Você está na página 1de 19

Biological Altruism

First published Tue Jun 3, 2003; substantive revision Tue Oct 28, 2008
In evolutionary biology, an organism is said to behave altruistically when its
behaviour benefits other organisms, at a cost to itself. The costs and benefits
are measured in terms of reproductive fitness, or expected number of offspring.
So by behaving altruistically, an organism reduces the number of offspring it
is likely to produce itself, but boosts the number that other organisms are like
to produce. This biological notion of altruism is not identical to the everyday
concept. In everyday parlance, an action would only be called altruistic if it
was done with the conscious intention of helping another. But in the biological
sense there is no such requirement. Indeed, some of the most interesting example
of biological altruism are found among creatures that are (presumably) not
capable of conscious thought at all, e.g. insects. For the biologist, it is the
consequences of an action for reproductive fitness that determine whether the
action counts as altruistic, not the intentions, if any, with which the action i
Altruistic behaviour is common throughout the animal kingdom, particularly in
species with complex social structures. For example, vampire bats regularly
regurgitate blood and donate it to other members of their group who have failed
to feed that night, ensuring they do not starve. In numerous bird species, a bre
pair receives help in raising its young from other helper birds, who protect the
nest from predators and help to feed the fledglings. Vervet monkeys give alarm c
to warn fellow monkeys of the presence of predators, even though in doing so the
attract attention to themselves, increasing their personal chance of being attac
ked. In
social insect colonies (ants, wasps, bees and termites), sterile workers devote
whole lives to caring for the queen, constructing and protecting the nest, forag
for food, and tending the larvae. Such behaviour is maximally altruistic: steril
workers obviously do not leave any offspring of their own so have personal
fitness of zero but their actions greatly assist the reproductive efforts of the
From a Darwinian viewpoint, the existence of altruism in nature is at first sigh
puzzling, as Darwin himself realized. Natural selection leads us to expect anima
ls to
behave in ways that increase their own chances of survival and reproduction, not
those of others. But by behaving altruistically an animal reduces its own fitnes
s, so
should be at a selective disadvantage vis-à-vis one which behaves selfishly. To se
this, imagine that some members of a group of Vervet monkeys give alarm calls
when they see predators, but others do not. Other things being equal, the latter
have an advantage. By selfishly refusing to give an alarm call, a monkey can red
the chance that it will itself be attacked, while at the sametime benefiting fro
m the
alarm calls of others. So we should expect natural selection to favour those mon
that do not give alarm calls over those that do. But this raises an immediate pu
How did the alarm-calling behaviour evolve in the first place, and why has it no
been eliminated by natural selection? How can the existence of altruism be
reconciled with basic Darwinian principles?
1. Altruism and the Levels of Selection
2. Kin Selection and Inclusive Fitness
2.1 A Simple Illustration: the Prisoner's dilemma
3. Conceptual Issues
3.1 Altruism, Co-operation, Mutualism
3.2 Weak and Strong Altruism
3.3 Short-term versus Long-term Fitness Consequences
4. Reciprocal Altruism
5. But is it Real Altruism?
1. Altruism and the Levels of Selection
The problem of altruism is intimately connected with questions about the level a
which natural selection acts. If selection acts exclusively at the individual le
favouring some individual organisms over others, then it seems that altruism
cannot evolve, for behaving altruistically is disadvantageous for the individual
organism itself, by definition. However, it is possible that altruism may be
advantageous at the group level. A group containing lots of altruists, each read
to subordinate their own selfish interests for the greater good of the group, ma
well have a survival advantage over a group composed mainly or exclusively of
selfish organisms. A process of between-group selection may thus allow the
altruistic behaviour to evolve. Within each group, altruists will be at a select
disadvantage relative to their selfish colleagues, but the fitness of the group
as a
whole will be enhanced by the presence of altruists. Groups composed only or
mainly of selfish organisms go extinct, leaving behind groups containing altruis
ts. In the example of the
Vervet monkeys, a group containing a high proportion of alarm-calling monkeys
will have a survival advantage over a group containing a lower proportion. So
conceivably, the alarm-calling behaviour may evolve by between-group
selection, even though within each group, selection favours monkeys that do not
give alarm calls.
The idea that group selection might explain the evolution of altruism was first
broached by Darwin himself. In The Descent of Man (1871), Darwin discussed
the origin of altruistic and self-sacrificial behaviour among humans. Such
behaviour is obviously disadvantageous at the individual level, as Darwin
realized: he who was ready to sacrifice his life, as many a savage has been, rath
than betray his comrades, would often leave no offspring to inherit his noble
nature (p.163). Darwin then argued that self-sarcrificial behaviour, though
disadvantageous for the individual savage , might be beneficial at the group level:
a tribe including many members who...were always ready to give
aid to each other and sacrifice themselves for the common good, would be
victorious over most other tribes; and this would be natural selection (p.166).
Darwin's suggestion is that the altruistic behaviour in question may have evolve
d by
a process of between-group selection.
The concept of group selection has a chequered and controversial history in
evolutionary biology. The founders of modern neo-Darwinism R.A. Fisher, J.B.S.
Haldane and S. Wright were all aware that group selection could in principle
permit altruistic behaviours to evolve, but they doubted the importance of this
evolutionary mechanism. Nonetheless, many mid-twentieth century ecologists and
some ethologists, notably Konrad Lorenz, routinely assumed that natural selectio
would produce outcomes beneficial for the whole group or species, often without
even realizing that individual-level selection guarantees no such thing. This
uncritical good of the species tradition came to an abrupt halt in the 1960s, due
largely to the work of G.C. Williams (1966) and J. Maynard Smith (1964). These
authors argued that group selection was an inherently weak evolutionary force,
hence unlikely to promote interesting altruistic behaviours. This conclusion was
supported by a number of mathematical models, which apparently showed that
group selection would only have significant effects for a limited range of param
values. As a result, the notion of group selection fell into widespread disreput
e in
orthodox evolutionary circles; see Sober and Wilson 1998, Okasha 2006 and
Segestrale 2000 for details of the history of this debate.
The major weakness of group selection as an explanation of altruism, according t
the consensus that emerged in the 1960s, was a problem that Dawkins (1976) calle
subversion from within ; see also Maynard Smith 1964. Even if altruism is
advantageous at the group level, within any group altruists are liable to be
exploited by selfish free-riders who refrain from behaving altruistically. These
free-riders will have an obvious fitness advantage: they benefit from the altrui
sm of
others, but do not incur any of the costs. So even if a group is composed exclus
of altruists, all behaving nicely towards each other, it only takes a single sel
mutant to bring an end to this happy idyll. By virtue of its relative fitness ad
within the group, the selfish mutant will out-reproduce the altruists, hence
selfishness will eventually swamp altruism. Since the generation time of individ
organisms is likely to be much shorter than that of groups, the probability that
selfish mutant will arise and spread is very high, according to this line of arg
Subversion from within is generally regarded as a major stumbling block for
group-selectionist theories of the evolution of altruism.
If group selection is not the correct explanation for how the altruistic behavio
found in nature evolved, then what is? In the 1960s and 1970s a rival theory
emerged: kin selection or inclusive fitness theory, due originally to Hamilton (19
. This theory, discussed in detail below, apparently showed how altruistic
behaviour could evolve without the need for group-level selection, and quickly
gained prominence among biologists interested in the evolution of social behavio
the empirical success of kin selection theory contributed to the demise of the g
selection concept. However, the precise relation between kin and group selection
a source of ongoing controversy. For in recent years, proponents of multi-level
selection theory have resuscitated a form of group-level selection - sometimes ca
lled new
group selection - and shown that it can permit altruism to evolve (cf. Sober and
Wilson 1998). But new group selection turns out to mathematically equivalent to
kin selection, as a number of authors have emphasized (Grafen 1984, Frank 1998,
West et al. 2007, Lehmann et al. 2007); this point was already appreciated by
Hamilton (1975). Since the relation between old and new group selection is itself
a point of controversy, this explains why disagreement about the relation betwee
kin and group selection should persist.
2. Kin Selection and Inclusive Fitness
The basic idea of kin selection is simple. Imagine a gene which causes its beare
r to
behave altruistically towards other organisms, e.g. by sharing food with them.
Organisms without the gene are selfish they keep all their food for themselves,
and sometimes get handouts from the altruists. Clearly the altruists will be at
fitness disadvantage, so we should expect the altruistic gene to be eliminated f
the population. However, suppose that altruists are discriminating in who they
share food with. They do not share with just anybody, but only with their relati
This immediately changes things. For relatives are genetically similar they shar
genes with one another. So when an organism carrying the altruistic gene shares
food, there is a certain probability that the recipients of the food will also c
copies of that gene. (How probable depends on how closely related they are.) Thi
means that the altruistic gene can in principle spread by natural selection. The
causes an organism to behave in a way which reduces its own fitness but boosts t
fitness of its relatives who have a greater than average chance of carrying the
gene themselves. So theoverall effect of the behaviour may be to increase the
number of copies of the altruistic gene found in the next generation, and thus t
incidence of the altruistic behaviour itself.
Though this argument was hinted at by Haldane in the 1930s, it was first made
explicit by William Hamilton (1964) in a pair of seminal papers. Hamilton
demonstrated rigorously that an altruistic gene will be favoured by natural sele
when a certain condition, known as Hamilton's rule, is satisfied. In its simples
version, the rule states that b > c/r, where c is the cost incurred by the altru
(the donor), b is the benefit received by the recipients of the altruism, and r
is the
co-efficient of relationship between donor and recipient. The costs and benefits
measured in terms of reproductive fitness. The co-efficient of relationship depe
on the genealogical relation between donor and recipient it is defined as the
probability that donor and recipient share genes at a given locus that are identi
by descent . (Two genes are identical by descent if they are copies of a single ge
in a shared ancestor.) In a sexually reproducing diploid species, the value of r
for full
siblings is ½, for parents and offspring ½, for grandparents and grandoffspring ¼,
for full cousins 1/8, and so-on. The higher the value of r, the greater the pro
that the recipient of the altruistic behaviour will also possess the gene for al
So what Hamilton's rule tells us is that a gene for altruism can spread by natur
selection, so long as the cost incurred by the altruist is offset by a sufficien
t amount
of benefit to sufficiently closed related relatives. The proof of Hamilton's rul
e relies on
certain non-trivial assumptions; see Frank 1998, Grafen 1985 or Boyd and
McIlreath 2006 for details.
Though Hamilton himself did not use the term, his idea quickly became known as
kin selection , for obvious reasons. Kin selection theory predicts that animals are
more likely to behave altruistically towards their relatives than towards unrela
members of their species. Moreover, it predicts that the degree of altruism will
greater, the closer the relationship. In the years since Hamilton's theory was
devised, these predictions have been amply confirmed by empirical work. For
example, in various bird species, it has been found that helper birds are much mor
likely to help relatives raise their young, than they are to help unrelated bree
pairs. Similarly, studies of Japanese macaques have shown that altruistic action
such as defending others from attack, tend to be preferentially directed towards
kin. In most social insect species, a peculiarity of the genetic system known as
haplodiploidy means that females on average share more genes with their sisters
than with their own offspring. So a female may well be able to get more genes in
the next generation by helping the queen reproduce, hence increasing the number
sisters she will have, rather than by having offspring of her own. Kin selection
theory therefore provides a neat explanation of how sterility in the social inse
may have evolved by Darwinian means. (Note, however, that the precise
significance of haplodiploidy for the evolution of worker sterility is a controv
question; see Maynard Smith and Szathmary 1995, ch.16.)
Kin selection theory is often presented as a triumph of the gene's-eye view of
evolution , which sees organic evolution as the result of competition among genes
for increased representation in the gene-pool, and individual organisms as mere
vehicles that genes have constructed to aid their propagation (Dawkins 1976, 1982)
The gene's eye-view is certainly the easiest way of understanding kin selection,
was employed by Hamilton himself in his 1964 papers. Altruism seems anomalous
from the individual organism's point of view, but from the gene's point of view
makes good sense. A gene wants to maximize the number of copies of itself that a
found in the next generation; one way of doing that is to cause its host organis
m to
behave altruistically towards other bearers of the gene, so long as the costs an
benefits satisfy the Hamilton inequality. But interestingly, Hamilton showed tha
kin selection can also be understood from the organism's point of view. Though a
altruistic behaviour which spreads by kin selection reduces the organism's perso
fitness (by definition), it increases what Hamilton called the organism's inclus
fitness. An organism's inclusive fitness is defined as its personal fitness, plu
s the sum
of its weighted effects on the fitness of every other organism in the population
, the
weights determined by the coefficient of relationship r. Given this definition,
selection will act to maximise the inclusive fitness of individuals in the popul
Instead of thinking in terms of selfish genes trying to maximize their future
representation in the gene-pool, we can think in terms of organisms trying to
maximize their inclusive fitness. Most people find the gene's eye approach to kin
selection heuristically simpler than the inclusive fitness approach, but
mathematically they are in fact equivalent (Michod 1982, Frank 1998, Boyd and
McIlreath 2006).
Contrary to what is sometimes thought, kin selection does not require that anima
must have the ability to discriminate relatives from non-relatives, less still t
calculate coefficients of relationship. Many animals can in fact recognize their
often by smell, but kin selection can operate in the absence of such an ability.
Hamilton's inequality can be satisfied so long as an animal behaves altruistical
towards others animals that are in fact its relatives. The animal might achieve
by having the ability to tell relatives from non-relatives, but this is not the
possibility. An alternative is to use some proximal indicator of kinship. For ex
if an animal behaves altruistically towards those in its immediate vicinity, the
n the
recipients of the altruism are likely to be relatives, given that relatives tend
to live
near each other. No ability to recognize kin is presupposed. Cuckoos exploit
precisely this fact, free-riding on the innate tendency of birds to care for the
in their nests.
Another popular misconception is that kin selection theory is committed to
genetic determinism , the idea that genes rigidly determine or control behaviour.
Though some sociobiologists have made incautious remarks to this effect,
evolutionary theories of behaviour, including kin selection, are not committed t
o it.
So long as the behaviours in question have a genetical component, i.e. are influ
to some extent by one or more genetic factor, then the theories can apply. When
Hamilton (1964) talks about a gene which causes altruism, this is really shorthand
for a gene which increases the probability that its bearer will behave altruisti
to some degree. This is much weaker than saying that the behaviour is geneticall
determined , and is quite compatible with the existence of strong environmental
influences on the behaviour's expression. Kin selection theory does not deny the
truism that all traits are affected by both genes and environment. Nor does it d
that many interesting animal behaviours are transmitted through non-genetical
means, such as imitation and social learning (Avital and Jablonka 2000).
The importance of kinship for the evolution of altruism is very widely accepted
today, on both theoretical and empirical grounds. However, kinship is really onl
y a
way of ensuring that altruists and recipients both carry copies of the altruisti
c gene,
which is the fundamental requirement. If altruism is to evolve, it must be the c
that the recipients of altruistic actions have a greater than average probabilit
y of
being altruists themselves. Kin-directed altruism is the most obvious way of
satisfying this condition, but there are other possibilities too (Hamilton 1975,
Sober and Wilson 1998). For example, if the gene that causes altruism also cause
animals to favour a particular feeding ground (for whatever reason), then the
required correlation between donor and recipient may be generated. It is this
correlation, however brought about, that is necessary for altruism to evolve. Th
point was noted by Hamilton himself in the 1970s: he stressed that the coefficie
nt of
relationship of his 1964 papers should really be replaced with a more general
correlation coefficient, which reflects the probability that altruist and recipi
share genes, whether because of kinship or not (Hamilton 1970, 1972, 1975). This
point is theoretically important, and has not always been recognized; but in pra
kinship remains the most important source of statistical associations between
altruists and recipients (Maynard Smith 1998, Okasha 2002, West et al. 2007).
2.1 A Simple Illustration: the Prisoner's dilemma
The fact that correlation between donor and recipient is the key to the evolutio
n of
altruism can be illustrated via a simple one shot Prisoner's dilemma game.
Consider a large population of organisms who engage in a social interaction in
pairs; the interaction affects their biological fitness. Organisms are of two ty
selfish (S) and altruistic (A). The latter engage in pro-social behaviour, thus
benefiting their partner but at a cost to themselves; the former do not. So in a
(S,A) pair, the selfish organism does better - he benefits from his partner's al
without incurring any cost. However, (A,A) pairs do better than (S,S) pairs - fo
r the
former work as a co-operative unit, while the latter do not. The interaction thu
s has
the form of a one-shot Prisoner's dilemma, familiar from game theory. Illustrati
payoff values to each player , i.e., each partner in the interaction, measured in u
of biological fitness, are shown in the matrix below.
Player 2
Altruist Selfish
Player 1 Altruist 11,11 0,20
Selfish 20,0 5,5
Payoffs for (Player 1, Player 2) in units of reproductive fitness
The question we are interested in is: which type will be favoured by selection?
make the analysis tractable, we make two simplifying assumptions: that
reproduction is asexual, and that type is perfectly inherited, i.e., selfish (al
organisms give rise to selfish (altruistic) offspring. Modulo these assumptions,
evolutionary dynamics can be determined very easily, simply by seeing whether
the S or the A type has higher fitness, in the overall population. The fitness o
f the S
type, W(S), is the weighted average of the payoff to an S when partnered with an
and the payoff to an S when partnered with an A, where the weights are determine
by the probability of having the partner in question. Therefore,
W(S) = 5 * Prob(S partner/S) + 20 * Prob(A partner/S)
(The conditional probabilities in the above expression should be read as the
probability of having a selfish (altruistic) partner, given that one is selfish
Similarly, the fitness of the A type is:
W(A) = 0 * Prob(S partner/A) + 11 * Prob(A partner/A)
From these expressions for the fitnesses of the two types of organism, we can
immediately deduce that the altruistic type will only be favoured by selection i
there is a statistical correlation between partners, i.e., if altruists have gre
ater than
random chance of being paired with other altruists, and similarly for selfish ty
For suppose there is no such correlation - as would be the case if the pairs wer
formed by random sampling from the population. Then, the probability of having a
selfish partner would be the same for both S and A types, i.e., P(S partner/S) =
partner/A). Similarly, P(A partner/S) = P(A partner/A). From these probabilistic
equalities, it follows immediately that W(S) is greater than W(A), as can be see
from the expressions for W(S) and W(A) above; so the selfish type will be favour
by natural selection, and will increase in frequency every generation until all
altruists are eliminated from the population. Therefore, in the absence of corre
between partners, selfishness must win out. This confirms the point noted in sec
2 that altruism can only evolve if there is a statistical tendency for the
beneficiaries of altruistic actions to be altruists themselves.
If the correlation between partners is sufficiently strong, in this simple model
, then
it is possible for the condition W(A) > W(S) to be satisfied, and thus for altru
ism to
evolve. The easiest way to see this is to suppose that the correlation is perfec
t, i.e.,
selfish types are always paired with other selfish types, and ditto for altruist
s, so P(S
partner/S) = P(A partner/A) = 1. This assumption implies that W(A)=11 and W(S)=5
so altruism evolves. With intermediate degrees of correlation, it is also possib
le for
the condition W(S) > W(A) to be satisfied, given the particular choice of payoff
values in the model above.
This simple model also highlights the point made previously, that donor-recipien
correlation, rather than genetic relatedness, is the key to the evolution of alt
What is needed for altruism to evolve, in the model above, is for the probabilit
y of
having a partner of the same type as oneself to be sufficiently larger than the
probability of having a partner of opposite type; this ensures that the recipien
ts of
altruism have a greater than random chance of being fellow altruists, i.e.,
donor-recipient correlation. Whether this correlation arises because partners te
to be relatives, or because altruists are able to seek out other altruists and c
them as partners, or for some other reason, makes no difference to the evolution
3. Conceptual Issues
Altruism is a well understood topic in evolutionary biology; the theoretical ide
explained above have been extensively analysed, empirically confirmed, and are
widely accepted. Nonetheless, there are a number of conceptual ambiguities
surrounding altruism and related concepts in the literature; some of these are p
semantic, others are more substantive. Three such ambiguities are briefly discus
below; for further discussion, see West et al. 2007, Sachs et al. 2004 or Lehman
and Keller 2006.
3.1 Altruism, Co-operation, Mutualism
According to the standard definition, a social behaviour counts as altruistic if
reduces the fitness of the organism performing the behaviour, but boosts the fit
of others. This was the definition used by Hamilton (1964), and by many subseque
authors. However, there is less consensus on how to describe behaviours that boo
the fitness of others but also boost the fitness of the organism performing the
behaviour. As West et al. (2007) note, such behaviours are sometimes termed
co-operative , but this usage is not universal; others use co-operation to refer to
behaviour that boosts the fitness of others irrespective of its effect on self;
while still
others use cooperation as a synonym for altruism. (Indeed, in the simple Prisoner'
dilemma game above, the two strategies are usually called co-operate and defect .)
To avoid this confusion, West et al. (2007) suggest the term mutual benefit for
behaviours that benefit both self and other, while Sachs et al. (2004) suggest
byproduct benefit .
Whatever term is used, the important point is that behaviours that benefit both
and others can evolve much more easily than altruistic behaviours, and thus requ
no special mechanisms such as kinship. The reason is clear: organisms performing
such behaviours thereby increase their personal fitness, so are at a selective
advantage vis-a-vis those not performing the behaviour. The fact that the behavi
has a beneficial effect on the fitness of others is a mere side-effect, or bypro
duct, and
is not part of the explanation for why the behaviour evolves. For example, Sachs
al. (2004) note that an action such as joining a herd or a flock may be of this
sort; the
individual gains directly, via his reduced risk of predation, while simultaneous
ly red
ucing the predation risk of other individuals. By contrast with an altruistic ac
there is no personal incentive to cheat , i.e., to refrain from performing the acti
for doing so would directly reduce personal fitness.
Also indicative of the difference between altruistic behaviour and behaviour tha
benefit both self and others is the fact that in the latter case, though not the
the beneficiary may be a member of a different species, without altering the
evolutionary dynamics of the behaviour. Indeed, there are numerous examples
where the self-interested activities of one organism produce an incidental benef
it for
a non-conspecific; such behaviours are sometimes called mutualistic , though again,
this is not the only way that the latter term has been used (West et al. 2007).
contrast, in the case of altruism, it makes an enormous difference whether the
beneificiary and the donor are con-specifics or not; for if not, then kin select
ion can
play no role, and it is quite unclear how the altruistic behaviour can evolve.
Unsurprisingly, virtually all the bona fide examples of biological altruism in t
living world involve donors and recipients that are con-specifics. (Cases of so-
reciprocal altruism are sometimes thought to be exceptions to this generalization;
but see section 4 below.)
3.2 Weak and Strong Altruism
A quite different ambiguity concerns the distinction between weak and strong
altruism, in the terminology of D.S. Wilson (1977, 1980, 1990). This distinction
is ab
out whether the altruistic action entails an absolute or relative fitness reduct
ion for
the donor. To count as strongly altruistic, a behaviour must reduce the absolute
fitness (i.e., number of offspring) of the donor. Strong altruism is the standar
d notion
of altruism in the literature, and was assumed above. To count as weakly altrui
an action need only reduce the relative fitness of the donor, i.e., its fitness
relative to
that of the recipient. Thus for example, an action which causes an organism to l
an additional 10 offspring, but causes each organism(s) with which it interacts
leave an additional 20 offspring, is weakly but not strongly altruistic. The act
boosts the absolute fitness of the donor , but boosts the absolute fitness of other
organisms by even more, thus reducing the donor's relative fitness.
Should weakly altruistic behaviours be classified as altruistic or selfish? This
question is not merely semantic; for the real issue is whether the conditions un
which weak altruism can evolve are relevantly similar to the conditions under
which strong altruism can evolve, or not. Many authors argue that the answer is
no , on the grounds that weakly altruistic behaviours are individually advantageous
so can evolve with no component of kin selection or donor-recipient correlation,
unlike strongly altruistic behaviours (Grafen 1984, Nunney 1985, West et al. 200
To appreciate this argument, consider a game-theoretic scenario similar to the
one-shot Prisoner's dilemma of section 4, in which organisms engage in a pair-wi
interaction that affects their fitness. Organisms are of two types, weakly altru
(W) and non-altruistic (N). W-types perform an action that boosts their own fitn
by 10 units and the fitness of their partner by 20 units; N-types do not perform
action. The payoff matrix is thus:
Player 2
Weak Altruist Non
Player 1 Weak Altruist 30,30 10,20
Non 20,10 0,0
Payoffs for (Player 1, Player 2) in units of reproductive fitness
The payoff matrix highlights the fact that weak altruism is individually
advantageous, and thus the oddity of thinking of it it as altruistic rather than
To see this, assume for a moment that the game is being played by two rational
agents, as in classical game theory. Clearly, the rational strategy for each ind
is W, for W dominates N. Each individual gets a higher payoff from playing W tha
N, irrespective of what its opponent does 30 rather than 20 if the opponent play
W, 10 rather than 0 if the opponent plays N. This captures a clear sense in whic
weak altruism is individually advantageous.
In the context of evolutionary game theory, where the game is being played by pa
of organisms with hard-wired strategies, the counterpart of the fact that W
dominates N is the fact that W can spread in the population even if pairs are fo
at random (cf. Wilson 1980). To see this, consider the expressions for the overa
population-wide fitnesses of W and N:
W(W) = 30 * Prob(W partner/W) + 10 * Prob(N partner/W)
W(N) = 20 * Prob(W partner/N) + 0 * Prob(N partner/N)
(As before, Prob(W partner/W) denotes the conditional probability of having a
weakly altruistic partner given that one is weakly altruistic oneself, and so-on
From these expressions, it is easy to see that W(W) > W(N) even if the there is
correlation among partners, i.e., even if Prob(W partner/W) = P(W partner/N) and
P(N partner/W) = P(N partner/N). Therefore, weak altruism can evolve in the
absence of donor-recipient correlation; as we saw, this is not true of strong al
So weak and strong altruism evolve by different evolutionary mechanisms, hence
should not be co-classified, according to this argument.
However, there is a counter argument due to D.S. Wilson (1977, 1980), who
maintains that weak altruism cannot evolve by individual selection alone; a
component of group selection is needed. Wilson's argument stems from the fact th
in a mixed (W,N) pair, the non-altruist is fitter than the weak altruist. More
generally, within a single group of any size containing weak altruists and
non-altruists, the latter will be fitter. So weak altruism can only evolve, Wils
argues, in a multi-group setting in which the within-group selection in favour o
N, is counteracted by between-group selection in favour of W. (On Wilson's view,
the evolutionary game described above is a multi-group setting, involving a larg
number of groups of size two.) Thus weak altruism, like strong altruism, in fact
evolves because it is group-advantageous, Wilson argues.
The dispute between those who regard weak altruism as individually advantageous,
and those like Wilson who regard it as group advantageous, stems ultimately from
differing conceptions of individual and group selection. For Wilson, individual
selection means within-group selection, so to determine which strategy is favour
by individual selection, one must compare the fitnesses of W and N types within
group, or pair. For other theorists, individual selection means selection based
differences in individual phenotype, rather than social context; so to determine
which strategy is favoured by individual selection, one must compare the fitness
of W and N types in the same social context, i.e., with the same partner. These
comparisons yield different answers to the question of whether weak altruism is
individually advantageous. Thus the debate over how to classify weak altruism is
intimately connected to the broader levels of selection question; see Nunney 198
Okasha 2005, 2006, Fletcher and Doebeli 2006, West et al. 2007, for further
3.3 Short-term versus Long-term Fitness Consequences
A further source of ambiguity in the definition of biological altruism concerns
time-scale over which fitness is measured. Conceivably, an animal might engage i
a social behaviour which benefits another and reduces its own (absolute) fitness
the short-term; however, in the long-term, the behaviour might be to the animal'
advantage. So if we focus on short-term fitness effects, the behaviour will seem
altruistic; but if we focus on lifetime fitness, the behaviour will seem selfish
- the
animal's lifetime fitness would be reduced if it did not perform the behaviour.
Why might a social behaviour reduce an animal's short-term fitness but boost its
lifetime fitness? This could arise in cases of directed reciprocation , where the
beneficiary of the behaviour returns the favour at some point in the future (cf.
et al. 2004). By performing the behaviour, and suffering the short-term cost, th
animal thus ensures (or raises the chance) that it will receive return benefits
in the
future. Similarly, in symbioses between members of different species, it may pay
organism to sacrifice resources for the benefit of a symbiont with which it has
a long-
term relationship, as its long-term welfare may be heavily dependent on the
symbiont's welfare.
From a theoretical point of view, the most satisfactory resolution of this ambig
is to use lifetime fitness as the relevant parameter (cf. West et al. 2007) Thus
action only counts as altruistic if it reduces an organism's lifetime fitness. T
stipulation makes sense, since it preserves the key idea that the evolution of a
requires statistical association between donor and recipient; this would not be
if short-term fitness were used to define altruism, for behaviours which reduce
short-term fitness but boost lifetime fitness can evolve with no component of ki
selection, or donor-recipient correlation. However, the stipulation has two
disadvantages: (i) it makes it harder to tell whether a given behaviour is altru
since lifetime fitness is notoriously difficult to estimate; (ii) it has the con
that most models of reciprocal altruism are mis-named.
4. Reciprocal Altruism
The theory of reciprocal altruism was originally developed by Trivers (1971), as
attempt to explain cases of (apparent) altruism among unrelated organisms,
including members of different species. (Clearly, kin selection cannot help expl
altruism among non-relatives.) Trivers' basic idea was straightforward: it may p
an organism to help another, if there is an expectation of the favour being retu
in the future. ( If you scratch my back, I'll scratch yours .) The cost of helping i
offset by the likelihood of the return benefit, permitting the behaviour to evol
ve by
natural selection. Trivers termed with evolutionary mechanism reciprocal altruism .
For reciprocal altruism to work, there is no need for the two individuals to be
relatives, nor even to be members of the same species. However, it is necessary
individuals should interact with each more than once, and have the ability to
recognize other individuals with whom they have interacted in the past.[1] If
individuals interact only once in their lifetimes and never meet again, there is
obviously no possibility of return benefit, so there is nothing to be gained by
helping another. However, if individuals encounter each other frequently, and ar
capable of identifying and punishing cheaters who have refused to help in the past
then the helping behaviour can evolve. A cheat who refuses to help will ultimately
sabotage his own interests, for although he does not incur the cost of helping o
he forfeits the return benefits too others will not help him in the future. This
evolutionary mechanism is most likely to work where animals live in relatively
small groups, increasing the likelihood of multiple encounters.
As West et al. (2007) note, if altruism is defined by reference to lifetime fitn
then Trivers' theory is not really about the evolution of altruism at all; for
behaviours that evolve via reciprocation of benefits, as described by Trivers, a
ultimately of direct benefit to the individuals performing them, so do not reduc
lifetime fitness. Despite this consideration, the label reciprocal altruism is
well-entrenched in the literature, and the evolutionary mechanism that it descri
is of some importance, whatever it is called. Where reciprocal altruism is refer
to below, it should be remembered that the behaviours in question are only
altruistic in the short-term.
The concept of reciprocal altruism is closely related to the Tit-for-Tat strateg
y in
the iterated Prisoner's Dilemma (IPD) from game theory. In the IPD, players
interact on multiple occasions, and are able to adjust their behaviour depending
what their opponent has done in previous rounds. There are two possible strategi
co-operate and defect; the payoff matrix (per interaction) is as in section 2.1
The fact that the game is iterated rather than one-shot obviously changes the
optimal course of action; defecting is no longer necessarily the best option, so
as the probability of subsequent encounters is sufficiently high. In their famou
computer tournament in which a large number of strategies were pitted against
each other in the IPD, Axelrod and Hamilton (1981) found that the Tit-for-Tat
strategy yielded the highest payoff. In Tit-For-Tat, a player follows two basic
rules: (i) on the first encounter, cooperate; (ii) on subsequent encounters, do
your opponent did on the previous encounter. The success of Tit-for-Tat was
widely taken to confirm the idea that with multiple encounters, natural selectio
could favour social behaviours that entail a short-term fitness cost. Subsequent
work in evolutionary game theory, much of it inspired by Axelrod and Hamilton's
ideas, has confirmed that repeated games permit the evolution of social behaviou
that cannot evolve in one-shot situations (cf. Nowak 2006). For an up-to-date
discussion of social behaviour that evolves via reciprocation of benefits, see S
et al. 2004.
Despite the attention paid to reciprocal altruism by theoreticians, clear-cut
empirical examples are relatively few (Hammerstein 2003, Sachs et al. 2004). Thi
is probably because the pre-conditions for reciprocal altruism to evolve- multip
encounters and individual recognition - are not especially common. However, one
possible example is provided by blood-sharing in vampire bats (Wilkinson 1984,
1990). It is quite common for a vampire bat to fail to feed on a given night. Th
is is
potentially fatal, for bats die if they go without food for more than a couple o
f days.
On any given night, bats donate blood (by regurgitation) to other members of the
group who have failed to feed, thus saving them from starvation. Since vampire
bats live in small groups and associate with each other over long periods of tim
the preconditions for reciprocal altruism are likely to be met. Wilkinson's stud
showed that bats tended to share food with their close associates, and were more
likely to share with others that had recently shared with them. These findings
accord with reciprocal altruism theory.
Trivers (1985) describes an apparent case of reciprocal altruism between non
con-specifics. On tropical coral reefs, various species of small fish act as clea
for large fish, removing parasites from their mouths and gills. The interaction
mutually beneficial the large fish gets cleaned and the cleaner gets fed. Howeve
Trivers notes that the large fish sometimes appear to behave altruistically towa
the cleaners. If a large fish is attacked by a predator while it has a cleaner i
n its
mouth, then it waits for the cleaner to leave before fleeing the predator, rathe
r than
swallowing the cleaner and fleeing immediately. Trivers explains the larger fish
behaviour in terms of reciprocal altruism. Since the large fish often returns to
same cleaner many times over, it pays to look after the cleaner's welfare, i.e.,
not to
swallow it, even if this increases the chance of being wounded by a predator. So
larger fish allows the cleaner to escape, because there is an expectation of ret
benefit getting cleaned again in the future. As in the case of the vampire bats,
is because the large fish and the cleaner interact more than once that the behav
can evolve.

5. But is it Real Altruism?

The evolutionary theories described above, in particular kin selection, go a lon
way towards reconciling the existence of altruism in nature with Darwinian
principles. However, some people have felt these theories in a way devalue
altruism, and that the behaviours they explain are not really altruistic. The grou
for this view are easy to see. Ordinarily we think of altruistic actions as
disinterested, done with the interests of the recipient, rather than our own int
in mind. But kin selection theory explains altruistic behaviour as a clever stra
devised by selfish genes as a way of increasing their representation in the gene
at the expense of other genes. Surely this means that the behaviours in question
only apparently altruistic, for they are ultimately the result of genic self-inter
Reciprocal altruism theory also seems to take the altruism out of altruism .
Behaving nicely to someone in order to procure return benefits from them in the
future seems in a way the antithesis of real altruism it is just delayed
This is a tempting line of argument. Indeed Trivers (1971) and, arguably, Dawkin
(1976) were themselves tempted by it. But it should not convince. The key point
remember is that biological altruism cannot be equated with altruism in the
everyday vernacular sense. Biological altruism is defined in terms of fitness
consequences, not motivating intentions. If by real altruism we mean altruism
done with the conscious intention to help, then the vast majority of living crea
are not capable of real altruism nor therefore of real selfishness either. Ants and
termites, for example, presumably do not have conscious intentions, hence their
behaviour cannot be done with the intention of promoting their own self-interest
nor the interests of others. Thus the assertion that the evolutionary theories r
above show that the altruism in nature is only apparent makes little sense. The
contrast between real altruism and merely apparent altruism simply does not
apply to most animal species.
To some extent, the idea that kin-directed altruism is not real altruism has been
fostered by the use of the selfish gene terminology of Dawkins (1976). As we have
seen, the gene's-eye perspective is heuristically useful for understanding the
evolution of altruistic behaviours, especially those that evolve by kin selectio
n. But
talking about selfish genes trying to increase their representation in the gene-po
is of course just a metaphor (as Dawkins fully admits); there is no literal sens
e in
which genes try to do anything. Any evolutionary explanation of how a
phenotypic trait evolves must ultimately show that the trait leads to an increas
e in
frequency of the genes that code for it (presuming the trait is transmitted gene
) Therefore, a selfish gene story can by definition be told about any trait, inclu
a behavioural trait, that evolves by Darwinian natural selection. To say that ki
selection interprets altruistic behaviour as a strategy designed by selfish genes
aid their propagation is not wrong; but it is just another way of saying that a
Darwinian explanation for the evolution of altruism has been found. As Sober and
Wilson (1998) note, if one insists on saying that behaviours which evolve by kin
selection / donor-recipient correlation are really selfish , one ends up reserving
word altruistic for behaviours which cannot evolve by natural selection at all.
Do theories of the evolution of biological altruism apply to humans? This is par
t of
the broader question of whether ideas about the evolution of animal behaviour ca
be extrapolated to humans, a question that fuelled the sociobiology controversy
the 1980s. All biologists accept that Homo sapiens is an evolved species, and th
that general evolutionary principles apply to it. However, human behaviour is
obviously influenced by culture to a far greater extent than that of other anima
ls, and
is often the product of conscious beliefs and desires (though this does not nece
mean that genetics has no influence.) Nonetheless, at least some human behaviour
seem to fit the predictions of the evolutionary theories reviewed above. In gen
humans behave more altruistically (in the biological sense) towards their close
than towards non-relatives, e.g. by helping relatives raise their children, just
as kin
selection theory would predict. It is also true that we tend to help those who h
helped us out in the past, just as reciprocal altruism theory would predict. On
other hand, numerous human behaviours seem anomalous from the evolutionary
point of view. Think for example of adoption. Parents who adopt children instead
of h
aving their own reduce their biological fitness, obviously, so adoption is an
altruistic behaviour. But it is does not benefit kin for parents are generally
unrelated to the infants they adopt and nor do the parents stand to gain much in
the form of reciprocal benefits. So although evolutionary considerations can hel
p us
understand some human behaviours, they must be applied judiciously.
Where human behaviour is concerned, the distinction between biological altruism,
defined in terms of fitness consequences, and real altruism, defined in terms of t
agent's conscious intentions to help others, does make sense. (Sometimes the lab
psychological altruism is used instead of real altruism.) What is the relationship
between these two concepts? They appear to be independent in both directions, as
Elliott Sober (1994) has argued. An action performed with the conscious intentio
of helping another human being may not affect their biological fitness at all, s
would not count as altruistic in the biological sense. Conversely, an action
undertaken for purely self-interested reasons, i.e., without the conscious inten
of helping another, may boost their biological fitness tremendously.
Sober argues that, even if we accept an evolutionary approach to human behaviour
there is no particular reason to think that evolution would have made humans int
egoists rather than psychological altruists. On the contrary, it is quite possib
le that
natural selection would have favoured humans who genuinely do care about helping
others, i.e., who are capable of real or psychological altruism. Suppose there is
evolutionary advantage associated with taking good care of one's children a quit
plausible idea. Then, parents who really do care about their childrens' welfare,
who are real altruists, will have a higher inclusive fitness, hence spread more of
genes, than parents who only pretend to care, or who do not care. Therefore,
evolution may well lead real or psychological altruism to evolve. Contrary to
what is often thought, an evolutionary approach to human behaviour does not impl
that humans are likely to be motivated by self-interest alone. One strategy by w
selfish genes may increase their future representation is by causing humans to be
non-selfish, in the psychological sense.