:--.:.in McLean, Virginia. The all-day session consisted of presentations, demonstrations, and discussions of several MITRE AI projects of potential interest to the Intelligence Community. The visit was hosted by Mr. George Steeg, Chief Engineer of MITRE's C31 Operations activity. An agenda is attached. The first speaker, described and demonstrated STAT the ANALYST system, a prototype expert system for intelligence fusion and situation assessment. The ANALYST system employs three -UNCLASSIFIED Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 levels of reasoning involving Tactical Intelligence Reports, Order of Battle Information, and Analysis of Critical Indicators. It uses (incomplete) information from COMINT, ELINT, and IMINT in a common report format. The system operates with the Battlefield Environment Model, an object-oriented simulation (written in ROSS on the VAX) in which all units act according to some pre-established doctrine. MITRE has invested 10 staff-years in develo in the ANALYST system, and so far it has 120 "rules." stimated that STAT an operating version might grow to 1500 rules which would require moving the system from a VAX to a 2MB Lisp Machine. The system has a menu-oriented user interface and can ask the user to supply data and confidence factors. The mathematics of how to combine confidence factors most effectively is still under consideration. There is a rule editor to facilitate adding or changing rules and an explanation capability to explain how the system arrived at an answer. A set of viewgraphs is attached. More detail is available from MITRE in a report on ANALYST: MTP-83W00002, February 1984. talk addressed the subject of Military STAT Planning. Planning systems generally are characterized by: A hierarchy of plans Deferred time ordering Introspection and modification of plans Deferred introduction of constraints. Military planning (and to a large extent business planning) is characterized by: No "right" plan Not able to choose the "best" plan Unknown constraints and data Erroneous constraints and data "Commander's prerogative." MITRE has developed a"system for logistics planning, but it, like most automated planning efforts, suffers the following deficiencies: Cannot produce partial plans Difficult to trace the results of following a plan Untimely response Hard to check= plans Not able to develop concurrent plans Difficult to modify plans or monitor their execution. MITRE has an IRAD project underway to address these deficiencies using an AI approach. They are using a joint military operation scenario which has been supplied by the Armed Forces Staff College. Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 UNCLASSIFIED Dr. Benoit suggested that there maybe intelligence applications of this planning approach in the area of indications and warning (either to detect deviations from an expected scenario or to fit a behavorial model to the observed data). discussed human factors in conjunction with the STAT design of expert systems. While acknowledging the importance of human factors issues in designing the end-user interface, Dr. Stech focused his talk on the process of knowledge acquisition and the need to systematically question knowledge before it g into a system. Several methods and tools were described and STAT emphasized a process called "chunking" of knowledge. Experiments have verified that chess masters "chunk" situation reports about a chess board status that enables them to reproduce a game board rapidly and accurately. MITRE is investigating how tactical analysts perceive a situation map and is setting up an analogous set of experiments to determine the key features of a map that an expert focuses on. A set of viewgraphs is attached. next discussed the subject of photo STAT interpretation and demonstrated .a PI softcopy workstation environment. The particular application task demonstrated was the recognition of significant changes of man-made objects in images of the same location taken at different times. The goal of this project is to develop a laboratory and a set of tools (as opposed to building a complete system) which can be used to experiment with developing aids for image analysts (not full automation). Future work will involve adding 3-D models, using shadow information, applying it to urban areas, and studying the use of specialized image processing hardware. A set of viewgraphs is attached. MITRE also has AI work going on at their Bedford, Massachusetts, operation, and described four AI projects at STAT MITRE/BEDFORD. The KNOBS project (which the AISG saw demonstrated at Symbolics in Cambridge last Fall) provides assistance in tactical air mission planning-. The KNOBS approach employs natural language processing, rules, frames, and templates. The general architecture has been reapplied to several other planning areas for NASA and the Navy. MITRE has also been working in the field of Natural Language (question-answering, inferencing, and command interpretation). Their current interests are moving toward multiple ways of presenting answers: using natural language and_:_g_aphics. Their Knowledge-Based Automatic Programming project employs an application knowledge base to facilitate rapid prototyping and debugging of a system. This project is similar to RADC's work in this area but is, perhaps, broader in scope. Next, described an Expert STAT System they are building for NASA to support mission GO/NOGO decisions by analyzing and validating sensor indications which monitor the liquid oxygen supply. Finally, he talked about their Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Future Generation Computer Architecture project. This work involves Multiple Instruction/Multiple Datastream machines and a "smart" memory interconnect scheme to link multiple processors (up to, perhaps, 1 million). summarized several general research issues which the STAT Bedford projects are attempting to address. Developing a KNOBS Replanning System - Resource allocation - Monitoring missions - Achieving interaction among missions Disambiguating natural language text and maintaining a conversational context Achieving incremental program specifications changes, removing temporary inconsistencies, and deciding which rule to apply Controlling mechanized reasoning for efficiency and for naturalness . Permitting continuous replanning . Monitoring the results of a plan. The last speaker was who presented a prototype STAT expert system for resolving aircraft conflicts (a project for the Federal Aviation Administration). The system, which was demonstrated, uses information which predicts violations of safe separation policies up to 30 minutes in advance. Twenty-seven expert rules are then applied to determine for the air traffic controller suggested strategies for resolving the forthcoming conflict (e.g., plane,A ascend and plane B turn right). The rules embody a set of doctrinal constraints about such things as right-of-way, restricted air space, and general aviation principles. A set of viewgraphs is attached. Executive Secretary AI Steering Group STAT Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 AI Briefing 10 May 1984 CAHIS Briefing Room Wilson Building MITRE C31/Washington Expert Systems for Intelligence Fusion and Situation Assessment The battlefield environment is used as.a model to develop expert systems to fuse and assess the dynamic information available. Adaptation to the changing environment and automatic correction of rules are being investigated. A demonstration will be presented. Artificial Intelligence Applied to Military Planning A demonstration Expert System has been developed which applies rules to the logistic planning of crisis responses. Operating bases and transportation facilities may be selected in this system. A more powerful system is under development and will be discussed. STAT STAT 11:30 STAT reinforce their strengths). Human, Factors Design of Expert. Systems The objective of this research is to develop a methodology to assess the strength and weaknesses of human experts and to use this information in the design of expert systems. A secondary objective of this research is to develop methods to test the value added by the system, (i.e., the degree to which the expert system does, in fact, compensate or correct the weaknesses of human experts and Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 The efforts discussed will include tactical planning, natural language understanding and automatic programming. AIRPAC - Advisor For Intelligent Resolution of Predicted Aircraft Conflicts An Expert System uses heuristic rules to develop and evaluate a list of feasible actions to avoid aircraft conflicts. These proposed solutions are presented to the air traffic controller for selection. A demonstration will be presented. Computer Vision as a Photographic Intelligence Aid This talk will briefly describe MITRE's current effort to adapt Computer Vision technology to aid the photo-interpreter, in a softcopy workstation environment, to identify the first signs of man-induced change of potential significance. Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 STAT STAT STAT Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 AN EXPERT SYSTEM FOR RESOLUTION OF AIRCRAFT CONFLICTS COMPUTER DEMONSTRATION MITRE/METREK IR&D 95690 PROJECT TEAM: CURTIS SHIVELY KARL SCHWAMB JILL NICOLA Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 WHY CONSIDER KBS TECHNIQUES FOR ATC? 0 ATC REQUIRES COMPLEX DECISIONS, CURRENTLY PERFORMED BY HUMAN EXPERTS 0 L'IiMITATIONS;;)OF ALGORITHM-BASED COMPUTERIZED DECISION AIDS - DIFFICULTY ESTABLISHING RELATIONSHIP TO HUMAN DECISION PROCESS - DIFFICULTY EXPLAINING RATIONALE UNDERLYING OUTPUT DECISIONS - SOME DIFFICULTY IN TESTING, CHANGING, MAINTAINING PROGRAM 0 KBS TECHNOLOGY ALLOWS EXPERT SYSTEM AUTOMATION.OF HUMAN DECISION PROCESS DATABASE OF ABSTRACT. DECISION RULES - COMPUTER MAKES INFERENCES FROM INPUTS AND RULE BASE - USER-FRIENDLY INTERFACE PROVIDES JUSTIFICATION FOR ANSWERS Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 PURPOSE OF CONFLICT RESOLUTION EXPERT SYSTEM - GIVEN: 0 PREDICTED CONFLICTS (POTENTIAL VIOLATIONS OF SAFE SEPARATION) BETWEEN AIRCRAFT PREDICTED 15-30 MINUTES IN ADVANCE FROM FLIGHT PLAN DATA CHARACTERIZED BY GEOMETRY, AIRCRAFT ALTITUDE, SPEED, INTENT, ETC. PROVIDE: 0 RESOLUTION ALTERNATIVES IN ORDER OF PREFERENCE - AIRCRAFT TO BE DEVIATED - TYPE OF MANEUVER, INCLUDING PARAMETERS AND TIME OF EXECUTION EXPLANATION OF RATIONALE UNDERLYING RECOMMENDATIONS Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 CONSIDERATIONS PERTINENT TO RESOLVING CONFLICTS Conflict Description Conflict situation Geometry--acosslsg,'`beadon. ovectfte. merging' Altitude coals--btgb er low Relative position (boils and wort). and velocity Timeldistance to point of first violation Neighboring aircraft Aircraft-specific Intent (at conflict and later)--accival, eta. speed (general maneuverability) Maneuver status (at conflict) turning. ate. Type--)et. piston, turboprop Category--OA. air cast/er, ate. around speed Metered? In queue[ Constraints on solution Rnviconsental Terrain winds and severe weather Bperial-use sic Paces Traffic flows. including I(-M altitudes. etc. Neighboring A/C ATC costclctions--$VA, IQA. eta. A/C-Bpecitic service limits--maximum altitude Performance limits--climb Cate, speed. etc. Pilot ability to comply (aircraft weight. etc.) A/C not subject to maneuvers Interlocks on maneuvers vs. currant maneuver Rules of the air Goals of solution Reduce number of conflicts Don't preclude fallback solution it this one Calls Preserve tight-of-way policy Maneuver A/C not in arrival flow or metering queue Minimise commands, no more vectors than one afloat at dogleg ' Preserve long-tots intent--route bends. descent. etc. Fuel efficiency Preserve arrival ordering consistent with speed ordering minimize new delay consistent with metering goals Try solutions in order of preference for given conflict Rationale should be relatively transparent to controller Provide lateral. long.. vent., or radar type separation Insensitivity to wind and A/C performance data uncertainty Avoid risky solutions requiring controller monitoring Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 ORGANIZATION OF CFR KNOWLEDGE 0 FACTS--DESCRIBE,CONFLICT SITUATION EXPERT RULES--GENERATE LIST OF RESOLUTIONS IN ORDER OF PREFERENCE 0 CONSTRAINTS--REJECT OR REORDER POSSIBLE SOLUTIONS BASED ON OTHER FACTORS - HIERARCHICAL ORDERING OF CONSTRAINTS -- SHOWS RELATIVE IMPORTANCE OF CONSTRAINT VIOLATIONS -- ALLOWS RELAXATION AND ACCEPTANCE OF MILD VIOLATIONS Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 FORM OF RESOLUTION EXPERT RULES 0 GENERATE LIST OF RESOLUTIONS (AIRCRAFT AND MANEUVER) IN ORDER OF PREFERENCE 0 INPUT FACTORS, TYPE OF CONFLICT (CROSSING, HEADON, MERGING, OVERTAKE) INTENT (ARRIVAL, DEPARTURE, OVERFLIGHT) SPEED TIME-TO-CROSSING (CROSSING CONFLICTS ONLY) - TIME-TO-CONFLICT (CROSSING CONFLICTS ONLY) - CLIMB RATE (DEPARTURE INTENT) 0 POSSIBLE MANEUVER CHOICES ARRIVAL--SPEED REDUCTION, RESTRICTED DESCENT, VECTOR q I DEPARTURE--RESTRICTED CLIMB, VECTOR OVERFLIGHT--SPEED REDUCTION, VECTOR, CLIMB/DESCEND TO NEW ASSIGNED ALTITUDE Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 FORM OF RESOLUTION EXPERT RULES (CONCLUDED) 0 TOTAL OF 27 RULES FOUR CONFLICT TYPES SIX INTENT COMBINATIONS (ARRIVAL-DEPARTURE, ETC.) FURTHER SUBDIVISION OF CROSSING OVF-OVF, MERGING OVF-OVF, AND OVERTAKE ARV-ARV CASES 0 EXAMPLE EXPERT RULE IF (CONFLICT-TYPE CROSSING -AC1 -AC2) AND (INTENT DEPARTURE.-AC1) AND (INTENT OVERFLIGHT -AC2) THEN (FACT *POSS* ((RESTR-CLIMB -AC1)(VECTOR -AC1) (VECTOR -AC2)(CHANGE-ALT -AC2))) 0 MANEUVER EXPANSION VIA SINGLE MAINTENANCE RULE (VECTOR -AC) EXPANDS TO (RIGHT-TURN -AC)(LEFT-TURN -AC) (CHANGE-ALT -AC) EXPANDS TO (CLIMB -AC)(DESCEND -AC) Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 RATIONALE FOR CURRENT SET OF CONFLICT RESOLUTION EXPERT RULES 0 CROSSING CONFLICT (AIRCRAFT IS PRIMARY FACTOR) AIRCRAFT--PREFERENCE BASED ON INTENT, IF SAME, MANEUVER THE ONE FARTHER AWAY (ARV, OVF) OR SLOWER CLIMBER (DEP) - MANEUVER--PREFER ALTITUDE CHANGE OR'RESTRICTION TO ACHIEVE CLEAR VERTICAL SEPARATION UNTIL AFTER HORIZONTAL ;::,`.CROSSING, EXCEPT PREFER SPEED REDUCTION, IF OVF-OVF AND TIME-TO-CONFLICT GREATER THAN 20 MINUTES 0 HEADON CONFLICT (MANEUVER IS PRIMARY FACTOR) - MANEUVER--PREFER VECTOR TO GAIN CLEAR-LATERAL SEPARATION - AIRCRAFT--PREFERENCE BASED ON INTENT, IF SAME. MANEUVER FASTER A/C 0 MERGING CONFLICT (AIRCRAFT IS PRIMARY FACTOR) - AIRCRAFT--PREFER SLOWER ONE TO ACHIEVE NECESSARY SPEED ORDERING - MANEUVER--PREFER VECTOR TO MERGE SLOWER IN BEHIND FASTER AIRCRAFT -;..,EXCEPTION--IF OVF-OVF AND TIME-TO-CONFLICT GREATER THAN 20 MINUTES PREFER SPEED REDUCTION FOR FASTER A/C 0 OVERTAKE CONFLICT (AIRCRAFT IS PRIMARY FACTOR) - AIRCRAFT--PREFERENCE BASED ON INTENT. IF SAME, MANEUVER FASTER AIRCRAFT - MANEUVER--PREFER VECTOR FOR CLEAR LATERAL SEPARATION, EXCEPT PREFER SPEED REDUCTION, IF SMALL SPEED DIFFERENCE (ARV, ARV) Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 CONSTRAINT RULES CURRENTLY USED IN CFR EXPERT SYSTEM 0 HIERARCHICAL ORDERING SEVERE (FOUR RULES) -- DON'T CLIMB/DESCEND IF SOMETHING ABOVE/BELOW -- DON'T LEVEL OFF IF DEPARTURE/ARRIVAL AND SOMETHING BELOW/ABOVE MODERATE (FIVE RULES) -- AVOID TURNING RIGHT/LEFT IF SOMETHING TO RIGHT/LEFT -- AVOID CLIMBING AIRCRAFT IF WITHIN 3,000 FEET OF SERVICE CEILING -- AVOID TURNING RIGHT/LEFT IF TIME-TO-CONFLICT LESS THAN 7 MINUTES - MILD (FOUR RULES) -- UNDESIRABLE TO MANEUVER AIRCRAFT IN METERING QUEUE -- UNDESIRABLE TO TURN RIGHT/LEFT IF ROUTE BENDS LEFT/RIGHT -- UNDESIRABLE TO CLIMB IF INTENT CHANGE OVERFLIGHT TO ARRIVAL 0 REORDERING OF SOLUTIONS - NO VIOLATIONS, ONLY MILD, AT MOST MODERATE, SOME SEVERE - ORIGINAL PREFERENCE ORDER IS PRESERVED AMONG SOLUTIONS WITH GREATEST VIOLATION IN SAME CATEGORY Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 CAPABILITIES OF CURRENT SOFTWARE IMPLEMENTED ON COMMAND AND MANAGEMENT INFORMATION SYSTEMS (CAMIS) LAB- 0 EXPERT SYSTEM FOR RESOLVING CONFLICTS APPLIES RULES TO FACTS DESCRIBING A TWO-AIRCRAFT CONFLICT SCENARIO RECOMMENDS SOLUTION ALTERNATIVES IN ORDER OF PREFERENCE 0 EXPLANATION OF RESULTS - WHICH RULES WERE ACTIVATED - HOW GIVEN FACTS LED TO INDICATED SOLUTION 0 GRAPHICAL PRESENTATION OF CONFLICT SCENARIOS - ROUTE AND ALTITUDE PROFILES ON RAMTEK COLOR GRAPHICS DISPLAY 0 RULE BASE DEVELOPMENT AIDS DEPICT RULE INTERACTIONS DUE TO COMMON INPUT AND OUTPUT CLAUSES - ALLOW RULE EDITING DIRECTLY IN LISP ENVIRONMENT Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 PHOTO INTELLIGENCE PROBLEM TOO MUCH IMAGERY ? TOO FEW TRAINED PHOTO INTERPRETERS AND INTELLIGENCE ANALYSTS ? COMPUTER VISION IS GOOD, HOW CAN IT BE USED: ? IN THE SOFTCOPY WORKSTATION ENVIRONMENT TO INCREASE PRODUCTIVITY RATHER THAN DISTRACT THE VIEWER ? PROBLEM DOMAIN ? FY84: DETECT NEW MAN-MADE STRUCTURES IN THE WILDERNESS ? FY85: CONSTRUCT BLUEPRINTS/MAPS/MODELS OF NEW BUILDINGS/TOWERS/SHIPS Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 SOLUTION IN FY84 WE EXPECT TO HAVE A SET'OF GENERAL PURPOSE COMPUTER VISION SOFTWARE TOOLS THAT CAN BE USED TO DETECT NEW MAN-MADE STRUCTURES IN THE WILDERNESS FROM PAIRS OF DIGITIZED AERIAL PHOTOS. IN FY85 WE PLAN TO USE MANY OF-THESE TOOLS TO COMPARE NEW IMAGERY WITH SYNTHETIC IMAGERY PREDICTED FROM A 3-DIMENSIONAL SCENE MODEL, FURTHERMORE, WE PLAN TO USE THE CHANGES WHICH ARE DETECTED TO UPDATE THE SCENE MODEL, Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 IMAGE PROCESSING TECHNOLOGIES IMAGE ENHANCEMENT (IMAGE IN - BETTER IMAGE OUT) USED TODAY --------------------------------------------------------------------------------- NOT USED TODAY IMAGE PROCESSING - PATTERN RECOGNITION (FEATURE EXTRACTION) COMPUTER VISION ASSIMILATION, MATCHING OF FEATURES TOWARD OBJECT RECOGNITION ARTIFICIAL INTELLIGENCE SCENE INTERPRETATION I RELATIONSHIPS BETWEEN OBJECTS IN A SCENE Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 TECHNICAL CHALLENGES OF IMAGERY DATA ? DIFFERENCES DUE TO VIEWING GEOMETRY ALTITUDE ? VIEWING ANGLE ? LAND TOPOGRAPHY ? DIFFERENCES DUE TO LIGHTING,EFFECTS ? SUN ANGLE ? SHADOWS ? GLARE BACK SCATTERED LIGHTING ? DIFFERENCES DUE TO OCCLUSION ? BUILDING - BUILDING ? BUILDING - GROUND ? CLOUD COVER Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 FY84 APPROACH NEW DIGITIZED IMAGE SOBEL EDGE OPERATOR MARYLAND EDGE THINNER HOUGH LINE FITTER GENERATOR PARAMETERS - GAIN - DECISION - CONTROL SOBEL EDGE OPERATOR MARYLAND EDGE THINNER ILLINOIS HOUGH GRAPH LINE oj~ MATCHER FITTER COLOR TINT Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 ADAPTIVE TRAINING EACH TOOL HAS SEVERAL GAIN/DECISION/CONTROL PARAMETERS ? DIFFI`CULT TO TUNE BY HAND FOR A GIVEN CLASS OF IMAGERY REQUIREMENTS ? GROUND TRUTH ? ERROR ANALYSIS PROGRAM ? OPTIMIZER OPTIMIZER Error Parameters Data TOOL Ground Truth ERROR ANALYSIS Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 FY85 APPROACH NEW IMAGERY EDGE DETECTOR EDGE THINNER DISPLAY OF 3-D MODELS COLOR TINTER MATCHER - FY84 SOFTWARE - FY85 SOFTWARE - DATA EDGE SHADOW PREDICTOR NEW EDGES RECONSTRUCTED OLD IMAGERY INTERACTIVE INPUT EDGE CONSTRAINT ENFORCEMENT OLD 3-D MODEL OLD MODEL MODEL E EDITOR Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 NEW MODEL Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 FY84 TECHNOLOGY ? EDGE OPERATOR ? LOW SENSITIVITY TO LIGHTING/VIEWING ANGLE ? FACTOR OF 10 DATA REDUCTION ? FAILS;TO DETECT 5% OF ACTUAL "GOOD" EDGES ? DETECTS "EXTRA" EDGES ? EDGE THINNER ? REMOVES 90% OF "EXTRA" EDGES PRODUCED BY THE EDGE DETECTOR ? FACTOR OF 3 DATA REDUCTION ? REMOVES 5% OF THE "GOOD" EDGES ? LINE FITTER ? SEPARATES MAN-MADE OBJECT EDGES FROM NATURAL EDGES ? RESYNTHESIZES SOME OF THE LOST "GOOD" EDGES ? FACTOR OF 5 DATA REDUCTION ? SNOW/CLOUD COVER DOES NOT PRODUCE STRAIGHT LINES ? GRAPH MATCHER ? DETECTS DIFFERENCES (IN THE STRAIGHT LINES PRODUCED FROM MAN-MADE OBJECTS) ? LOW SENSITIVITY TO EXTRA/LOST EDGES Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 ADDITIONAL TECHNOLOGY ? DISPLAY OF 3-D MODELS WITH HIDDEN. SURFACE REMOVAL ? SOFTWARE (MODIFIED MOVIE BYU) ? INPUT:. 3-D MODEL OUTPUT: IMAGE ? EDGE/SHADOW PREDICTOR ? ALGORITHMS EXIST (MUST BE PROGRAMMED) ? INPUT: 3-D MODEL OUTP-UT: EDGES ? EDGE CONSTRAINT ENFORCEMENT ? ALGORITHMS EXIST (MODIFIED STRAIGHT LINE FITTER) ? INPUT: EDGES OUTPUT: STRAIGHT/CURVED LINES ? FITS EDGES TO LINES AND CURVES ? MAKES EDGES PARALLEL, VERTICAL, EQUAL LENGTH, ETC. ? MODEL EDITOR ? PARTIAL SOFTWARE (MODIFIED GRAPH EDITOR) ? INPUT: INTERACTIVE + STRAIGHT/CURVED LINES + MODEL ? OUTPUT: NEW MODEL ? CONSTRUCT SIMPLE SOLID MODELS FROM PARALLEL EDGES ? CAN COMBINE TWO SOLID MODELS TO FORM A NEW SOLID MODEL Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 EXPECTED RESULTS ? FACTOR OF 10 INCREASE-IN PHOTO INTERPRETER-TASK- THROUGHPUT ? A MITRE DEMONSTRATION SYSTEM THAT USES COMPUTER VISION TOOLS TO ASSIST AN INTELLIGENCE ANALYST TO INTERACTIVELY UPDATE A 3-DIMENSIONAL MODEL FROM NEW IMAGERY ? SYNTHESIS OF IMAGERY FROM OBLIQUE ANGLES NOT AVAILABLE TO THE.SENSOR Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 GRADIENT BASED EDGE DETECTORS ? ESTIMATE THE GRADIENT OF THE IMAGE FUNCTION USING DISCRETE SPACIAL CONVOLUTION ? HYPOTHESIZE THAT EDGES PASS THROUGH POINTS WHERE THE MAGNITUDE OF THE GRADIENT VECTOR Q a,b IS LARGER THAN A THRESHOLD t. ? SOBEL T - 1=-n 1=-n U = a ? PARAMETERS r, a, t 2 0 1 0 Vi,1 P (a-i), (b-1) 1 2 1 0 0 0 -1 -2 -1 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Ui,J P (a-i ), (b'=1) T I i=-n 1=-n Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 HOUGH TRANSFORM AND STRAIGHT LINE FITTING ? PARTITION SPACE INTO n by n CELLS ? HOUGH TRANSFORM x, y r, 8 r = x? cos9 + Y. sing ? ACCUMULATOR ARRAY ? THRESHOLD t ? PARAMETERS n, t, L~ 8 Straight Line r Detected Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 FEATURE GRAPH MATCHING Input > Feature Graphs PROTOTYPE FORMATION DISTANCE RECALCULATION NORMALIZATION CONSTRAINT ENFORCEMENT OUTPUT PRUNING Relaxation Loop Output Matched/Unmatched Features Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 HUMAN FACTORS DESIGN OF EXPERT SYSTEMS IR&D UPDATE Frank Stech Joseph Schmu l l e r The MITRE Corporation 10 May 1984 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 HUMAN FACTORS DESIGN OF EXPERT SYSTEMS ? Issue: Expert Systems (ES) are designed to capture, reproduce.. and distribute human expertise repackaged as computer programs. ? Problems: How can the ES designer capture human expertise accurately, completely, and efficiently? How can the ES designer assess components of human expert i?se to improve subobt ima l components? How can the ES designer reassemble the human and enhanced components to represent expertise precisely? Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 TECHNIQUES AND TOOLS ? Expert Modeling Methods: (Applicable to I R&D) Expert-Novice Comparisons Expert Protocol Analysis Information Processing Experiments Simulations Field Observations Developmental. Studies Cross-cultural Comparisons Historical Comparisons Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 TECHNIQUES AND_ TOOLS (CONCLUDED) ? Tools Chronometric measures (e. g., reaction time, p rocess i ng;,,k.t ime, chunking) Multidimensional categorization analysis (e. g, scaling, clustering) Subjective probability analysis (e.g., Non-Bayes i a n probability revision) Problem-solving (e.g., accuracy, sequences, relationships) Multiple regression (e.g.,, input-output trace, information integration modeling) Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 PROTOTYPE ASSESSMENT: TACTICAL MAP BATTLEFIELD READING ? Methodology: Chunking (Chess Experts) applied to tactical maps Copying maps Reconstructing maps from memory Estimating classes and overall situation ? Stimuli Meaningful, semi-meaningful, and anomalous tactical maps Varied map symbology; e.g., unit symbols, unit designation, unconventional symbols, chess symbols ? Data: I me rp l acement times, glances, gaze, symbol placement order, protocol Is Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Battlemap Appears on Screen for 2 min. Subject Studies _.-._--- Battlemap Battlemap Disappears 'Copy Map' & Palette of Features Appear Subject Reconstructs as much of Battlemap as he can, using Copy-Map, Palette and Light-Pen Subject presses a key, Copy-Map and Palette disappears, original Battlemap reappears, for a maximum of one minute Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 SQ FQ S & F 0 V Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 I I I I I Elements Correctly Recalled vs. Trials -- F Q Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Elements Correctly K-suilea vs. Trials -- F a S Q Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Types of Elements Correctly Recalled vs. Trials -- -0 Types of Elements Correctly Recalled vs. Trials -- F & S Q a Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 a Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 ^ 19 11 El Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 PILOT STUDIES ? Study 1: Map Reading Attention affects chunking Chunks reflect pattern recogfn i t i on - Memory/Learning curves ? Study 2: Map Assessment Chunks relate to categories Categories relate to summaries Summaries relate to assessments Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 PILOT STUDIES (CONCLUDED) ? Study 3: Summary Composition - Report chunking - Cova riat ion assessment - Availability effects Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 EXPERIMENTAL SERIES- Tactical battlemap reading: What are the basic elements and relationships? ? Tactical battlemap assessment: What is the transfer function from-symbolic map input to verbal/textual assessment output? ? Tactical summaries composition: Now are tactical reports transformed and integrated into tactical assessments and summaries? ? Map and report integration and assessment: How are tactical reports integrated with ba tt l emap data and transformed into situation assessments? ? D namic assessment: How does chang inj information (reports and maps impact on analyst s assessments? Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 . ANALYST A Knowledge-Based Intelligence Support System JTF Demonstration Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 What Is ANALYST? ANALYST is an Intelligence Support System. . . ? Targeted for Army Corps and Allied Tactical Operations Centers (Air Force) ? Supports All Source Sensor Fusion and Primary Information Request Processing ? Is Capable of Dynamic Refinement of Findings from New Data ... Composed of Integrated Expert .System Techniques ? Both Data Driven and Request Driven ? Uses Both Hierarchical and Spatial Data Structures ? Employs Procedural, Propositional, and Iconic (Image) Knowledge Structures ? Maintains Truth Justifications ? Includes Knowledge-Engineering Tools Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Data Bases Expert Systems Triggers Updates Control Mechanism Knowledge Bases ? Use a Knowledge-Base of -Rules-of-Thumb Provided by Experts ? Capture Fundamental Principles of the Problem Domain as Well as General Methodology ? Solve Complex Problems Well ? Can Interact Intelligently with the User ? Are Useful for Interpretation, Diagnosing, Predicting, Analyzing, Consulting Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 KVM Knowledge Engineering PIRIIR Handler ANALYST Architecture Questions Truth Maintenance Answers/Critical Information Gaps Sensor Fusion Module Intelligence Reports Sensor Control Tasking Battlefield Environment Model (BEM) Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Commander G2 .,Terminal SITMAP Display 'Collection Management . Terminal , Scenario Controller Terminal Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Sensor Coverage Symbology Moving Target Indicator (MTI) COMINT ELINT Counter Mortar/ Counter Battery (CMCB) IMINT (Photo) Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 MITRE Threat Modeling ? Experimental Boundaries - Soviet/WP Division from SCORES - Company/Battery Resolution - Stylized Terrain ? Emphasis on Activities - Movement - Emissions ? Behavior Modeling - Object-Oriented Programming - Activities are Rule-Based Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 MITRE Sensor Modeling ? Current Experimental Boundaries - Corps/Division Orientation - Continuous Resource Availability Optimal Communications - No Formal Models ? Emphasis on Geographic Coverage - Platform Flight Paths, Time-on-Station - Ranges, Coverage Angles ? Interactive Tasking -- Sensor Control Station - View "TAKE" - Reposition Sensors Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Level 3 (?) Level 2 (FY84) Level 1 (FY83) Level 0 (FY83) Threat Units & Hierarchies Levels of ANALYST Reasoning Activity Groups SSensor Reports Establish Activity Groups - _ - - -~~ ~ Analyze \% 1 `I Threat l4.40.1 1 ~ t Intentions Analyze Logical Critical )4-10) Indicator Indicators J Network Merge Rules Refinement Rules Templates Purge Rules Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 MITRE Level 3 (?) Level 2 (FY84) Level 1 (FY83) Level 0 (FY83) Levels of ANALYST Reasoning Analyze t t 1t I ~---t Threat IMI Intentions I I Threat Units & Hierarchies Analyze Logical Critical H Indicator Indicators Network Hypothesize Single r e M R Sensor Reports Establish Activity Groups t ou c erge u es Pattern Refinement Rules Rules Templates Purge Rules Single Source Cluster Rules Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Results With 35% coverage ? ANALYST locates over 70% of the active enemy battalions ? Accuracies average 750 meters circular error probable (CEP) ? General force movement trends are detectable ? System tracked individual movement of 15% of the regimental and battalion command posts With 20% coverage ? Over 50% of the active battalions located ? General force movement trends still discernible Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 MITRE ANALYST System Characteristics Capabilities ? Knowledge-Based . - Can View On-Going Analytical Processes - Can Edit Rule-Base On-Line - Straightforward Architecture - Keys on Patterns of Activity ? Processing - Stays Abreast of. Battlefield Dynamics,,,-, - Real-Time Speed - Multiple-Source - Does Not Require a Mainframe Environment Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9 MITRE ANALYST System Characteristics Limitations ? Idiot-Savante Nature - Totally Dependent on "Expert" Rules - Does Not Introspect ? Single Scenario Development ? Needs Operational Knowledge Instead of Demonstration Knowledge Approved For Release 2008/08/22 : CIA-RDP86M00886R000500050006-9