Altruistic Punishment and Human Cooperation

1
Altruistic Punishment and Human Cooperation

• Urs Fischbacher
• University of Zurich
• NWO, Utrecht 2004
• Fischbacher, Gächter and Fehr, Are People
  Conditionally Cooperative? Evidence from a Public
  Goods Experiment, Economics Letters 2001.
• Fehr and Gächter, Altruistic Punishment in
  Humans, Nature 2002.
• Fehr and Fischbacher, Third Party Punishment and
  Social Norms, Evolution and Human Behavior 2004.
• Fehr and Fischbacher, The Nature of Human
  Altruism, Nature 2003.
• De Quervain, Fischbacher, Treyer, Schellhammer,
  Schnyder, Buck, and Fehr, The Neural Basis of
  Altruistic Punishment, Science 2004.

2
Overview

• Human cooperation and strong reciprocity
• Experimental evidence for strong reciprocity
• Proximate models of strong reciprocity
• Altruistic punishment activates reward related
  areas in the brain
• Ultimate models of strong reciprocity

3
Humans Large-Scale Cooperation

• Humans societies are a huge anomaly in the animal
  world. They are based on a detailed division of
  labor and cooperation of genetically unrelated
  individuals in large groups.
• In most animal species there is little division
  of labor and cooperation is limited to small
  groups.

4
Why do Humans cooperate?

• Strategic cooperation (cooperation to induce
  cooperation by the other players) in the form of
• Reciprocal altruism, i.e. self-interested
  exchanges in repeated interactions, at a scale
  and in domains of behavior that is unprecedented
  in the animal world.
• Reputation-based cooperation is also a powerful
  force among humans and differs in scale and in
  kind from what has so far been observed in
  animals.
• However, human altruism even goes beyond
  reciprocal altruism and reputation-based
  cooperation, taking the form of strong
  reciprocity.

5
Strong Reciprocity

• Is a combination of altruistic rewarding (strong
  positive reciprocity) and altruistic punishment
  (strong negative reciprocity).
• Altruistic rewarding A readiness to incur costs
  to reward others for cooperative, norm-abiding
  behaviors in the absence of any individual
  economic benefit for the rewarding individual.
• Altruistic punishment A readiness to incur costs
  to punish others for norm violations in the
  absence of any individual economic benefits for
  the punishing individual.

6
Public-Goods Experiment

• N players get an endowment.
• Decide simultaneously how many point of they
  contribute to the public goods.
• The contributions are summed up, multiplied with
  a factor F (e.g. 2) and distributed equally
  between all players.
• If Fgt1, it is efficient to contribute
  (cooperate).
• If F/Nlt1, it is a dominant strategy not to
  contribute (defect).
• Structure mimics the logic of many important real
  world examples. Whenever individual actions have
  positive or negative effects on other individuals
  a similar situation arises Pollution problems,
  over-fishing the seas, cooperative production and
  food-sharing in small-scale societies,
  cooperative hunting and warfare, etc.

7
Altruistic Rewarding

• (Fischbacher et al. 2001, see also FKR 93 or BDM
  95)
• Standard public goods situation (endowment 20, N
  4, F1.6) played only once
• Subjects can make a conditional contribution to
  the project, i.e. they fill out a contribution
  table in which they can condition their
  contribution on every possible contribution of
  the others

8
Predictions

• Selfish subjects (e.g. subjects who cooperate for
  strategic reasons only) always put in zero into
  the schedule.
• Strongly reciprocal subjects contribution
  increases in the average contribution of the
  other group members.
• The other subjects contribution is a cooperative
  act which deserves altruistic rewarding.

9
Average schedulesFischbacher, Gächter, Fehr 2001
20
18
Strong reciprocators 50
16
14
12
Own contribution
10
Mean (N44)
8
6
4
Hump-shaped 14
2
Selfish 30
0
0
2
4
6
8
10
12
14
16
18
20
Average contribution level of other group members
10
Altruistic Punishment (Fehr Gächter, American
Economic Review 2000, Nature 2002)

• Public goods game as above.
• Six periods to allow for learning and to study
  the stability of cooperation. At the end of each
  period group members are informed about
  individual contributions of other group members
  without revealing their identities.
• No repeated interaction with the same subjects.
  In each period each subject faces new group
  members.
• Nobody knows the previous actions of the other
  group members.

11
Altruistic Punishment Treatments

• Control treatment exactly as described above.
• Punishment treatment adds the opportunity to
  punish other group members after being informed
  about their investments. Two Stages in each
  period
• The first stage is identical to the control
  treatment.
• At the second stage each group member can
  allocate punishment points to the other members.
• The first stage payoff of the punished
  individuals is reduced.
• Punishing is costly for the punisher. Each
  "invested" into punishment reduces the payoff of
  the sanctioned player by 3.

12
Predictions with selfish individuals

• Since punishment is costly for the punisher and
  yields not material benefits no selfish subject
  will ever punish.
• If nobody punishes in the punishment condition
  then the cooperation behavior in the punishment
  condition is predicted to be identical to the
  behavior in the control condition.
• ?In both treatments cooperation should be zero.

13
Cooperation without and with punishmentSource
FehrGächter Nature 2002
without punishment
20
18
16
14
12
Mean contribution
10
8
6
4
2
0
1
2
3
4
5
6
7
8
9
10
11
12
Period
14
Cooperation without and with punishmentSource
FehrGächter Nature 2002
with punishment
without punishment
20
18
16
14
12
Mean contribution
10
8
6
4
2
0
1
2
3
4
5
6
7
8
9
10
11
12
Period
15
Cooperation with and without punishment Source
FehrGächter Nature 2002
16
Punishment
17
Is Punishment an altruistic act?

• The presence of punishers establishes a credible
  threat that deters non-cooperation - all group
  members benefit from this threat.
• Punished subjects contribute more in the next
  periods - future interaction partners of the
  punished subjects benefit from the punishment.
• Punishing subjects bear costs.

18
Strong reciprocity is documented in dozen of
studies

• It has been documented in a wide variety of
  situations
• It applies among strangers. Virtually all
  experiments implement anonymous interactions
  among subjects.
• Confirmed under experimenter-subject anonymity
  (Berg et al. 1995, Bolton and Zwick 1995, Abbink
  et al. 1997, etc.)
• Confirmed under rather high stake levels (Cameron
  1999, Fehr, Tougareva Fischbacher 2002, three
  months' income)
• Confirmed under one-shot repetitions (Roth et al.
  1991, Fehr et. al 1998, Charness 1996, etc.)
• Strong variation across different small-scale
  societies (Henrich, Boyd, Bowles, Camerer,
  Gintis, Fehr McElreath 2001)

19
Proximate Motives behind Strong Reciprocity

• Inequity aversion
• Fehr Schmidt 1999, Bolton Ockenfels 2000.
• Ui pi ai pi pj
• Intention based reciprocity
• Rabin 1993, Levine 1998, Dufwenberg
  Kirchsteiger 2004, Falk Fischbacher
  forthcoming.
• Ui pi ri kindnessj-gti pj
• All theories assume a fairness motive in addition
  to self interest.

20
Neural Basis of Altruistic punishment De
Quervain, Fischbacher, ....., and Fehr, Science
2004

• There is well documented evidence for reward
  related areas in the brain (Nucleus Accumbens,
  Nucleus Caudate). These areas are activated when
  subjects get reward in the form of
• Money
• Beautiful faces
• Cocain
• Fairness theories assume that people derive
  utility from altruistic rewarding and from
  altruistic punishment.
• Are reward related areas in the brain also
  activated when subjects have the opportunity to
  punish?

21
The basic game

• Two traders, A and B, are matched anonymously.
  The good possessed by A is four times more worth
  for trader B. Thus, if A gives the good to B and
  B pays A a fair share of the gains from trade
  both traders can benefit.
• However, trade takes place sequentially, i.e., A
  first has to give the good to B, then B pays A.
  Thus, A has to trust B and B can abuse A's trust
  by not paying.
• Both are endowed with 10 MUs. A can send his 10
  MUs to B. The experimenter quadruples this amount
  so that B has, in total, 50 MUs. Then B can send
  back 25 MUs to A.
• After B has made his payment decision A has the
  opportunity to punish B. By spending 1 MU on
  punishment he can reduce B's income by 2 MUs. A
  can spend up to 20 MUs on punishment.

22
Behavioral Results

• The vast majority of A sends the 10 MUs.
• Roughly 50-60 of the B's send back nothing.
• Roughly 80 of the A's punish those B's who abuse
  their trust.
• Average payoff reduction for the B's is 23 MUs.

23
Treatment conditions

• Punishment is costly for both A and B (Costly,
  IC). A is hypothesized to experience a desire to
  punish cheating and he can in fact punish.
• Punishment is only symbolic, i.e., A and B have
  no costs of punishing (Symbolic IS). A is also
  hypothesized to experience a desire to punish
  cheating but he cannot punish.
• Punishment is free for A but costly for B (Free,
  IF). A is hypothesized to experience a desire to
  punish cheating and he can in fact punish - even
  without a cost.
• We scanned the brain of player A (with PET) in
  the sequential trading game when A's trust was
  abused and A decided whether (and how much) to
  punish B.

24
Hypothesis

• The possibility for punishing unfair behavior
  activates reward-related neural circuits.
  (Nucleus Accumbens, Nucleus Caudate).
• IF - IS is hypothesized to activate reward
  related brain regions.
• IC - IS is also hypothesized to activate reward
  related brain regions

25
IF-IS and IC-IS do activate the caudate nucleus
26
Individuals with higher caudate activation punish
more I

• Is the activation caused by the punishment act?

27
Individuals with higher caudate activation punish
more II

• Those with high caudate activation in IF
  treatment punished more in the IC treatment.
• Caudate activation has to do with expected
  satisfaction of punishment.

28
Overview

• Human cooperation and strong reciprocity
• Experimental evidence for strong reciprocity
• Proximate models of strong reciprocity
• Altruistic punishment activates reward related
  areas in the brain
• Ultimate models of strong reciprocity

29
Prevailing Evolutionary Theories of Human
Cooperation

• Kin Selection (Hamilton 1964) - Individuals are
  genetically related
• Reciprocal Altruism (Trivers 1971, Axelrod and
  Hamilton 1981) - Individuals are engaged in
  repeated interactions. Helping today yields
  benefits from the other individual in the future.
  
• Indirect Reciprocity (Alexander 1987, Nowak and
  Sigmund 1998) - Helping creates a good reputation
  in the group. Individuals with a good reputation
  are more likely to receive help from others in
  the future.
• Signaling (Zahavi and Zahavi 1997) - Cooperative
  acts signal personal qualities that are not
  directly observable like, e.g., good genes. The
  signals generate some benefits in the future.

30
Problem of the Theories in Explaining Large-Scale
Cooperation

• Kin selection Cooperation limited to close kin.
  Subjects in experiments are unrelated strangers.
• Reciprocal altruism, indirect reciprocity
  Cooperation limited to situation in which
  reputation can be formed, cooperation in
  experiments also in one-shot situations.
• Signaling theory In the absence of selection
  between groups it is hard to understand why the
  signal is pro-social.
• Moreover, all these theories apply, in principal,
  equally well to many animal species. They do not
  answer the question, why humans are such an
  outlier.

31
Maladaption

• Theories above can rationalize strong reciprocity
  only as a maladaptive trait.
• i.e., the proximate mechanisms driving human
  behavior are not yet fine-tuned to interactions
  among unrelated people in non-repeated
  interactions where reputational gains are small
  or absent.
• Problem of the maladaption hypothesis
• Humans are capable to distinguish between
  situations in which reputation can be gained and
  situation in which this is impossible.

32
Ultimatum game(Güth et al. 1982)

• A proposer and a responder are matched
  anonymously. The proposer receives 10 money units
  and must make one proposal how to allocate the
  money between the two players.
• If the responder accepts, the proposal is
  implemented. If he rejects, both get nothing.

33
Ultimatum game with reputation

• Treatment condition without reputation
• Normal ultimatum game. Repeated with different
  players.
• Treatment condition with reputation
• Proposers get to know which offers were rejected
  in the past by the responder they are matched
  with. Repeated with different players.
• Maladaption prediction Subject cannot
  distinguish between situations in which they can
  build up reputation and situation in which they
  cannot. Therefore Whether responders can build
  up reputation for being tough or not, they have
  the same threshold for accepting.

34
Average Rejection Threshold in Ultimatum Game
with and without Reputation Formation(Source
Fehr and Fischbacher, NATURE 2003)
35
Rejection Threshold in Ultimatum Game with and
without Reputation Formation(Source Fehr and
Fischbacher, NATURE 2003)
5
4.5
4
3.5
3
Threshold with reputation
2.5
2
1.5
1
0.5
0
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Threshold without reputation
36
Rejection Threshold in Ultimatum Game with and
without Reputation Formation(Source Fehr and
Fischbacher, NATURE 2003)
5
4.5
4
3.5
3
Threshold with reputation
2.5
2
1.5
1
0.5
0
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Threshold without reputation
37
The Evolution of Altruistic PunishmentBoyd,
Bowles, Gintis and Richersen, PNAS 2003

• Types of behavior
• Contributors incur cost c to produce total
  benefit b, which is shared equally among n group
  members.
• Defectors incur no costs and produce no
  benefits.
• Altruistic Punishers contribute and punish all
  those who defect at cost k for themselves and
  cost p for each defector.
• If there are no punishers, individual selection
  favors defectors over contributors.
• If punishers are frequent, defectors do worse
  than altruistic punishers and contributors.
• However, contributors do always better than
  altruistic punishers.

38
The Evolution of Altruistic Punishment IIBoyd,
Bowles, Gintis and Richersen, PNAS 2003

• Evolutionary dynamics
• Individual selection Individuals imitate more
  successful individuals within the group.
• Migration between groups.
• Group selection mechanism With some probability
  unsuccessful groups are extinct and replaced by
  successful groups.

39
Simulation Results Fehr/Fischbacher, Nature
2003 based on Boyd et al. PNAS 2003
1
0.9
0.8
0.7
0.6
Average cooperation rate
0.5
0.4
0.3
no punishment possible
0.2
0.1
0
2
4
8
16
32
64
128
256
Group size
40
Simulation Results Fehr/Fischbacher, Nature
2003 based on Boyd et al. PNAS 2003
1
0.9
0.8
0.7
0.6
Average cooperation rate
0.5
0.4
punishment of defectors
0.3
possible
no punishment possible
0.2
0.1
0
2
4
8
16
32
64
128
256
Group size
41
Why Does Selection not Remove Altruistic
Punishers?

• If punishers are frequent and defectors are rare,
  punishers rarely incur the cost of punishment.
  Thus, in the absence of mutant defectors
  punishers would do equally well as pure
  contributors.
• In the presence of mutant defectors punishers
  have a small disadvantage relative to pure
  contributors.
• Selection among groups can outweigh this
  disadvantage of altruistic punishers.
• Remark Group selection without punishment does
  not work Without punishment cooperators have a
  fitness disadvantage independent of their
  frequency.

42
Simulation Results Fehr/Fischbacher, Nature
2003 based on Boyd et al. PNAS 2003
43
Why does Migration not Undermine Group Selection?

• Because it is based on a cultural process of
  payoff-biased imitation. Those who have a high
  payoff are imitated.
• Traditionally, in genetic models of group
  selection migration and within-group selection
  remove between-group differences in the share of
  defectors. Thus, group selection cannot become
  operative.
• Payoff-biased imitation maintains group
  differences. In groups with a low share of
  altruistic punishers defectors do best and they
  are imitated. In groups with a high share of
  punishers, contributors do best and they are
  imitated and defectors do worst.

44
Summary

• Human cooperation represents a spectacular
  outlier in the animal world. This is probably due
  to human forms of altruism that are unique in
  kind and in scope.
• Reciprocal altruism and reputation-seeking are
  powerful forces of cooperation in dyadic
  interactions.
• However, humans exhibit even strong reciprocity,
  a combination of altruistic rewarding and
  altruistic punishment that is associated with net
  costs for the altruist.
• Altruistic punishment is key for understanding
  cooperation in multi-lateral interactions.
  Without altruistic punishment cooperation
  unravels if opportunities for altruistic
  punishment exist cooperation flourishes.
• Humans seem to experience altruistic punishment
  as psychologically rewarding. Caudate nucleus is
  a key component in the neural circuits involved
  in altruistic punishment.
• Reciprocal altruism and reputation-seeking are
  powerful forces of cooperation in dyadic
  interactions but they have difficulties in
  explaining the evolution of cooperation in
  N-person public goods situations.

45
The end
46
Conditional cooperation design

• (Fischbacher et al. 2001, see also FKR 93 or BDM
  95)
• Standard public goods situation (endowment 20, N
  4, F1.6) played only once
• Subjects have to make two decisions
• An unconditional contribution to the project
• A conditional contribution to the project
  (conditional on every possible contribution of
  the others called contribution table)
• For 3 subjects, their unconditional contribution
  is relevant. For a randomly selected group member
  his/her contribution schedule is relevant for the
  decision.

47
Testing Evolutionary Models
Environment Game(s)
Types of behavior

• Are the types complete?
• Are there no type who can invade the population?
• Does the type distribution correspond the
  distribution which is actually observed?
• This question can be addressed with experiments.

• Is the environment representative?
• Does the game correspond the the interaction how
  it actually took place in the relevant time
  period?

Evolutionary Dynamics
48
Simulation Results Fehr/Fischbacher, Nature
2003 based on Boyd et al. PNAS 2003
49
Typical experimental outcome Isaac, Walker,
Thomas (1984)

• There is cooperation.
• Cooperation declines over time.

10HN10, F7.5 4H N4, F3 10L N10, F3 4L
N4, F1.2