The Hitchhiker's Guide to Altruism: Gene-culture Coevolution, and

12 Pages·2002·0.18 MB·English

J. theor. Biol. (2003) 220, 407–418
doi:10.1006/jtbi.2003.3104, available online at http://www.idealibrary.com on
The Hitchhiker’s Guide to Altruism: Gene-culture Coevolution,
and the Internalization of Norms
Herbert Gintisn
Santa Fe Institute and Department of Economics, University of Massachusetts, 15 Forbes Avenue,
MA 01060, Northampton, U.S.A.
(Received on 5 December 2001, Accepted in revised form on 4 June 2002)
An internal norm is a pattern of behavior enforced in part by internal sanctions, such as
shame,guiltandlossofself-esteem,asopposedtopurelyexternalsanctions,suchasmaterial
rewards and punishment. The ability to internalize norms is widespread among humans,
although in some so-called ‘‘sociopaths’’, this capacity is diminished or lacking. Suppose
there is one genetic locus that controls the capacity to internalize norms. This model shows
that if an internal norm is fitness enhancing, then for plausible patterns of socialization, the
allele for internalization of norms is evolutionarily stable. This framework can be used to
model Herbert Simon’s (1990) explanation of altruism, showing that altruistic norms can
‘‘hitchhike’’ on the general tendency of internal normsto be personally fitness-enhancing. A
multi-level selection, gene-culture coevolution argument then explains why individually
fitness-reducing internal norms are likely to be prosocial as opposed to socially harmful.
r 2003 Elsevier Science Ltd. All rights reserved.
Introduction human agents have socially programmable ob-
jective functions. Human behavior thus depends
An internal norm is a pattern of behavior
notonlyonbeliefs,whichconcernconstraintson
enforced in part by internal sanctions, including
action(takingactionXwillleadtoresultY),but
shame, guilt and loss of self-esteem, as opposed
values, which are the very goals of action.
to purely external sanctions, such as material
Suppose there is one genetic locus that
rewards and punishments. Humans internalize
controls the capacity to internalize norms. I
norms through socialization by parents (vertical
develop models of gene-cultural coevolution to
transmission) and extraparental conspecifics (ob-
show that if an internal norm is fitness enhan-
lique and horizontal transmission). The capacity
cing, then the allele for internalization of norms
to internalize norms is widespread among hu-
is evolutionarily stable. Moreover, if the fitness
mans, although in some so-called ‘‘sociopaths’’,
payoff to the internalized norm is sufficiently
this capacity is diminished or lacking (Mealey,
large, or if there is a sufficiently high rate of
1995). Human behavior is commonly modeled
phenotypiclevelassortativemating,theallelefor
assuming agents have objective functions which
internalization is globally stable.
they maximize subject to constraints. In these
Basic sociological theory holds that society’s
terms, the capacity to internalize norms means
values are transmitted through the internaliza-
tion of norms (Parsons, 1967; Grusec &
nTel.: [phone]; fax:[phone].
Kuczynski, 1997). Successful societies tend to
E-mailaddress: [email protected] (H.Gintis).
URL: http://www-unix.oit.umass.edu/Bgintis. foster internal norms that enhance personal
0022-5193/03/$35.00 r2003Elsevier Science Ltd. Allrights reserved.
408 H.GINTIS
fitness,suchasfuture-orientation,goodpersonal by helping those in distress and punishing anti-
hygiene, positive work habits, and control of social behavior) at personal cost.w Adding an
emotions, as well as altruistic norms that altruism norm allows us to model Herbert
subordinate the individual to group welfare, Simon’s (1990) explanation of altruism. Simon
fostering such behaviors as bravery, honesty, suggested that altruistic norms could ‘‘hitch-
fairness, willingness to cooperate, and empathy hike’’ on the general tendency of internal norms
for others (Brown, 1991). People follow internal to be personally fitness-enhancing. Of course,
norms because they value this behavior for its internal norms may persist even if they are
own sake, in addition to, or despite, the effects fitness-reducing both for individuals and the
the behavior has on personal fitness and/or group (Boyd & Richerson, 1992; Edgerton,
perceived well-being. For instance, an individual 1992). I develop a multi-level gene-culture
who has internalized the value of ‘‘speaking coevolutionary model to elucidate the process
truthfully’’ will dosoeven in caseswhere the net whereby altruistic internal norms will tend to
payofftospeakingtruthfullywouldotherwisebe drive out norms that are both socially harmful
negative.It followsthatwhere peopleinternalize and individually fitness-reducing.
a norm, the frequency of its occurrence in the
population will be higher than if people follow
Socialization and Fitness-Enhancing
the norm only instrumentally; i.e. when they
Internal Norms
perceive it to be in their interest to do so.
Why does the capacity to internalize Suppose there is a norm C that can be
norms have adaptive value? It might be argued internalized by a new member of society. Norm
that if a norm is fitness enhancing, a non- C confers fitness 1þt41; while the normless
internalizing agent could simply mimic the phenotype, denoted by D, has baseline fitness 1.
behavior of an internalizer. But this assumes There is a genetic locus with two alleles, {a} and
that agents maximize fitness. In general, how- {b}. Allele {a}, which is dominant, permits the
ever,inanyspecies,individualsdonotmaximize internalization of norms, whereas {b} does not.
fitness but rather a objective function that has We assume that possessing at least one copy of a
evolved to reflect biological fitness more or less imposesafitnesscostuAð0;1Þ;onthegroundsthat
accurately for a given environment. If the Homo there are costly physiological and cognitive pre-
sapiens objective function were perfectly requisites for the capacity to internalize norms.z
adapted, internalization would not be fitness- We assume ð1þtÞð1(cid:5)uÞ41; so the cost of the
enhancing. But the rapid cultural evolution and internalization allele is more than offset by the
highly variable environments (Richerson et al., benefit of the norm C. There are five phenogeno-
2001) that characterized the period in which types, whose fitnesses are listed in Fig.1.y
Homo sapiens developed doubtless led to a Families are formed by random pairing, and
situation in which the unsocialized human offspring genotypes obey the laws of Mendelian
objective function deviated strongly from fitness segregation. Thus, there are six familial
maximization.The internalization of normsthus genotypes, aaaa, aaab, aabb, abab, abbb, and
permitted rapid cultural adaptation towards bbbb. We assume also that only the phenotypic
fitness-maximization, while a purely genetic
adaptive process would have taken orders of wFor reviews of the evidence on the importance of
altruism in human societies, see Sober & Wilson (1998),
magnitude longer in time. In short, non-
Gintis(2000a,b), andFehr &Ga¨chter(2002).
internalizers fail to mimic internalizers not zAssuming u40 is conservative, in that it biases the
because they cannot, but because they model against the global stability of the internalization
donotwantto(sociopaths,forinstance,dooften allele. However, the contrasting assumption uo0 is also
mimic internalizersFdisplaying empathy and plausible. I will point out the implications of uo0 where
appropriate.
helpfulness, for exampleFbut only as long as yFeldman et al. (1985) develop a model similar to ours.
it suits them to do so). Their model, however, is haploid, assumes uniparental
transmission, and the phenotypic trait is kin-altruistic.
Altruism is the tendency of individuals to
Ours, by contrast, is diploid, assumes biparental transmis-
behaveprosociallytowardsunrelatedothers(e.g. sion, andabstractsfrom kin-altruism.
HITCHHIKER’SGUIDETOALTRUISM 409
Individual Individual be the fraction of the population carrying the C
Phenogenotype Fitness phenotype, and let gA½0;1(cid:8): We assume a
fraction gp of aa-types and a fraction np of
aaC (1-u)(1+t) C C
ab-types who have not internalized C through
aaD (1-u)
parental transmission, are influenced by extra-
abC (1-u)(1+t)
parental transmission to switch to C. bb
abD (1-u)
types are not affected by extraparental trans-
bbD 1
mission.
Fig. 1. Fitnesses of the five phenogenotypes. Here u is The resulting system consists of four equa-
thefitnesscostofpossessingtheinternalizationallele,andt tions in four unknowns (bbC cannot occur, and
is the excess fitness value of possessing the norm C. Note
one offspring phenogenotype is dropped, since
thatbbC cannotoccur.
the sum of phenogenotypic frequencies equals
unity). It is straightforward to check that there
traits of parents, and not which particular are three pure equilibria (i.e. equilibria in which
parent expresses them, are relevant to the the whole population bears a single phenogen-
transmission process. Therefore, there are three otype). These are aaC, in which all agents
familial phenotypes, CC, CD, and DD, and 18 internalize the fitness enhancing norm, aaD, in
familial phenogenotypes, of which only 14 can which the internalization allele is present but
occur. The frequency of familial phenogeno- the phenotype C is absent, and bbD, in which
types are as shown in Fig.2, where pðiÞ neither the internalization allele nor the norm
represents the frequency of phenogenotype i ¼ is present.
aaC;y;bbD: A check of the eigenvalues of the Jacobian
The rules of gene-culture transmission are as matrix of the dynamical system shows that the
follows. If familial phenogenotype is xyzwXY, aaD equilibrium is unstable. Eigenvalues of the
where x,y,z,wA{a,b}, X, YA{C,D}, an offspring system at the aaC equilibrium are given by
is equally likely to inherit xz, xw, yz, or yw. An
offspring whose genotype includes a copy of the (cid:1) 1(cid:5)g 1(cid:5)g(cid:2)
a allele is equally likely to inherit X or Y.8 But 0;1; ; :
2ð1þtÞ 1þt
an offspring of genotype bb always has the
normless phenotype D. The transition table is
The unit eigenvalue is semisimple,ww so the
shown in Fig.3.z
linearization of the equilibrium aaC, in which
The above accounts only for parental trans-
the fitness-enhancing norm is internalized, is
mission. In addition, extraparental transmission
stable. However, we cannot conclude that the
is ubiquitous in human society, in the form of
nonlinear model itself is stable. Extensive simu-
social pressure (rumor, shunning, and ostra-
lations fail to find a case in which the aaC
cism), rituals (dancing, prayer, marriage, birth,
equilibrium is unstable.zz Moreover, in the case
and death), and in modern societies, formalized
where the a allele is incompletely dominant, the
institutions (schools and churches).** To
unit root disappears, so stability is assured (this
account for extraparental transmission, let p
C remark applies as well to all cases in which unit
roots appear, some of which are discussed
8Simulations show thatthe assumptionthat the a allele
below).
is dominant is not critical. The stability results described
belowcontinuetohold,andindeedmorestrongly,whena
isless thanfullydominant. wwAn eigenvalue is semisimple if its algebraic and
zBiased parental transmission, in which heterogeneous geometric dimensions are equal. Semisimple unit roots of
familial phenotypes are more likely to transmit one lineardynamical systems arestable.
phenotype to offspring that the other (Cavalli-Sforza & zzThe process of coding this and the other models
Feldman,1981) is discussedbelow. presented in this paper is tedious and error-prone. To
nnExtraparental transmission is generally individually ensure accuracy I wrote the simulations in two completely
costlyandthebenefitsaccruetounrelatedothers.Henceit different languages, one Lisp-like (Mathematica) and the
is a form of altruistic behavior, and ideally should not be other procedural (C++), and verified that the results
introduced until our analysis of altruism is completed. We agreedtosixdecimalplacesoverthousandsofgenerations
introduceit nowpurelyfor expositionalpurposes. ofsimulation.
410 H.GINTIS
Familial Frequencyin
Phenogenotype ReproductivePool
aaaaCC p(aaC)2(1_ u)2(1+ t)2/p
aaaaCD 2p(aaC)p(aaD)(1_ u)2(1+ t)/p
aaaaDD p(aaD)2(1_ u)2/p
aaabCC 2p(aaC)p(abC)(1_ u)2(1+ t)2/p
aaabCD 2(p(aaC)p(abD)+ p(aaD)p(abC))(1_ u)2(1+ t)/p
aaabDD 2p(aaD)p(abD)(1_ u)2/p
ababCC p(abC)2(1_ u)2(1+ t)2/p
ababCD 2p(abC)p(abD)(1_ u)2(1+ t)/p
ababDD p(abD)2(1_ u)2/p
_
aabbCD 2p(aaC)p(bbD)(1 u)(1+ t)/p
_
aabbDD 2p(aaD)p(bbD)(1 u)/p
_
abbbCD 2p(abC)p(bbD)(1 u)(1+ t)/p
_
abbbDD 2p(aaC)p(aaD)(1 u)(1+ t)/p
bbbbDD 2p(bbD)2/p
Fig. 2. Frequencies of phenogenotypes. Here, p% is chosen so the sum of the frequencies is unity. Note that aabbCC,
abbbCC, bbbbCC, andbbbbCD are notlisted,sincebbC cannotoccur.
assume this is the case in this paper. It follows
Familial OffspringPhenogenotypicFrequency
Type aaC aaD abC abD bbD thatfort42=ð1(cid:5)uÞ(cid:5)1;aaCisagloballystable
aaaaCC 1 equilibrium.yy
aaaaCD 1/2 1/2 Therearefourplausibleconditionsthatrender
aaaaDD 1
the bbD equilibrium unstable, in which case aaC
aaabCC 1/2 1/2
will be globally stable. The first is uo0; which
aaabCD 1/4 1/4 1/4 1/4
means that the apparatus upon which internali-
aaabDD 1/2 1/2
aabbCD 1/2 1/2 zation depends has net positive (pleiotropic)
aabbDD 1 fitness effects independent from its contribution
abbbCD 1/4 1/4 1/2
to the internalization of norms. The second is
abbbDD 1/2 1/2
that t is sufficiently large that ð1þtÞð1(cid:5)uÞ42:
ababCC 1/4 1/2 1/4
Third, if parental transmission is sufficiently
ababCD 1/8 1/8 1/4 1/4 1/4
ababDD 1/4 1/2 1/4 biased in favor of C, the internalization equili-
bbbbDD 1 brium is globally stable.
The fourth condition leading to the global
Fig. 3. Phenotypic inheritance is controlled by geno-
stability of the aaC equilibrium is that there is
type. Note that aabbCC, abbbCC, bbbbCC, and bbbbCD
are notlisted,sincebbC cannotoccur. some assortative mating that overcomes the
tendency of the internalization allele to become
‘‘diluted’’.Supposeeachtypemateswithanother
The eigenvalues of the Jacobian matrix of the
of its type with probability z; and with an
unnormed equilibrium bbD are given by
random member of the population with
probability 1(cid:5)z: Then the eigenvalues of the
(cid:3)ð0;0;1(cid:5)u;1ð1þtÞð1(cid:5)uÞ(cid:4):
2
Therefore this equilibrium, in which no inter-
nalization occurs, is locally stable if ð1þtÞð1(cid:5) yyThe above result depends on our assumption of
unbiased parental transmission. Suppose, however, that a
uÞo2;andunstablewhentheoppositeinequality
fraction of offspring who would acquire norm C under
holds. There may exist equilibria involving more
unbiased transmission in fact acquire D. In this case,
thanonetypeofbehavior,althoughthesystemis inspection of the eigenvalues of the Jacobian tells us that
too complex to determine whether or not this is theaaCequilibriumislocallystableprovideddotandthe
bbDequilibriumisstableprovidedð1(cid:5)dÞð1þtÞð1(cid:5)uÞo2:
the case. Extensive simulations suggest that if
Thus, parental transmission biased against the internaliz-
such equilibria exist, they are not stable. I shall ablenormCis hostile tointernalization.
HITCHHIKER’SGUIDETOALTRUISM 411
bbD equilibrium become phenogenotypes, which we can write as
aaaaAACC,y,bbbbBBDD, only 36 of which
f1zð1(cid:5)uÞ;1zð1þtÞð1(cid:5)uÞ;1(cid:5)u;
2 2 can occur. We write the frequency of familial
1zð1þtÞð1(cid:5)uÞð1þzÞg: ð1Þ phenogenotype j as pðjÞ; and we assume the
2
populationissufficientlylargethatwecanignore
Therefore,thereis alwaysadegreeof assortative randomdrift.Forillustrativepurposes,herearea
mating that renders the bbD equilibrium un- few of the phenogenotypic frequencies:
stable. Thus, it is plausible that some combina-
pðaaaaAACCÞ
tion of assortativemating, parental transmission
biased towards the internal norm, and high
¼ pðaaACÞ2ð1(cid:5)sÞ2ð1þtÞ2ð1(cid:5)uÞ2=p%;
returns to the internal norm, assures the global
stability of the aaC equilibrium.88
pðaaaaAACDÞ
Altruism
¼ pðaaACÞpðaaADÞð1(cid:5)sÞ2ð1þtÞð1(cid:5)uÞ2=p%;
We now add a second dichotomous pheno-
pðaaaaAACDÞ ¼ 2pðabACÞpðabBDÞ
typic trait with two variants. Internal norm A is
altruistic in the sense that its expression benefits
pðabADÞpðabBCÞÞð1(cid:5)sÞð1þtÞð1(cid:5)uÞ2=p%;
the group, but imposes fitness loss sAð0;1Þ on
those who adopt it. The normless state, B, is pðbbbbBBDDÞ ¼ pðbbBDÞ2=p%;
neutral, imposing no fitness loss on those who
adopt it, but also no gain or loss to other and so on, where p% is chosen so the sum of the
members of the social group. frequencies is unity:
p% ¼ pðaaaaAACCÞþ?þpðbbbbBBDDÞ:
We assume A has the same cultural transmis-
sion rules as C: individuals who have a copy of
The rules of cultural transmission are as before.
allele a inherit their phenotypes from their
If familial phenogenotype is xyzwXYZW, where
parents, while bb individuals always adopt the x,y,z,wA{a,b},X,YA{A,B}, and Z,WA{C,D}, an
normless phenotype BD. In addition, there is
offspringisequallylikelytoinheritxz,xw,yz,or
extraparental transmission, as before. There are
yw.Anoffspringwhosegenotypeincludesacopy
now three genotypes and four phenotypes,
of the a allele is equally likely to inherit X or Y,
giving rise to nine phenogenotypes that can
andequallylikelytoinheritZorW.Offspringof
occur, which we denote by aaAC, aaAD, aaBC,
bb geneotype always have the normless pheno-
aaBD, abAC, abAD, abBC, abBD, and bbBD,
type BD, unless they are socialized extraparen-
and three that cannot occur, bbAC, bbAD, and tally.zz The transition table is shown in Fig.4.
bbBC. We represent the frequency of pheno-
We assume both genotypic and phenotypic
genotype i by pðiÞ; for i ¼ aaAC;y;bbBC:
fitness, as well as their interactions, are
We maintain the assumption that families are
multiplicative. Thus, the fitness of the nine
formed by random pairing and the offspring
phenogenotypes that can appear with positive
genotype obeys Mendelian segregation. We
frequency are as shown in Fig.5. The resulting
assume also that only the phenotypic traits
system consists of eight equations in eight of the
of parents, and not which particular parent
nine offspring phenogenotypes. One offspring
expresses them, are relevant to the transmission
phenogenotype is dropped, since the sum of
process. Therefore there are nine family pheno-
phenogenotype frequencies must be unity.
types, which can be written as AACC, AACD,
It is straightforward to check that there are
AADD, ABCC, ABCD, ABDD, BBCC, BBCD,
fivepureequilibria.TheseareaaAC,inwhichall
and BBDD. It follows that there are 54 familial
agents internalize both the altruistic and fitness
enhancing norms, aaAD, in which only the
88The same results hold for a haploid version of the
altruistic norm is internalized, aaBC, in which
model, except that there is no unit root in the aC
equilibrium. Moreover, if a is not completely dominant in
the diploid model, the unit root disappears and the zzExtensive simulations show that if a is incompletely
equilibrium is unambiguouslystable. dominant,the results describedbelow continue to hold.
412 H.GINTIS
Familial OffspringPhenogenotypicFrequency
type aaAC aaAD aaBC aaBD abAC abAD abBC abBD bbBD
aaaaAACC 1
aaaaABCC 1/2 1/2
aaaaBBCC 1
aaaaAACD 1/2 1/2
aaaaABCD 1/4 1/4 1/4 1/4
aaaaBBCD 1/2 1/2
aaaaAADD 1
aaaaABDD 1/2 1/2
aaaaBBDD 1
aaabAACC 1/2 1/2
aaabABCC 1/4 1/4 1/4 1/4
aaabBBCC 1/2 1/2
aaabAACD 1/4 1/4 1/4 1/4
aaabABCD 1/8 1/8 1/8 1/8 1/8 1/8 1/8 1/8
aaabBBCD 1/4 1/4 1/4 1/4
aaabAADD 1/2 1/2
aaabABDD 1/4 1/4 1/4 1/4
aaabBBDD 1/2 1/2
aabbABCD 1/4 1/4 1/4 1/4
aabbBBCD 1/2 1/2
aabbABDD 1/2 1/2
aabbBBDD 1
abbbABCD 1/8 1/8 1/8 1/8 1/2
abbbBBCD 1/4 1/4 1/2
abbbABDD 1/4 1/4 1/2
abbbBBDD 1/2 1/2
ababAACC 1/4 1/2 1/4
ababABCC 1/8 1/8 1/4 1/4 1/4
ababBBCC 1/4 1/4 1/2
ababAACD 1/8 1/8 1/4 1/4 1/4
ababABCD 1/16 1/16 1/16 1/16 1/8 1/8 1/8 1/8 1/4
ababBBCD 1/8 1/8 1/4 1/4 1/4
ababAADD 1/4 1/2 1/4
ababABDD 1/8 1/8 1/4 1/4 1/4
ababBBDD 1/4 1/2 1/4
bbbbBBDD 1
Fig. 4. Culturaland biological transition parameters.
only the fitness-enhancing norm is internalized, Individual Individual Individual Individual
aaBD, in which agents carry the gene for Phenogenotype Fitness Phenogenotype Fitness
internalization of norms, but no norms are in aaAC (1-u)(1-s)(1+t) aaAD (1-u)(1-s)
aaBC (1-u)(1+t) aaBD (1-u)
fact internalized, and bbBD, in which internali-
abAC (1-u)(1-s)(1+t) abAD (1-u)(1-s)
zationisabsent,andneitheraltruisticnorfitness- abBC (1-u)(1+t) abBD (1-u)
enhancing norms are transmitted from parents bbBD 1
to offspring. A check of the eigenvalues of the
Fig. 5. Payoffsto ninephenogenotypes.
Jacobian matrix shows that the aaAD and the
aaBD equilibria are unstable. inequalities. However, many simulations under
The Jacobian of the altruistic internalization varying parameter sets have failed to turn up an
equilibrium aaAC has eigenvalues instance of instability.***
(cid:1) The eigenvalues of the Jacobian for the aaBC
1 1 1(cid:5)g 1(cid:5)g
0;1; ; ; ; ; equilibrium are
2ð1þtÞ ð1þtÞ 1(cid:5)s 2ð1(cid:5)sÞð1þtÞ
(cid:1)
(cid:2) 1 1 1
1(cid:5)n 1(cid:5)n
0;1; ; ; ;
4ð1(cid:5)sÞð1þtÞ;4ð1(cid:5)sÞð1þtÞ : 2ð1þtÞ ð1þtÞ 2ð1þtÞ2ð1(cid:5)uÞ2
(cid:2)
1 1(cid:5)s
It is easy to check that the linearization of this ð1(cid:5)sÞ; ;1þg(cid:5)s :
2 4ð1þtÞ
equilibriumisstableifsog;sincetheunitrootis
semisimple. We cannot conclude that the equili-
nnnWhenalleleaisincompletelydominant,theunitroot
brium itself is necessarily stable for all para-
disappears, so the aaAC equilibrium is unambiguously
meters s; t; u; g; and n satisfying the above stable.
HITCHHIKER’SGUIDETOALTRUISM 413
Thus, aaBC is stable when gos; and unstable phenotype, since only such types are capable of
when the opposite inequality holds. internalizing a norm, and non-internalizers will
Finally,theeigenvaluesoftheJacobianforthe not desire to mimic internalizers.
bbBD equilibrium are We assume an agent with the a allele and
phenotype XY meets an agent of type WZ with
f0;0;0;0;1(cid:5)u;1ð1þtÞð1(cid:5)uÞ; probability ap ; where p is the fraction
2 WZ WZ
of the population with phenotype WZ, and
1ð1(cid:5)sÞð1þtÞð1(cid:5)uÞ;1ð1(cid:5)sÞð1(cid:5)uÞg:
4 2 switches to WZ if that type has higher fitness
than XY. The parameter a is the measure of the
Asinthesinglephenotypecase,thisisunstableif
strength of the tendency to shift to high-payoff
uo0 or ð1þtÞð1(cid:5)uÞ42; and is stable if either
phenotypes.
of the opposite inequalities hold. Moreover, it
It is easy to see that adding a replicator
can be shown that adding assortative mating
dynamic does not change the single phenogen-
leads to the instability of the bbBD equilibrium
otype equilibria. By checking the eigenvalues of
under the same conditions as in the single
the Jacobian matrix, we find that the aaAD and
phenotype case, shown in eqn (1).
aaBD equilibria remain unstable, and the repli-
In sum, under plausible conditions, one inter-
cator dynamic does not affect the conditions
nalization equilibrium is stableFthe altruism
for stability of the unnormed equilibrium bbBD.
equilibrium when g4s and the nonaltruism
The condition g4s for stability of the altruism
equilibrium when s4g: Since we expect s to be
equilibrium aaAC now becomes
small, whereas the ubiquity of extraparental
transmission favors a high g; the altruism g(cid:5)s
ao ; ð2Þ
equilibrium appears the more plausible of the 1(cid:5)g
two. Under not implausible conditions, either a
high return to the internal norm, assortative so a sufficiently strong replicator dynamic can
mating of agents that internalize, or pleiotropism undermine the stability of the aaAC equili-
in the form uo0; the only stable equilibrium of brium.www The condition s4g for stability of
thesysteminvolvesinternalization.Thedynamics, the non-altruism internalization equilibrium
which we present below, support this conclusion. aaBC when the replicator dynamic is included
now becomes
Copying High-Fitness Phenotypes:
g(cid:5)s
The Replicator Dynamic a4 ;
1þg(cid:5)s
The above models of cultural transmission
have been strongly criticized in the literature for andthisequilibriumisunstablewhenthereverse
suggesting that agents adopt norms independent inequality holds. Thus in this case, s4g con-
of their perceived payoffs. In fact, people do not tinues to ensure that aaBC is stable, but there is
always blindly follow the norms that have been now for sufficiently large a; this equilibrium is
inculcated in them, but at least at times treat stable even when g4s:zzz
compliance as a strategic choice (Wrong, 1961; In sum, adding a replicator dynamic changes
Gintis, 1975). The ‘‘oversocialized’’ model of the the stability properties of the model in only one
individual developed above may be improved by important way: a sufficiently strong replicator
adding a phenotypic copying process reflecting process can render the non-altruistic yet
the fact that agents shift from lower to higher internalized equilibrium aaBC, rather than the
payoff strategies. We represent this process as
wwwThis model also has a semisimple unit root, so
a replicator dynamic (Taylor and Jonker, 1978;
stability was checked by extensive simulations. A similar
Samuelson, 1997; Nowak and Sigmund, 1998;
resultholds forthe haploid model.
Gintis, 2000b). In the current context, there are zzzThis model also has a semisimple unit root, so
four phenotypes whose relative fitness ranks stability was checked by extensive simulations. A similar
result holds for the haploid model, except the relevant
them as BC4AC4BD4AD, and only agents
inequalityforstabilityofaaBCbecomesthemuchstronger,
with a copy of the a allele will copy another andhence implausible, inequality a4ðg(cid:5)sÞ=sð1þg(cid:5)sÞ:
414 H.GINTIS
altruistic equilibrium, aaAC, stable. Realisti- within groups, a high ratio of between-group to
cally, while the replicator process is key to within-group variance on the phenotypic trait is
understanding social change, in general we easily maintained, and hence high-payoff groups
expect this to be a relatively weak force, and quickly outpace low-payoff groups. Because
certainly too weak to undermine altruistic altruism hitchhikes, the mechanism that gener-
norms,unlesstheyincursubstantialfitnesscosts. ally undermines group selection, a high rate
Norms do not come labeled ‘‘altruistic norm’’, of inter-group migration (Maynard Smith, 1976;
‘‘instrumental practice’’. Rather, they are inex- Boorman & Levitt, 1980) does not undermine
tricably intermingled. It is quite common to internalization, as long as altruistic individuals
believe that immoral acts lead to disease, for adopt the A norms of the groups to which they
instance, just as does poor hygiene. Human migrate.
psychology has to be uncritical absorbers of To test this argument, I created an agent-
hosts of beliefs and values, only a small fraction based model of society with the following
of which can be seriously questioned by an characteristics (the specific assumptions made
individual member of society conditions us. are not critical, unless otherwise noted). The
society consists of 256 groups, each initially
comprising100members,arrangedspatiallyona
Why is Altruism Predominantly Prosocial?
torus (a 16(cid:11)16 grid with the opposite edges
Internal norms may be either pro- or anti- identified). Each group was seeded with ten
social. Indeed, there are many accounts of social aaACtypes,tenaaBCtypes,74bbBDtypes,and
norms that are severely socially costly, such as one each of the other possible types. In all
those involving invidious displays of physical groups, t ¼ 0:3 and u ¼ 0:05: Each group was
prowess (Edgerton, 1992). The reason for the then randomly assigned a value of g between 0
feasibility of anti-social norms is that once the and 0.60, a value of a between 0 and 0.5, a value
internalization gene has evolved to fixation, of s between 0.01 and 0.10, and a value of z (the
there is nothing to prevent group-harmful degree of assortative mating) between 0 and
phenotypic norms, such as our A, from also 0.70. Each group was also assigned an A
emerging, provided they are not excessively phenotype with a fitness effect between (cid:5)1 and
costly in comparison with the strength of the 1, such that if a group’s A-fitness effect is q; and
replicator process. The evolution of these if a fraction f of the group exhibit the A
phenotypes directly reduces the overall fitness phenotype, then each member of the group has
of the population. its fitness augmented by fq:
Yet as Brown (1991) and others have shown, In each round, for each of the 256 groups, I
there is a tendency in virtually all successful simulated the model as described in the previous
societies for cultural institutions to promote sections, and update the frequencies of
prosocial and eschew anti-social norms. The the various types in each group, according to
most reasonable explanation for the predomi- the fitness effect of their A phenotype and
nanceof prosocialnormsis gene-culturecoevolu- the fraction of the group that exhibits this
tionary multi-level selection: societies that phenotype. A fraction of each group (typically
promote prosocial norms have higher survival 5%) then migrated to a neighboring group. If
rates than societies that do not (Parsons, 1964; altruistic migrants adopt the A norm of their
Cavalli-Sforza & Feldman, 1981; Boyd & new groups, migration never undermines the
Richerson,1985;Boyd&Richerson,1990;Soltis stability of the altruism equilibria. To be
et al., 1995). Note that the usual arguments conservative, we assume here that agents take
againsttheplausibilityofgeneticgroupselection their genes with them to a new group, and they
do not apply to our model. This is because take their C phenotype to this new group (since
altruism is (a) phenotypic, and (b) ‘‘hitchhikes’’ C is the same for all groups), but they abandon
on the fitness-enhancing phenotypic norm. Be- their A phenotype when they migrate (e.g. an
cause altruism is phenotypic, and because a high aaAC type becomes a aaBC type in the new
degree of cultural uniformity can be maintained group). This assumption is maximally geared to
HITCHHIKER’SGUIDETOALTRUISM 415
undermine the altruism phenotype since immi- then more likely to move to high-fitness pheno-
grants never exhibit altruism. We then allow for types, it hurts the groups they are in, and on
some random drift in the individual groups balance lowers group fitness.
parameters s; a; g; and z; as well as the payoff of (g) The altruistic equilibrium was attained as
altruism to the group, q: long as the initial average assortative mating
Our final modeling assumption is that when probability z was at least 12.5%. An increase in
group size drops below a minimum (generally, I the rate of assortative mating led to more
set this to zero or ten agents), it is replaced by a globally stable altruism equilibria, but had no
copy of a randomly chosen other group. measurable effect on the equilibrium values of
I ran this model many times with varying other variables.
numbers of rounds, and varying the parameters (h)TheemergenceoftheaaACagentsandthe
described above. The system always stabilized elimination of anti-social internal norms were
by 100 periods, and the specific assumptions both due to population growth alone. The
concerning the parameters were never critical. extinction and replacement of groups by more
The following conclusions hold for these successful groups accounted only for the change
simulations: in the frequency of g and a (extinctions began to
occur after 20 rounds, and more than 1500
(a) Groups exhibiting the non-internalization extinctions typically occurred in a 100 round
equilibrium bbBD were quickly driven from the simulation).
population, except under the joint assumptions
thataltruisticagentsbecomenon-altruisticwhen These simulations thus strongly support the
they migrate, and the migration rate is over basic arguments of this paper. In particular, a
20%. high level of migration does not undermine the
(b) The equilibrium fraction of groups for altruistic equilibrium, since most of the effects
which the non-internalization equilibrium was occur on the cultural rather than the genetic
stable was highly variable and dependent upon level.Moreover,plausiblepatternsofpopulation
the specifics of the initial distribution of groups. growth and migration account for the prosoci-
Thusinequilibrium,thoughallgroupswerenear ality of the altruism phenotype A. The critical
the internalization equilibria (aaAC and aaBC), assumption that drives the model is simply that
in some this was globally stable and in others, there is a fitness-enhancing effect of the selfish C
only locally. phenotypic norm sufficiently strong to ensure
(c) The equilibrium fraction of the population that C can invade a population of D agents. The
exhibiting the altruistic phenotype was greater ability of the altruism phenotype A to ‘‘hitch-
than 85% (the mean at the start of each hike’’ on C is quite robust.
simulation was 10%).
(d) All but the highest prosocial A phenotypes
were eliminated from the population, so that the Altruistic Punishment Can Sustain Cooperation
mean fitness effect of the altruistic phenotypes When Altruistic Participation Cannot
was greater than 0.9 (the maximum possible was
Consider the following ‘‘social dilemma’’.888
1.0, and the mean at the start of the simulation
Each member of a group can either cooperate or
was approximately zero). In particular, no
shirk.Shirkingcostsnothing,butaddsnothingto
groups with anti-social norms ever survived in
the payoffs of the group members. Cooperating
equilibrium.
costs sn40; but contributes an amount fn4sn
(e) The mean level of extraparental socializa-
shared equally by the other members. Selfish
tion,gwasalsoveryhigh,beingatleast45%(the
individuals will always shirk in this situation, so
mean at the start of the simulation was 30%).
the potential gains from cooperating will be
(f) The mean strength of the replicator
forgone. If the situation is repeated sufficiently
dynamic, a; was 9% (the mean at the start of
frequentlywiththesameplayers,andifthegroup
the simulation was 25%). This shows that while
a higher a helps individuals, because they are 888ThissectionisanelaborationonBoydetal.(2001).
416 H.GINTIS
is sufficiently small, cooperation can be sustained small (as is likely to be the case except under
even with selfish players (Trivers, 1971; Axelrod extremeconditions,sincenoonehasanincentive
& Hamilton, 1981). However, with large groups toshirk),thenthisvalueofswillbeclosetozero
and/or infrequent repetition, universal shirking is for each altruistic punisher. But then the
virtually inevitable (Boyd & Richerson, 1988), as altruism equilibrium will be stable according
has been confirmed repeatedly in experiments to eqn (2), even when it would be violated for
with humans (Ledyard, 1995). s ¼ sn:
Given the potential gains to society of people Since the fitness costs of altruistic punishment
internalizing the altruistic norm A¼‘‘always are low, a replicator dynamic is unlikely to
cooperate’’ in the above situation, the absence render the altruism equilibrium unstable in this
ofthisnorminsocietysuggeststhatthecostsn is case. Moreover, there is evidence that altruistic
simply too high to sustain as an equilibrium of actsserveascostlysignalsofagentfitness(Gintis
the aaAC form. However, experimental results et al., 2003), in which case the altruistic
(Fehr & Ga¨chter, 2002) and ethnographic data phenotype is cannot be undermined by the
(Boehm 1993, 2000), not to mention everyday tendency to shift from lower to higher payoff
observation, suggest that the threat of being phenotypes.
punished by other group members for shirking
may serve to sustain cooperation where the
Conclusion
internalized value of cooperating does not.
Punishment can succeed where the norm of We have developed a plausible model of
cooperation cannot because the expected cost per altruistic cooperation and punishment that does
period of punishing shirkers is typically much not depend on repeated interaction, reputation
smaller that the cost of cooperating. This is effects, or multi-level selection. The latter
because (a) punishment such as shunning and obtains because there is no net within-group
ostracizing are inherently low cost, yet effective penalty to either the altruistic gene or the
when directed by large numbers against a few altruistic norm, even though there is a penalty
transgressors; and (b) the punishment need be to individuals carrying the gene and behaving
carried out only when shirking occurs, which is according to the norm.
likely to be infrequent in comparison with the One shortcoming of our model is that payoffs
number of times cooperationmustbecarriedout. areassumedconstant,whereasinmanycases,we
Nevertheless, altruistic punishment likely has would expected payoffs to be frequency depen-
strictly positive cost, so a selfish individual still dent, as when group members are engaged in a
will refrain from engaging in this activity. While non-cooperative game. For instance, the payoff
a genetic group selection model can explain the tobeingself-interestedmayincreasewhenagents
evolutionary stability of altruistic punishment are predominantly altruistic. In a related paper
(Gintis, 2000b), such models are sensitive to (Gintis, 2003), I show that such a situation gives
group size and migration rates (Eshel, 1972; rise to a heterogeneous equilibrium, in which
Rogers, 1990). The gene-culture coevolutionary both altruists and self-interested types partici-
model presented in this paper, by contrast, pate.Sincethepayoffstothetwotypesareequal
suffers less from these problems. in equilibrium, once again we can dispense with
To see this, suppose a fraction p of a group multi-level selection in specifying an equilibrium
with n members consists of altruistic punishers. with a positive level of altruism.
Topreventintentionalshirking byselfishagents, There are two objections that biologists
each must be prepared to inflict a punishment naturally raise to this model of altruism. First,
sn=pn on a shirker. Suppose a fraction q of the if the C norm is individually fitness enhancing
group nevertheless shirks (or perhaps is simply while the A is not, why is there not a genetic
perceived to shirk under conditions of imperfect mutation (for instance at another genetic locus)
information). Then the total amount of punish- that allows the individual to distinguish between
ment per altruistic punishment is s ¼ qsn=p: altruistic and fitness-enhancing behaviors, and
If p is large (as in our simulations) and q is hence to eschew the former? The answer is that