Groups as adaptive devices: Free-rider problems, the wisdom of crowds, and evolutionary games
Post on 05-Jan-2016
19 Views
Preview:
DESCRIPTION
Transcript
Groups as adaptive devices: Free-Groups as adaptive devices: Free-rider problems, the wisdom of rider problems, the wisdom of
crowds, and evolutionary gamescrowds, and evolutionary games
Tatsuya KamedaTatsuya KamedaHokkaido UniversityHokkaido University
Center for Experimental Research in Social Sciences
Center for the Sociality of Mind
July 23July 23Invited addressInvited address
1
Imagine you live in a tropical rain forest.
Your survival tasks include….
2
Gathering
3
Hunting
4
Avoiding predatory risks
5
Handling enemies
6
Groups as adaptive devices• Obviously, you cannot
survive in such a natural, uncertain environment by yourself.– Your life is highly dependent
on your group.
• In this sense, the image of groups as adaptive devices looks quite natural.– “Groups are wise and
intelligent.”7
• On the other hand, such a positive image of groups has been rather rare in the psychological literature on groups.
• Instead, for many years, psychologists have been keen on identifying various “biases” leading to group inefficiencies (cf. Krueger & Funder, 2004, Beh. Brain Sci.).
• Examples include (just to name a few):– Groupthink– Group polarization– Conformity with false group judgments – Etc., etc. 8
A caricature of such group imageA caricature of such group image
JapanJapan
KoreaKorea ChinaChina
the Houses of Parliament of Japan 9
Purpose of this talk: “Mind the gap!”
• How can we reconcile the two drastically different views?– “Group as an intelligent (adaptive) device”
vs.– “Group as a vehicle for various biases”
• In this talk, I’d like to approach this issue from a behavioral ecological perspective.
10
• Behavioral ecology (= study of animal behavior from the adaptationist perspective) is also concerned with group life.– Why do some animals form groups? What
functions are served by groups?– How efficient is group foraging as compared to
solitary foraging? – How is risk-monitoring conducted in groups?
• Although these are, in principle, the same kind of questions that have interested psychologists for many years, conversation with behavioral ecologists has been extremely rare.
11
Shared core questions• Group efficiency (Steiner, 1972)
– How efficient is group performance as compared to performance by isolated individuals?
• Collective wisdom (Surowiecki, 2004) – Can a group of individuals achieve a collective
wisdom beyond any single individual in the group including the best and brightest member?
– What cognitive, motivational, and ecological factors must exist for the collective wisdom to emerge?
12
Behavioral ecological literature“Group decision making”
by honey bees
• Seeley (1995) “The wisdom of the hive”
13
Searching a new nest• In a late spring or early summer, a colony
of honey bees often divides itself.
• The queen leaves with about 2/3 of the worker bees, and a daughter queen stays behind with the rest.
• How does the swarm that has left the colony find a new home?
14
“Search committee” composed of several hundred bees
• These “scout bees” fly out to inspect potential nest sites, and then perform waggle dances to advertise any good sites they have discovered.
• The duration of the dance depends on bee’s perception of the site’s quality: the better the site, the longer the dance.
• Other bees are more likely to visit and inspect the sites advertised by others.
• Thus, high-quality sites receive more advertisement and are visited by more scout bees.
15
• This process eventually leads to a group consensus.
• The striking empirical fact:- When different possible nest sites vary in quality,
the bees usually choose the best one.
- Seeley & Buhrman (2001). Beh. Ecol. Sociobiol.
- Seeley, Visscher, & Passino (2006). Amer. Sci.
16
Q. How do the bees solve the problem of interdependency?
• Communication among the bees via waggle dance could create sequential interdependencies between decision-makers.– Carry-over and amplification of initial
errors, such as seen in fads.– The honey bee GDM system may be
susceptible to the erroneous informational cascades (Bikhchandani et al., 1992, J. Polit. Econ.)
Errors in sequential communication
17
List, Elsholtz, & Seeley (in press). Phil. Trans. Roy. Soc. B.
• Computer simulation model assuming that:– Scout bees are interdependent in that they give
more attention to nest sites strongly advocated by others (i.e., conformity in nest-site search).
– Simultaneously, they are independent when assessing the quality of nest sites (i.e., independence in the preference formation).
• Duration of the dance is determined solely by own perception of the site’s quality.
• Such a right mix of independence and interdependence can yield a high-quality GDM.
18
So, what have we learned from the behavioral ecological literature?
• The honey bee “group decision making” provides a beautiful example of good coordination among members.
• Honey bees have built-in cognitive/behavioral systems to enable such coordination, which yields their collective wisdom in GDM.
19
Viewing this from human group psychology
• Steiner (1972) “Group process and productivity”– Group performance often suffers from two
sources of inefficiencies– Coordination problems
• Inefficiencies accruing from poor coordination among members
– Motivation problems• Social loafing (Latané et al., 1979, JPSP) “Many
hands make light the work.”
20
Motivation problems• Collective action, whereby members’ inputs are
pooled into a group performance while group outcomes are shared by all members, can cause motivational loss.– Individual costs vs. Shared group outcome
• Social dilemma (Dawes, 1980, An. Rev. Psych.)– Free-rider problem
• How do honey bees cope with the free-rider problem?
21
Reply from behavioral ecology
• Yes, free-rider problem is a serious threat to collective action.
• Fortunately, honey bees are basically free from the problem, because individuals in the same nest are kin. – Helping your kin is essentially helping your clones.
• But, generally, such strong kinship does not hold for human societies.
• So, given the free-rider problem, “collective wisdom” may not be guaranteed in human GDM. It’s your job to study it!
22
Are humans as smart as honey Are humans as smart as honey bees in GDM?bees in GDM?
(Kameda, Tsukasaki, & Hastie, in prep.) • Research question
– By computer simulations and an experiment, we (Hastie & Kameda, 2005, Psych Rev) have shown that Majoritarian Group Decision Making (as used by honey bees) works extremely well in locating resources in an uncertain environment.
– The Majoritarian GDM beats the best/brightest member in the group in terms of performance quality.
– But, the motivation problems were NOT handled explicitly in the study. 23
• So, here, we ask the following questions.
– When the free-rider problem exists, how efficient is the human Majoritarian GDM?
– If the logic of social dilemma (Dawes, 1980) applies, the Majoritarian GDM can easily degrade into a mob rule, where no member works for the group seriously.
– Can we overcome the free-rider problem in the collective action?
24
25
Experiment
• Purpose – Testing the ecological rationality of the
Majoritarian Group Decision Making when incentives for free-riding exist.
– Comparison to groups guided by the best/brightest dictator (Hastie & Kameda, 2005, Psych Rev)
– Test bed: “Foraging under uncertainty” setting created in a laboratory
26
Laboratory simulation of “foraging under uncertainty”
---- Brunswikian Paradigm --Brunswikian Paradigm --
Forager
Location j’s resource value, Qj
Environmental Events
Proximal Stochastic Cues
C1
C3
C2
error
error
error27
Procedure
28
• Participants– 180 (127 males and 53 females) Hokkaido
University undergraduates
• Six participants were called for each hourly session.
• Upon arrival, each participant was seated in a private cubicle connected by LAN.
• They received further instructions individually on computer displays.
29
• Individual practice session (20 trials)- Opportunities to learn about how to
use the 3 stochastic cues for making choices.
- Feedback about choice accuracies.- Participants could learn cue validities.
30
6-person team• Participants were then
instructed that they were a 6-person “hunting team.”
• Group Decision Task:Group Decision Task: Choosing the most profitable patch from 10 patches. – The resource in the chosen patch is shared
evenly among all members. – However, individual cooperation for GDM is
optional and costly. Free-rider problemFree-rider problem
31
Cooperation costs: A metaphorNOTICE: We’ll decide where to hunt for the next week. We’ll meet at 8:00 am on Sunday.
Sunday morning? Must I be there?
Voting (meeting) cost
Need to prepare for the meeting? Must I really search information?
Information-search cost
32
• Both types of cooperation costs are well-recognized in the political science literature about public choice (e.g., Downs, 1957, “An economic theory of democracy”).
– Voting (meeting) cost
Voter’s paradox
– Information search cost
Problem of ignorant voters
33
To recap,• 6-person foraging teams working in a
stochastic environment– Brunswikian choice task– Parallel to the honeybee GDM situation
• Individual cooperation for the team foraging is costly and optional.– Voting cost– Information search cost– Absent in the honeybee case (kinship)
• On the other hand, the resource in the chosen patch is shared evenly among ALL members.– Incentives for free-riding– No sanctioning opportunity was allowed in the experiment.
34
2 experimental conditions:Majority rule vs. Smart Dictatorship
(e.g., Hastie & Kameda, 2005, Psych Rev)
• Majority/plurality rule condition– The group follows the majority/plurality opinion
among voters about where to hunt.
Whether or not to incur:
- Information search cost?
and/or
- Voting cost?
“Patch 3”
“Patch 5”
“Patch 3”
Suppose 3 members appeared in the meeting (i.e., incurred the voting cost)
Patch 3
35
• Best member rule condition– The group follows the opinion of the best
member among voters (best as determined by the performance level during the practice session)
“Patch 3”
Suppose 2 members appeared in the meeting (i.e., incurred the voting cost)
Whether or not to incur:
- Information search cost?
and/or
- Voting cost?
“Patch 5”
Patch 3
36
• In the experiment, the aggregation via majority rule or the best member rule was done automatically by a computer program, after each participant decided whether or not to cooperate (i.e., to vote and/or to search information).
37
Results
• Theoretical prediction:– If the logic of social dilemma applies, we can
expect no cooperation no cooperation in a group.• No one engages in costly information-search.• No one incurs cost for voting.
– As a result, the Majoritarian Group Decision Making should degrade into a mob rule mob rule where nobody works seriously for the group.
– The collective wisdom, as displayed by the honey bees, may not be observed in human GDM.
39
Q. Are there any cooperative Q. Are there any cooperative members in groups?members in groups?
Mean frequencies of cooperative members (who search information AND vote) across trials.
Best Member Rule
Majoritarian GDM
0
1
2
3
4
5
6
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
A. Yes, cooperation persisted, A. Yes, cooperation persisted, and was stabilized over time at and was stabilized over time at about 3 out of 6 members.about 3 out of 6 members.
40
3 out of 6
Q. Which condition yielded greater mean Q. Which condition yielded greater mean individual net profit, the Majoritarian GDM or individual net profit, the Majoritarian GDM or
the Best Member Rule?the Best Member Rule?Mean individual net profits (in Yen)
61.3
66.3
56.8
45.7
56.7
45.6
0
10
20
30
40
50
60
70
80
1st 2nd 3rd
Block
F(1, 28) = 11.90, p<.01
Majoritarian GDMMajoritarian GDM
Best Member RuleBest Member Rule
41
A. The Majoritarian GDM.A. The Majoritarian GDM.
42
The human Majoritarian GDM The human Majoritarian GDM works well even when incentives works well even when incentives for free-riding exist !?for free-riding exist !?
Why did the Majoritan GDM work?
• “We know that some people behave altruistically in social dilemmas. This is a typical and robust finding in behavioral economics (e.g., Gintis, 2007, Beh. Brain Sci.). So, cooperation in human GDM is no news at all!”
• Some cooperation observed here may have originated from such purely “altruistic” motives.
• But, we don’t believe that the purely “altruistic” motive is the core reason for the stable cooperation in the collective action.
43
What is the core reason for stable cooperation in the collective action?
The Majoritarian GDM under uncertainty is NOT a social dilemma!
To see this, let us revisit the payoff structure in social dilemmas.
44
Individual payoff function in social dilemmas
0 1 2 3 4 5
Expe
cted
net
ret
urn
to e
go
Number of other cooperators in a group
When ego cooperates
When ego defects
Defection is a dominant strategy.
45
Rewriting this into Group Production Function: Group profit (per member) is a linear function of the number of cooperators in a group.
δ1
δ2
δ3
δ4
δ5
δ6
• Each increment by cooperation, δ (δ1 = δ2 = δ3 = δ4 = δ5 = δ6 ), is smaller than cost for cooperation.
• So, nobody cooperates. Social dilemmasSocial dilemmas
δ : increment in profit with an additional cooperator
46
Does this linear group production function hold for the Majoritarian Group Decision Making?
• Quality of the Majoritarian GDM improves with more cooperators who incur costs for information search and voting, but diminishes in margin.
Statistical property of the aggregation rule (law of large numbers)
No, group production function is NOT linear!
47
0
0.5
1
1.5
2
2.5
3
3.5
4
0 1 2 3 4 5
Number of other cooperators in a group
Exp
ecte
d ne
t ret
urn
to e
go
When ego cooperates
When ego defects
Individual Payoffs in the Majoritarian GDM
• Different from the social dilemma, no dominant (pure) strategy exists.
• When there are MANY OTHER cooperators, you’re personally better off defecting. • But, when there are only FEW OTHER cooperators, you’re personally better off cooperating.
Mixed Mixed equilibriumequilibrium
48
To recap,• The Majoritarian GDM under uncertainty is not a
social dilemma!– Marginally diminishing group production– Neither cooperation nor defection is dominant.
• A mixed equilibrium thus emerges where cooperators and defectors coexist in a stable manner.
• Thanks to those “rational cooperators”, the Majoritarian GDM can outperform the smart dictatorship by the best and brightest member.
49
Evolutionary computer simulations
50
• The experiment has demonstrated the effectiveness of Majoritarian GDM under uncertainty.
• But, how robust is this observation?– Parametric constraints?
An agent-based evolutionary computer simulations to see robustness of the Majoritarian GDM, while varying key parameters systematically.
51
What is evolutionary simulation?• Most famous example:
– Axelrod (1984) “Evolution of cooperation”– Tit-for-tat strategy in repeated Prisoner’s
Dilemma
• Interactions of agents with various strategies
• A strategy performing better than the other strategies in net profits proliferates gradually in the population -- Analogous to biological evolution.
• Does a stable equilibrium (“evolutionary equilibrium”) emerge in the population?
52
Simulation procedure
53
Infinite Infinite populationpopulation
Cooperator (vote and search information)
DefectorSampling many 12-Sampling many 12-person groups (“hunting person groups (“hunting teams”)teams”)
Each group makes decisions by majority rule about where to hunt. Only cooperators incur costs for information search and voting. But, the resource in the chosen patch is shared equally by all 12 members.
Calculate average net profits for cooperators and defectors across the groups. Agents with more fit strategies produce slightly more offspring for next generation (replicator dynamic).
SelectioSelectionn
54
Infinite Infinite populationpopulation
CooperatorDefector
Sampling of many12-Sampling of many12-person groups person groups (“hunting teams”)(“hunting teams”)
SelectioSelectionn
The simulation repeats these steps for The simulation repeats these steps for many “generations” until an equilibrium many “generations” until an equilibrium emerges in the population.emerges in the population.
55
Simulation results
56
Equilibrium proportions of cooperators and defectors in the population as a function of cooperation cost
Cooperator (searcher/ voter)
Defector (non-searcher/ abstainer)
Mixed Mixed equilibriumequilibrium 57
0.0
3
0.0
6
0.0
9
0.1
2
0.1
5
0.1
8
0.2
1
0.2
4
0.2
7 2
5
8
11
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
Differencein
individualpayoff
(Majority -Best)
Cooperation costs
Number ofchoice alts(patches)
Majoritarian Majoritarian GDM yields GDM yields greater greater individual net individual net payoffs than payoffs than the Best the Best Member RuleMember Rule..
0
58
Q. Which rule yielded a greater profit to individuals?
Both the experiment and the simulation results suggested:
• The Majoritarian GDM is ecologically rational and robust under uncertainty.– Despite the inherent free-rider problem, the
Majoritarian GDM functions as a “fast-and-frugal” decision rule (Gigerenzer, Todd & ABC, 1999).
– This may explain the immense popularity of the rule in many societies, including tribal (Boehm, 2000) as well as industrialized (Kameda, Tindale & Davis, 2003; Kerr & Tindale, 2005) societies.
59
GDM Under Uncertainty: A Historic Example
Lewis and Clark’s expedition of the American West (1804-1806) “the Corps of Discovery”“the Corps of Discovery”
Winter of 1805
St. Louis 1804
60
November 24, 1805
• To make the crucial decision of where to spend the winter, the captains decide to put the matter the captains decide to put the matter to a voteto a vote. Significantly, in addition to the others, Clark’s slave, York, is allowed to vote – nearly 60 years before slaves in the U.S. would be emancipated and enfranchised. Sacagawea, the Indian woman, votes too – more than a century before either women or Indians are granted the full rights of citizenship.
The majoritymajority decides to cross to the south side of the Columbia, near modern-day Estoria, Oregon, to build winter quarters.
(quoted from PBS Online)61
Captain Clark’s meta-decision about how to decide (use of the Majoritarian GDM) had a solid adaptive ground under uncertainty!
62
63
ConclusionConclusion
• This talk aimed to explore applicability of behavioral ecological notions in the study of human group behavior.– I used GDM (by honey bees and humans) as a test case
to see the applicability.
• The exploration suggests, I hope, that the behavioral ecological notions are highly useful for the study of human group behavior.– These notions free us from the rather fragmentary, overly
pessimistic images of human groups (i.e., the “biases” approach).
64
Lessons from behavioral ecology
(1) Analysis of group tasks is quite important to understand human group behavior.– These tasks should not be arbitrary or contrived ones,
but must represent our ecological (social as well as natural) environments.
– Examples of such representative group tasks include:• Foraging• Risk-monitoring• Group defense• Etc., etc.
65
(2) Given the task analysis, the marginally-diminishing group performance curve seems quite universal in our ecological environments.
66
67
• Truth-finding “Eureka” situations– Risk-monitoring against predators and enemies
(e.g., Kameda & Tamura, 2007, JESP)
– Searching survival resources (e.g., Kameda, Ishibashi & Hastie, in prep; Kameda & Nakanishi, 2002, 2003, Evol. Hum. Beh.)
– Group memory (e.g., Hinsz, 1990, JPSP; Weldon et al., 2000, JEP: Learn. Mem. Cog.)
• Pooling members’ physical inputs– Tug of war (Latane’ et al., 1979, JPSP)
– “Group effort” (Karau & Williams, 1993, JPSP)
(3) Social behavior is fundamentally strategic.
Game theory is thus indispensable for better understanding of human group behavior.
68
0
0.5
1
1.5
2
2.5
3
3.5
4
0 1 2 3 4 5
Number of other cooperators in a group
Exp
ecte
d ne
t ret
urn
to e
go
Mixed Mixed equilibriumequilibrium
This way of thinking may eventually bridge the seemingly formidable gap between the images of “group as an intelligent/adaptive device” and “group as a vehicle for various biases” toward a fuller and more comprehensive understanding of human group behavior.
69
Thank you very much!Thank you very much!
70
Network Lab at the Center for Experimental Research in Social Sciences 16 cubicles and a control room connected by LAN.
71
Cooperation and coordination are fundamentally inseparable.
72
• The number of “rational” cooperators at the equilibrium depends on the steepness of the group production curve.
• Improvement in group coordination More cooperation
73
• Therefore, the problem of cooperation in natural groups cannot be separated from the problem of coordination.
• These two features are intertwined inseparably in ecological environments.– Overlooked in the cooperation literature as
well as in the group literature.
0
0.5
1
1.5
2
2.5
3
3.5
4
0 1 2 3 4 5 6
Gro
up
pro
fit p
er
me
mbe
r
Number of cooperators in a group
δ1
δ2
δ3
δ4
δ5
δ6
δ : increment in profit with an additional cooperator
• Each increment by cooperation, δ, is not constant: δ1 > δ2 > δ3 > δ4 > δ5 > δ6 .• In such a marginally-diminishing payoff structure, your personal benefit from your cooperation can exceed your cost, when there are only few other cooperators.
E.g., δ1 > δ2 > cooperation cost 74
top related