Download presentation
Presentation is loading. Please wait.
Published byれいな たておか Modified over 6 years ago
1
Irregular Warfare Analysis and Validation with the Social Impact Model
Dr. Deborah Duong Mr. Gerald Pearman CPT Richard Brown
2
analysis and validation
Purpose and Agenda Purpose: To show the usefulness of the Social Impact Model (SIM) for Irregular Warfare (IW) analysis and validation Agenda SIM background. Practical applications of the SIM. SIM’s enabling technologies. Proof of concept application to Tactical Wargame (TWG). Results. Summary. Social Impact Model
3
SIM Background Social Impact Model
4
Background The Social Impact Model (SIM) performs adjudication, analysis and validation of social impact in US Army TRADOC Analysis Center’s (TRAC) Tactical Wargame (TWG). TWG and the models within the SIM are focal points of TRAC’s Irregular Warfare Analysis Capability (IWAC) initiative. IWAC is the one of the largest IW analysis efforts in the Department of Defense. The SIM is a federation of stand alone models and tools. The Cultural Geography (CG) model at the Population level (TRAC-Monterey). The Nexus Network Learner Model (NNL) at the Individual level (OSD,TRAC- Monterey). More models to be added in 2011. SIM middleware integrates and analyzes the models. SIM middleware extends the eXtensible Behavioral Model (XBM) framework (OUSDI, TRAC-Monterey, CTTSO). XBM implements basic versions of TRAC’s IW analysis designs. The models and tools of the SIM are designed to solve problems in IW adjudication, analysis, and validation. Each SIM component applies advanced artificial intelligence technologies to solve problems. Social Impact Model
5
SIM Components The models that comprise the SIM take a principled approach to IW modeling, enabled by advanced technologies. CG and NNL both include: Cognitive Agent Based Modeling so that social phenomena is computed from first principles and is emergent rather than hard coded. Bayesian Networks so that data can be read into and tracked in the model in a flexible manner. Reinforcement Learning so that agents learn new behaviors from motivations over time, the true causes of new social structure, rather than being pre-determined towards desired structures. Social Networks to emphasize the relational aspect of social structure. Representation of the major schools of social theory (interpretive, materialist). SIM middleware applies TRAC’s analysis and validation methods to the models, including: Combinatorial Game Theory. Probabilistic Ontologies (with logical and Bayesian inference engines). Information Theory for model comparison and validation. Markov Processes. Social Impact Model
6
Practical Applications of the SIM
Social Impact Model
7
SIM enables you to… Apply wargame data to rigorous simulation analysis. Model an IW game of perceptions. Translate disparate data between IW models. Social Impact Model
8
SIM Can Apply Wargame Data to Rigorous Simulation Analysis
SIM enables wargames to be run multiple times. Most IW analyses incorporate wargames, but human-in-the-loop wargames are resource intensive and cannot span the realm of possibilities. Risk based analysis requires that the analyst considers the realm of possibilities and their likelihoods. Models can cover the possibilities if, when they are run out many times stochastically, moves are entered as a human player would enter them. Human players respond to enemy moves and to the environment. SIM enters moves using human player strategies from an actual wargame. Both interviews and statistics on actual moves are used. SIM can apply player strategies to disparate models. Strategies in SIM are represented separately from model moves and then translated to models. SIM can also translate strategic moves to models that were not used in the wargame. SIM offers a rigorous method to improve wargames. SIM enables measurement of the measurement space. SIM models player perception, enabling analysts to test the effect of what human players can (or cannot) observe. SIM exposes how to “game the game” . Identifies design errors in models, helping analysts to improve models for adjudicating human games and for multiple stochastic runs. Social Impact Model
9
SIM can model an IW “Game of Perceptions”
SIM enables the modeling of IW moves and countermoves. SIM turns multiple model runs into games of strategy. Kilcullen described IW conflict as “co-evolutionary,” capturing the importance of moves and countermoves. His phrase, “Perception is Reality” depicts the idea that IW is a game of perception. The movie, “The Battle of Algiers,” captures the importance of the timing of moves and countermoves in IW showing how insurgents went from unpopular to widespread public support by forcing the hand of the host nation and making the nation appear oppressive. SIM enables the modeling of deception in Information Operations. Automated players have mental models of the world that may be incorrect, so that one automated player may deceive another. SIM enables the testing of IW doctrine and strategy. Commander’s Intent, Goals, Decision Points, Branches and Sequels, can be tested and compared. Social Impact Model
10
SIM Translates Disparate Data between IW Models
SIM enables analysts to apply IW data from one unique situation in the world to a different but unique situation in a model. Input data, data traded between federated models, calibration data, and testing set data all have uncertainty of match as well as translation difficulties because of the different concepts used to arrange different data. SIM translates data to ensure one model “means the same thing” as another model. If there is no exact match, SIM translates data probabilistically. SIM enables hybrid modeling. Analysts can couple models “loosely” , at the level of general patterns, or tightly, at the level of details. SIM can integrate models at different levels of aggregation, for example, “tactical” and “operational/strategic”. SIM enables data to be compared “apples to apples”. Validation needs to be at the level of statistical patterns rather than single outcomes. The real world is just one possible world, and simulations model many possible worlds. We know a simulation is good if what is rare in the simulation is rare in the world, and what is common in the simulation is common in the world (under right circumstances). SIM enables analysts to express data in statistical patterns and dynamics. After SIM translates data to a common lexicon, SIM compares data at the level of statistical patterns and dynamics to calculate a ‘distance’ score that measures the statistical distance between dynamic patterns. SIM’s distance score allows comparison of different versions of the same model, different scenarios, and models against data for a calibration or a validation score. SIM can validate a model against multiple real-world data sources with uncertain matches to the model. Social Impact Model
11
SIM’s Enabling Technologies
Social Impact Model
12
SIM Applies Advanced Technologies to IW Problems
Combinatorial Game Theory models strategic games. Probabilistic Ontologies arrange and translate data. Information Theory finds relations in data and compares data. Markov Processes help analyze and validate models. Social Impact Model
13
Combinatorial Game Theory (Game Trees)
SIM uses Strategic Data Farming (SDF), an application of combinatorial game theory to optimize player moves according to player’s goals and strategies to assess best courses of action (COA). Role players are modeled as automated agents that look ahead to results of moves assuming players are trying to achieve goals, in a simple, general cognitive model. Social Impact Model
14
Strategic Data Farming (SDF)
SDF models aspects of the Military Decision Making Process (MDMP). Automated role players have the following as part of their COA strategy: Decision Points: Points at which players will consider a change to its COA (nodes of the game tree). COA Options: Options to consider at each decision point, specifying conditions under which it will be exercised and possible moves (branches of the game tree). Goals: Ways to evaluate the situation within the move selection algorithm. Goals can also be interpreted as measures of effectiveness (MOEs) (game tree leaf evaluation). Mental Models: Presumed strategy of other players (i.e., belief levels, decision points, COA options, move selection). Social Impact Model
15
Game Tree Example – Africa Use Case
Popular support levels (from Nexus): G = 0.57 R = 1.0 Evaluation Function GE – (each side attempts to maximize their evaluation function): GE = ((1-R)+G)/2 = 0.28 RE = 1-GE = 0.72 Disrupt alliance between tribe J and tribe D. Conduct Civil Affairs. Green Action 1. 2. GE: 0.5 GE: 0.35 (after looking ahead) GE: 0.5 GE: 0.25 (after looking ahead) Make tribe O, a green ally, appear to harm tribe J. Make green appear to harm tribe J. Red Reaction 3. 4. 3. 4. RE: 0.35 RE: 0.65 RE: 0.75 RE: 0.5 Without looking ahead, Green’s actions seem the same (both are .5). But by looking ahead to how Red would react, he finds action Disruption (action 1) (GE=1-.65=.35) is better than CA (action 2) (GE=1-.75=.25). Social Impact Model
16
Proof of Concept Application to TWG
Social Impact Model
17
Player Strategy, Goals, Decision Points
As a first step to determine player strategy/intent, goals, decision points, and moves, the study team assessed all role player surveys and actual moves. The matrix below is an excerpt of the assessment. The study team identified over 20 viable strategies from the surveys and moves. Role Player Intent (from surveys and actual moves) Goal (Measurable evaluation function in SDF based on goal). Source is surveys or most likely common sense from knowledge of TWG. Decision Point (measurable) Moves Don't Do Moves Perceptions CF Company Security Maximize Green and Blue OAB and decrease Red OAB Green OAB decreases one step or if Red OAB increases one step. Provide Security Provide Legal Red: Red trying to disrupt infrastructure, intimidate the population and prove that Blue and Green cannot protect population centers. Dismounted Patrol Hero Payments Condolence Payments Green OAB remains the same or Red OAB remains the same. The study team also assessed point-wise mutual information (PMI) scores to determine strategy, goals, and decision points. The study team calculated PMI scores to determine the co-occurrence of kinetic or non-kinetic strategies and popular support levels. Social Impact Model
18
Using Point-wise Mutual Information (PMI)
PMI is a concept from Information Theory. PMI tells how uniquely a sign, such as popular support score, is associated with another sign, such as an action. We used PMI to tell how players actually reacted to popular support scores. PMI quantifies the discrepancy between the probability of coincidence of two outcomes given their joint distribution and the probability of their coincidence given only their individual distributions, applying the equation1: Tally of Green Moves versus Green popular support category. Y-variable Move Categories (# of moves) Blue kinetic Blue non-kinetic Green kinetic Green non-kinetic Red kinetic Red non-kinetic Totals X-variable popular support observations green OAB is PA 1958 776 2734 green OAB is PP 450 112 562 green OAB is NS 200 2674 2874 green OAB is NP 151 753 904 green OAB is NA This category never occurred. 2759 4315 7074 Resulting PMI Scores [-1, +1]. -1 = never occurs together, 0 = independent, +1 = complete co-occurrence. Pointwise mutual information told us how uniquely related moves were to certain popular support levels and changes in those levels. Our PMI analysis revealed that players only looked at their own popularity, and not at anyone elses. Y-variable PMI Scores Blue kinetic Blue non-kinetic Green kinetic Green non-kinetic Red kinetic Red non-kinetic X-variable green OAB is PA green OAB is PP green OAB is NS green OAB is NP 1 Social Impact Model
19
Scenario Definition Scenario 1 – TWG 2010 implementation.
Scenario 1 mimics TWG 2010 implementation. Specifically, blue, green, and red role players attempted to maximize their own popularity. Blue and green wanted to become as popular as possible. Red was satisfied not to become unpopular. Scenario 2 – Excursion. Same as TWG 2010 implementation, except blue attempted to maximize green popularity vice blue popularity. Scenario 1. Role Player Goal Decision Point Strategy Blue Maximize blue popularity Blue popularity increases Kinetic Action Blue popularity decreases Non-kinetic Action Green Maximize green popularity Green popularity increases Green popularity decreases Red Maximize red popularity Red popularity increases Red popularity decreases Note that one’s popular support is just an indicator and we now have a framework to measure the measurement space by studying the effect of player access to different indicators. We do so, with a set of runs that mimics the twg 2010 player strategies, where each player attempted to maximize their own popularity, and a set that does an excursion where blue maximizes greens popularity rather than its own. Scenario 2. Role Player Goal Decision Point Strategy Blue and Green Maximize green popularity Green popularity increases Kinetic Action Green popularity decreases Non-kinetic Action Red Maximize red popularity Red popularity increases Red popularity decreases Social Impact Model
20
Iterative Hub and Spoke Architecture
Hybrid Model Hybrid Model Inference Engine Inference Engine Pave/CG Mediation Ontology (Inference Engines: Probabilistic Inference (Bayesian networks) Logical Inference (Jena Micro OWL)) Pave/Nexus Mediation Ontology (Inference Engines: Probabilistic Inference (Bayesian networks) Logical Inference (Jena Micro OWL)) Updated Indicators Pave Hub Ontology Updated Indicators CG Move CG Adjudication Nexus Adjudication Nexus Move CG Ontology (Cultural Geography Model) Nexus Ontology (Nexus Model) Legend: Input / Output Ontology Social Impact Model
21
TWG 2010 Probabilistic Ontologies
CG ontology. Defines CG moves. Nexus ontology. Defines Nexus moves. PAVE ontology. Hub ontology for model. Contains PAVE moves and role player strategies, goals, and decision points. PAVE CG Mediation ontology. Performs dynamic translation of PAVE tasks to CG moves. PAVE Nexus Mediation ontology. Performs dynamic translation of PAVE tasks to Nexus moves. Tactical Wargame 2010 ontology. Maintains states of the automated role player, such as OAB level/popularity and state of individual move. Multi-Resolutional Bayesian ontology. Defines the macro and micro agents that are used to integrate multi-resolutional models. TEO ontology. Defines events and outcomes. Design of Experiment (DOE) ontology. Abstracts the concept of strategies, goals, and decision points in a doctrinal manner. ProbOnt ontology. Holds the representation of the Bayesian networks that determine selection of events and outcomes. Pakaf ontology. Holds the moves to the Helmand/PAKAF scenario. PakafCgMediation. Automatically translates CG TWG moves to Helmand/PAKAF moves. Social Impact Model
22
Probabilistic Ontology Relationships
DOE TEO PAVE ProbOnt MultiResolutionalBayes PaveCgMediation ProbOnt MultiResolutionalBayes PaveNexusMediation CG Nexus Inheritance Hierarchy and Relationship Structure of SDF Probabilistic Ontologies Social Impact Model
23
Logical Inference Indicators in Action
The logical inference engine (Jena Micro OWL) classified when decision points were triggered. Yellow portion in the ontologies below indicate inferred states for coalition forces player, calculated from indicator definitions. Social Impact Model
24
Implementation of Event Probabilities
The study team implemented a probabilistic translation from the moves of one model to another using Bayesian networks. SIM added Bayesian Networks, such as the one below, directly to the probabilistic ontology representation. Social Impact Model
25
Probabilistic Inference in Action
The example below illustrates two probabilistic translations from the same PAVE move (‘CS_CF’ – cordon and search by coalition force). Micro agent #3 generated three CG events from the PAVE task, whereas micro agent #1 generated two CG events from the same PAVE task. Social Impact Model
26
Results Social Impact Model
27
Scenario 1 (TWG 2010). Blue Moves/Green Popularity
Level of Violence (blue player actions) K = Kinetic M = Medium Kinetic N = Non-Kinetic Popularity of Green P = Popular U = Unpopular Level of Violence Kinetic Medium Kinetic Non-Kinetic .50 .43 .10 K P M P N P Start Scenario .14 .10 .20 .07 .70 Unpopular Popular Level of Green Support .36 .50 .40 .10 .29 .14 K U M U N U .21 .40 .36 9/17/2018 Social Impact Model
28
Scenario 2 (Excursion). Blue Moves/Green Popularity
Level of Violence (blue player actions) K = Kinetic M = Medium Kinetic N = Non-Kinetic Popularity of Green P = Popular U = Unpopular Level of Violence Kinetic Medium Kinetic Non-Kinetic .14 .10 K P M P N P Start Scenario .10 .50 .14 .28 .70 Unpopular Popular Level of Green Support .50 .14 .10 .71 .14 K U M U N U .57 .14 .28 .43 Probabilistic Distance from Scenario 1: 0.09 9/17/2018 Social Impact Model
29
Validation – Markov Processes from Model and Real World
Popularity of Green P = Popular U = Unpopular Level of Violence K = Kinetic M = Medium Kinetic N = Non-Kinetic State Definition Model State Transitions Real World (Afghan Nationwide Quarterly Assessment Review) Apply Kullback-Leibler (KL) divergence to measure probabilistic distance between Markov processes: Score of 0 means exact same Markov process. Score of 1 means the most different Markov process possible. Example above generated KL score = 0.21. Social Impact Model
30
Summary Social Impact Model
31
SIM IW Analysis and Validation Capabilities
Flexibility to define state spaces (i.e. type of actions by role player –popularity level) that align with the study questions based on doctrine and cognitive-based measurement spaces. Ability to run out the wargame with perception based moves and countermoves keeping track of the likelihood of outcomes for risk based analysis. Ability to examine model dynamics vice simply end of run output via Markov processes to assess tipping points. Ability to compare data based on statistical patterns vice single outcomes. Flexibility to translate, compare, and surrogate data. Social Impact Model
32
Questions and Comments
POC: Deborah Duong Social Impact Model
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.