“Explaining Altruism: A New Defence of Group Selection” By Keyana C. Sapp

Explaining Altruism: A New Defence of Group Selection

By Keyana C. Sapp

MScR Philosophy at the University of Edinburgh

Keyana C. Sapp

1. Abstract

When providing explanations of cooperation in nature, evolutionary biologists draw upon three distinct selective pressures. They are direct benefits, reciprocal altruism and kin selection. They are useful for the explanation of both human and non-human examples of cooperative behaviour. However, their scope of applicability is much more significant with respect to non-human cooperation. The aim of this paper is to assess the extent to which the selective pressures listed above are relevant to our best explanation for the super-cooperative behaviour demonstrated by human beings. Ultimately, I show that they are incomplete. In brief, this is because human cooperative behaviour is unique from other lifeforms’ in so far as it is based in cognitive abilities which other species cannot be said to possess (or at least cannot be said to possess to the same extent). Our use of language, our participation with social institutions, and the internalisation of cultural norms all play a role in the evolutionary progress of human beings. These sorts of uniquely human traits are central to the purpose of this paper; namely, to provide a new defence of group selection as the operative level of selection when it comes to altruism as an adaptive trait.

2. Introduction

The pursuit of understanding human cooperative behaviour is fraught with semantic inconsistency. As a result, it is especially important to clarify some definitions. To help maintain regularity, I will employ the explanatory framework described by Lehmann & Keller (2006. 1365-1376). I shall describe any behaviour which increases the relative fitness of a recipient as helping behaviour. Existing within the set of helping is cooperation, which refers to any behaviour that improves the fitness of another individual as well as the focal individual (FI) herself. In this sense, all cooperative behaviour can be considered a form of mutualism, whereby both parties benefit. Altruism, on the other hand, refers to behaviour that improves the fitness of another individual, at a detriment to the FI’s own fitness. Therefore, altruistic behaviour can be thought of as a subset of helping behaviour, but distinct from cooperation. There are two kinds of benefits gained from cooperative behaviour. They are direct and indirect benefits. When an action benefits the FI directly, that is to say the benefit is gained immediately. When an action benefits the FI indirectly, we are describing expected future benefits that come as a result of a present behaviour through repeated interactions. Applying the three selective pressures for cooperation to the categorical definitions above, we will find that cooperation is parsable into direct benefits and reciprocation, while altruism reduces to kin selection. With the fundamental principles clarified, we move onto the problem of altruism.

“He who was ready to sacrifice his life, as many a savage has been, rather than betray his comrades, would often leave no offspring to inherit his noble nature.”

– Charles Darwin
The Descent of Man, p. 163.

The problem of altruism has posed issues for Darwinian theory since its original conception. Given what we know about natural selection and the importance of individual competition, it seems impossible for evolution to favour cooperative behaviour over selfish, exploitative behaviour. However, examples of cooperative behaviour are abundant in nature. Sterile worker bees, blood-sharing vampire bats, human social institutions and many others, all serve as evidence of this fact. If evolution operates solely at the level of the individual, then the operative evolutionary pressure must be for self preservation, and pro-social behaviours would be selected against. But they are too prevalent for this to be plausible.

The first attempt to deal with the problem was a provisional form of group selection. This position suggested that whilst altruistic behaviour is detrimental to the fitness of the individual, it improves the relative fitness of the group as a whole. This higher level increase in fitness is said to displace the individual selective pressures to act selfishly. Cooperative behaviour, according to early group selectionists, evolved in this way. However, this kind of explanation has been dismissed for two main reasons. First, is the free-rider problem. This is the claim that in a group containing exclusively altruists, it would take only one selfish mutant to infiltrate the ranks to be at a huge fitness advantage relative to the rest of the group. Should a mutant arise, then selection will favour her reproduction over the rest of the group, and the altruistic condition would slowly cease to exist. This criticism of early group selection was made famous by Richard Dawkins, dubbing the problem ‘subversion from within’ (Dawkins, 1976. 72). Second, mathematical models have shown that selection at the level of groups is a relatively weak evolutionary force in comparison to selection at the level of the individual (Maynard Smith, 1964. 1145-1147). As a consequence, the idea of group selection as a significant evolutionary force fell out of fashion in mainstream biology. Even today, the consensus is largely that group selection is too weak an evolutionary pressure to be a realistic explanation of altruistic behaviour (see Earnshaw, 2015. and Jeler, 2016). I aim to demonstrate that group selection, has a very specialised and important role to play in our explanatory arsenal. I hope to convince you that it is our best explanation of human altruistic behaviour. In any case, the widespread dismissal of group selection as a viable prospect for the explanation of cooperative behaviour led biologists to begin working on new hypotheses. The products of this endeavour are the three selective pressures which I highlighted above. The aim of the immediately following section is to give a thorough explanation of each mechanism, and to relevantly link them to examples from nature. Achieving this, I demonstrate why these models do not satisfactorily fit with human cooperative behaviour. By the end of the paper, attention shifts to demonstrating why group selection is the operative evolutionary pressure with respect to human altruistic behaviour. To this end, I argue that incremental improvements in linguistic capabilities during human evolutionary history facilitated increased competition between human cultural groups; and that this increased competition makes human groups a legitimate level of selection.

3. The Selective Pressures Promoting Cooperation

As mentioned above, non-human cooperative behaviour can be justifiably explained with appeal to one or more of the following selective pressures: direct benefits, reciprocation and kin selection. In this section, explanations of each mechanism alongside examples of the mechanism acting in nature are provided. The mechanisms are discussed in turn.

In situations where direct benefits are accrued from helping behaviour, we mean that the FI improves the relative fitness of a second agent as well as her own. Behaviours which induce this mechanism are regarded as mutualistic since both the FI and the second individual receive fitness benefits during the same interaction. They are the only kind which produce benefits directly to the agents involved without a temporal lag or a requirement that the helping behaviour be performed to genetic kin. An effective metaphor for explaining direct benefits is the free market. According to capitalists at least, when the market is left to its own devices, interactions within it will only occur if they produce net benefits for both interacting parties. A trade will only be agreed if both parties are expected to accrue some sort of direct benefit from interaction. As a result, when a trade is made, a mutualistic agreement is reached, in which both parties receive immediate, direct benefits. Examples of directly beneficial interactions in nature are common. One such example is exhibited in interactions between oxpeckers and rhinoceroses. The oxpecker is a species of bird, whose main food source is lice and ticks. These lice and ticks live on the skin of rhinos, and their chosen habitat results in dermatological discomfort. Consequently, oxpeckers are often seen following rhinos, perching on their backs and relieving them of their microscopic pests (Stutterheim, 1980. 21-25). Much like the buyer and seller within a free market, the oxpecker and the rhinoceros provide solutions to each other’s problems, and each party accrues direct benefits, resulting in an increase in relative fitness for both agents.

Having explained cooperative behaviour as a result of direct benefits, discussion now moves onto cooperation as a result of indirect benefits. The first mechanism which falls under this category is reciprocal altruism, a concept popularised by Robert Trivers in the early 1970’s. According to Trivers’ model, altruistic behaviour could have evolved as a result of an expectation for future fitness improvements. In other words, selection could favour cooperative behaviour, if the fitness improvement which the FI bestows on another agent is to be reciprocated in the future (Trivers, 1971. 36-37). Due to its requirement for repeated interactions, an iterated prisoner’s dilemma is a useful way of demonstrating reciprocal altruism in practice. In 1980 Axelrod held a tournament, the aim of which was to find the most effective strategy for maximising payoffs in 200 repeated games. Many strategies were submitted such as random cooperation, always defect and always cooperate, and each strategy was put up against every other. In the end, the winning strategy was tit for tat, according to which the FI will cooperate in the first round, then always do what the opponent did in the previous round. The importance of this discovery was that it made clear the environmental requirements for reciprocation to be able to take place. First, both agents must have the ability to recognise previous defectors and cooperators. Second, reciprocation is more likely to occur in smaller groups where the likelihood of repeat interactions is obviously very high. Finally, reciprocation is more likely to occur in groups where cooperation is already common and social ‘trust’ is high (Axelrod, 1980. 389-401). This model can be applied to a variety of cooperative behaviours found in nature. The welfare state is a prime example of reciprocation at play amongst human beings. Each of us incurs a cost by paying taxes towards public provisions such as healthcare and police protection. This cost is incurred with the expectation of a future fitness benefit if it is required. As such, people cooperate with one another by paying taxes with the expectation that one’s initial loss in fitness will be reciprocated by others at a later date. For a non-human example consider the vervet monkey who gives alarm calls in order to warn her troop. Initially, this behaviour is evolutionarily counterintuitive. The monkey who spots the eagle is surely fitter if she stays quiet and hides, thus allowing the predator to catch one of her peers, rather than blatantly giving away her own position. This kind of behaviour could be explained by reciprocation. The monkey who gives the alarm call when she sees the predator does so with the expectation that the same behaviour will be repeated by other troop members when they encounter a predator. By engaging in the cooperative behaviour, it is possible that the monkey would actually gain a long run advantage, providing she is a member of a mostly cooperative troop.

Having said this, it is my view that in virtually all instances of non-human reciprocation it is better to appeal to kin selection. This is because repeated interactions outside of kin groups are relatively uncommon outside of humankind. The greater cognitive capacities which human beings possess are what make sustained repeated interactions possible. Communication, facial recognition and technological interconnectedness are all unique features of human psychology which allow for a greater opportunity to cooperate through reciprocation. As a consequence, biologists draw a distinction between the kinds of reciprocal mechanisms we find amongst non-human cooperative interactions, and those we find amongst humans due to increased cognitive faculties. They are distinguished using the terms weak, and strong altruism, respectively (Gintis, 2000. 169-179). This distinction will play a significant role in my later evolutionary explanation for human cooperation. For now, we examine the mechanism for explaining cooperative behaviour known as kin selection.

Kin selection, when put in basic terms, is a relatively simple idea. Let us suppose that there is a gene which causes its host to altruistically share its resources with other individuals. As uncovered by the problem of altruism, natural selection operating at the level of the individual could not select for this altruistic gene. This is because bearers of the gene will be at a fitness disadvantage to individuals who do not share resources, and act in their own self-interest. But what if the bearers of the altruistic gene actively chose which individuals they behave altruistically towards? If altruistic individuals only acted altruistically towards towards kin (family members, or more specifically, individuals who are similar genetically), then selection could favour altruistic behaviour as kin are more likely to carry copies of the altruistic gene as well. Providing the net effect of the individual fitness loss to the altruist is less than the relative fitness improvement of family members who are more likely to carry the same altruistic gene, natural selection will favour altruistic kin groups over non-altruistic ones, and the altruistic gene will be propagated in the next generation. Though Darwin fleeted with the idea in his study of sterile social insect behaviour (Darwin, 2009. Ch.8), W.D. Hamilton is considered the originator of the concept in its complete form. Hamilton earned this accolade due to his mathematical demonstration that a gene for altruistic behaviour will be selected for providing it satisfies ‘Hamilton’s Rule’. The rule is:


r = Genetic relatedness of recipient to altruist (coefficient of relatedness).
B = Fitness benefit gained by recipient.
C = Cost to altruist.

(Hamilton, 1964. 21-23)

The coefficient of relatedness is dependant on the level of genealogical relatedness between the two interactors. In a diploid species like human beings, the coefficient is 1/2 for siblings and parents/offspring, 1/4 for grandparents/grandchildren, and so on. It characterises the probability that the two interactors will share an identical gene at any particular loci, and is easily calculable for our purposes due to the constant 1:1 ratio of haploid cell recombination during human zygote production. The importance of Hamilton’s rule in the context of solving the problem of altruism, is held in the fact that it shows that a gene promoting altruistic behaviour can be favoured by natural selection, providing that the benefit caused by the altruistic behaviour on related individuals is greater than the relative cost incurred by the individual bearer of the altruistic gene.

The prosocial behaviour of sterile, worker honeybees is a useful example of the kin selection mechanism in action. Within a functioning honeybee hive there is only one fertile female which is known as the queen. The male bees are known as drones, and their primary function within the population is simply to fertilize a receptive queen. The rest of the female bees are called workers, and are completely sterile. Their lives consist in completing a multiplicity of prosocial chores for the benefit of the hive, ranging from wax production, to looking after the brood. This kind of behaviour could not be selected for if the selection operated at the level of the individual alone. Kin selection provides an explanation for the sterility of worker honeybees, since each worker on average shares more genetic information with their sisters than with offspring of their own. This is a genetic system known as haplodiploidy; a consequence of which is that a female worker’s time would be best spent helping the queen bee in any way possible if she wishes to ensure the propagation of her genes in future generations. In this sense, kin selection is considered a triumph for gene selectionists as it shows that arguments inciting selection at the level of individuals alone are not complete. Later in this paper, we will evaluate whether the triumph is completely justified when we consider arguments for human cooperation.

If my analysis to this point is sound, then we have uncovered the major mechanisms used in evolutionary biology to explain cooperative behaviour in nature. The aim of the following section of the thesis is to demonstrate why the three mechanisms outlined above are insufficient when it comes to explaining the human cooperative condition. Once this has been achieved, attention will turn to providing an evolutionary explanation for human cooperation given our inability to appeal to the three mechanisms alone.

4. The Applicability of the Selective Pressures in Favour of Cooperation to the Human Altruistic Condition

It is my opinion that the mechanisms which I have outlined provide us with all of the necessary tools to explain non-human cooperative behaviour. Proof of this claim would require work that extends beyond the scope of this essay. Nevertheless, these mechanisms appear to be sufficient for the explanation of all canonical examples of non-human cooperation in nature. As a result, the burden of proof falls on the sceptic to provide an example which the mechanisms cannot accommodate.

In any case, the important question for our purposes is the extent of applicability when it comes to the human cooperative condition. My claim is not that the three mechanisms never apply to human beings. Our preferential treatment of family members over strangers is explained by appeal to kin selection. Our participation in market dynamics can still be explained with appeal to the mutual benefits gained through joint transaction. Our reluctance to steal the property of others can be explained with appeal to reciprocal altruism and our expectation that our property not be stolen in the future. However, human beings seem to display levels of altruism that transcend the explanatory power of all of the mechanisms for explaining cooperation. Consider the act of anonymously donating to charity. There is no direct benefit for the focal individual. There is no expected future reciprocal benefit, since an anonymous donation is a one shot interaction with no potential for recognition. Further, we cannot appeal to kin selection, because the genetic makeup of the beneficiary of the donation is unknown. An argument could be made that donating anonymously to charity has the effect of promoting social stability within a society, and that improvements in social factors such as crime rates might provide an individual benefit for charitable donations. But this is not sufficient in cases where the donations are provided internationally to societies with which the donor has no direct contact. The adoption of orphaned children from across the planet is another example of altruistic behavior which cannot be explained with appeal to the three mechanisms for explaining cooperation. Likewise, the preparedness to die in battle for one’s nation. If we are to give a justified evolutionary account of human cooperative behaviour, we must look further than the general framework outlined above. We must look to uniquely human information transmission mechanisms, which in turn, allow for culture to evolve alongside our genetics by a process called gene-culture coevolution. The significance of this claim in answering the question of the origins of human altruism is that our inability to appeal to biological adaptation is remedied by appealing to cultural adaptation instead. There is much more to say on the topic of cultural adaptation in support of our conclusions. However, it is necessary to first explain how human culture is a valid target of natural selection.

In The Origin of Species, Darwin outlined three necessary conditions for natural selection to occur. They are variation among the targets of selection, the ability to inherit traits, and competition amongst interactors within the environment (Darwin, 2009. 61). Natural selection at the time of Darwin was said to act at the level of the individual. Variation amongst phenotypes allowed for discrepancies in the level of adaptiveness displayed by individuals in competition, resulting in the proliferation of the traits found in the most competitive individuals. However, as we saw from attempting to solve the problem of altruism amongst non-human species, processes such as kin selection show that the selective pressure does not have to occur at the level of the individual trait alone. For the purposes of explaining the evolutionary origins of human altruism, I am going to apply Darwin’s necessary conditions for natural selection to an entirely different locus: culture.

For the condition of heredity to serve a function, there must be a process by which adaptive information is passed cross-generationally. In the vast majority of instances, this transfer of information is facilitated by the nucleic acids, DNA and RNA. Without going into too much detail, DNA stores the genetic instructions for every living organism. Its nucleotides along with RNA are coded to produce proteins with particular biological functions. Therefore, it is the information within the DNA which determines which phenotypes are produced within an organism (Watson, 1970. Ch.2, § 3 & 4). However, with the dawn of language, human beings evolved a remarkable new adaptation – the ability to transmit environmentally adaptive information outside of a purely genetic system. Language became a catalyst for evolutionary progress as it allowed natural selection to occur at the level of ideas. Early humans with minds that were receptive to new adaptive information in the form of ideas were more likely to survive, and pass on their linguistic ability to their offspring. As such, the ability to use language became a genetically adaptive phenotype, but more importantly, the products of that phenotype, namely the adaptive ideas themselves became sources of competition between early human groups. Before the advent of language, competition was dominated by adaptation at the level of the gene. The human ability to process information cross-generationally allowed for adaptive information to be shared through a new medium; namely, culture. This is what is meant by the term gene-culture coevolution. Before getting on to the validity, and ultimate significance of competition at the level of groups in explaining altruistic behaviour in humans, we must demonstrate how human beings might have originally developed the ability to adapt culturally through their use of language and social learning. It is important to do so since the human mind is still a product of biological evolution. An explanation of how the ability to adapt culturally could be born out of a purely biological process is required in order for any conclusions about the origins of human altruism based on information transfer through language to be satisfactory. We must demonstrate the evolutionary process by which linguistic capabilities could have evolved from communication at the level of chimps, for example, to the higher level capabilities of human beings. To provide such an explanation we must look to anthropology, and the lifecycles of our early hominin ancestors.

5. The Origins of Social Learning

In order to explain the evolutionary progress from rudimentary communication mechanisms of early hominin ancestors to the linguistic mechanisms that allow for cultural adaptation through social learning which we find in human beings today, I will employ a lineage explanation. Borrowing from the likes of Calcott and Sterelny, explanations of this kind seek to understand particular adaptations by making use of incremental analysis, where each incremental stage of phenotypic adaptation contributes to a large leap in the overall adaptive trajectory (Calcott, 2009, 51. Sterelny, 2012, 25-28. Sterelny, 2016, 271-273). The adaptive trajectory in question for our purposes is the evolution of language as a mechanism for cultural adaptation.

An understanding of the way in which human language differs from the communication mechanisms displayed by vervet monkeys, for example, is required if we are to provide a good picture of how our language evolved. Non-human communication mechanisms can be categorised into three distinct kinds (Pinker, 1995, 334). First, a finite repertoire of calls. One distinct call for warning of predators, another for sharing locations of food stores, and so on. The second kind are continuous physical signals or gestures, where variations in intensity of the gesture are indicative of the magnitude of the state in question. This mechanism is exemplified in asian honey bee predator signalling. Asian honeybee drones when discovering a threat from a predator signal to their hive in the form of a brief vibrational pulse, the fundamental frequency of which is indicative of the level of threat from the predator (Tan et al. 2016, 1-4). The third communicative mechanism is random variation on a specific theme. Birds communicate through this mechanism, where small alterations in melody and tempo of a specific birdsong communicate particular states (Thorpe, 1961. Ch.1). By contrast, human language has several distinct features. Most fundamentally, human language allows for virtually infinite reference. The rules of grammar and syntax result in a communicative system where simple referent clauses can be combined to produce new meaning. It is arguable that the communicative behaviour demonstrated by the asian honeybees is infinite also (in so far as the predator signal has a multitude of different fundamental frequencies depending on the level of threat), but there is a difference. The honey bee predator signal operates along a single continuum whereas the infinite possible complex words and sentences found in human language is due to the human ability to rearrange distinct elements in specific orders through the use of syntax. Second, language is compositional. This means that anyone with relevant knowledge of the syntax and grammatical rules can deduce the meaning of complex sentences.

At this point, we have begun to unpick the differences between human language and animal communication mechanisms. It is essential to note at this point that the evolution of language must be intimately linked with the evolution of cooperation, since the ability to communicate is necessary for cooperation. In this sense, language has coevolved with the ability to cooperate, and it is no surprise that the most cooperative species to walk this earth is also the one with the most developed faculties of communication. But how did these faculties develop in the first place?

Our early human ancestors lived in environments that were constantly changing. For instance, the use of tools made out of specific and scarce material required that early hominins be aware of the various resource distributions through space and time. As a result early hominins had to develop methods of passing on information about various environments to their peers. This is not true of chimpanzees, for instance, who gain knowledge of their environments predominantly through genetic inheritance, since their environments remain largely unchanged over many generations. A chimpanzee does not need to learn by cross-generational social learning that their social groups are organised in a linear dominance hierarchy with one dominant male (Goldberg & Wrangham, 1997. 559). The constancy of this social structure in chimpanzee life means that this knowledge is inherent in their standard behaviour, and is embellished in the chimpanzee genome (ibid. 561-567).

To explain how the capacity for social learning could become an adaptive trait, it is necessary to show how social learning itself preceded adaptations for social learning. This is possible due to phenotypic plasticity. This refers to the ability of a particular genotype to produce various phenotypic effects (behaviours/traits etc…) based on environmental change (West-Eberhard, 1989. 249). This concept is important for our purposes as it demonstrates how cross-generational social learning may have been established in organisms who had not yet gained any particular genetic adaptations for social learning. The phenotypic effects of genes are often dependant on the state of the interactor’s environment, and once human environments began changing at swift rates, so too did the effects of our existing genotype, and thus, social learning itself became an adaptive trait favoured by natural selection.

We have postulated that phenotypic plasticity has allowed for social learning to become an adaptive trait amongst early hominins as a result of growing environmental change. Now, we must account for the expanded scope of social learning, and track its evolutionary progress to modern language. It is likely that early hominins received the majority of their socially learned information exclusively from the parent generation. As a part of our lineage explanation for modern human language, we must show how information began to be passed between individuals outside of a parent-offspring relationship.

The key to the expansion of social learning as an adaptive trait is the development and increased application of mental tools. Just as the increased use of physical tools resulted in expressed phenotypes for social learning due to phenotypic plasticity; the use of mental tools (of which language is a developed form) has resulted in a transition from early hominids as species adapted for basic trial and error learning, to species with several adaptations for social learning. The term ‘mental tool’ is borrowed from Daniel Dennett and Andy Clark (Dennett, 2001, 133-143. Clark, 2002). It finds its use within a view of the mind as composed of ‘tools for thinking’ which we obtain from our social environment through social learning. In this sense, the tools themselves stay in the environment, and are propagated not through genetic transmission, but rather, the cultural environment. An example of a basic mental tool would be simple induction. Putting all philosophical worries about the permissibility of induction as a reasoning process aside, an individual who has the ability to recognise that future events often resemble the past would be at an advantage to peers who did not. To understand the progress of the increased scope of social learning within the context of a lineage explanation, we must show how the use of less developed mental tools than language resulted in an increase in relative fitness for the bearer, and thus became a trait selected for by natural selection. In the case of induction, an individual with the capacity to induce future effects from similar past causes would bear a strong selective advantage. If Uga knows the dangers of sitting under coconut trees after seeing Iga be killed by a falling coconut, then she is at a selective advantage when compared to Aga who saw Iga die, but could not infer the cause from the effect.

We can begin to see now how adaptations for basic mental tools quickly compound each other and become favoured by natural selection. Similarly, once communicative tools begin to morph from simple calls or gestures, to basic referent syntax, to fully fledged language with syntax and grammar, the ability for human beings to learn socially is amplified. The progress from species where adaptive information spread mostly through genetic transmission to a species where it is predominantly spread culturally was kickstarted by the use of basic mental tools. These basic mental tools aided the development of human being’s linguistic capacity, and it is this linguistic capacity that has lead to the ability to learn socially and to adapt culturally, rather than genetically. The intention for the remainder of this paper is to show how cultural adaptation has resulted in the human altruistic condition by opening up a new level on which selection operates: groups.

6. Cultural Adaptation and Inter-group Competition

To provide an evolutionary explanation for the altruistic behaviour demonstrated by humans, I apply multi-level selection theory. As mentioned previously, multi-level selection proposes that natural selection operates on several levels depending on the trait one is trying to explain. For our purposes the trait in question is human altruism, and I posit, that it was not selection at the level of the the individual or the gene that drove its proliferation, but rather the high amount of variation between altruistic and non-altruistic groups. My strategy for supporting this claim with relevant evidence is to revisit the original criticisms of group selection (which we glossed over in section 2) and to refute them by application of the principles discussed in the previous section; namely, cultural adaptation and social learning.

Recall the two major criticisms of early group selection. First, altruistic groups are vulnerable to ‘subversion from within’. This is the claim that altruism at a group level could not be an evolutionary stable strategy because altruistic groups are at risk of individual exploitation. Critics attacking group selection from this angle argue that the selective advantage of altruism at the group level could not possibly displace the huge selective advantage at the individual level that a selfish mutant would experience when they receive all of the altruistic benefit without enduring any cost in return (Dennett, 1994, 617-618. Dawkins, 1976, 7-10). The second criticism of group selection arose through the use of mathematical modelling. These models seem to support the view that selection at the level of groups is too weak an evolutionary force to account for the origination of altruistic behaviour, and as such should be dismissed in favour of lower level explanations. We shall address these criticisms in turn, and show how the higher cognitive capacities of human beings which allow for the transfer of adaptive information through culture aid in their refutation.

One does not need to look far to find examples of individual exploitation of cooperative group dynamics in nature. We shall call the practice cheating. Cheating is often found within the mutualistic relationship between cleaner fish and their larger clients which we mentioned earlier. The cleaner fish, under normal circumstances, feed on microscopic ectoparasites. However, it has been shown that the cleaner fish will begin to feed on the client fish’s tissue if the larger fish is unable to manipulate the cleaner fish’s behaviour through aggression (Bshary & Grutter, 2002. 547-555). The cheating fish will accrue a greater level of relative fitness from each interaction when compared to other cleaner fish who do not cheat. As a result, one would expect to see a proliferation in cleaner fish with a tendency to cheat. To apply this idea in the terms of the subversion from within criticism; the level of fitness benefits gained from cooperating at the group level is not significant enough to outweigh the fitness benefits that would be gained from exploiting the cooperative dynamics and feeding of client tissue instead.

Let us consider another example of cheating that occurs within a species specific group. Recall the vervet monkeys discussed in section 3. Vervet monkey groups display cooperative behaviour in the form of alarm calling when spotting a predator to warn other members of the group. The behaviour is considered altruistic in the sense that it reduces the relative fitness of the individual (since giving an alarm call alerts the predator of the caller’s position) while at the same time increasing the relative fitness of all members of the group who hadn’t yet spotted the predator. According to critics of group selection, this behaviour could not have evolved through selective pressures at the level of the group, because cheating would bestow a much higher fitness benefit. Imagine a vervet monkey mutant who lives within a cooperative group containing altruistic alarm callers, but does not herself cooperate. When compared to the altruistic members of the group, she is at a significant fitness advantage by never calling when spotting a predator, because she is still alerted whenever one of her peers give an alarm call, whilst at the same time consistently remaining inconspicuous to any predators. This relative fitness increase would result in a proliferation of the tendency to cheat, and as such one must look elsewhere from group selection to provide a satisfactory explanation for the evolution of cooperative behaviour (Dawkins, 1976. 72-74).

I am willing to grant that group selection does not satisfactorily explain non-human cooperative behaviour. A study of vervet monkey behaviour by Cheney & Seyfarth supports this view. They found that adult female monkeys alarm call at much higher rates when accompanied by offspring (1985. p.160). Furthermore, adult males tend to alarm call more frequently when in the presence of female monkeys (ibid. p.159). These findings demonstrate two things. In the case of adult females, kin dynamics are much more effective in explaining the propensity of a monkey to cooperate. If she has more offspring in her group, then she is more likely to act cooperatively since more copies of her specific genes are a part of her group. In the case of adult males, cooperative behaviour is displayed as a form of fitness signalling to potential mates. In this case, explaining the evolutionary stability of cooperative behaviour is better achieved by employing the handicap principle (the view that non-adaptive behaviours or traits might be selected for due to their indication of fitness in the organism which displays them)1 than group selection. Having said that, I do contend that subversion from within does not threaten the permissibility of multi-level selection theory when it comes to the evolutionary explanation for the super-cooperative behaviour demonstrated by human beings. It is my aim now to convince you of this.

1For more on the handicap principle, see: Zahavi & Zahavi (1999)

It is the case that cheating undermines cooperation. It is also the case that the punishment of defectors serves as a stabilising force for the proliferation of cooperative behaviour (Bowles & Gintis, 2013. 24-26). My argument that human beings are uniquely placed to benefit from group level adaptation rests, in part, on our heightened ability to regulate cheating through punishment. This heightened ability is due to the greater cognitive capacities displayed by humans which we discussed in section 4. Our incremental cognitive adaptations allow for cheating regulation by altruistic punishment due to our capacity for strong reciprocity. Public good and punishment games are useful in supporting these claims. Fehr and Fischbacher examined the social dynamics of a third-person punishment game, and their findings are helpful for my conclusions. The game in question consists of three players; A, B and C. In the game between players A and B, player A is given 100 tokens, and she is instructed to give any amount that she wishes to player B. Player B has no say in the matter. Player C is a third party who stands to gain no benefit from the game between players A and B. However, she is given 50 tokens, and is able to observe the game. Upon witnessing the interaction, Player C is able to punish player A. The punishment comes in the form of a 1 token loss for player C, but this loss translates to a 3 token loss for player A. Fehr and Fischbacher found that punishments were not administered when 50 or more tokens were transferred from player A to B. However, when fewer than 50 tokens were transferred, punishments were administered at a rate inversely proportional to the number of tokens transferred. The transfer of fewer tokens resulted in a greater level of punishment from player C, even though she is in no way affected by the original interaction between players A and B (Fehr & Fischbacher, 2004. 63-87). The findings can be represented graphically as:

third party punishment
Fig.1. Graph showing the pattern of third-party punishment. Graph from: Fehr, Ernst, and Urs Fischbacher. “Third-party punishment and social norms.” Evolution and human behavior 25, no. 2 (2004): Sec. 2.2.

This is important for our purposes as it shows the resilience of human groups when it comes to the threat of subversion from within. Even when people are not themselves the subject of group exploitation, they still exhibit a tendency to punish the non-cooperator. The reason for using a third-party punishment example is that it is more surprising that people will choose to punish (at a cost to themselves) even when they were not personally affected by the selfish behaviour. Having said that, punishment mechanisms are more prominent in situations where the punishers are themselves victims of selfishness2. The findings of these studies suggest that it is not justified to criticise multi-level selectionist principles when applied to explanations for human cooperation by arguing that human groups are susceptible to subversion from within; because human beings are uniquely well adapted to deal with cheating in the form of altruistic punishment. Perhaps the criticism that group level selection is too weak an evolutionary force to offset the effects of individual selection will do a better job. It is my aim to show that it does not.

2For public goods/punishment games of this vein, see: Dawes et al. (1986): 1171-1185. Also, Fehr et al. (2002): 137-140.

In order for altruism to be a trait favoured by selection at the level of the group, altruists within a group must be able to choose to associate with other altruists. As discussed above, this is to ensure that the group is not at risk of subversion from within. When employing mathematical models to understand the evolution of specific traits, it is helpful to assume that the trait arose by random mutation, and as a result, the trait must have existed with very limited frequency when it first appeared. A difficulty arises. Altruists do very poorly when modelled at very low frequencies. This is because they are rarely in contact with other altruists to form cooperative groups with. This is what Wilson and Dugatkin refer to as the problem of origination (Wilson & Dugatkin, 1997. 336-338). The problem is further complicated when we consider the mechanisms by which altruistic tendencies might be recognised by other altruists. Critics of group selection have argued the problem of origination in addition to worries about the ability of altruists to recognise other altruists within large populations requires that we look elsewhere for answers about the origins of altruism. Typically, the alternative explanation is kin selection. This is because kin selection does not suffer from the problem of origination. An altruistic mutant who mates with a non-altruist creates a kin group made up of 50% altruists. Furthermore, kin groups are obviously not vulnerable to the same issues of recognition as larger population groups.

One attempt to deal with the problem of origination is to view individual variation continuously rather than discretely (Sober & Wilson, 1999. 136-142). This is to say that accurate models must account for differing levels of altruism amongst individuals in a population rather that binarily ascribing either an altruistic or non-altruistic tendency. Let us return to the vervet monkeys to exemplify. To combat the problem of origination, group selectionists might advocate a system of alarm call behaviour modelling where individual monkeys within a group vary based on their altruistic tendencies rather than attempting to model the origination of alarm calling behaviour by way of a single altruistic mutant. In this way, natural selection could favour a greater tendency to give alarm calls rather than favouring a specific altruistic trait. Continuous reasoning of this sort is useful when trying to understand particular traits that are controlled by genes at multiple loci (Boyd & Richerson, 1980. 7506-7509). In order to assess the applicability of continuous reasoning to the behaviour of vervet monkeys, one would have to provide proof of the fact that the alarm calling behaviour is a result of multiple genetic inputs, since this is the only way that individual variation occurs continuously. If it does not, then it would certainly be advisable to appeal to kin selection rather than group selection in this case. Reasoning to this end is beyond the scope of this paper. However, we have already done much of the legwork for the human case.

In section 5 we argued that the plasticity of human phenotypes allowed our ancestors to gain adaptations for social learning, and ultimately for full linguistic mechanisms to allow for the super-cooperative behaviour demonstrated by humans. Should variation amongst human altruism be continuous rather than discrete (which is likely true since it is not the case that an individual is either altruistic or not), then it is best to appeal to selection at the level of groups, since it no longer falls foul of the problem of origination. Having said this, there are more reasons to favour group selectionist analysis over kin selection. One example is that human groups are uniquely placed in nature in so far as they do not largely consist of kin. As we saw in the case of adoption, human altruism is not contingent simply on kin groups. In fact, humans often make great sacrifice for members of non-kin groups. The fact that inter-group conflict is so prevalent in human history is a possible indication of this (Jung-Kyoo & Bowles, 2007. 636-640). It would be a fruitless endeavour to attempt to explain a soldier’s willingness to go to war for her country with appeal to genetic kin dynamics. A gene selectionist argument could be made that human altruistic behaviour is simply evolutionary error. In other words, our tendencies to behave in the interest of our social group is a consequence of our ancestors living in isolated kin groups. Altruism is an adaptation, but no longer adaptive. However, I struggle to accept this. Not only can we show that altruism results in significant variation in the fitness of human groups (Bowles, 2006. 1569-1572), but anthropological evidence shows that human inter-group interaction dates back to as early as the late holocene (Jung-Kyoo & Bowles, 2007. 636-640). This suggests that the propensity to cooperate has never been specific to kin. Consequently, a hypothesis which attempts to deal with the origins of human altruism with appeal to evolutionary error, appears to be telling a just so story to aid in the specific application of kin selection mechanisms to the issue.

I admit that the scope of group selection within the context of a multi-level selection theory is limited. In order for it to apply legitimately, three primary conditions must be satisfied. First, there must be significant variation amongst groups as a result of a specific trait. Second, cognitive capacities within the group must be high enough to regulate behaviour within the group. This is to ensure a relatively uniform distribution of the adaptive trait within the population. Third, the adaptive trait in question must be continuous. This has led me to the conclusion that selection at the level of the group is the best method for explaining altruistic behaviour at the level of human beings. The propagation of linguistic mechanisms aided cooperation to such an extent that variation amongst the level of fitness of human groups was hugely significant. The resultant ability for human groups to adapt to their environments especially quickly only served to catalyse this effect. Furthermore, punishment mechanisms based in the heightened cognitive capacities discussed above prevented these altruistic groups from being infiltrated by cheaters. This meant that human groups were especially homogenous with respect to the trait of altruism, thus strengthening the case for the application of group selection in explaining altruistic behaviour. Finally, phenotypic plasticity amongst human beings requires that selective variation amongst altruistic behaviours must be continuous rather than discrete. As a result group selectionist reasoning with respect to the origins of human altruism does not fall foul of the common criticism of the problem of origination. All of this leads me to the conclusion that the common explanatory mechanisms for the evolution of cooperation in nature do not satisfactorily account for cooperation at the level of human beings, and that group selection within the context of a multi-level selection framework is where we must look for its best explanation.

7. Conclusion

This paper has defended a multi-level selectionist account of the evolution of human cooperation, where natural selection favoured the proliferation of altruistic behaviour due to variations in the level of fitness between groups. It began by assessing the applicability of current theories (direct benefits, kin selection and reciprocal altruism) of the evolution of cooperation in nature to the human cooperative condition, and found them to be lacking. In their place, this paper has advocated group selection as the operative mechanism for the best explanation of human altruism. This is due to characteristics of human development that cannot be found elsewhere in nature. Namely, the ability to adapt culturally through social learning, and linguistic systems which developed as a result of phenotypic plasticity and rapidly variable environments throughout our anthropological history.


Axelrod, Robert. “More Effective Choice in the Prisoner’s Dilemma.” The Journal of Conflict Resolution 24, no. 3 (1980).

Bowles, Samuel, and Herbert Gintis. A Cooperative Species. Princeton University Press, (2013).

Bowles, Samuel. “Group competition, reproductive leveling, and the evolution of human altruism.” Science 314, no. 5805 (2006): 1569-1572.

Boyd, Robert, and Peter J. Richerson. “Effect of phenotypic variation on kin selection.” Proceedings of the national academy of sciences 77, no. 12 (1980): 7506-7509.

Bshary, Redouan, and Alexandra S. Grutter. “Asymmetric cheating opportunities and partner control in a cleaner fish mutualism.” Animal Behaviour 63, no. 3 (2002): 547-555.

Calcott, Brett. “Lineage explanations: explaining how biological mechanisms change.” The British Journal for the Philosophy of Science 60, no. 1 (2009).

Cheney, Dorothy L., and Robert M. Seyfarth. “Vervet monkey alarm calls: manipulation through shared information?.” Behaviour 94, no. 1 (1985): 150-166.

Choi, Jung-Kyoo, and Samuel Bowles. “The coevolution of parochial altruism and war.” Science 318, no. 5850 (2007): 636-640.

Clark, Andy. “Minds, brains and tools.” (2002).

Darwin, Charles. “The origin of species by means of natural selection: or, the preservation of favored races in the struggle for life.” (2009).

Dawes, Robyn M., John M. Orbell, Randy T. Simmons, and Alphons JC Van De Kragt. “Organizing groups for collective action.” American Political Science Review 80, no. 04 (1986): 1171-1185.

Dawkins, Richard. The selfish gene. Oxford university press, (2016).

Dennett, Daniel. “Things about things.” The Foundations of Cognitive Science. Oxford: Clarendon (2001): 133-143.

Dennett, Daniel C. “E pluribus unum?.” Behavioral and Brain Sciences 17, no. 04 (1994): 617-618.

Earnshaw, Eugene. “Group selection and contextual analysis.” Synthese 192.1 (2015): 305-316.

Fehr, Ernst, and Simon Gächter. “Altruistic punishment in humans.” Nature 415, no. 6868 (2002): 137-140.

Fehr, Ernst, and Urs Fischbacher. “Third-party punishment and social norms.” Evolution and human behavior 25, no. 2 (2004): 63-87.

Gintis, Herbert. “Strong reciprocity and human sociality.” Journal of theoretical biology 206, no. 2 (2000): 169-179.

Goldberg, Tony L., and Richard W. Wrangham. “Genetic correlates of social behaviour in wild chimpanzees: evidence from mitochondrial DNA.” Animal Behaviour 54, no. 3 (1997).

Hamilton, William D. “The genetical evolution of social behaviour. II.” Journal of theoretical biology 7, no. 1 (1964).

Jeler, Ciprian. “Multi-level selection and the issue of environmental homogeneity.” Biology & Philosophy 32.5 (2017): 651-681.

Lehmann, Laurent, and Laurent Keller. “The evolution of cooperation and altruism–a general framework and a classification of models.” Journal of evolutionary biology 19, no. 5 (2006).

Pinker, Steven. The language instinct: The new science of language and mind. Vol. 7529. Penguin UK, (1995).

Smith, J. Maynard. “Group selection and kin selection.” Nature 201 (1964).

Sober, Elliott, and David Sloan Wilson. Unto others: The evolution and psychology of unselfish behavior. No. 218. Harvard University Press, (1999).

Sterelny, Kim. The evolved apprentice. MIT press, (2012).

Sterelny, Kim. “Deacon’s challenge: From calls to words.” Topoi 35, no. 1 (2016).

Stutterheim, C. J. “Symbiont selection of redbilled oxpecker in the Hluhluwe-Umfolozi game reserve complex.” Lammergeyer 30 (1980).

Tan, Ken, Shihao Dong, Xinyu Li, Xiwen Liu, Chao Wang, Jianjun Li, and James C. Nieh. “Honey bee inhibitory signaling is tuned to threat severity and can act as a colony alarm signal.” PLoS Biol 14, no. 3 (2016).

Thorpe, William Homan. “Bird-song: the biology of vocal communication and expression in birds.” (1961).

Trivers, Robert L. “The evolution of reciprocal altruism.” The Quarterly review of biology 46, no. 1 (1971).

Watson, James D. “Molecular biology of the gene.” Molecular biology of the gene. 2nd edn (1970).

West-Eberhard, Mary Jane. “Phenotypic plasticity and the origins of diversity.” Annual review of Ecology and Systematics 20, no. 1 (1989).

Williams, George Christopher. Adaptation and natural selection: A critique of some current evolutionary thought. Princeton university press, (2008).

Wilson, David Sloan, and Lee A. Dugatkin. “Group selection and assortative interactions.” The American Naturalist 149, no. 2 (1997): 336-351.

Zahavi, Amotz, and Avishag Zahavi. The handicap principle: A missing piece of Darwin’s puzzle. Oxford University Press, 1999.

One thought on ““Explaining Altruism: A New Defence of Group Selection” By Keyana C. Sapp

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.