Docstoc

Communication Commitment and Cooperation in Social Dilemmas

Document Sample
Communication Commitment and Cooperation in Social Dilemmas Powered By Docstoc
					                                                                         1




Evolution, Trust, and
    Reciprocity

     Robert Kurzban


                      Trust in Groups from Cross-Societal Perspectives
                      Hokkaido University, Sapporo, Japan,
                      September 26-28, 2003
                                              2



   The Evolution of Cooperation:
      Constraints on Models
• No model of the evolution of cooperation
  should predict promiscuous cooperation
• Cue-based models (“green beard”) have
  profound difficulties, in particular the
  problem of mimics.
• Plausible models of the evolution of
  cooperation must provide a basis by which
  exploitation is minimized, or otherwise
  eliminate the problem
                                                           3



    The Evolution of Cooperation:
      Minimizing Exploitation
• Exploitation can be minimized through:
    Punishment (but note the 2nd order problem)
       • Boyd et al., 2003
    Withdrawing cooperation, changing groups…
       • Aktipis, in prep.
    Indirect reciprocity models: limiting transactions?
       • Panchanathan & Boyd, in press (JTB)
• Exploitation irrelevant in:
    kin selection models, honest signaling models
                                                     4



   The Evolution of Cooperation:
      Cognitive Mechanisms
• A well-designed cognitive system should be
  fashioned to reap benefits of cooperation
  while minimizing exposure to exploitation:
  Process cues that inform computations
   regarding cooperation (costs/benefits, features
   of others, etc)
  Decide whether/how much to trust (endure cost
   with expectation of future benefit)
                                                               5

      Models of the evolution of cooperation

• Kin selection doesn’t generalize because
    r falls off fast
    independent assortment of alleles makes “similarity”
    irrelevant.
• Reciprocal altruism doesn’t generalize easily
  because (e.g., Axelrod, 1986)
    When common, “tolerant” reciprocators are replaced
    by defectors.
    When rare, “intolerant” reciprocators find insufficient
    conspirators (unless there is a lot of assortment).
    Punishment presents a 2nd order problem (e.g., Boyd
    & Richerson, 1988)
                                                6




   Models of the evolution of cooperation

• Some models suggest that striving for
  uniqueness/irreplaceability will lead to
  mechanisms that cause individuals to
  deliver benefits to others.
• Note: KS and RA are very different, and
  suggest that it’s not the number of
  interactants that matters, but the relevant
  selection pressures (pace Caporael et al.)
                                                            7




          Reciprocity in Groups?
• “Mean matching” strategies are too “tolerant” (Miller &
  Andreoni, 1991) because marginally less cooperative
  group members will out-replicate more cooperative group
  members.
• But, empirically, contributions in public goods game
  correlate with expectations about others’ cooperation
      Bornstein & Ben-Yossef, 1994
      Braver & Barnett, 1974
      Croson, 1998
      Dawes, McTavish, & Shaklee, 1977
      Komorita, Parks, & Hulbert, 1992
      Messick et al., 1983
      Yamagishi & Sato, 1986
                                                   8




      Incremental Reciprocity
• Strategies that cooperate as long as every
  other group member does so.
• Intolerant
• Enforces equity
• Preliminary agent-based simulations show
  conditions broaden under which cooperation
  in groups can evolve. (Panchanathan et al., in
  prep., similar to Boyd & Richerson, 1988)
                                               9



   Assessing Cooperation: The
      Public Goods Game
1. Participants are given an allotment of
  Tokens

2. Task: Divide tokens between two accounts:
 •Private Account: Returns $1 for 1 Token
 •Public Account: Returns $2 per Token,
 divided equally among players (>2)
                                            10



Assessing Cooperation: The
   Public Goods Game

Creates a tension between individual and
            aggregate outcomes.


Cooperation is indexed by contribution to
            Public Account.
                                                                              11



 Public Goods Game:
Typical Results (Kurzban 1998)
                                Public Goods Baseline Results - Males

                                60
  Contribution to Group (%) .



                                50
                                40
                                30
                                20
                                10
                                 0
                                     1   2   3   4   5   6   7   8   9   10
                                                     Round
                                                      12




        Puzzling findings?
In round 10, one’s best move is to contribute zero.
So, by backward induction, one ought to
contribute zero every time. Why are there any
contributions at all? (Economics)
                                                    13




            Communication
If players are allowed to discuss the dilemma
before the game, we observe much more
cooperation. (Dawes, McTavish, and Shaklee, 1977)
                                                  14




          Puzzling findings?
Communication is cheap talk. Why does this have
  any effect on cooperation? (Economics)

How does communication work its magic?
 (Psychology)
                                                   15




               Hypotheses
• People are willing to contribute to the extent
  others are, but have limited trust.
• Communication influences beliefs about
  others’ contributions, especially when
  others are committed.
• To be effective, commitment must not
  expose the player to “being free ridden.”
                                                     16




      Distinguishing Accounts
If cooperation is due to…    Commitment should…
 Confusion or Altruism       have no effect.
 Social identity             have no effect.
 Reciprocity or              increase cooperation.
 Fear of being free ridden
                                                  17




                 Testing
IF people are willing to contribute to the
  extent others are, THEN
  Observing others’ commitments will induce
   reciprocal cooperation
  Incremental commitment minimizes the cost of
   committing by limiting exposure
                                                       18




               Experiment 1
Claim
  Players in public goods games are trying to play a
  reciprocal strategy but this has been obscured by
  uncertainty about others’ contributions.

Hypothesis
  Providing players the ability to publicly
  communicate their incremental commitment will
  elicit contributions.
                                        19



   Experiment 1: Implementing
    Incremental Commitment
• Real time updating of contributions
• Manipulation of commitment
  INCREASE ONLY condition
     • Commitment
  INCREASE/DECREASE condition
     • Cheap talk
• Prediction: Higher contributions in
  INCREASE ONLY condition.
                                                  20




      Experiment 1: Parameters
•   Five players per group
•   50 Tokens per round
•   1 Token = $.01 (MPCR = .33)
•   10 Rounds (known)
•   90 seconds per round
•   ALL players’ contributions displayed during
    round
     Updated ~3 times/second.
21
                                                                                      22

                                50

Average Contribution (Tokens)            Increase/Decrease
                                40
                                         Increase Only

                                30


                                20


                                10


                                 0
                                     1    2      3       4   5   6   7   8   9   10
                                                             Round
                 23




Real Time Play
                                          24




   Experiment 1: Conclusions
• Cooperation increase with incremental
  commitment mechanism.
• Reciprocity? (R = .74)
• Imperfect cooperation.
                                                    25




             Experiment 2
• Claim.
  Players are reciprocators, but concerned about
   being free ridden (by the least cooperative
   group member).
• Claim 2.
  Players know this about others.
                                            26




      Experiment 2: Design
• Consider 2 independent variables
  IO/ID, as in experiment 1
  Highest vs. Lowest current contributor
                                            27




   Experiment 2: Predictions

                            Information

                         LOWEST   HIGHEST

             Increase
                          High       Low
               Only
Mechanism
            Increase &
                          Low        Low
             Decrease
                                  28




      Experiment 2: Parameters
•   Five players per group
•   50 Tokens per round
•   1 Token = $.01 (MPCR = .33)
•   10 Rounds (known)
•   90 seconds per round
                                   29




A look at the real time dynamics
                                                                                          30




Average Contribution (Tokens)   50
                                         Low Info & Increase/Decrease
                                         Low Info & Increase Only
                                40
                                         High Info & Increase/Decrease
                                         High Info & Increase Only
                                30


                                20

                                10

                                0
                                     1     2     3     4     5       6   7   8   9   10

                                                             Round
                                                                                  31

Contributions by Round in the Increase Only/Low
             Information Condition

                                    50
Avetrage Contribution (in Tokens)

                                    45
                                    40
                                    35
                                    30
                                    25
                                    20
                                    15
                                    10
                                     5
                                     0
                                         1   2   3   4   5   6   7   8   9   10

                                                         Round
                                                  32



   Experiment 2: Summary of
           findings
• Lowest info IO can improve cooperation
  rates.
• Individual differences seem to cause big
  changes in group dynamics.
• There is a close (r =.9) relationship between
  low info and player i’s contribution
                                              33




             Conclusions
• Aversion to being free ridden
• Excessive trusting (in Exp 1) may inhibit
  reciprocal cooperation
• Points to “inequity aversion” models
• Role for individual differences in group
  dynamics
                                                    34



           Experiment 3:
        Information Seeking
• IF it is true that players care about the LEAST
  cooperative member of a group,
• THEN when given the opportunity to learn one
  other group member’s contribution,
• Participants can be predicted to choose the
  LOWEST current contributor…
                                                              35




       Experiment 3: Method
• “Circular” Public Goods game
   Players make initial contribution
   Players, in turn, can observe ONE contribution of other
    players, listed lowest to highest.
   After observing this value, player may update their own
    contribution
   Round ends with p = .04 each update
                                                36




                 Results
How often do players look at the:
• Low?
  42%               Take with grain of salt:
• Middle?            Sessions run in last 2
  25%               weeks add noise to
                     these findings.
• High?
  33%
                                                                         37




                   Results (con’t)
Does one’s contribution depend on the information one observes?



                  Information Seen      Contribution     Old Aggregate
   Calculation          Mean               Mean              Mean
   Request low          25.84              33.43            142.14
   Request mid          28.60              24.63            112.11
  Request high          30.29              21.29             91.27
                                            38




         Results (con’t)
200
          Aggregate
150

100

 50

  0
      1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
                      Game
                                                39




    Experiment 3: Conclusions
• There is a tendency to prefer observing the
  LOWEST current contributor.
• The circular game, even under heavy
  information restriction conditions, affords
  cooperation
• The circular game might be useful in
  understanding reciprocal play.
                                              40



      Experiment 4: Individual
            Differences
• Circular public goods games, with
  parameters similar to previous experiment
• Except: Players have access to only
  aggregate information of others’
  contributions
• This allows us to plot a “contribution
  profile” for each player (CP)
                                                 41




       Individual Differences
• Regress contribution on information
  observed.
• This gives an intercept and slope.
• Intercept ~ how much player i contributes
  when others aren’t contributing much
• Slope ~ player i’s responsiveness to others’
  contributions
                                              42




       Individual Differences
• Free Rider = CP everywhere below 25 (1/2)
  20% of sample (N = 84)
• Cooperator = CP everywhere above 25
  13%
• Recriprocator = positive slope, and CP is
  both above and below the 50% line.
  63%
• Small percentage unclassifiable
                                                       43




       Individual Differences
• We use some rounds to see if typing scheme
  captures something stable.
• If so, we should be able to predict (in a
  hold-out sample) the dynamics of play
  given the makeup of the constituted groups.
  Groups are assigned a “Cooperativeness
   Score,” 2 for a Cooperator, 1 for a Reciprocator,
   0 for a Free Rider…
                                                                                                                                  44




                                           Group Cooperativeness and Mean Final Group Contribution
                                                   First Seven Games and Holdout Sample
                            200
Total Public Contribution




                                                                                                                      (N=1)
                            160
                                                                                             (N=6)        (N=2)       (N=2)
                                                                                             (N=25)       (N=9)
                            120
                                                                                (N=7)
                                                                   (N=36)       (N=34)
                             80                                    (N=7)
                                                      (N=7)
                                          (N=1)
                             40           (N=3)       (N=17)

                              0
                                  0   1           2            3            4            5            6           7           8
                                                              Cooperativeness Score
                                                               First Seven         Holdout
                                                                              45




                             Mean Contribution Path
                              Groups with Score = 2
               200
Contribution



               150
               100
                50
                 0
                                 10

                                      13
                                           16

                                                19

                                                     22

                                                          25

                                                               28

                                                                    31

                                                                         34
                     1

                         4

                             7




                                           Round

                             First Seven Games       Hold-out sample
                                                                                46




                             Mean Contribution Path
                              Groups with Score = 3
               200
Contribution



               150
               100
                50
                 0
                                 10

                                      13
                                           16

                                                  19

                                                       22

                                                            25

                                                                 28

                                                                      31

                                                                           34
                     1

                         4

                             7




                                           Round

                              First seven games        Hold-out sample
                                                                              47




                             Mean Contribution Path
                              Groups with Score = 4
               200
Contribution




               150
               100
                50
                 0
                                  10

                                       13

                                            16

                                                  19

                                                         22

                                                              25

                                                                   28

                                                                         31
                     1

                         4

                             7




                                            Round

                              First seven games        Hold-out sample
                                                                              48




                             Mean Contribution Path
                              Groups with Score = 5
               200
Contribution



               150
               100
                50
                 0                10

                                       13

                                            16

                                                  19

                                                         22

                                                              25

                                                                   28

                                                                         31
                     1

                         4

                             7



                                            Round

                              First seven games        Hold-out sample
                                 49




          Cross-Cultural Data
• NOTE:
  I did none (0) of this work
                                                                          50



     Public goods game results -
            Machiguenga        Are the
                               Machiguenga
                                                            Homo
            0.4                                             economicus?
                                       Machiguenga
            0.3
                      American
Frequency




            0.2



            0.1



             0
                  0   0.2        0.4        0.6       0.8   1

                       Fraction of Money Not Contributed
                                                                                               51




Orma public goods game results
Orma
•East-African
Pastoralists                0.4


                            0.3
                                                                       mean = 58%
                Frequency




                            0.2


                            0.1


                             0
                                  0   10   20    30   40    50    60     70   80    90   100
                                                % Contributed to Common Pot

Researcher: Jean Ensminger
                                                                     52


Tsimané Public Goods behavior is
different from both Westerners and
Machiguenga

                0.45
                 0.4
                0.35
   proportion




                 0.3
                0.25
                 0.2
                0.15
                 0.1
                0.05
                   0
                       0   13   27     40     53    67     80   93
                            % Contribution to common pot
                                                                   53




                   Conclusions
  Why do people cooperate in groups, and
what influences when and how much they do?
• People cooperate to induce others to & because others do
(reciprocity)
• What they know about others’ rates of cooperation is critical.
• The lowest contributor is important
•“It depends.” There seem to be reliable individual differences.
•Weak and Wild Speculation: Types recur. But, cultures differ
in the mix of Types, which emerge via a mysterious
ontogenetic process.
                                            54




         Research Agenda
               Information        ?
                 Seeking


               Group Size &   Individual
Asymmetrical
               Information    Differences
  Groups


                Reciprocity
                                       55




          Thank you

Collaborators: Athena Aktipis, Peter
  Descioli, Daniel Houser, Kevin
McCabe, Vernon Smith, Bart Wilson

				
DOCUMENT INFO