Evolution and the Social Contract

Brian Skyr ms

The Tanner Lectures on Human Values

Delivered at

The University of Michigan

November 2, 2007

Peterson_TL28_pp i-250.indd 47 11/4/09 9:26 AM

 Brian Skyrms is Distinguished Professor of Logic and Philosophy of
Science and of Economics at the University of California, Irvine, and Pro-
fessor of Philosophy and Religion at Stanford University. He graduated
from Lehigh University with degrees in Economics and Philosophy and
received his Ph.D. in Philosophy from the University of Pittsburgh. He
is a Fellow of the American Academy of Arts and Sciences, the National
Academy of Sciences, and the American Association for the Advancement
of Science. His publications include The Dynamics of Rational Delibera-
tion (1990), Evolution of the Social Contract (1996), and The Stag Hunt and
the Evolution of Social Structure (2004).

Peterson_TL28_pp i-250.indd 48 11/4/09 9:26 AM

 Dewey and Darwin
Almost one hundred years ago John Dewey wrote an essay titled “The In-
fluence of Darwin on Philosophy.” At that time, he believed that it was
really too early to tell what the influence of Darwin would be: “The exact
bearings upon philosophy of the new logical outlook are, of course, as yet,
uncertain and inchoate.” But he was sure that it would not be in provid-
ing new answers to traditional philosophical questions. Rather, it would
raise new questions and open up new lines of thought. Toward the old
questions of philosophy, Dewey took a radical stance: “Old questions are
solved by disappearing . . . while new questions . . . take their place.”
I don’t claim that the old philosophical questions will disappear, but
my focus is the new ones. Evolutionary analysis of the social contract—be
it cultural or biological evolution—does not tell you what to do. Rather, it
attempts to investigate how social conventions and norms evolve—how
social contracts that we observe could have evolved and what alternative
contracts are possible.
The tools for a Darwinian analysis of the social contract are those of
evolutionary game theory. From the theory of games it takes the use of
simple stylized models of crucial aspects of human interaction; from evo-
lution it takes the use of adaptive dynamics. The dynamics need not have
its basis in genetics; it may as well be a dynamic model of cultural evolu-
tion or of social learning (Björnerstedt and Weibull 1996; Schlag 1998;
Samuelson 1997; Weibull 1995). No part of what follows implies any kind
of genetic determinism or innateness hypothesis. Problems of coopera-
tion may be solved by genetic evolution in some species and by cultural
evolution in others.
Here are three features of the evolutionary approach to bear in mind in
the ensuing discussion:

1. Different social contracts have evolved in different circumstances.

2. Existing social contracts are not altogether admirable.
3. We can try to change the social contract.

Correlation and the Evolution of Cooperation

Cooperation may be easy or hard, or somewhere in between. Here is a
game-theory model of an easy problem of cooperation; I call it Prisoner’s
Delight:

[49]

Peterson_TL28_pp i-250.indd 49 11/4/09 9:26 AM

 50 The Tanner Lectures on Human Values

Prisoner’s Delight
Cooperate Defect
Cooperate 3 1
Defect 2 0

The entries in the table represent the payoffs of Row’s strategy when
played against Column’s strategy. If your partner cooperates, you are bet-
ter off if you do as well, for a payoff of 3 rather than 2. If your partner
defects, you are still better off cooperating for 1 rather than 0, although
your partner does even better with 2. So no matter what your partner does,
you are better off cooperating. Your partner is in the same situation, and
reasons likewise. It is easy to cooperate.
We can change the story a little bit to fit a different kind of situation.
Perhaps if your partner defects, your attempt at cooperation is counter-
productive. You are better off defecting. This change gives us the game
known as the Stag Hunt:

Stag Hunt
Cooperate Defect
Cooperate 3 1
Defect 2 2

In the Stag Hunt, what is best for you depends on what your partner
does. If you both cooperate, you are both doing the best thing given your
partner’s action—likewise if you both defect. Cooperation is more dif-
ficult. It is an equilibrium, but not the only one.
Another modification of the situation calls for a different game. Sup-
pose that defecting against a cooperator actually pays off. Then we have
the Prisoner’s Dilemma:

Prisoner’s Dilemma
Cooperate Defect
Cooperate 3 1
Defect 4 2

Your optimal act again no longer depends on what your partner does.
Now it is to defect. Cooperation is hard.
Each of these games is a reasonable model of some social interactions.
Two men sit in a rowboat, one behind the other. Each has a set of oars.

Peterson_TL28_pp i-250.indd 50 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 51

They have been out fishing, and a hot dinner awaits them across the lake. If
one doesn’t row for some reason, the other will row to get them there; if one
does row, the other prefers to row also to get home faster. This is Prisoner’s
Delight. Cooperation is easy. Now change the picture. They sit side by side,
and each has one oar. One man rowing alone just makes the boat go in
circles. This is a Stag Hunt.1 Back to the first rowboat with two sets of oars,
but take away the hot dinner on the opposite shore and suppose that the
men are very tired. They could camp on this shore for the night, although
they would prefer to get back to the opposite shore. But either prefers not
to row no matter what the other does. This is a Prisoner’s Dilemma.
Consider any reasonable adaptive dynamics operating on a large popu-
lation of individuals paired at random to play one of these games. Then in
Prisoner’s Delight cooperation takes over the population, in Prisoner’s Di-
lemma defection goes to fixation, and in Stag Hunt one or the other may
prevail depending on the initial composition of the population.
There is an enormous literature devoted to explaining the evolution
of cooperation in the Prisoner’s Dilemma, while the other two games are
relatively neglected. Everyone wants to crack the hardest problem—the
evolution of altruism.2
All these accounts of cooperation in Prisoner’s Dilemma either (1) use
an interaction that is not really a Prisoner’s Dilemma or (2) use pairings to
play the game that are not random. It must be so.3 Suppose the interaction
is a Prisoner’s Dilemma, and pairings are random in a large population.
Then cooperators and defectors each have the same proportion of part-
ners of each type. Defectors must on average do better than cooperators.
Replicator dynamics increases the proportion of the population of the
type that does better, and that’s all there is to it.
But if nature somehow arranges for positive correlation—for coopera-
tors to meet cooperators and defectors to meet defectors more often than
they would with random matching—then it is possible for cooperators to
do better than defectors. The point is obvious if we consider perfect correla-
tion. Then the relevant comparison in payoffs is not vertical but diagonal:

1.  This is David Hume’s famous example from the Treatise of Human Nature. For more
game theory in Hume, see Vanderschraaf 1998.
2.  Binmore (1994) devotes a chapter titled “Squaring the Circle in the Social Sciences” to
attempts to justify cooperation in the one-shot Prisoner’s Dilemma.
3.  This is the “Iron Rule of Selfishness” of Bergstrom (2002). See also Eshel and Cavalli-
Sforza 1982.

Peterson_TL28_pp i-250.indd 51 11/4/09 9:26 AM

 52 The Tanner Lectures on Human Values

Prisoner’s Dilemma
Cooperate Defect
Cooperate 3 1
Defect 4 2

Every explanation of the evolution of cooperation in real Prisoner’s

Dilemmas—kin selection, group selection, repeated games, spatial inter-
action, static and dynamic interaction networks, and all the rest—works
by providing a mechanism that induces correlation in plays of the Pris-
oner’s Dilemma. This was clear to William Hamilton and to George Price
back in the 1960s (Hamilton 1964, 1995; Price 1970; Eshel and Cavalli-
Sforza 1982; Frank 1995). (A version of Hamilton’s rule for kin selection
can be derived just from the positive correlation.)
But to say that a mechanism can sometimes generate enough positive
correlation to maintain cooperation in the Prisoner’s Dilemma is not to say
that it can always do so. In some circumstances correlation may fall short.
Thus, in each of these accounts, an examination of specific correlation
mechanisms is of interest. And often a scenario can be analyzed both as Pris-
oner’s Dilemma with correlation and as a larger game in which the plays of
Prisoner’s Dilemma are embedded. I will illustrate with two examples.
Axelrod (1984) directs our attention to the shadow of the future. Co-
operation may be maintained not by immediate payoffs but by the conse-
quences of current actions on future cooperative behavior of partners. In
this he follows Thomas Hobbes and David Hume:
hobbes: He, therefore, that breaketh his Covenant, and consequently
declareth that he think that he may with reason do so, cannot be re-
ceived into any society that unite themselves for Peace and Defense,
but by the error of them that receive him.
hume: Hence I learn to do a service to another, without bearing him
any real kindness; because I foresee, that he will return my service, in
expectation of another of the same kind, and in order to maintain the
same correspondence of good offices with me and with others.

Axelrod, following John Nash and other founding fathers of game

theory,4 analyzes the shadow of the future using the theory of indefinitely

4.  The use of discounted repeated games to explain cooperation in Prisoner’s Dilemma is
already to be found in Luce and Raiffa 1957, with no claim of originality. John Nash is reported
to have invented the explanation in conversation.

Peterson_TL28_pp i-250.indd 52 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 53

repeated games. Suppose that the probability that the Prisoner’s Dilemma
will be repeated another time is constant. This (somewhat far-fetched)
idealization of geometrical discounting of the future allows us to sum an
infinite series and compute the expected payoffs of strategies in the large
repeated game. For simplicity, we consider only two strategies in the re-
peated game, Always Defect and Tit for Tat. A Tit for Tat player initially
cooperates and then does what was done to him in the preceding round,
and Always Defect is self-explanatory. Two players remain matched for
the whole repeated game, but this restrictive assumption can be relaxed
in more complicated “community enforcement” models (Sugden 1986;
Kandori 1992; Milgrom, North, and Weingast 1990; Nowak and Sigmund
1998).
Since the payoffs on each individual play are those of Prisoner’s Di-
lemma, the strategies in the repeated games must induce a correlation be-
tween individual plays of Cooperate and Defect if cooperation is not to
be driven to extinction. The presence of Tit for Tat players in the popula-
tion is this correlation device. They always cooperate with each other and
quickly learn to defect against defectors.
What is the larger game in which plays of Prisoner’s Dilemma are em-
bedded? Using the version of Prisoner’s Dilemma given before and prob-
ability of another trial as six-tenths, we get:

Tit for Tat Always Defect

Tit for Tat 7.5 4
All Defect 7 5

This is a Stag Hunt. There are two stable equilibria, one where everyone
always plays Tit for Tat and one where everyone always plays Always De-
fect. Which one you get depends on initial population proportions. In our
example an initial population split equally evolves to universal defection.
Sober and Wilson (1998) direct our attention to group selection. Con-
sider the haystack model of Maynard Smith (1964). In the fall, farmers cut
hay and field mice randomly colonize the haystacks. In the haystacks they
play the Prisoner’s Dilemma and reproduce according to the payoffs. In
the spring the haystacks are torn down, the mice scatter, and the cycle is
continued. If there are enough generations of mice in the life of a haystack,
then it is possible for cooperators to do better on average than defectors.
This is because differential reproduction within haystacks creates positive
correlation in the population. In haystacks colonized by cooperators and

Peterson_TL28_pp i-250.indd 53 11/4/09 9:26 AM

 54 The Tanner Lectures on Human Values

defectors, defectors take over. Then cooperative haystacks outreproduce

noncooperative ones.
What is the larger game within which the plays of the Prisoner’s Di-
lemma are embedded? Following Ted Bergstrom (2002), we consider the
game played by founding members of a haystack. Their payoff is the num-
ber of descendants in the spring, at the end of the life of a haystack. Analy-
sis of this founders game shows that it too is a Stag Hunt. There are two
equilibria, one with all cooperators and one with all defectors. Which you
get depends on where you start.
Each of these models explains how a population of cooperators can
be at a stable equilibrium, but neither explains the origin of cooperation.
That is because noncooperation is also a stable state, and the transition
from noncooperation to cooperation is left a mystery.
We are left with the question of how it is possible to evolve from the
noncooperative equilibrium to the cooperative one in interactions with
the structure of the Stag Hunt. Axelrod and Hamilton (1981) raise this
question and take kin selection and cooperation in family groups as an
origin of cooperative behavior. Beyond this there are a few other good
answers available. I will focus on three of them.
The first, due to Arthur Robson (1990), is the use of a signal as a secret
handshake. Consider a population of defectors in the Stag Hunt game.
Suppose that a mutant (or innovator) arises who can send a signal, coop-
erate with those who send the same signal, and defect against those who
don’t. The new type behaves like a native against the natives and like a
cooperator against itself, and so can (slowly) invade. (The signal needn’t
really be a secret, but it shouldn’t be in current use by defectors for other
purposes.) Once cooperators are established, it does not matter if the sig-
naling system somehow falls apart, because in Stag Hunt—unlike Prison-
er’s Dilemma—no one can do better against a population of cooperators.
The second involves a special kind of local interaction with neighbors.
Instead of the random encounters of the usual model, individuals interact
with their neighbors on a spatial grid (or some other spatial structure).5
They play a Stag Hunt game with each neighbor, and cultural evolution

5.  There are pioneering papers by Pollock (1989), Nowak and May (1992), and Hegsel-
mann (1996). These are models where the interaction is Prisoner’s Dilemma, or in the case of
Nowak and May at a bifurcation between Prisoner’s Dilemma and Hawk-Dove. Ellison (1993,
2002) discusses local interaction models of the Stag Hunt in which the <Defect, Defect> equi-
librium is risk-dominant. In his models it is the noncooperators who can invade and quickly
take over. The differences between Ellison’s models and ones in which cooperators can invade
and take over are discussed in Skyrms 2004.

Peterson_TL28_pp i-250.indd 54 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 55

proceeds by imitation of the most successful (on average) strategy in

the neighborhood. For biological evolution, there is an alternative inter-
pretation in which success translates into reproduction in the neighbor-
hood. We thus have both neighborhood interaction and neighborhood
imitation.
Eshel, Sansone, and Shaked (1999) point out that these two neighbor-
hoods need not be the same. For a biological example, take a plant that
interacts locally but disperses its seeds widely. On the cultural side we can
consider cases where the flow of information allows an individual to ob-
serve success beyond the confines of immediate interactions. This is the
crucial modification to local interaction models that allows robust evo-
lution of cooperation: the imitation neighborhood must be sufficiently
larger than the interaction neighborhood. Then a small clump of contigu-
ous cooperators will grow and eventually take over the population. That
is because defectors can see the success of the internal cooperators who
interact only with cooperators, and imitate them.
Kevin Zollman (2005) shows that the secret handshake and lo-
cal interaction work especially well together. The signal used as a secret
handshake now needs only to be a local secret to work. It could be used
elsewhere in the population to mean all sorts of other things. This makes
the hypothesis of the existence of an unused signal much more plausible.
The local secret handshake can then facilitate the initial formation of a
large-enough clump of contiguous cooperators to allow cooperation to
spread by imitation.
The third answer involves dynamic networks. Instead of constraining
individuals to interact with neighbors on a fixed structure, we can allow the
interaction structure to evolve as a result of individuals’ choices (Skyrms
and Pemantle 2000; Bonacich and Liggett 2003; Pemantle and Skyrms
2004a, 2004b; Liggett and Rolles 2004; Santos, Pacheco, and Lenaerts
2006; Pacheco, Traulsen, and Nowak 2006; Skyrms 2007; Skyrms and
Pemantle forthcoming). Cooperators want to interact with each other. In
Stag Hunt—unlike Prisoner’s Dilemma—noncooperators do not much
care. Cooperators and defectors may not wear their strategies on their la-
pels, but even if it is not so easy to tell cooperators from noncooperators, it
is not hard to learn. Robin Pemantle and I show how even naive reinforce-
ment learners will form cooperative associations in Stag Hunt provided
the social-network structure is sufficiently fluid (Skyrms and Pemantle
2000; Pemantle and Skyrms 2004a, 2004b). The conclusion is robust to
various variations in the learning dynamics (Skyrms 2004, 2007). These

Peterson_TL28_pp i-250.indd 55 11/4/09 9:26 AM

 56 The Tanner Lectures on Human Values

cooperative associations could then be the focal points for imitation to

spread cooperation, as in the foregoing discussion.6
Theory is borne out in laboratory studies of human interactions. There
is experimental evidence that there are many different types of individuals
(Page, Putterman, and Unel 2005; Burlando and Guala 2005; Fischbacher
and Gächter 2006) and that given the opportunity and the requisite in-
formation, cooperators will learn to associate with one another to their
own benefit.7
It is evident that the three preceding accounts of transition from the
noncooperative equilibrium of the Stag Hunt to the cooperative one also
rely on the establishment of correlated interactions. They share two dis-
tinctive features that are missing in many accounts of cooperation:

1. Sufficient positive correlation can be established by a few coopera-

tors in a large population of noncooperators.
2. Once the cooperative equilibrium is reached, it can be maintained
even if the correlation fades away.

Negative Correlation and Spite

If social structure can create positive correlation of encounters, it can also
create negative correlation. Negative correlation can overturn the conclu-
sions of conventional game theory just as radically as positive correlation.
Consider the effect of perfect negative correlation—of always meeting the
other type—on our three games. We are now comparing the other diagonal:

Prisoner’s Dilemma
Cooperate Defect
Cooperate 3 1
Defect 4 2

Stag Hunt
Cooperate Defect
Cooperate 3 1
Defect 2 2
6.  For other models of correlation induced by partner choice, see Wright 1921, where the
literature begins; Hamilton 1971; Feldman and Thomas 1987; Kitcher 1993; Oechssler 1997;
Dieckmann 1999; and Ely 2002.
7.  For an experiment in a public goods–provision game with voluntary association, see
Page, Putterman, and Unel 2005.

Peterson_TL28_pp i-250.indd 56 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 57
Prisoner’s Delight
Cooperate Defect
Cooperate 3 1
Defect 2 0

Not only is defection restored in the Prisoner’s Dilemma and favored

in the Stag Hunt, but it is also imposed on Prisoner’s Delight. In this last
case an individual hurts himself by defecting, but hurts his partner more.
This is a case of spiteful behavior. Hamilton and Price also showed how
negative correlation is the key to the evolution of spite.
Both spite and altruism on their face appear to violate the rational-
choice paradigm. Both have an evolutionary explanation in terms of
correlated interactions. Yet spite rarely receives the attention given to
altruism. A search on Google Scholar for “Evolution of Altruism” gets
1,570 hits, while one for “Evolution of Spite” gets 32. One cannot help
asking whether this is due to a Pollyanna bias. It’s enjoyable to write about
the sunny side of human nature. But the world is full of spiteful behavior:
feuds, vendettas, senseless wars. It is as important to study it as to study
altruism.
The way to study spite is to study endogenous correlation mechanisms.
In some kinds of repeated interactions, the shadow of the future may sus-
tain spite. Johnstone and Bshary (2004) recently analyzed the persistence
of spite in a repeated-game setting. In repeated contests, a reputation for
fighting too hard—to one’s own detriment as well as the greater detriment
of the opponent—may enable one to win future contests more easily. The
application need not be restricted to animal contests.
It might be remarked that this is not really spite in the larger game but
only self-interest, just as Hobbes claimed that cooperative behavior in a
repeated Prisoner’s Dilemma may just be self-interest in the longer view.
It is useful to be able to look at the phenomenon from both perspectives.
Successful invasion of a spiteful type into a nonspiteful population
can be sustained by local interaction. A strain of E. coli bacteria produces,
at some reproductive cost to itself, a poison that kills other strains of
E. coli—but to which it is impervious. It cannot invade a large random-
mixing population because a few poisoners can’t do that much damage
to the natives and the natives outreproduce the poisoners. But in a spa-
tial, local interaction setting, a clump of poisoners can take over. These
phenomena have been observed in the laboratory, with the random en-
counters taking place in a well-stirred beaker and local interactions taking

Peterson_TL28_pp i-250.indd 57 11/4/09 9:26 AM

 58 The Tanner Lectures on Human Values

place in a petri dish. Theoretical analysis by Durrett and Levin (1994) and
Iwasa, Nakamaru, and Levin (1998) is a complement to the local interac-
tion model of evolution of cooperation. If we put this case in the frame-
work of Eshel, Shaked, and Sansone of the last section, we find that a large
interaction neighborhood and a small imitation neighborhood robustly
favor the evolution of spite (see Skyrms 2004).
Group-selection models are not without their spiteful aspects. Sup-
pose that the farmer never tears down the haystacks—the population is-
lands that they represent remain isolated. Then for a mouse, its haystack
becomes its world, and this makes all the difference (see Gardner and
West 2004). Its haystack’s population becomes a population unto itself,
within which evolution takes place. The carrying capacity within a hay-
stack is limited, so we are dealing with a small, finite population. This,
in itself, induces negative correlation even if pairs of individuals form at
random within the haystack, because an individual does not interact with
herself. This effect is negligible in a large population but can be significant
in a small population. (For a transparent example, consider a population
consisting of four individuals, two C and two D. Population frequencies
are 50-50, but each type has probability 2/3 of meeting the other type and
1/3 of meeting its own.)
If a defector is introduced into a haystack full of cooperators—a mu-
tant or a migrant—he can cause problems. If the interaction is Prisoner’s
Dilemma, defectors will, of course, take over. But in small populations,
with some versions of Stag Hunt and even Prisoner’s Delight, defectors
can still invade as a result of negative correlation.
For any positive value of e, the following is a version of Prisoner’s De-
light—an individual prefers to cooperate no matter what the other does:

Prisoner’s Mild Delight

Cooperate Defect
Cooperate 2 + e e
Defect 2 0

For any finite population, there is an e—yielding some version of Prison-

er’s Mild Delight—such that a spiteful defector can invade.8

8.  For example, suppose 1 defector is introduced into a population of N cooperators. In-
dividuals pair at random. Since the defector cannot interact with himself, he always pairs with
a cooperator for a payoff of 2. Cooperators pair with the defector with probability (1/N) and
with other cooperators with probability (N-1)/N, for an average payoff of [(N-1)/N]*2 + e. So
if e < (2/N), a spiteful mutant does better than the native cooperators.

Peterson_TL28_pp i-250.indd 58 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 59

These three examples serve to indicate that the evolution of spite is an

aspect of the evolution of the social contract that is worthy of more de-
tailed study. There is no reason to believe that they exhaust the mechanisms
for negative correlation that may be important in social interactions.
To stop here would be to represent the social contract as a neat and
simple package of problems. But the social contract is not neat and simple.

Bargaining
Prisoner’s Delight, Stag Hunt, and Prisoner’s Dilemma are not the only
games that raise issues central to the social contract. We could separate
the issues of cooperating to produce a public good and deciding how that
good is to be divided. This is the philosopher’s problem of distributive
justice, and it brings bargaining games to center stage (Braithwaite 1955;
Rawls 1957; Sugden 1986; Binmore 1994, 1998, 2005; Gauthier 1985, 1986).
Consider the simplest Nash Bargaining game. Two players have bot-
tom-line demands for a share of a common good. If the demands exceed
the available good, agreement is impossible and players get nothing. Oth-
erwise, they get what they demand. We simplify radically by assuming
that there are only three possible demands: one-third, one-half, and two-
thirds. Evolutionary dynamics in a large random-encounter environment,
with and without persistent random shocks, is well studied.
Allowing differential reproduction to carry a population to equilib-
rium, there are two possibilities. The population may reach an egalitar-
ian consensus, where all demand one-half. Or it may reach a polymorphic
state, where half the population demands two-thirds and half the popu-
lation demands one-third (Sugden 1986). Greedy players get their two-
thirds half the time; modest players get their one-third all the time. This
inefficient polymorphism wastes resources, but it is evolutionarily stable
and has a significant basin of attraction. Persistent shocks can allow a
population to escape from this polymorphic trap and favor the egalitarian
norm in the very long run, but the inefficient polymorphism remains a
possibility for medium-run behavior.9
However, the effect of correlation mechanisms is rarely discussed in
connection with Nash Bargaining. If correlation plays an important role
in producing a surplus to be divided, might it not also play an important
role in deciding how the division takes place? Positive correlation of de-
mand types obviously favors the egalitarian solution. Those who ask for

9. Most of what we know about this is due to Peyton Young. See Young 1993a, 1993b,
1998; and Binmore, Samuelson, and Young 2003.

Peterson_TL28_pp i-250.indd 59 11/4/09 9:26 AM

 60 The Tanner Lectures on Human Values

equal shares do best when they meet each other. Negative correlation is
more complicated. Greedy players who demand two-thirds do very well
if paired with modest players who ask for one-third, but not well if paired
with those who ask for half. If negative correlation initially allows greedy
players to outreproduce all others, it cannot be maintained because they
run out of modest players. But a greedy-modest polymorphism is a real pos-
sibility if the negative correlation is of the kind that sufficiently disadvan-
tages the egalitarians. Possibilities are multiplied if we allow more demand
types. It should be of interest to look at specific correlation mechanisms.
If we allow individuals to bargain with neighbors on a spatial grid,
islands of egalitarianism spontaneously form. This generates positive cor-
relation, and if individuals emulate their most prosperous neighbors, egal-
itarianism takes over the population. This is a very rapid process, as other
types who interact along the edges of the egalitarian islands are quickly
converted to asking for half ( J. Alexander and Skyrms 1999; J. Alexander
2007; Skyrms 2004). Equal sharing is contagious.
If we add costless signaling to a large-population random-encounter
evolutionary model of bargaining, complicated correlations arise and then
fade away. Cooperators establish a positive correlation with cooperators.
Greedy types establish a negative correlation with themselves. Although
these correlations are transient, their effect is that the basin of attraction of
the egalitarian equilibrium is greatly enlarged (see Skyrms 2004).
Axtell, Epstein, and Young (2006) investigate a related model where
individuals have one of two “tags” and can condition their action on the
tag of their partner in a bargaining game. But there is a different dynamics.
Instead of evolution by replication or imitation, they consider a rational-
choice model. Things may fall out in various ways—here is one. When
interacting with those having the same tag, individuals share alike. But
in interactions between tags, one tag type becomes greedy and always de-
mands two-thirds and the other becomes modest and always demands
one-third. In this equilibrium tags are used to set up both positive and
negative correlations between behaviors. Both correlations are perfect: de-
mand-half behaviors always meet themselves, and the other two demand
behaviors always meet each other. The result is egalitarianism within tag
types and inegalitarian distribution between types. Axtell, Epstein, and
Young see this as a spontaneous emergence of social classes.
We can also see the spontaneous emergence of social classes in a dy-
namic social network (Skyrms 2004). The classes are stabilized by rational
choice but destabilized by imitation. Depending on the details and timing

Peterson_TL28_pp i-250.indd 60 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 61

of the dynamics, the social network may end up egalitarian or with a class
structure.

Division of Labor
So far, negative correlation has played a rather sinister role in this story.
It is not always so. In cooperating to produce a common good, organisms
sometimes discover the efficiency of division of labor, and find a way to
implement it. Modern human societies are marvels in their implementa-
tion of division of labor; so are the societies of cells in any multicellular or-
ganism. On the most elementary level, we can suppose that there are two
kinds of specialists that an individual can become, A and B, and that these
specialists are complementary. On the other hand, an individual might
not specialize at all but rather, less efficiently, do what both specialists do.
This gives us a little division-of-labor game:10

Division of Labor
Specialize A Specialize B Go It Alone
Specialize A 0 2 0
Specialize B 2 0 0
Go It Alone 1 1 1

In a random-encounter setting, specialists do badly. Positive corre-

lation makes it worse. What is required to get division of labor off the
ground is the right kind of negative correlation. Not all the correlation
mechanisms that we have discussed here do the trick.11 What works best is
dynamic social-network formation, where the network structure evolves
quickly. Specialists quickly learn to associate with the complementary spe-
cialists, and then specialists outperform those who go it alone. The effect
of correlation depends on the nature of the interaction.

Groups Revisited
Individuals sometimes form groups that have a permanence and unifor-
mity of interaction with other groups that qualified them to be thought of
as individuals. This happens at various levels of evolution. We ourselves are
such groups of cells. And humans participate in various social corpora—
teams, states, ideological groups—that interact with others.
10.  For analysis of different division-of-labor games, motivated by evolution of coviruses,
see Wahl 2002.
11. One population signaling models face is when an individual interacts with another
who sends the same signal. See Skyrms 2004.

Peterson_TL28_pp i-250.indd 61 11/4/09 9:26 AM

 62 The Tanner Lectures on Human Values

How such superindividuals are formed and hold together (or don’t)
is a central issue of both biology (R. Alexander 1979, 1987; Buss 1987;
Maynard Smith and Szathmary 1995; Frank 1998, 2003) and social sci-
ence. There is no one answer, but answers may involve both elements of
cooperation and elements of spite. One important factor is the punish-
ment of individuals who act against the interests of the group. A large
experimental literature documents the willingness of many individuals to
pay to punish “free riders” in public goods–provision games, and shows
that such punishment is able to stabilize high levels of cooperation.12 This
is also an important finding of Ostrom’s field studies of the self-organized
government of commons (1990). Costly punishment is, from an evolu-
tionary point of view, a form of spite—although it is not called by that
name in the literature.
The name may strike the reader as overly harsh in the case of the mod-
est, graduated punishments found in Ostrom’s field studies of successful
cooperative collective management. But in tightly organized superindi-
viduals, punishment can be draconian. Totalitarian regimes or ideologies
classify those who violate social norms as traitors or heretics. They may be
stoned to death. They have been burned at the stake. The righteous people
carrying out such acts no doubt believe that they are engaged in “altruistic
punishment.” We need also to think about the dark side of punishment.
When groups that can operate more or less like superindividuals have
been formed, the interactions of the superindividuals themselves are also
liable to the effects of positive and negative correlation described above.
They can cooperate to produce a common good, or not. Their interactions
may exemplify spite—not only in behavior, like the bacteria, but also in
the full psychological sense of the word.
Repeated interactions, alliances, local interaction on a geographical
landscape, signals, tags, and network formation all play a role. Division
of labor is facilitated by trade networks, and trade may promote both the
good and the bad sides of negative correlation.
The negative correlation conducive to spite in small populations
may take on a larger significance when we consider interactions between
groups. A local population of six interacting nations is perhaps more plau-
sible than a local population of six interacting mice.

12.  For instance, see Ostrom, Walker, and Gardner 1992; and Fehr and Gachter 2000,
2002. Costly punishment is already implicit in the behavior of receivers in ultimatum-game
experiments from Güth, Schmittberger, and Schwartze (1982) to Henrich et al. (2004).

Peterson_TL28_pp i-250.indd 62 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 63
Evolution and the Social Contract
An evolutionary theory of the social contract stands in some contrast with
social-contract theory as practiced in the contemporary philosophical
treatment of John Harsanyi and John Rawls. They assume that everyone
is, in some sense, rational. And they assume that in the relevant choice
situation—behind a veil of ignorance—the relevant choosers are all basi-
cally the same. They all have the same rational-choice rule,13 they all have
the same basic values, and therefore they all make the same choice.14 Cor-
relation of types plays no part because it is assumed that there is only one
relevant type.
Evolutionary game theory brings different types of individuals into the
picture from the beginning. Evolutionary game theory is full of contin-
gency. There are typically many equilibria; there are many possible alter-
native social contracts. The population might never get to equilibrium but
rather cycle or describe a chaotic orbit. Mutation, invention, experimenta-
tion, and external environmental shocks add another layer of contingency.
Evolutionary game theory has some affinity with rational-choice the-
ory in the absence of correlation.15 This vanishes when interactions are
correlated. But correlation, positive and negative, is the heart of the social
contract. Correlation gets it started. Correlation lets it grow and develop
more complex forms. Social institutions and networks evolve to enable
and maintain correlation. Correlation explains much of what is admirable
and what is despicable in existing social contracts—what we would like
to keep and what we would like to change. A better understanding of the
dynamics of correlation should be a central concern for Darwinian social
philosophy.

13.  But theorists disagree about the nature of rational choice. Rawls minimizes the maxi-
mum loss; Harsanyi maximizes the expected payoff.
14.  The theorist tells you what that choice will be.
15.  In large populations, expected fitness can then be calculated using population propor-
tions in place of the subjective probabilities of rational choice theory.

Peterson_TL28_pp i-250.indd 63 11/4/09 9:26 AM

 64 The Tanner Lectures on Human Values

Notes

I would like to thank my discussants at the Tanner symposium, Eleanor

Ostrom, Michael Smith, and Peyton Young, for valuable comments on
the lecture. I would also like to thank Jeffrey Barrett, Louis Narens, Don
Saari, Rory Smead, Elliott Wagner, Jim Woodward, and Kevin Zollman
for comments on earlier drafts, which much improved the lecture.

References

Alexander, J. M. 2000. “Evolutionary Explanations of Distributive Justice.” Phi-

losophy of Science 67: 490–516.
———. 2007. The Structural Evolution of Morality. Cambridge: Cambridge Uni-
versity Press.
Alexander, J. M., and B. Skyrms. 1999. “Bargaining with Neighbors: Is Justice
Contagious?” Journal of Philosophy 96: 588–98.
Alexander, R. D. 1979. Darwinism and Human Affairs. Seattle: University of
Washington Press.
———. 1987. The Biology of Moral Systems. New York: de Gruyter.
Axelrod, R. 1981. “The Emergence of Cooperation among Egoists.” American Po-
litical Science Review 75: 306–18.
———. 1984. The Evolution of Cooperation. New York: Basic Books.
Axelrod, R., and W. D. Hamilton. 1981. “The Evolution of Cooperation.” Science
211: 1390–96.
Axtell, R., J. M. Epstein, and H. P. Young. 2006. “The Emergence of Classes in
a Multi-agent Bargaining Model.” In Generative Social Science: Studies in
Agent-Based Computational Modeling, 177–95. Princeton: Princeton Univer-
sity Press.
Bergstrom, T. 2002. “Evolution of Social Behavior: Individual and Group Selec-
tion Models.” Journal of Economic Perspectives 16: 231–38.
Bergstrom, T., and O. Stark. 1993. “How Altruism Can Prevail in an Evolution-
ary Environment.” American Economic Review 83: 149–55.
Binmore, K. 1994. Game Theory and the Social Contract I: Playing Fair. Cam-
bridge: MIT Press.
———. 1998. Game Theory and the Social Contract II: Just Playing. Cambridge:
MIT Press.
———. 2005. Natural Justice. Oxford: Oxford University Press.
Binmore, K., L. Samuelson, and H. P. Young. 2003. “Equilibrium Selection in
Bargaining Models.” Games and Economic Behavior 45: 296–328.
Björnerstedt, J., and J. W. Weibull. 1996. “Nash Equilibrium and Evolution by
Imitation.” In The Rational Foundations of Economic Behavior, edited by K. J.
Arrow et al. New York: St. Martin’s Press.

Peterson_TL28_pp i-250.indd 64 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 65

Bonacich, P., and T. Liggett. 2003. “Asymptotics of a Matrix-Valued Markov

Chain Arising from Sociology.” Stochastic Processes and Their Applications
104: 155–71.
Braithwaite, R. B. 1955. The Theory of Games as a Tool for the Moral Philosopher.
Cambridge: Cambridge University Press.
Burlando, R. M., and F. Guala. 2005. “Heterogeneous Agents in Public Goods
Experiments.” Experimental Economics 8: 35–54.
Buss, L. W. 1987. The Evolution of Individuality. Princeton: Princeton University
Press.
Dewey, J. 1910. The Influence of Darwin on Philosophy, and Other Essays in Con-
temporary Thought. New York: Henry Holt.
Dieckmann, T. 1999. “The Evolution of Conventions with Mobile Players.” Jour-
nal of Economic Behavior and Organization 38: 93–111.
Durrett, R., and S. Levin. 1994. “The Importance of Being Discrete (and Spa-
tial).” Theoretical Population Biology 46: 363–94.
Ellison, G. 1993. “Learning, Local Interaction, and Coordination.” Econometrica
61: 1047–71.
———. 2000. “Basins of Attraction, Long-Run Stochastic Stability, and the
Speed of Step-by-Step Evolution.” Review of Economic Studies 67: 17–45.
Ely, J. 2002. “Local Conventions.” Advances in Theoretical Economics 2, no. 1.
http://www.bepress.com/.
Epstein, J. M. 2006. Generative Social Science: Studies in Agent-Based Computa-
tional Modeling. Princeton: Princeton University Press.
Eshel, I., and L. L. Cavalli-Sforza. 1982. “Assortment of Encounters and the Evo-
lution of Cooperativeness.” Proceedings of the National Academy of Sciences of
the USA 79: 331–35.
Eshel, I., E. Sansone, and A. Shaked. 1999. “The Emergence of Kinship Behavior
in Structured Populations of Unrelated Individuals.” International Journal of
Game Theory 28: 447–63.
Fehr, E., and S. Gachter. 2000. “Cooperation and Punishment in Public Goods
Experiments.” American Economic Review 90: 980–94.
———. 2002. “Altruistic Punishment in Humans.” Nature 415: 137–40.
Feldman, M., and E. Thomas. 1987. “Behavior-Dependent Contexts for Repeated
Plays in the Prisoner’s Dilemma II: Dynamical Aspects of the Evolution of
Cooperation.” Journal of Theoretical Biology 128: 297–315.
Fischbacher, U., and S. Gächter. 2006. “Heterogeneous Social Preferences and
the Dynamics of Free-Riding in Public Goods.” Working paper, University
of Zurich.
Frank, S. A. 1995. “George Price’s Contributions to Evolutionary Genetics.” Jour-
nal of Theoretical Biology 175: 373–88.
———. 1998. Foundations of Social Evolution. Princeton: Princeton University
Press.

Peterson_TL28_pp i-250.indd 65 11/4/09 9:26 AM

 66 The Tanner Lectures on Human Values

———. 2003. “Perspective: Repression of Competition and the Evolution of Co-

operation.” Evolution 57: 693–705.
Fudenberg, D., and D. Levine. 1998. A Theory of Learning in Games. Cambridge:
MIT Press.
Gardner, A., and S. A. West. 2004. “Spite and the Scale of Competition.” Journal
of Evolutionary Biology 17: 1195–1203.
Gauthier, D. 1985. “Bargaining and Justice.” Social Philosophy and Policy 2: 29–47.
———. 1986. Morals by Agreement. Oxford: Oxford University Press.
Gibbard, A. 1990. Wise Choices, Apt Feelings: A Theory of Normative Judgement.
Cambridge: Harvard University Press.
Grafen, A. 1984. “Natural Selection, Kin Selection, and Group Selection.” In Be-
havioral Ecology: An Evolutionary Approach, edited by J. R. Krebs and N. B.
Davies, 62–84. Sunderland, Mass.: Sinauer.
———. 1985. “A Geometric View of Relatedness.” In Oxford Surveys in Evolu-
tionary Biology, edited by R. Dawkins and M. Ridley, 2:28–89. Oxford: Ox-
ford University Press.
Greif, A. 1989. “Reputations and Coalitions in Medieval Trade.” Journal of Eco-
nomic History 49: 857–82.
———. 2006. Institutions and the Path to the Modern Economy: Lessons from Me-
dieval Trade. Cambridge: Cambridge University Press.
Güth, W., R. Schmittberger, and B. Schwartze. 1982. “An Experimental Analysis
of Ultimatum Bargaining.” Journal of Economic Behavior and Organization
3: 367–88.
Hamilton, W. D. 1963. “The Evolution of Altruistic Behavior.” American Natu-
ralist 97: 354–56.
———. 1964. “The Genetical Evolution of Social Behavior I and II.” Journal of
Theoretical Biology 7: 1–52.
———. 1971. “Selection of Selfish and Altruistic Behavior in Some Extreme
Models.” In Man and Beast, edited by J. F. Eisenberg and W. S. Dillon, 59–91.
Washington, D.C.: Smithsonian Institution Press.
———. 1995. Narrow Roads of Gene Land. Vol. 1, Evolution of Social Behavior.
New York: W. H. Freeman.
Hampton, J. 1996. Hobbes and the Social Contract Tradition. Cambridge: Cam-
bridge University Press.
Harms, W. 2001. “Cooperative Boundary Populations: The Evolution of Co-
operation on Mortality Risk Gradients.” Journal of Theoretical Biology 213:
299–313.
———. 2004. Information and Meaning in Evolutionary Processes. New York:
Cambridge University Press.
Harms, W., and B. Skyrms. 2007. “Evolution of Moral Norms.” In Oxford Hand-
book in the Philosophy of Biology, edited by Michael Ruse. Oxford: Oxford
University Press.

Peterson_TL28_pp i-250.indd 66 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 67

Harsanyi, J. 2007. Essays on Ethics, Social Behaviour, and Scientific Explanation.

Dordrecht: Reidel.
Hegselmann, R. 1996. “Social Dilemmas in Lineland and Flatland.” In Frontiers
of Social Dilemmas Research, edited by W. B. G. Liebrand and D. Messick,
337–62. Berlin: Springer Verlag.
Henrich, J., R. Boyd, S. Bowles, C. Camerer, E. Fehr, and H. Gintis. 2004. Foun-
dations of Human Sociality: Economic Experiments and Ethnographic Evi-
dence from Fifteen Small-Scale Societies. New York: Oxford University Press.
Hofbauer, J., and K. Sigmund. 1998. Evolutionary Games and Population Dynam-
ics. Cambridge: Cambridge University Press.
Iwasa, Y., M. Nakamaru, and S. A. Levin. 1998. “Allelopathy of Bacteria in a
Lattice Population: Competition between Colicin-Sensitive and Colicin-
Producing Strains.” Evolutionary Ecology 12: 785–802.
Johnstone, R. A., and R. Bshary. 2004. “Evolution of Spite through Indirect Rec-
iprocity.” Proceedings of the Royal Society of London B 271: 1917–22.
Kandori, M. 1992. “Social Norms and Community Enforcement.” Review of Eco-
nomic Studies 59: 63–80.
Kavka, G. 1986. Hobbesian Moral and Political Theory. Princeton: Princeton Uni-
versity Press.
Kitcher, P. 1993. “The Evolution of Human Altruism.” Journal of Philosophy 10:
497–516.
Liggett, T. M., and S. W. W. Rolles. 2004. “An Infinite Stochastic Model of Social
Network Formation.” Stochastic Processes and Their Applications 113: 65–80.
Luce, R. D., and H. Raiffa. 1957. Games and Decisions. New York: Wiley.
Maynard Smith, J. 1964. “Group Selection and Kin Selection.” Nature 201: 1145–
47.
———. 1982. Evolution and the Theory of Games. Cambridge: Cambridge Uni-
versity Press.
Maynard Smith, J., and E. Szathmary. 1995. The Major Transitions in Evolution.
Oxford: Oxford University Press.
Milgrom, P., D. North, and B. Weingast. 1990. “The Role of Institutions in the
Revival of Trade: The Law Merchant, Private Judges, and the Champagne
Fairs.” Economics and Politics 2: 1–23.
Nowak, M. A., and R. M. May. 1992. “Evolutionary Games and Spatial Chaos.”
Nature 359: 826–29.
Nowak, M. A., and K. Sigmund. 1998. “Evolution of Indirect Reciprocity by Im-
age Scoring.” Nature 393: 573–77.
Oechssler, J. 1997. “Decentralization and the Coordination Problem.” Journal of
Economic Behavior and Organization 32: 119–35.
Ostrom, E. 1990. Governing the Commons. Cambridge: Cambridge University
Press.
Ostrom, E., J. Walker, and R. Gardner. 1992. “Covenants with and without a

Peterson_TL28_pp i-250.indd 67 11/4/09 9:26 AM

 68 The Tanner Lectures on Human Values

Sword: Self-Governance Is Possible.” American Political Science Review 86:

404–17.
Pacheco, J. M., A. Traulsen, and M. A. Nowak. 2006. “Active Linking in Evolu-
tionary Games.” Journal of Theoretical Biology 243: 437–43.
Page, T., L. Putterman, and B. Unel. 2005. “Voluntary Association in Public
Good Experiments: Reciprocity, Mimicry, and Efficiency.” Economic Journal
115: 1032–53.
Pemantle, R., and B. Skyrms. 2004a. “Network Formation by Reinforcement
Learning: The Long and the Medium Run.” Mathematical Social Sciences 48:
315–27.
———. 2004b. “Time to Absorption in Discounted Reinforcement Models.”
Stochastic Processes and Their Applications 109: 1–12.
Pollock, G. B. 1989. “Evolutionary Stability in a Viscous Lattice.” Social Networks
11: 175–212.
Price, G. R. 1970. “Selection and Covariance.” Nature 227: 520–21.
Ratnieks, F., and K. Visscher. 1989. “Worker Policing in the Honeybee.” Nature
342: 796–97.
Rawls, J. 1957. “Justice as Fairness.” Journal of Philosophy 54: 653–62.
———. 1971. A Theory of Justice. Cambridge: Harvard University Press.
Robson, A. J. 1990. “Efficiency in Evolutionary Games: Darwin, Nash, and the
Secret Handshake.” Journal of Theoretical Biology 144: 379–96.
Samuelson, L. 1997. Evolutionary Games and Equilibrium Selection. Cambridge:
MIT Press.
Santos, F. C., J. M. Pacheco, and T. Lenaerts. 2006. “Cooperation Prevails When
Individuals Adjust Their Social Ties.” PLoS Computational Biology 2, no. 10:
1–6.
Scanlon, T. 1998. What We Owe to Each Other. Cambridge: Harvard University
Press.
Schelling, T. 1960. The Strategy of Conflict. Cambridge: Harvard University Press.
Schlag, K. H. 1998. “Why Imitate and If So, How? A Boundedly Rational Ap-
proach to Multi-armed Bandits.” Journal of Economic Theory 78: 130–56.
Skyrms, B. 1996. Evolution of the Social Contract. Cambridge: Cambridge Uni-
versity Press.
———. 2001. “The Stag Hunt.” Proceedings and Addresses of the American Philo-
sophical Association 75: 31–41.
———. 2004. The Stag Hunt and the Evolution of Social Structure. Cambridge:
Cambridge University Press.
———. 2007. “Dynamic Networks and the Stag Hunt: Some Robustness Con-
siderations.” Biological Theory 2, no. 1: 1–3.
Skyrms, B., and R. Pemantle. 2000. “A Dynamic Model of Social Network
Formation.” Proceedings of the National Academy of Sciences of the USA 97:
9340–46.

Peterson_TL28_pp i-250.indd 68 11/4/09 9:26 AM

 [Brian Skyrms]   Evolution and the Social Contract 69

———. forthcoming. “Learning to Network.” In Probability in Science, edited by

E. Eells and J. Fetzer. Chicago: Open Court.
Sober, E., and D. S. Wilson. 1998. Unto Others: The Evolution and Psychology of
Unselfish Behavior. Cambridge: Harvard University Press.
Sugden, R. 1986. The Economics of Rights, Co-operation, and Welfare. Oxford: Ba-
sil Blackwell.
Trivers, R. 1971. “The Evolution of Reciprocal Altruism.” Quarterly Review of Bi-
ology 46: 35–57.
Vanderschraaf, P. 1998. “The Informal Game Theory in Hume’s Account of Con-
vention.” Economics and Philosophy 14: 215–47.
———. 2006. “War or Peace: A Dynamical Analysis of Anarchy.” Economics and
Philosophy 22: 243–79.
Vanderschraaf, P., and J. M. Alexander 2005. “Follow the Leader: Local Interac-
tion with Influence Neighborhoods.” Philosophy of Science 72: 86–113.
Wahl, L. M. 2002. “Evolving the Division of Labor: Generalists, Specialists, and
Task Allocation.” Journal of Theoretical Biology 219: 371–88.
Weibull, J. 1995. Evolutionary Game Theory. Cambridge: MIT Press.
Wright, S. 1921. “Systems of Mating III: Assortative Mating Based on Somatic
Resemblance.” Genetics 6: 144–61.
———. 1945. “Tempo and Mode in Evolution: A Critical Review.” Ecology 26:
415–19.
Young, H. P. 1993a. “An Evolutionary Model of Bargaining.” Journal of Economic
Theory 59: 145–68.
———. 1993b. “The Evolution of Conventions.” Econometrica 61: 57–84.
———. 1998. Individual Strategy and Social Structure. Princeton: Princeton Uni-
versity Press.
Zollman, K. 2005. “Talking to Neighbors: The Evolution of Regional Meaning.”
Philosophy of Science 72: 69–85.

Peterson_TL28_pp i-250.indd 69 11/4/09 9:26 AM