SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015....
Transcript of SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015....
![Page 1: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/1.jpg)
model results comments conclusion
Smart Transformations: or, The Evolution of ChoicePrinciples
A Talk About Rationality
Paolo Galeazzi
ILLC, University of Amsterdam
Joint work with M. Franke
![Page 2: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/2.jpg)
model results comments conclusion
Outline
modelresultscommentsconclusion
![Page 3: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/3.jpg)
model results comments conclusion
Outline
1 model
2 results
3 comments
4 conclusion
![Page 4: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/4.jpg)
model results comments conclusion
Introduction
The model:
a population of playersa class of possible games
Interpretation: different types in the population representdifferent reasoning/player types, or choice principles.
![Page 5: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/5.jpg)
model results comments conclusion
Game theory
Definition (Games)
A game G is a tuple G = ⟨N,S1, ...,Sn, π1, ..., πn⟩
As usual, S ∶= S1 × ... × Sn.
Definition (Solution concepts)
Given a game G, a solution concept is a function F s.t. F (G) ⊆ S
![Page 6: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/6.jpg)
model results comments conclusion
Nash equilibrium
Definition (Nash equilibrium)
A Nash equilibrium (NE) of a game G is a strategy profile s ∈ S s.t.∀i ∈ N,∀s ′i ∈ Si , πi(s) ≥ πi(s ′i , s−i).
![Page 7: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/7.jpg)
model results comments conclusion
Regret
Definition (Regret minimization [4])
Given a game G, the regret of a strategy si for player i is defined asfollows: regi(si) ∶= maxs−i πi(s∗i , s−i) − πi(si , s−i), where s∗i is i’sbest reply to s−i .A strategy s ′i is a regret minimization strategy (regmini ) of player iif regi(s ′i ) = minsi maxs−i πi(s∗i , s−i) − πi(si , s−i).
It will be useful to define a pairwise notion of regret.
Definition (Pairwise regret)
Given a game G, the regret of player i at profile (si , s−i) is:regi(si , s−i) ∶= πi(s∗i , s−i) − πi(si , s−i).
![Page 8: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/8.jpg)
model results comments conclusion
the Travellers’ dilemma
the Travellers’ dilemma (TD)
N = Ann,Bob
SAnn = SBob = {2, ...,100}
πi =⎧⎪⎪⎪⎪⎨⎪⎪⎪⎪⎩
si if si = s−i
si + 2 if si < s−i
s−i − 2 if s−i < si
![Page 9: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/9.jpg)
model results comments conclusion
the Travellers’ dilemma case
In the Travellers’ dilemma there is only one NE: (2,2).
All the other solution concepts grounded on expected utility thatsolve the game (e.g., rationalizability, iterated elimination of weaklydominated strategies, ...) return the outcome (2,2), given perfectrationality of the players.
In the TD, iterated regret minimization gives the outcome (97,97).
However, in what follows we do not use the iterated version ofregret minimization (epistemically not well-founded).
![Page 10: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/10.jpg)
model results comments conclusion
The pairwise notion defines a transformation of the game G intoanother game Greg .
1,5
4,2
3,2
1,4
L R
U
DAnn
BobG
3,0
0,2
0,3
2,0
L R
U
DAnn
BobGreg
![Page 11: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/11.jpg)
model results comments conclusion
The pairwise notion defines a transformation of the game G intoanother game Greg .
1,5
4,2
3,2
1,4
L R
U
DAnn
BobG
3,0
0,2
0,3
2,0
L R
U
DAnn
BobGreg
![Page 12: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/12.jpg)
model results comments conclusion
Inside regret minimization
Regret minimization may be seen as a security strategy.
Definition (Maximin)
Let us define mini(si) = mins−i πi(si , s−i). Then we can say thatstrategy s ′i is a maximin strategy of player i (maximini ) ifmini(s ′i ) = max si mins−i πi(si , s−i).
Remark: the maximin solution to the TD is also (2,2).
![Page 13: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/13.jpg)
model results comments conclusion
Regret minimization as maximin
-3,0
0,-2
0,-3
-2,0
L R
U
DAnn
Bob-Greg
Finding the regret minimizing strategies in G corresponds to findingthe maximin strategies in -Greg .
![Page 14: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/14.jpg)
model results comments conclusion
Regret minimization as maximin
-3,0
0,-2
0,-3
-2,0
L R
U
DAnn
Bob-Greg
Finding the regret minimizing strategies in G corresponds to findingthe maximin strategies in -Greg .
![Page 15: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/15.jpg)
model results comments conclusion
Evolutionary analysis of choice principles: the intuition
So far we have seen that regret defines a specific transformation ofthe game. This gives an intuition on how we might understanddifferent player types (or choice principles).
![Page 16: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/16.jpg)
model results comments conclusion
Evolution of preferences
An objective fitness game G is played in the population andeach type θ is a preference type, i.e., θ ∶ S1 × ... × Sn → R.The regret type has the subjective preference/utility functionθreg ∶ S1 × ... × Sn → R s.t. θreg(si , s−i) = −regi(si , s−i).The π-type has the subjective utility functionθπ ∶ S1 × ... × Sn → R s.t. θπ(si , s−i) = πi(si , s−i).Θ is the set of all preference types.Indirect evolutionary approach.
![Page 17: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/17.jpg)
model results comments conclusion
Meta-game
Tabella : A coordination game
G I III 1,1 0,0II 0,0 2,2
Tabella : Chicken game
G’ I III 0,0 -1,1II 1,-1 -2,-2
![Page 18: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/18.jpg)
model results comments conclusion
Meta-game
Tabella : A coordination game
G I III 1,1 0,0II 0,0 2,2
Tabella : Chicken game
G’ I III 0,0 -1,1II 1,-1 -2,-2
![Page 19: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/19.jpg)
model results comments conclusion
A bit more formal: player types or choice principles
Given a class of fitness games G, we call player type a functionτ ∶ G→ Θ.Intuition: a player type is interpreted as a way of thinkingacross games, it is a thread, a red line that relates differentpreference types across different games.The regret type τ reg is defined s.t. for any G ∈ G,τ reg(G) = θreg .Similarly, for any G ∈ G, τπ(G) = θπ.Let T denote the set of player types.
![Page 20: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/20.jpg)
model results comments conclusion
Evolutionary analysis of choice principles: the model
G: class of symmetric two-by-two gamesPlayers in the population are player types (i.e., τ -functions)Radical uncertainty, i.e., no probabilistic beliefs (because oflack of information, lack of cognitive capabilities, ...)Because of their radical uncertainty we assume that playersplay a security strategy wrt their player type (secure players).At each time a game G is randomly selected from G andplayers are randomly matched to play G.The evolutionary fitness of each type is determined by thefitness games G via the objective fitness function πG .The players’ action choices across games depend on theirplayer type τ .
![Page 21: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/21.jpg)
model results comments conclusion
Evolutionary analysis of choice principles: the model
G: class of symmetric two-by-two gamesPlayers in the population are player types (i.e., τ -functions)Radical uncertainty, i.e., no probabilistic beliefs (because oflack of information, lack of cognitive capabilities, ...)Because of their radical uncertainty we assume that playersplay a security strategy wrt their player type (secure players).At each time a game G is randomly selected from G andplayers are randomly matched to play G.The evolutionary fitness of each type is determined by thefitness games G via the objective fitness function πG .The players’ action choices across games depend on theirplayer type τ .
![Page 22: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/22.jpg)
model results comments conclusion
Evolutionary analysis of choice principles: the model
G: class of symmetric two-by-two gamesPlayers in the population are player types (i.e., τ -functions)Radical uncertainty, i.e., no probabilistic beliefs (because oflack of information, lack of cognitive capabilities, ...)Because of their radical uncertainty we assume that playersplay a security strategy wrt their player type (secure players).At each time a game G is randomly selected from G andplayers are randomly matched to play G.The evolutionary fitness of each type is determined by thefitness games G via the objective fitness function πG .The players’ action choices across games depend on theirplayer type τ .
![Page 23: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/23.jpg)
model results comments conclusion
Evolutionary analysis of choice principles: the model
G: class of symmetric two-by-two gamesPlayers in the population are player types (i.e., τ -functions)Radical uncertainty, i.e., no probabilistic beliefs (because oflack of information, lack of cognitive capabilities, ...)Because of their radical uncertainty we assume that playersplay a security strategy wrt their player type (secure players).At each time a game G is randomly selected from G andplayers are randomly matched to play G.The evolutionary fitness of each type is determined by thefitness games G via the objective fitness function πG .The players’ action choices across games depend on theirplayer type τ .
![Page 24: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/24.jpg)
model results comments conclusion
Evolutionary analysis of choice principles: the model
G: class of symmetric two-by-two gamesPlayers in the population are player types (i.e., τ -functions)Radical uncertainty, i.e., no probabilistic beliefs (because oflack of information, lack of cognitive capabilities, ...)Because of their radical uncertainty we assume that playersplay a security strategy wrt their player type (secure players).At each time a game G is randomly selected from G andplayers are randomly matched to play G.The evolutionary fitness of each type is determined by thefitness games G via the objective fitness function πG .The players’ action choices across games depend on theirplayer type τ .
![Page 25: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/25.jpg)
model results comments conclusion
Evolutionary analysis of choice principles: the model
G: class of symmetric two-by-two gamesPlayers in the population are player types (i.e., τ -functions)Radical uncertainty, i.e., no probabilistic beliefs (because oflack of information, lack of cognitive capabilities, ...)Because of their radical uncertainty we assume that playersplay a security strategy wrt their player type (secure players).At each time a game G is randomly selected from G andplayers are randomly matched to play G.The evolutionary fitness of each type is determined by thefitness games G via the objective fitness function πG .The players’ action choices across games depend on theirplayer type τ .
![Page 26: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/26.jpg)
model results comments conclusion
Evolutionary analysis of choice principles: the model
G: class of symmetric two-by-two gamesPlayers in the population are player types (i.e., τ -functions)Radical uncertainty, i.e., no probabilistic beliefs (because oflack of information, lack of cognitive capabilities, ...)Because of their radical uncertainty we assume that playersplay a security strategy wrt their player type (secure players).At each time a game G is randomly selected from G andplayers are randomly matched to play G.The evolutionary fitness of each type is determined by thefitness games G via the objective fitness function πG .The players’ action choices across games depend on theirplayer type τ .
![Page 27: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/27.jpg)
model results comments conclusion
Outline
1 model
2 results
3 comments
4 conclusion
![Page 28: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/28.jpg)
model results comments conclusion
Evolutionarily and neutrally stable strategies
Definition (ESS)
In evolutionary game theory a strategy s is said to be evolutionarystable (ESS) if for all the other strategies t:
1. (s vs s) > (t vs s) or2. (s vs s) = (t vs s) and (s vs t) > (t vs t)
Definition (NSS)
A strategy s is said to be neutrally stable (NSS) if for all the otherstrategies t:
1. (s vs s) > (t vs s) or2. (s vs s) = (t vs s) and (s vs t) ≥ (t vs t)
![Page 29: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/29.jpg)
model results comments conclusion
Remark 1: the assumption of radical uncertainty is necessary todistinguish τπ from τ reg . I.e., with probabilistic beliefs maximizingexpected utility would be the same as minimizing expected regret,for any G.
Formally,
MaxExp(-Greg , µ) = argmaxi ∑j(-Gregij ⋅ µj) (by def)
= argmaxi ∑j(Gij −maxk(Gkj)) ⋅ µj= argmaxi ∑j(Gij ⋅ µj −maxk(Gkj) ⋅ µj)= argmaxi ∑j(Gij ⋅ µj)=MaxExp(G, µ).
![Page 30: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/30.jpg)
model results comments conclusion
Evolutionary analysis of choice principles
Just for fun, we define two other possible player types.
Definition (Competitive)
θcom ∶ S1 × S2 → R s.t. θcom(si , s−i) = πi(si , s−i) − π−i(si , s−i).
Definition (Altruistic)
θalt ∶ S1 × S2 → R s.t. θalt(si , s−i) = πi(si , s−i) + π−i(si , s−i).
Then we have: τ com(G) = θcom and τ alt(G) = θalt .And the corresponding security strategies aremaxsi mins−i θ
com(si , s−i) and maxsi mins−i θalt(si , s−i).
![Page 31: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/31.jpg)
model results comments conclusion
Two-by-two symmetric games: matrix of average payoffs
6.302
6.128
6.305
6.131
τ reg τπ
τ reg
τπ
τ com τ alt
τ com
τ alt
5.561
5.750
5.226
6.682
6.331
6.5366.3395.974
5.588 5.413 5.029 6.018
τ reg turns out to be the only evolutionary stable (secure) type.
![Page 32: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/32.jpg)
model results comments conclusion
More in detail
More in detail:
Coordination games: a > c ,d > b
Anti-coordination games: a < c ,d < b
Strong dominance games: a > c ,d < b or a < c ,d > b
Weak dominance games: aut a − c = 0 aut d − b = 0Indifferent games: a − c = 0 and d − b = 0
![Page 33: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/33.jpg)
model results comments conclusion
Average payoff matrix for coordination games
7.676
6.430
5.914
6.367
τ reg τπ
τ reg
τπ
τ com τ alt
τ com
τ alt
5.794
6.341
6.346
6.436
5.905
5.9206.4346.462
5.441 4.281 4.237 5.833
![Page 34: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/34.jpg)
model results comments conclusion
Average payoff matrix for anti-coordination games
4.751
5.275
6.202
5.288
τ reg τπ
τ reg
τπ
τ com τ alt
τ com
τ alt
4.713
5.405
3.426
6.141
5.314
7.0167.0745.342
5.207 5.303 5.387 5.270
![Page 35: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/35.jpg)
model results comments conclusion
Average payoff matrix for strong dominance games
7.212
7.212
7.212
7.212
τ reg τπ
τ reg
τπ
τ com τ alt
τ com
τ alt
6.704
6.704
6.365
7.749
7.749
7.4466.9096.909
6.588 6.588 6.080 7.197
![Page 36: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/36.jpg)
model results comments conclusion
Average payoff matrix for weak dominance games
6.402
6.155
6.342
6.099
τ reg τπ
τ reg
τπ
τ com τ alt
τ com
τ alt
5.187
5.120
4.706
7.168
6.735
6.7876.1575.959
5.710 5.492 4.532 6.486
![Page 37: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/37.jpg)
model results comments conclusion
Average payoff matrix for indifferent games
5.833
5.833
5.75
5.75
τ reg τπ
τ reg
τπ
τ com τ alt
τ com
τ alt
3.541
3.541
3.541
7.166
7.166
7.1665.755.833
5.833 5.75 3.541 7.166
![Page 38: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/38.jpg)
model results comments conclusion
These results from simulations can also be proven analytically.
Proposition 1
Fix T = {τπ, τ reg , τ com, τ alt} and G the class of symmetric 2 × 2games with i.i.d. sampled payoffs from a finite or compact andconvex set of values. Then τ reg is the only evolutionary stable(secure) type in the population.
![Page 39: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/39.jpg)
model results comments conclusion
With probabilistic beliefs
Suppose now that players can have both probabilistic beliefs (insome situations) and radical uncertainty (in other situations).Moreover, suppose that the information (beliefs) available issymmetric. Then, from Proposition 1 and Remark 1 it follows that:
Corollary 1
Fix T = {τπ, τ reg} and G the class of symmetric 2 × 2 games withi.i.d. sampled payoffs from a finite or compact and convex set ofvalues. Then τ reg is the only evolutionary stable type in thepopulation.
![Page 40: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/40.jpg)
model results comments conclusion
Digression: Ellsberg’s urn, ambiguity, and security strategies
R Y B
fR 100 0 0fY 0 100 0fRB 100 0 100fYB 0 100 100
90 marbles in the urn, 30 red marbles.
The standard pattern is: fR ≻ fY and fRB ≺ fYB .
Preferences that satisfy this pattern are called uncertainty, orambiguity, averse.The DM is not representable by a probability measure as inSavage’s theorem.
![Page 41: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/41.jpg)
model results comments conclusion
Digression: Ellsberg’s urn, ambiguity, and security strategies
R Y B
fR 100 0 0fY 0 100 0fRB 100 0 100fYB 0 100 100
90 marbles in the urn, 30 red marbles.The standard pattern is: fR ≻ fY and fRB ≺ fYB .Preferences that satisfy this pattern are called uncertainty, orambiguity, averse.The DM is not representable by a probability measure as inSavage’s theorem.
![Page 42: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/42.jpg)
model results comments conclusion
Digression: Ellsberg’s urn, ambiguity, and security strategies
R Y B
fR 100 0 0fY 0 100 0fRB 100 0 100fYB 0 100 100
90 marbles in the urn, 30 red marbles.The standard pattern is: fR ≻ fY and fRB ≺ fYB .Preferences that satisfy this pattern are called uncertainty, orambiguity, averse.The DM is not representable by a probability measure as inSavage’s theorem.
![Page 43: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/43.jpg)
model results comments conclusion
Digression: Ellsberg’s urn, ambiguity, and security strategies
R Y B
fR 100 0 0fY 0 100 0fRB 100 0 100fYB 0 100 100
90 marbles in the urn, 30 red marbles.The standard pattern is: fR ≻ fY and fRB ≺ fYB .Preferences that satisfy this pattern are called uncertainty, orambiguity, averse.The DM is not representable by a probability measure as inSavage’s theorem.
![Page 44: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/44.jpg)
model results comments conclusion
Digression: Ellsberg’s urn, ambiguity, and security strategies
It has been shown [3] that ambiguity averse preferences arerepresentable by a closed convex set P of probability measurestogether with the maximin rule: Maxmin Expected Utility withNon-unique Prior.
Working example: P = [(13 ,
23 ,0), (
13 ,0,
23)].
minµ∈P ∑s∈S u(fR(s)) ⋅ µ(s) = 1003
minµ∈P ∑s∈S u(fY (s)) ⋅ µ(s) = 0
minµ∈P ∑s∈S u(fRB(s)) ⋅ µ(s) = 1003
minµ∈P ∑s∈S u(fYB(s)) ⋅ µ(s) = 2003
![Page 45: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/45.jpg)
model results comments conclusion
Digression: Ellsberg’s urn, ambiguity, and security strategies
It has been shown [3] that ambiguity averse preferences arerepresentable by a closed convex set P of probability measurestogether with the maximin rule: Maxmin Expected Utility withNon-unique Prior.
Working example: P = [(13 ,
23 ,0), (
13 ,0,
23)].
minµ∈P ∑s∈S u(fR(s)) ⋅ µ(s) = 1003
minµ∈P ∑s∈S u(fY (s)) ⋅ µ(s) = 0
minµ∈P ∑s∈S u(fRB(s)) ⋅ µ(s) = 1003
minµ∈P ∑s∈S u(fYB(s)) ⋅ µ(s) = 2003
![Page 46: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/46.jpg)
model results comments conclusion
Digression: Ellsberg’s urn, ambiguity, and security strategies
It has been shown [3] that ambiguity averse preferences arerepresentable by a closed convex set P of probability measurestogether with the maximin rule: Maxmin Expected Utility withNon-unique Prior.
Working example: P = [(13 ,
23 ,0), (
13 ,0,
23)].
minµ∈P ∑s∈S u(fR(s)) ⋅ µ(s) = 1003
minµ∈P ∑s∈S u(fY (s)) ⋅ µ(s) = 0
minµ∈P ∑s∈S u(fRB(s)) ⋅ µ(s) = 1003
minµ∈P ∑s∈S u(fYB(s)) ⋅ µ(s) = 2003
![Page 47: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/47.jpg)
model results comments conclusion
Games with (some) uncertainty
In our framework, secure players are maxmin expected utilityplayers (à la Gilboa-Schmeidler) with P =∆(S−i).
Remark 2. τ reg (secure) players outperform τπ (secure) players forany non-singleton set P .
![Page 48: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/48.jpg)
model results comments conclusion
Outline
1 model
2 results
3 comments
4 conclusion
![Page 49: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/49.jpg)
model results comments conclusion
On the model: Theoretical biology and behavioral ecology
Behavioral gambit in theoretical biology: the evolutionarymodels put all the focus on the expressed behavior, and neglectthe underlying mechanisms that generate that behavior.
[...] we should expect animals to have evolved a setof psychological mechanisms which enable them toperform well on average across a range of differentcircumstances. [2]
Psychological mechanisms and subjective conceptualizations asthe phenotype under selection.Our model:
retains the indirect evolutionary approachworks on a class of possible games
![Page 50: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/50.jpg)
model results comments conclusion
On the result: What is rational?
This talk takes the cue from a paper by J.Halpern and R.Pass [4]introducing iterated regret minimization as a new solution concept.
The notion of regret had already been introduced in decision theoryas an alternative to expected utility theory.
Regret theory
[...] we shall challenge the idea that the conventionalaxioms constitute the only acceptable basis for rationalchoice under uncertainty. We shall argue that it is no lessrational to act in accordance with regret theory, and thatconventional expected utility theory therefore representsan unnecessarily restrictive notion of rationality. [5]
![Page 51: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/51.jpg)
model results comments conclusion
On the result: What is rational?
Rationality is a descriptive notion: maxEU fails.(Btw, regret can explain and describe observed behavior betterthan maxEU [4],[5].)
Rationality is a normative notion: apparently, EU-maximizerscan be outperformed by regret minimizers.
![Page 52: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/52.jpg)
model results comments conclusion
On the result: What is rational?
Rationality is a descriptive notion: maxEU fails.(Btw, regret can explain and describe observed behavior betterthan maxEU [4],[5].)
Rationality is a normative notion: apparently, EU-maximizerscan be outperformed by regret minimizers.
![Page 53: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/53.jpg)
model results comments conclusion
Outline
1 model
2 results
3 comments
4 conclusion
![Page 54: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/54.jpg)
model results comments conclusion
Conclusion
We found subjective transformations of the games (”playertypes”) that can perform better than the maximization of theobjective payoff.
![Page 55: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/55.jpg)
model results comments conclusion
Evolution of preferences
These papers [i.e., [6];[1]] highlight the dependence ofindirect evolutionary models on observable preferences,posing a challenge to the indirect evolutionary approachthat can be met only by allowing the question ofpreference observability to be endogenously determinedwithin the model. [8]
The indirect evolutionary approach with unobservablepreferences gives us an alternative description of theevolutionary process, one that is perhaps less reminiscentof biological determinism, but leads to no new results. [7]
![Page 56: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/56.jpg)
model results comments conclusion
Future work
more general Gdifferent/asymmetric beliefs and theory of mindother player typesadding learning/adaptation process
![Page 57: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/57.jpg)
model results comments conclusion
Thanks for your attention.
![Page 58: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/58.jpg)
model results comments conclusion
References:J.C. Ely and O. Ylankaya.Nash Equilibrium and the Evolution of Preferences.Journal of Economic Theory, 97:255–272, 2001.
Tim W. Fawcett, Steven Hamblin, and Luc-Alain Giraldeaub.Exposing the behavioral gambit: the evolution of learning anddecision rules.Behavioral Ecology, 24(1):2–11, 2013.
I. Gilboa and D. Schmeidler.Maxmin Expected Utility with Non-unique Prior.Journal of Mathematical Economics, 18:141–153, 1989.
J. Halpern and R. Pass.Iterated Regret Minimization: a New Solution Concept.Games and Economic Behavior, 74:1:194–207, 2012.
G. Loomes and R. Sugden.
![Page 59: SmartTransformations: or,TheEvolutionofChoice Principlesfitelson.org/bridges2/galeazzi.pdf · 2015. 9. 20. · modelresultscommentsconclusion SmartTransformations: or,TheEvolutionofChoice](https://reader036.fdocuments.us/reader036/viewer/2022071214/604391abf7f9066c8e6a4eb6/html5/thumbnails/59.jpg)
model results comments conclusion
Regret Theory: An Alternative Theory of Rational ChoiceUnder Uncertainty.The Economic Journal, 92:368:805–824, 1982.
E. Ok and F. Vega Redondo.On the Evolution of Individualistic Preferences: An IncompleteInformation Scenario.Journal of Economic Theory, 97:231–254, 2001.
A. Robson and L. Samuelson.The Evolutionary Foundations of Preferences.In J. Benhabib, M. Jackson, and A. Bisin, editors, Handbook ofSocial Economics, 1A. North-Holland, 2011.
L. Samuelson.Introduction to the Evolution of Preferences.Journal of Economic Theory, 97:225–230, 2001.