UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting MATHEMATICAL REPRESENTATIONS OF JUDGMENT AND DECISION MAKING IN MILITARY CONTEXTS M ARC C ANELLAS *, K AREN F EIGH , P H D, AND R ACHEL H AGA C OGNITIVE ENGINEERING C ENTER G EORGIA I NSTITUTE OF T ECHNOLOGY *P RESENTER , MARC . C . CANELLAS @ GATECH . EDU “ S IMULATION OF M ULTIPLE S TRATEGIES WITHIN A D ECISION P ROCESS M ODEL : A P ATHWAY TO I MPROVED D ECISION S UPPORT , ” O FFICE OF N AVAL R ESEARCH , C OMMAND D ECISION M AKING P ROGRAM , #N00014-14-1-0136
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting OBJECTIVE: BETTER MODELING, SIMULATION, AND SUPPORT OF MILITARY OPERATORS 1. Military operators use naturalistic heuristics (simple decision algorithms, pattern recognition, etc.) based on their experiences and expertise to make quick and accurate decisions – especially when faced with limited time, information, resources, or cognition. 2. Our new general linear model of judgment and decision making can mathematically and transparently represent these types of strategies (and more) while accounting for expertise and incomplete information. 3. Leveraging the perspectives of naturalistic heuristics and the mathematics of the GLM will enable new ways to • Model and simulate military operators • Develop prescriptive strategies for better performance • Design support tools and interfaces
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting DEVELOPMENT, MODELING, AND SIMULATION OF NATURALISTIC HEURISTICS 1 N AT U R A L I S T I C D E C I S I O N M A K I N G + FA S T - A N D - F R U G A L H E U R I S T I C S 1 Keller et al., 2010
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting NATURALISTIC DECISION MAKING “ God does not play dice. ” “ Neither do people [in the wild] …” • People do not generate and compare option sets • People use prior experience to Cognitive Continuum 1 rapidly categorize situations • People rely on a synthesis of their Skill ’ s, Rules, and experiences Knowledge-Based Behavior 2 • People do not passively await outcomes of their gambles and Naturalistic Decision Making 3 bets; they actively shape events 1 Hammond et al., 1987; 2 Rasmussen, 1983; 3 Klein, 2008; Lipshitz et al., 2001
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting SOLDIERS AND COMMANDERS USE AND ARE TAUGHT NATURALISTIC DECISION MAKING Single-Option Model: Multiple-Options Model: Evaluate the first COA and use if Evaluate 3 courses of action (COAs), satisfactory (else reevaluate situation compare each, then select the best. and generate a new COA) COA 1 COA 1 COA 1.1 COA 2 vs vs COA 3 Multiple-attribute utility analysis Based on Recognition-primed decision making Natural and comfortable strategy 1 ; Rarely fully-implemented Used by military In 2014 US Army Field Manual 3 Reduced planning by 20% 2 or 30% 1 ; Research results Better COAs 1 ; Degrades under time-pressure; Requires advanced programming while Limitations Not fully-representative still not implementing all aspects 1 Ross et al. , 2004: Fort Leavenworth Battle Command Battle Lab; 2 Thundholm, 2003: Members of Swedish Staff Officer Program
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting FAST-AND-FRUGAL HEURISTICS • Heuristics: Simple search, stopping, and decision rules • People use the bounds on rationality of simplicity, speed, and frugality as a mechanism for simple, robust, and accurate decisions Question • Example: Fast-and-frugal trees - 1 + • Binary predictors and outcomes No Question • One exit at each level, two at final level 2 - + • Derived from data and qualitative methods 1 or from simplifying random forests 2 Question No Accurate and robust 3 in military, finance, and • 3 - + medical domains No Yes • Transparency, consistency, and communicability lead to more acceptance than actuarial methods 4 1 Katsikopoulos et al., 2008; Martignon et al. , 2008; 2 Deng, 2014; 4 Katsikopoulos et al., 2008; Elwyn et al., 2001; Green & Mehr, 1997;
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting SOLDIERS AND COMMANDERS USE AND ARE TAUGHT FAST-AND-FRUGAL HEURISTICS Escalation of Force 1 in Vehicle Checkpoints 2 Hostile act (continued) Increase ? Level of Yes Force No Nothing Use Force Escalation of Force Level Description 1 – SHOUT Warning to stop 2 – SHOW Weapon and intent to use it 3 – SHOOT Warning shot at ground/air 4 – SHOOT One shot placed in grill/tires 5 – SHOOT Destroy vehicle/eliminate threat 1 Bagwell, 2008 2 Keller and Katsikopoulos, 2014
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting MILITARY OPERATORS USE NATURALISTIC HEURISTICS Military operators often use simple decision algorithms, pattern recognition, etc. based on their experiences and expertise to make quick and accurate decisions – especially when faced with limited time, information, resources, or cognition. Recognize Generate Hostile act Increase Level (continued)? of Force No Yes Modify Evaluate Nothing Use Force Act
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting A GENERAL LINEAR MODEL OF JUDGMENT AND DECISION MAKING W H AT I F A L I N E A R M O D E L C O U L D R E P R E S E N T T H E S E N AT U R A L I S T I C H E U R I S T I C S ?
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting SIMULATING NATURALISTIC DECISION MAKING – LIMITED BY COMPLEXITY • Simulating naturalistic decision making has often required complex computations • Fuzzy logic 1 , episodic recognition memory 2 , system dynamics 4 • Designers and users prefer the simpler analytic models because they are easier to use despite their difficulty to account for environmental and psychological issues. 4 There are many shared components with fast-and-frugal heuristics 5 , so can ’ t we use the computational models of the fast-and-frugal heuristics? 1 Ji et al., 2007; 2 Mueller, 2009; 3 Patterson et al., 2009; 4 Katsikopoulos and Fasolo, 2006; Katsikopoulos et al., 2008; 5 Keller et al., 2010
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting SIMULATING HEURISTIC DECISION MAKING – LIMITED BY LACK OF UNIFIED MODEL 1 Take-the-Best Incomplete information Take Two Cue weights Estimates of missing information Cutoff values Weighted-Additive Effort Redundancy Tallying Minimalist Distribution of Weights Dominance DEBA Equal-Weighting Expertise and Experience Utility Functions Thresholds Variability Predictability Time Pressure Cue directions Strategies Environmental Parameters Components Context 1 Two major books from the ABC Research Group which developed the fast-and-frugal heuristics research program: Gigerenzer et al., 1999; Todd et al., 2012
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting SIMPLE DECISION MODELS – CLASSES OF JUDGMENT AND DECISION MAKING PROBLEMS • Utility of any option is defined by the function: 𝑉 𝑌 = 3𝑏 1 𝑌 + 2𝑏 2 𝑌 + 𝑏 3 (𝑌) • Two options with cue values: a 1 A = 1; a 2 𝐵 = 0; 𝑏 3 𝐵 = 1 ⟶ 𝐵 = 1,0,1 a 1 B = 0; a 2 𝐶 = 0; 𝑏 3 𝐶 = 1 ⟶ 𝐶 = {0,0,1} Decision Making: Judgment: Select the option with the highest utility. Categorize individual option. 𝑏𝑠𝑛𝑏𝑦 𝑗 { 𝑉 𝐵 = 2, 𝑉 𝐶 = 1 } 𝐼𝑗ℎ, 𝑉 ≥ 2 𝐷𝑏𝑢𝑓𝑝𝑠𝑧 = 𝑀𝑝𝑥, 𝑉 < 2 𝑽 𝑩 = 𝟒 = 𝑰𝒋𝒉𝒊 𝑽 𝑩 > 𝑽 𝑪 → 𝑩
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting TOWARD A GENERAL LINEAR MODEL Recognize Generate 𝑜 Components Strategies Modify Evaluate 𝑤 𝐷 𝑗 = 𝑥 𝑘 ∙ 𝑉 𝑘 𝑏 𝑗,𝑘 Environmental Context Parameters 𝑘=1 Act Diversity and Processes of Simple linear models component structure of naturalistic decision from decision theory fast-and-frugal making heuristics
UNCLASSIFIED – Marc Canellas – Dec. 6, 2016 MORS Experimental Techniques Special Meeting GENERAL LINEAR MODEL – BINARY FORM 1 𝒐 𝒘 − 𝒇 𝒌 𝒜 𝒋,𝒌 𝑫 𝒋 = 𝒙 𝒌 ∙ 𝑰 𝒌 𝒆 𝒌 𝒇 𝒌 + 𝒃 𝒋,𝒌 − 𝒆 𝒌 𝒅 + 𝚬 𝒌 𝒌=𝟐 Cue direction (d): Incomplete Criterion (C): Cutoff value (c): Sign of the information overall score of an Threshold correlation (z): 1 if option comparison for between the scores known, 0 if binary. Cue weight (w): and criterion unknown relative importance of the cue Estimate of Threshold ( 𝚬 ): The missing info difference between cue Heaviside utility function (e): Assumed values is large enough to (H): Convert cue values into value of a be meaningfully different 1 ’ s or 0 ’ s missing piece of information 1 Canellas and Feigh, 2016
Recommend
More recommend