MODELING THE FUNCTIONAL ARCHITECTURE OF HUMAN DECISION MAKING - - PowerPoint PPT Presentation

MODELING THE FUNCTIONAL ARCHITECTURE OF HUMAN DECISION MAKING

(11RH08COR) PI: Dr. Robert E. Patterson (AFRL)

Senior Personnel: Dr. Alan Boydstun

• Dr. Christine Covas-Smith
• Dr. Lisa Tripp

AFOSR Program Review: Cognition, Decision, and Computational Intelligence Program (Jan 28-Feb 1, 2013, Washington, DC) Distribution A 88ABW-2013-0162; CLEARED on 16 Jan 2013

Modeling Human Decision Making (Patterson)

Objective: Understand functional architecture of human decision making and determine the degree to which that architecture will support decision priming DoD Benefit: Increased knowledge about appropriate designs

• f

human-machine systems, training systems, decision-support systems Technical Approach: Combine analytical and intuitive decision making within a behavioral double- factorial paradigm; stochastically model data and provide formal tests of functional architecture Budget:

Actual/ Planned $K

FY11 FY12 FY13

200/245 230/280 250/292

Annual Progress Report Submitted? Yes No No Project End Date: 9/30/2013 Distribution A

List of Project Goals 1. Combine analytical and intuitive decision making in

• ne paradigm; calibrate paradigm

2. Determine the functional architecture of human decision making 3. Determine the degree to which early activation of the analytical or intuitive process can prime human decision making 4. Create a system dynamics model of decision priming

Distribution A

Progress Towards Goals

June 2011: FY11 funds received; August 2011: FY11 funds put on contract w L-3 Comm; September 2011: Initiated set up of hardware/software at Wright State University; December 2011: Began collecting preliminary data; January 2012: Successfully combined analytical and intuitive decision making in one paradigm; achieved goal #1 February 2012-present: Determining functional architecture of human decision making; making progress achieving Goal #2 July 2012-present: Determining degree to which early activation of analytical or intuitive process can prime human decision making; making progress achieving Goal #3

Distribution A

BACKGROUND: DUAL-PROCESSING MODEL OF DECISION MAKING

(derived from Evans, 2008; partial list)

References “System 1” “System 2” (Sit pattern recog) (Deliberation) Schneider & Schiffrin (1977) Automatic Controlled Epstein (1994), Epstein & Pacini (1999) Experiential Rational Chaiken (1980); Chen & Chaiken (1999) Heuristic Systematic Reber (1993), Evans & Over (1996) Implicit/Tacit Explicit Evans (1989, 2006) Heuristic Analytic Sloman (1996) Associative Rule based Hammond (1996, 2007) Intuitive Analytic Hogarth (2001) Tacit Deliberative Evans (2008) Implicit Capacity-limited Implicit/intuitive process vs. deliberative (working memory) capacity-limited process

Distribution A

Analytical Versus Intuitive Decision Making

(adapted from Hogarth, 2005; Patterson et al., 2009):

WORKING MEMORY (CONSCIOUS) (Sys 2)

DECISION

PATTERN RECOGNITION (Sys 1)

RESPONSE LONG-TERM MEMORY

ENCODING STIMULUS Distribution A

INTUITIVE DECISION MAKING:

Distribution A

Experiment 1: Methods Finite State Algorithm

(From Knowlton & Squire 1996)

Distribution A

ANALYTICAL DECISION MAKING: Perform mental calculation that requires working memory and deliberation: Pre-trial exposure: 5 numbers (memory set) During trial exposure: 3 numbers matched to memory array (high saliency); or add the 3 numbers and have sum matched to memory array (low saliency)

Distribution A

RT (seconds)

Target Set (2.22 sec) 2 1 4 Memory Set (3.1 sec) 2 1 4 3 6

Analytical Intuitive

(NO OBJECTS) 8 OBJECT SEQUENCE (3.3 sec) Simulated Movement Begins

Trial Structure PRE-TRIAL TRIAL

STOCHASTIC MODELING

(Townsend & Ashby, 1983; Townsend & Nozawa, 1995) Goal #2: Determine computational architecture of human decision-making Double factorial paradigm: During each trial participant will get either: Interaction contrast: meanRTl,l - meanRTh,l - meanRTl,h + meanRTh,h = x, If x = 0, then the system is additive; if x is negative, there is underadditivity; and if x is positive, there is overadditivity Affirmative Affirmative Affirmative Negative Affirmative Negative Affirmative Affirmative Analytical LOW SALIENCE Pres Ab Present Absent Intuitive LOW SALIENCE Analytical LOW SALIENCE Intuitive HIGH SALIENCE Intuitive LOW SALIENCE Intuitive HIGH SALIENCE Analytical HIGH SALIENCE Affirmative Affirmative Affirmative Negative Negative Affirmative Affirmative Affirmative Analytical HIGH SALIENCE Pres Ab Present Absent Pres Ab Pres Ab Present Absent Present Absent

Distribution A

STOCHASTIC MODELING (Cont’d)

Mean RT interaction contrast together with Survivor function interaction contrast will determine functional architecture of decision making system (e.g., parallel or serial system) Survivor function: Additive inverse of cumulative distribution function of reaction time, F(t) Survivor function S(t): S(t) = 1-F(t), …probability that a process has not been completed by some time t (for positive t) Survivor function interaction contrast: Sll(t) - Slh(t) - Shl(t) + Shh(t) = y(t), This expression is calculated at every time bin for which the survivor function is estimated. Concepts of additivity (y(t) is 0), underadditivity (y(t) is negative), and overadditivity (y(t) is positive) apply at any given value of time t.

Distribution A

STOCHASTIC MODELING (Cont’d)

Mean RT Inter Contr Survivor Func Inter Contr Underadd Add Overadd Underadd Add Overadd Parallel Channel Sys; ST Rule X X Parallel Channel Sys; ES Rule X X Serial Channel Sys; ST Rule X X Serial Channel Sys; ES Rule X X (small t) X (other t) Parallel Channel Coactive Sys X X (small t) X (other t) ST = Self-Terminating; ES = Exhaustive Stopping

Distribution A

Baselines: Achieved Goal #1

5 10 15 20 25 30 35 Frequency RT (seconds)

Intuitive Baseline P1 High Saliency

5 10 15 20 25 30 35 Frequency RT (seconds)

Intuitive Baseline P2 Low Saliency

5 10 15 20 25 30 35 Frequency RT (seconds)

Analytical Baseline P2 Low Saliency

5 10 15 20 25 30 35 Frequency RT (seconds)

Analytical Baseline P1 High Saliency

Dual Task (Analytical & Intuitive Signal Present) Achieving Goal #2

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Low High RT (seconds) Analytical Saliency Low High Intuitive Saliency

Trial Structure A-I (1 sec) SOA (0 sec)

• 2
• 1

1 2 3 4

• 3

SOA (seconds) Intuitive Intuitive Analytical Analytical

0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00

• 1

Reaction Time (seconds) SOA (seconds) Dual Task Intuitive Baseline Analytical Baseline

Priming Achieving Goal #3

STOCHASTIC MODELING (Cont’d):

Goal #4: Create a system dynamics model of decision priming

(model from Patterson, Fournier, Williams, Amann, Tripp & Pierce (2012). System dynamics modeling of sensory-driven decision priming. Journal of Cognitive Engineering and Decision Making, in press.)

Distribution A

Analytical Intuitive

Simulated Movement Begins

Dual Task Attentional Switching

Attentional Switching RT (seconds)

Target Set (2.22 sec) 2 1 4 Memory Set (3.1 sec) 2 1 4 3 6 (NO OBJECTS) 8 OBJECT SEQUENCE (3.3 sec)

PRE-TRIAL TRIAL

Analytical Intuitive

Simulated Movement Begins

Hypothetical Dual-Task Attentional Switching

Analytical Signal Present/Intuitive Signal Present

RT (seconds)

Target Set (2.22 sec) 2 1 4 Memory Set (3.1 sec) 2 1 4 3 6 (NO OBJECTS) 8 OBJECT SEQUENCE (3.3 sec)

No Yes Yes No

RT (seconds) Attentional Switching

Analytical Intuitive

Simulated Movement Begins

Hypothetical Dual-Task Attentional Switching

Analytical Signal Not Present/Intuitive Signal Present

RT (seconds)

Target Set (2.22 sec) 4 3 2 Memory Set (3.1 sec) 2 1 4 3 6 (NO OBJECTS) 8 OBJECT SEQUENCE (3.3 sec)

No Yes Yes No

RT (seconds) Attentional Switching

Interaction with Other Groups and Organizations

• Collaborating with Wright State University (with Drs.

Flach and Watamaniuk)

• Collaborating with Washington State University (with Dr.

Fournier)

• Collaborating with AFRL’s Warfighter Interface Research

Division (with Drs. Liggett, Blaha, and Havig)

• Collaborating with AFRL’s Human-Centered ISR Division

(with Maj Stuart Lloyd)

Distribution A

List of Publications Attributed to the Grant

Patterson, R., Pierce, B., Boydstun, A., Park, L., Shannon, J., Tripp, L. & Bell, H. (2013). Training intuitive decision making in a simulated real- world environment. Human Factors, in press. Patterson, R., Fournier, L., Williams, L., Amann, R., Tripp, L. & Pierce, B.P. (2012). System dynamics modeling of sensory-driven decision

• priming. Journal of Cognitive Engineering and Decision Making, in

press.

Distribution A

QUESTIONS?

Distribution A

REFERENCES

Evans, J.S.B.T. (2008). Dual-processing accounts of reasoning, judgment and social

• cognition. Annual Review of Psychology, 59, 255-278.

Fiser, J., & Aslin, R. N. (2002). Statistical learning of higher-order temporal structure from visual shape sequences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(3), 458-467. Fiser, J., & Aslin, R. N. (2001). Unsupervised statistical learning of higher-order spatial structures from visual scenes. Psychological Science, 12, 499-504. Hogarth, R. M. (2005). Deciding analytically or trusting your intuition? The advantages and disadvantages of analytic and intuitive thought. In T. Betsch & S. Haberstroh (eds.), The routines of decision making (pp. 67-82). Mahwah, NJ: Erlbaum. Hogarth, R. M. (2001). Educating Intuition. Chicago, IL: University of Chicago Press. Klein, G. (1998). Sources of power: How people make decisions. Cambridge, MA: MIT Press. Patterson, R., Pierce, B. P., Bell, H. H., Andrews, D., & Winterbottom, M. (2009). Training robust decision making in immersive environments. Journal of Cognitive Engineering and Decision Making, 3, 331–361. Patterson, R., Pierce, B. J., Bell, H. H., & Klein, G. (2010). Implicit learning, tacit knowledge, expertise development, and naturalistic decision making. Journal of Cognitive Engineering and Decision Making, 4, 289-303.

Distribution A

Perruchet, P., & Pacton, S. (2006). Implicit learning and statistical learning: One phenomenon, two approaches. Trends in Cognitive Sciences, 10, 233-238. Reber, A. S. (1967). Implicit learning of artificial grammars. Journal of Verbal Learning, and Verbal Behavior, 6, 855-863. Reber, A. S. (1989). Implicit learning and tacit knowledge. Journal of Experimental Psychology: General, 118, 219-235. Townsend, J.T. & Ashby, F.G. (1983). The stochastic modeling of elementary psychological processes. Cambridge, UK: Cambridge University Press. Townsend, J. T. & Nozawa, G. (1995). Spatio-temporal properties of elementary perception: An investigation of parallel, serial, and coactive theories. Journal of Mathematical Psychology, 39, 321-359. Zsambok, C. E., & Klein, G. (1997). Naturalistic Decision Making. Mahwah, NJ: Lawrence Erlbaum Ass.

REFERENCES

Distribution A

Experiment 1- Baselines, Naïve Participants

RUN INTER-STIMULUS INTERVAL (seconds) INTUITIVE RTs ANALYTICAL RTs N INTUITIVE ANALYTICAL LOW HIGH LOW HIGH 1 0.41 0.44 3.81 3.1 2.92 1.69 3 2 0.27 0.44

• 3.08

3.84 2.59 4 3 0.27 1.13 2.8 2.51 4.01 2.68 7 4 0.41 1.38 3.6 6.3 4.4 4.3 4 5 0.41 1.13 3.83 3.73 4.56 3.56 7

Experiment 1 Method

Stimuli & Apparatus: Structured sequences of

• bjects, positioned along the ground of a

simulated real-world scene, and sequences of numbers presented auditorially. Design: Intuitive task saliency (high or low) x Analytical task saliency (high or low).

• Intuitive task- Forced choice recognition of

structured object sequences vs. random

• bject sequences.

–High saliency-Training and testing objects were consistent –Low saliency- Training and testing objects were inconsistent

• Analytical task- Presentation of a set of five

values followed by a set of three values which must be matched to values in the first set. –High saliency- Second set must be matched, in order, to first set. –Low saliency - Second set must be added together and then matched to first set. Procedure: –Training: Passive observation of 20 structured sequences (15 presentations each). –Baseline Test:

• Intuitive Test- Recognition of novel

structured sequences vs. randomly- generated sequences.

• Analytical Test- Presentation of a set of five

values followed by a set of three values which must be matched to values in the first set. –Test: Intuitive and analytical baseline tasks presented simultaneously. Forced choice recognition of both objects and numbers.

• “OR” response rule-

– Positive response if both stimuli are correct. – Positive response if only one stimuli is correct. – Negative response if neither stimuli is correct.

Experiment 2 Method

Stimuli & Apparatus: Structured sequences of

• bjects, positioned along the ground of a

simulated real-world scene, and sequences of numbers presented auditorially. Design: Stimulus onset asynchronicity, or SOA (-1 sec, 0 sec, +1 sec).

• Intuitive task- Forced choice recognition of

structured object sequences vs. random

• bject sequences.
• Analytical task- Presentation of a set of five

values followed by a set of three values which must be matched, in order, to values in the first set. Procedure:

• Training: Passive observation of 20

structured sequences (15 presentations each).

• Baseline Test:
• Intuitive Test- Recognition of novel

structured sequences vs. randomly- generated sequences.

• Analytical Test- Presentation of a set of

five values followed by a set of three values which must be matched to values in the first set.

• Test: Intuitive and analytical baseline

tasks presented. Forced choice recognition of both objects and numbers.

• “AND” response rule-
• Positive response if both stimuli are correct.
• Negative response if only one stimuli is correct.
• Negative response if neither stimuli is correct.

STOCHASTIC MODELING (Cont’d):

Capacity: comparisons among conditions involving presentation of a single target pattern (intuitive or analytical) with conditions involving presentation of two target patterns (intuitive and analytical). Data across saliency levels are collapsed Townsend and Nozawa (1995): capacity coefficient, C(t) (capacity in combined target- pattern condition versus that of the sum of single intuitive and analytical target-pattern conditions). Coefficient is greater than, equal to, or less than, 1.0 if capacity is super, unlimited, or limited, respectively, at a particular value of t. For survivor function S(t), SA,I(t) = (SA(t) * SI(t))C(t) , where subscript A is analytical target pattern condition, I is intuitive target pattern condition (* is used here as the symbol for multiplication, not convolution). Unlimited capacity: SA,I(t) = (SA(t) * SI(t)), and C(t) = 1.0 Limited capacity: SA,I(t) > SA(t) * SI(t) and C(t) < 1.0 Supercapacity: SA,I(t) < SA(t) *SI(t) and C(t) > 1.0 These expressions deal with survivor functions--additive inverse of cumulative probability density function; thus inequalities are reversed from what would occur with the original cumulative probability density functions

Distribution A

STOCHASTIC MODELING (Cont’d):

Two boundary conditions (Townsend & Nozawa, 1995): Limited capacity, Grice's inequality (Grice, Canham, & Gwynne, 1984) Super-capacity, Miller's inequality (Miller, 1982) With survivor functions, Grice bound and Miller bound reverse themselves: Grice survivor function bound: upper bound above which capacity is very limited; Miller survivor function bound: lower bound below which capacity is extreme super capacity Grice upper bound (limited capacity): Violation occurs when survivor function of combined target-pattern condition is greater than survivor function corresponding to the least single target-pattern condition, SAI(t) = [SA(t) * SI(t)]C(t) > SMIN(t), the latter of which is defined as MIN {SA(t), SI(t)}. (In cumulative probability terms, very limited capacity occurs when cumulative probability in combined target-pattern condition is smaller than cumulative probability associated with maximum responding channel)

Distribution A

STOCHASTIC MODELING (Cont’d):

Miller's lower bound (super capacity): Violation occurs when observed survivor function in combined target-pattern condition is less than sum of survivor functions associated with the two single target-pattern conditions minus 1, SAI(t) = [SA(t) * SI(t)]C(t) < SA(t) + SI(t) - 1, (In cumulative probability terms, super capacity occurs when cumulative probability in combined target-pattern condition is greater than sum of cumulative probabilities in the two single target-pattern conditions minus 1) If Miller's inequality is violated at some value of t, then the system is super-capacity at that t. If the system is everywhere super-capacity, then Miller's inequality will be violated for some values of t. Super-capacity is a feature of coactivated systems.

Distribution A