contextuality in strategic games · 2018. 11. 17. · tp = total probability cooperate after...
TRANSCRIPT
CONTEXTUALITY IN STRATEGIC GAMES
Jerome R. Busemeyer Indiana University
Zheng Joyce WangThe Ohio State University
Supported by AFOSR and NSF
1. Violations marginal invariance for category-decision making task, prisoner dilemma game, and centipede game
2. Quantum cognition explanations violations of marginal invariance
3. Hypothetical example of CBD type contextually for game of chicken.
4. Psychological importance of contextually
ORGANIZATION OF THIS TALK
INTERFERENCE OF CATEGORIZATION ON DECISION
Busemeyer, Wang, Mogiliansky-Lambert (2009, J. of Mathematical Psychology)
Psychological version of a double slit experiment
Wang & Busemeyer(2016, Cognition)
!Categoriza*on,
!
1000/2000ms, 1000ms, 10s,
!Decision,
!
10s,
Feedback,on,C,and,D,
10s,
Participants shown pictures of faces
Categorize as “good” guy or “bad” guyDecide to act “friendly” or “aggressive”
Bad Guys Good Guys
!Categoriza*on,
!
1000/2000ms, 1000ms, 10s,
!Decision,
!
10s,
Feedback,on,C,and,D,
10s,
1000/2000ms, 1000ms, 10s,
!Decision,
!
10s,
Feedback,on,D,
C-then-D: Categorize face first and then decide
D-alone: Decide without categorization
Two Conditions:
• Pr(Bad | Narrow) = .60
• Pr(Good | Wide) = .60
• Pr(Reward Attack | Bad ) = .70
• Pr(Reward Withdraw | Good) = .70
Contingencies ( Learned from Experience)
LAW OF TOTAL PROBABILITY
p(A)= ?=p(G)p(A|G)+ p(B)p(A|B)
D alone Condition C-then-D Condition
G =good guy, B=Bad guy, A=Attack
Total Probability
MARKOV MODEL
Face
Good
Bad
Attack
Withdraw
p(G|F)
p(B|F)
p(A|G)
p(A|B)
P(A|F)=p(G|F)·p(A|G)+p(B|F)·p(A|B)(total probability satisfied)
RG ∩ RW
RG ∩ RA RB ∩ RA
RB ∩ RW
Face
Signal Detection Model
p(A|F) = p(RG ∩ RA|F) + p(RB ∩ RA|F)
= p(RG|F)p(RA|RG,F) + p(RB|F)p(RA|RB,F)
(total probability satisfied)
Face p(G) p(A|G) p(B) p(A|B) TP P(A)
Good 0.84 0.35 0.16 0.52 0.37 0.39
Bad 0.17 0.41 0.82 0.63 0.59 0.69
RESULTS
PRISONER DILEMMA GAME
OD OC
PD O: 10P: 10
O:5P: 25
PC O: 25P: 5
O:20P: 20
Croson, 1999, Organizational Behavior and Human Decision Processes
Game p(c) p(C|c) p(d) p(C|d) TP P(C)
PD-1 0.45 0.83 0.55 0.32 0.55 0.775
PD-2 0.42 0.53 0.58 0.35 0.425 0.625
c = Predict opponent will cooperated = Predict opponent will defectC = Player decides to cooperateTP = total probability cooperate after predictionp(C) = Prob player cooperate with no prediction
Sequentialsocialdilemmagame
Blanco,etal.(2014,GamesandEconomicBehavior)
N = 40 60 60
Centipede6-stagegame,linearincreasingpayoff
EXPERIMENTAL DESIGN
• N = 195 participates paid based on decision outcomes
• Each participant played 30 computer agents Programmed to respond according to base rates from previous research on similar tasks
• All received two types of trials Predict then act Act alone
• For each game/agent, randomly assigned to play first or second Human plays first vs. second
Humanplaysfirst
.731 .93 .684 ..73 .67
.703 .732 .414
Pred-Act
Act-alone
.386
Probability to Continue
Humanplayssecond
.93 .703 .73 .546 .67
.656 .581
Pred-Act
Act-alone
.325
.345
Probability to continue
QUANTUM MODEL OF INTERFERENCE
p(PD) = PPDS2 = PPD ⋅ I ⋅S
2
= PPD ⋅ POD + POC( ) ⋅S 2
= PPD ⋅POD ⋅S + PPD ⋅POC ⋅S2
= PPD ⋅POD ⋅S2 + PPD ⋅POC ⋅S
2 + Int
Int = S | POCPPDPPDPOD | S + S | PODPPDPPDPOC | S
GAME OF CHICKEN
James Dean in “Rebel without
a Cause”
CHICKEN PAYOFF MATRIX
Yield Continue
Yield Chicken, Chicken Chicken, Hero
Continue Hero, Chicken Crash, Crash
Nash equilibrium: choose different action than opponent
HYPOTHETICAL EXPERIMENTAL RESULTS
B-Y B-CA-Y 0 0.5
A-C 0.5 0
2 x 2 contingency tables of relative frequency
C-Y C-C
B-Y 0 0.5
B-C 0.5 0
C-Y C-C
A-Y 0 0.5
A-C 0.5 0
Contextual effects should occur, as expected by traditional game theory
n = 3delta = 0sodd = 3
contextually = sodd - (n-2) - delta = 2 > 0
ADVANTAGE OF QUANTUM MODELS
• Single model that accounts for
• violations of marginal invariance
• order effects
• contextual effects
• Key ingredient required is incompatible (non-commutative) events
CONCLUSIONS
• Violations of marginal invariance can provide strong challenges to major psychological and economic theories
• Contextuality effects can occur, but can (sometimes) be easily explained by existing psychological and economic theories
• Importance of either effect depends on extent to which the effect challenges traditional psychological theory