Female vervet monkeys fine-tune decisions on tolerance versus conflict in a communication network

Abstract

Group living promotes opportunities for both cooperation and competition. Selection on the ability to cope with such opposing social opportunities has been proposed as a driving force in the evolution of large brains in primates and other social species. However, we still know little about the degree of complexity involved in such social strategies. Here, we report advanced social strategies in wild vervet monkeys. Building on recent experimental evidence that subordinate females trade grooming for tolerance from higher-ranking individuals during foraging activities, we show that the audience composition strongly affects this trade. First, tolerance was lower if the audience contained individuals that outranked the subordinate partner, independently of audience size and kinship relationships. Second, we found a significant interaction between previous grooming and relative rank of bystanders: dominant subjects valued recent grooming by subordinates while intermediate ranked subjects valued the option to aggress subordinate partners in the presence of a dominant audience. Aggressors were also more likely to emit coalition recruitment calls if the audience contained individuals that outranked the subordinate partner. In conclusion, vervet monkeys include both recent grooming and knowledge about third-party relationships to make complex decisions when trading grooming for tolerance, leading to a finely balanced trade-off between reciprocation and opportunities to reinforce rank relationships.

1. Introduction

In an evolutionary context, the social complexity of group living has been considered one of the main factors selecting for cognitive capacities and correspondingly increased brain size [1–3]. Social complexity arises in stable groups due to a high frequency of encounters between members, where the same pairs can be partners, neutral interactors, or competitors, and the outcome of each encounter depends on the independent decisions of both. In order to successfully manage their relationships and social challenges under such circumstances, individuals have to keep track of their own relationships with other group members. An additional level of complexity arises when interactions take place within a communication network, where bystanders can extract valuable information about an observed interaction for future use or decide to intervene according to their own interests. Communication networks are indeed a widespread phenomenon described in all major vertebrate clades and even in invertebrates [4]. As a consequence of being observed, interaction partners should adjust decisions to the presence of bystanders, leading to so-called ‘audience effects’ (see [5]). There is widespread evidence for audience effects, in competitive as well as cooperative contexts (for competitive: [6–8]; for cooperative: [9,10]). To date, many studies on various taxa have focused on the modulation of the communication system in response to audience effects (chickens, Gallus gallus domesticus: [11]; house sparrow, Passer domesticus: [12]; marmots, Marmota flaviventris: [13]; and various primates species: chimpanzees, Pan troglodytes: [14–16]; brown capuchins, Cebus paella: [17]; vervet monkeys: [18]) but only a few have studied a modification of behaviour (siamese fish, Beta splendens: [19]; cichlids, Astatotilapia burtoni: [20] and primates: rhesus macaques, Macaca mulatta: [21,22]; orangutans, Pongo pygmaeus: [23]). However, experimental studies manipulating the audience composition to test such effects remain scarce [10,19–21,24].

While audience effects are widespread, it is important to realize that the complexity of underlying decision rules may vary between species and situation, and that complexity may be particularly high in species that live in stable groups for at least three reasons. First, in groups with stable group members the same pairs may alternate between being competitors and cooperators. Second, as stable groups are characterized by a social relationship structure, decisions about helping/support may often be linked to either genetic relatedness [25,26] and/or relationship quality [27]. Therefore, it may become important to adjust one's response to an audience to the specific audience composition in relation to the self or to the interaction partner. Finally, stable groups are often characterized by a stable hierarchy, meaning decisions relating to an audience should incorporate both interactors' relative rank relationships with bystanders. The basis for the latter is known as third-party relationship knowledge, and has been demonstrated in a variety of species [28–32]. If it could be shown that individuals in stable groups integrate all the aforementioned information in response to an audience in a self-serving way [7,33] such results would provide important support for the strategic aspects of the Machiavellian intelligence hypothesis [34,35]. This hypothesis builds on previous suggestions that cognitive processes mainly evolved to allow individuals to cope with the challenges of the rather unpredictable social environment [1,36]. While it covers social behaviour and cognition very broadly, for example by also including the scope for social learning and traditions, our study focuses on the earlier more specific use of the term ‘Machiavellian’ as aiming for power [37].

We studied wild adult vervet monkey females' decisions to tolerate or lower ranking group members in an experimental foraging situation. We employed wooden boxes provisioned with food to allow for controlled foraging. Before the experiments, each female had been trained to approach her personal wooden box, recognizable by a unique colour pattern, in order to obtain high-quality food inside. Boxes were opened by remote control only when the owner touched the top. In the experiments, two boxes were placed in close proximity to each other. As vervet monkeys have a strict linear hierarchy and limited tolerance around food sources [18], the set up allowed us to test factors that might impact high-ranking individuals’ decisions to aggress their subordinate box partner (i.e. box partner). During conflicts, subjects often produce vocalizations associated with other aggressive behaviours such as stares, attacks, or chases [18]. It has been suggested that these calls are emitted to solicit potential support from nearby individuals [18,38]. In a previous study, we showed that recent grooming caused reciprocation via a selective reduction of aggression towards the grooming partner [39]. However, in that study we did not analyse audience effects. Therefore, the aim of the current paper was to investigate the impact of audience composition on subject's decisions to aggress or tolerate another individual.

Various studies have shown that in vervet monkeys third-party interventions in a conflict typically involve high-ranking individuals and only occur to support the aggressor against the victim [18,39]. As a consequence, high-ranking vervet monkey females (1) do not have to worry about the audience joining a conflict to oppose them, and (2) are more likely to receive coalitionary support in an aggression if the audience contains individuals that outrank the subordinate box partner. Coalitionary support increases the benefits of aggression in two ways. First, it increases the chance of victory while decreasing the risk of injury [40]. Second, it may help strengthen the bond with the coalition partner and thereby reinforce the existing hierarchy [18]. Thus, if vervet monkeys have some awareness of these benefits we predicted that subjects should be more likely to aggress against the box partners in the presence of suitable coalition partners. Suitable coalition partners may be identified via more or less sophisticated knowledge. Simple decision rules would include an increased probability of aggression if the audience contains individuals that outrank self or closely bonded individuals/kin. A more sophisticated decision rule would take advantage of third-party relationship knowledge and hence increase the probability of aggression if the audience contains any individuals that outrank the subordinate box partner. As female vervet monkeys are known to possess such triadic awareness [18,32], we predicted that the presence of dominant or intermediate ranked individuals (i.e. ranked between the subject and the box partner; called intermediate audience hereafter in contrast to dominant and subordinate audience) would lead to a higher probability of aggression. In the case of an intermediate audience, we predicted that such increase of aggression would occur independently of the kin relationship between the audience and the subject. Furthermore, we investigated how much the effects of recent grooming on reciprocation interacted with audience composition. If recent grooming promotes tolerance and a dominant audience promotes aggression, do they counteract each other or is one factor more important? Furthermore, does the way these two factors are integrated depend on the relative rank constellation? For example, one could predict that dominant subjects give more weight to recent grooming while subjects of intermediate rank give more weight to the opportunity to acquire a more dominant coalition partner. As such, we predicted an interaction between relative rank and recent grooming events as an explanatory variable for the probability of aggression during the experiments. Finally, we examined whether low tolerance between subjects and box partners in the presence of a dominant audience might actually be due to the latter enforcing such behaviour rather than due to strategic interests of the subject. Such a scenario might arise if higher-ranking individuals benefit from interrupting interactions that would strengthen bonds between lower ranking individuals. If this was the case, we predicted that tolerance at the boxes would regularly lead to aggression by dominant bystanders towards the tolerant individual.

2. Material and methods

Experiments were run from December 2012 until December 2014 at the Inkawu Vervet Project, Mawana Game Reserve, KwaZulu Natal, South Africa. Subjects were 17 adult females from three habituated groups of vervet monkeys. In 2014, without infants, the first group, Ankhase, included 27 individuals (4 adult males, 10 adult females, 13 sub adults and juveniles). The second group, Baie Dankie, included 38 individuals (4 adult males, 11 adult females, 23 sub adults and juveniles) and the third group, Noha, included 35 individuals (4 adult males, 11 adult females, 20 sub adults and juveniles).

(a). Experimental procedure

Research at the study site is based on experiments involving high-quality food presented in artificial containers [41,42]. For the current study, we adapted a method developed by Fruteau et al. [43], training females to approach and get access to a reward in a wooden box that could be opened from a distance with a remote control by the experimenter (see [39] for more details about methodology). Each box was covered with a colour and shape-coded cover so that a monkey could learn that she could access only one box (her personal box chosen by the experimenter). We trained the maximum number of females possible, including the alpha female in each group as well as some intermediate ranking and subordinate females to get a representative sampling. In the Ankhase group, 6 females were trained, in the Baie Dankie group, 5 females and in the Noha group, 7 females. Hierarchy was assessed by analysing matrices based on dyadic aggressive interactions (i.e. winner–loser) which occurred either in a natural context or around various food experiments. Rank relationships of females and juveniles were assessed through the ‘de Vries’ [44] methodology. This allowed us to classify individuals according to dominance. The audience could include any individual older than one year as juveniles are part of the hierarchy and usually inherit their mother's rank. From a young age, juveniles can aggress subordinates and intervene in conflicts [18]. Adult males were not considered within our analyses as they rarely intervene in a female–female conflict [18]. In this study, we never had any adult male intervention. We, therefore, did not take into consideration data points where only adults males were present as an audience.

The experiments involved a 2 × 3 design, in which we controlled for two grooming situations and three audience situations. Experimental trials involved placing two personal boxes next to each other in order to induce an interaction through forced proximity. A total of 40 different female pairs were tested. The maximum number of female pairs was tested in each of the two different grooming situations: after the box partners had not groomed each other (no grooming) and after they had (post-grooming). In order to increase the power of our analyses, as we had many categories of grooming and audience situations, we decided to create our ‘No grooming’ control situation by integrating the ‘Dominant groomed’ (i.e. dominant subject was groomed by a random individual) situation with the ‘No grooming at all’ (i.e. dominant did not receive any grooming within the last 60 min) situation as these categories were not different from each other [39]. After a grooming session, a time window of maximum 60 min was allowed in which to run a trial. The outcome of each interaction was recorded as conflict or tolerance. If the individuals remained around the experiment location, the boxes were reloaded a maximum of two times to assess a potential change in behaviour. Multiple experiments involving the same subject could be run per day if an individual got involved in a grooming bout with other individuals or had not been involved in a grooming bout for at least 60 min. For each trial, we recorded who was within 10 m, the outcome of the interaction and in case of conflict, the exact behaviour of each box partner.

To test if there was an effect of audience on aggressiveness, we compared the ratio of conflict between a pair in the two grooming situations in the three audience conditions: dominant (i.e. individual in audience outranks both box partners), intermediate (i.e. individual in audience outranks the subordinate box partner), and subordinate audience (i.e. individual in audience is subordinate to both box partners). Given that considering multiple individuals and ranks within a single model was not possible, we determined the audience condition in one analysis as a function of the highest ranked individual within the audience, and in a second analysis, we only considered trials in which the audience consisted of a single individual. Furthermore, to increase the dataset and prevent a lack of variance, we also decided to integrate the subordinate audience with the no audience at all as these conditions subsequently proved to not be significantly different from each other (proportion of conflicts within each pair with subordinate and no audience; post hoc from generalized linear mixed model (GLMM): Z = 0.41; p = 0.97). To test if the audience had an effect on the probability of conflicts, we analysed all results from all trials. In case of conflict, in order to test if the audience had an effect on the production of aggressive calls (i.e. threat vocalizations) we compared the number of vocal and non-vocal conflicts between a pair in the three audience and two grooming situations. The maximum number of pairs was tested in each situation although it was not possible to get a complete dataset by testing all possible pairs. This was because, for example, some females were never seen grooming each other or it was not possible to have the pair and an appropriate audience together within the right time window of 60 min after grooming. To test for a potential kinship effect, we assessed genetic relatedness between the audience and the subject. Finally, to test the effect of kinship in the presence of an intermediate audience we categorized the audience into two categories: more closely related to the subject or more closely related to the subordinate box partner. To do this we compared the absolute difference between the intermediate ranked audience individual and subject and between the audience individual and the box partner. The largest value was used to classify them according to one of the two categories mentioned above.

(b). Analyses of genetic relatedness

DNA was extracted from tissue, hair, and faecal samples (see electronic supplementary material for methods of extraction).

To estimate genetic relatedness, we used 16 autosomal microsatellite markers in four multiplex reactions. Thirteen of these microsatellites were previously published human-derived microsatellite markers [44], while three were derived from vervet monkeys (N Arora 2013, personal communication). Marker details and PCR conditions are given in the electronic supplementary material.

The choice of the most accurate estimator depends on inherent population structure and history [45,46]. We used the program Coancestry 1. 0.1.5 [47] to calculate dyadic relatedness for seven different estimators and we found the best performing estimator for this specific population to be the triadic likelihood (TrioML) estimator [48]. For each of the three groups, we constructed a symmetric pairwise relatedness matrix using the TrioML estimator.

(c). Statistical analyses

Statistics were performed in R (v. 3.0.1). First, each model was compared to a null model consisting of the random factors, confirming the robustness of all our models and for each test, we ran a GLMM with an analysis of deviance (Type II) using the function ‘glmer’ of the R package ‘car’ and ‘lme4’. For the conflict analyses, we coded the outcome of each interaction for each pair as tolerance or conflict (dependent variable and binomial data) over all trials. The fixed effects in the model were: type of ‘audience’ (dominant, intermediate, subordinate), ‘situation’ (post-grooming, no grooming), and ‘group identity’. We also tested the interaction between the audience and the grooming situation. To account for pseudoreplication, as individuals were tested within different pairs multiple times across situations (sessions) and sometimes multiple times a day, we introduced the random factors for individuals inside each pair and pairs inside each session as well as the number of interactions the dominant individual (i.e. the one making the decision) experienced per day. To analyse the impact of audience effect within the two grooming situations, we ran a paired comparison with the package ‘lsmeans’ and across the grooming situations we ran a post hoc test of Linear Hypothesis with the package ‘multcomp’.

We could not control for the audience size within the model as both rank and audience size were highly collinear. Therefore, to distinguish between the effect of audience rank and audience size on the likelihood of aggression we ran extra analyses on a subset of data with only one individual as the audience. To keep a large enough sample size we did not test the effect of grooming as we had few data points when partners groomed each other (post-grooming situation). However, this does not seem to bias our results as even within the ‘post-grooming’ situation we found that focal individuals were less likely to be aggressive with their box partner in the presence of a subordinate audience (dominant audience: 5 conflicts versus 7 tolerance; intermediate audience: 5 conflicts versus 5 tolerance; subordinate audience: 1 conflict versus 13 tolerance). Therefore, we only considered for each pair the outcome of each interaction as tolerance or conflict (dependent variable and binomial data) and the rank of the audience as a fixed effect. As in the previous model, we also introduced random factors for individuals inside each pair and pairs inside each session. To analyse the differences between the three types of audience, we ran a post hoc test of Linear Hypothesis to compare each audience with each other using the function ‘ghlt’ of the package ‘multcomp’.

As both audience rank and kinship were confounded factors we ran a separate GLMM to test the effect of the presence of kin within the audience with ‘kinship’ (degree of relatedness) as a fixed effect and the same random factors as above. To distinguish between a third-party rank relationship knowledge and kinship effects within the middle-ranking audience, we ran a GLMM with the outcome of each interaction as tolerance or conflict (dependent variable and binomial data) over all trials. We introduced the random factors for individuals inside each pair and pairs inside each session. Finally, for the vocalizations (as the dependent variable classified as binomial data-vocalizations/no vocalizations), we ran a GLMM with the type of ‘audience’ (dominant, intermediate, subordinate), grooming ‘situation’ (post-grooming, no grooming), and ‘group identity’ as fixed effects. We also introduced random factors for individuals inside each pair and pairs inside each session. To analyse the differences between the three types of audience, we ran a post hoc test of Linear Hypothesis using the function ‘ghlt’ of the package ‘multcomp’.

3. Results

Seventeen adult females participated in the experiments as subjects and 66 different individuals were part of the audience. For our analyses, we tested the rank of 27 different individuals which were the highest ranking individuals within the audience. Out of 500 trials, aggression occurred 314 times. Audience size ranged from 0 to 10 individuals but consisted mostly of 0 to 3 individuals with an average size of 1.6. There could be both adults and juveniles within the audience but we had 205 data points where at least one adult female was present.

(a). Overall audience effect on aggressiveness

We found that subjects were on average 1.5 times more aggressive in the presence of a dominant audience across all grooming situations (GLMM: χ² = 24.71; d.f. = 2; p < 0.001; figure 1). This effect was similar in all three groups (group identity: χ² = 0.53; d.f. = 2; p = 0.77). However, there was an interaction between audience and the grooming situations (χ² = 9.82; d.f.= 2; p = 0.007).

Figure 1.

Proportion of conflicts within all trials with a dominant, intermediate, and subordinate audience in the two situations: partners did not groom (no grooming), partners groomed each other (post-grooming). The average proportion is based on the proportion of conflict for each pair. Bars represent +/− the standard error.

(b). Audience and grooming effect on aggressiveness

When running a post hoc analysis to compare the audience effect within the two grooming situations, we found that subjects were 4.5 times more aggressive within the post-grooming situation (Z = 4.72; p < 0.001) and about 1.5 times more aggressive within the no grooming situation (Z = 4.36; p < 0.001) in the presence of a dominant audience than in the presence of a subordinate one. When the box partners had not groomed each other, there was no difference in the proportion of conflicts in the presence of a dominant audience in comparison to an intermediate audience (Z = 0.23; p = 0.97), while subjects were about 1.5 times more likely to be aggressive in the presence of an intermediate compared to a subordinate audience (Z = 3.51; p = 0.001). However, when box partners had groomed each other, subjects were about 1.5 times more aggressive in the presence of a dominant audience than in the presence of an intermediate audience (Z = 3.44; p = 0.002). Furthermore, there was no difference in the proportion of conflicts between an intermediate and a subordinate audience (Z = 1.93; p = 0.13). As visualized in figure 1, the presence of a dominant bystander seems to largely mitigate the positive effect of grooming on tolerance. With a supplementary post hoc test, we found that across grooming situations the proportion of conflicts was about two times higher in the presence of an intermediate audience than a subordinate one (Z = 3.37; p = 0.003).

(c). Audience effect of a single individual on aggressiveness

Because rank effect and audience were collinear, we could not analyse them within the same model. When analysing a subset of data with only one individual in the audience, we found that subjects were on average two times more aggressive in the presence of a dominant audience (Z = −6.22; p < 0.001) and an intermediate one (Z = −6.21; p < 0.001) than in the presence of a subordinate audience (figure 2). There was, however, no difference in the proportion of conflict between a dominant and an intermediate audience (Z = 0.39; p = 0.92).

Figure 2.

Proportion of conflicts within all trials with a single individual as an audience and according to its rank (dominant, intermediate, and subordinate). The average proportion is based on the proportion of conflict for each pair. Bars represent +/− the standard error.

(d). Effect of kinship

We found that individuals were more aggressive with close kin in the audience (GLMM: χ² = 5.83; d.f. = 1; p = 0.016). This effect was similar in all three groups (group identity: χ² = 3.92; d.f. = 2; p = 0.14). There was as expected an effect of the grooming situation with subjects being more tolerant towards a subordinate that groomed them (χ² = 18.71; d.f.= 1; p < 0.001) but there was no interaction between the grooming situation and kinship (χ² = 1.71; d.f. = 1; p = 0.19; figure 3).

Figure 3.

Proportion of conflicts according to the relatedness between the subject and the highest ranking individual within the audience. Lines represent the trend line and the confidence intervals.

(e). Distinction between kinship effect and the use of third-party relationship knowledge

Additional analyses about the presence of an intermediate ranked audience indicated that the probability of conflict does not depend on whether the potential supporter within the audience is more closely related to the subject or to the box partner (χ² = 0.39; d.f. = 1; p = 0.53)

(f). Audience effect on emission of vocalizations during a conflict

In the presence of an audience that outranked the box partner conflicts were on average two times more vocal than in the presence of a subordinate audience (χ² = 11.77; d.f.= 2; p = 0.003; figure 4). Using a post hoc test, we found that there was no difference between the presence of a dominant and intermediate ranked audience (Z = 1.50; p = 0.29), while subjects were 2.3 times more vocal in the presence of a dominant audience in comparison to a subordinate audience (Z = 3.41; p = 0.002). There was only a tendency for subjects to be more vocal (1.8 times) in the presence of an intermediate in comparison to a subordinate audience (Z = 2.31; p = 0.055). Furthermore, there was no effect of grooming situation (χ² = 0.72; d.f. = 1; p = 0.39) nor group identity (χ² = 3.48; d.f. = 2; p = 0.18).

Figure 4.

Proportion of conflicts where vocalizations were produced within all trials with a dominant, intermediate, and subordinate audience in the two situations: partners did not groom (no grooming), partners groomed each other (post-grooming). The average proportion is based on the proportion of conflict for each pair. Bars represent +/− the standard error.

(g). Punishment from the audience?

Out of all the tolerance data points with a dominant audience (n = 30), we had only one clear instance of aggression from the audience towards a tolerant individual at the boxes.

4. Discussion

The aim of this study was to examine the extent to which female vervet monkeys are influenced by recent personal grooming interactions and the composition of the current audience in their decisions to tolerate or aggress a partner during foraging. Our results provide unprecedented evidence suggesting that these monkeys can make sophisticated decisions about aggression by incorporating personal and third-party relationships with recent grooming. As such, vervet females are able to make strategic choices between tolerance as a means of reciprocation and aggression to induce coalitionary support to reinforce rank relationships. In addition, it seems that such decision rules are not simply based on an individual's kin relationships with the audience. Previous analyses have shown that female vervet monkeys are more tolerant towards bonded partners than towards non-bonded partners, and more tolerant towards partners who had groomed them recently [39]. Our current results showing that the effect of recent grooming on tolerance is strongly modulated by the composition of the audience in a communication network [4] lead us to various insights regarding individual decision rules and underlying cognitive abilities.

An obvious conclusion is that vervet monkey females are aware of the general rule that bystanders that join a conflict support the higher ranking and use it to their advantage. This appears to be a rather opportunistic strategy as such audience members may join the conflict, invariably against lower ranking individuals [18,39], similarly to many other species ([49,50]; but see [38]). Such behaviour should increase the probability of winning and minimize the risk of injury [40]. More importantly, the data show that females make use of their third-party rank relationship knowledge [32] during conflicts. While many studies provide evidence for some knowledge about third-party relationships [28–32], only a few studies focused on how such knowledge may be used for personal benefits. So-called strategic triadic awareness has been demonstrated in chimpanzees where victims exaggerate their screams in the presence of an audience that outranks or equals the rank of the aggressors [7]. Similarly, male bonnet macaques appear to recruit coalition partners in a way that is best explained with triadic awareness [33]. By contrast, a study on capuchins [51] looking at their recruitment strategies did not exclude simpler explanations such as ‘solicit an ally that outranks yourself’ or ‘solicit the highest ranking available individual’. Here, we could rule out such simpler explanations because subjects react in a similar way to the presence of a dominant audience and to an intermediate audience that only outranks the box partner. Importantly, this conclusion persists when the potentially confounding effect of genetic relatedness (see [52,53]) between subject and audience is taken into consideration, excluding the possibility that the full dataset can be explained by a decision rule such as ‘increase the probability of being aggressive towards a partner if in the presence of your kin’. The similar results found on the subset of data where only one individual formed the audience show that the results cannot be explained by individuals only taking audience size into account. This latter hypothesis would have yielded a simpler explanation as the probability of finding an appropriate supporter would indeed be higher with a large audience. Furthermore, in the presence of an appropriate audience, there was a tendency for aggression to be accompanied by aggressive calls that serve as recruitment (see also [18]), providing additional evidence that subjects incorporate audience composition in their decisions. Overall, our results are thus quite similar to the chimpanzee and macaque studies.

The interaction we found between audience and grooming effects reflects the outcome of the most basic social challenge that individuals face during social interactions, i.e. to cooperate or to compete. While they still consider audience effects, dominant subjects seem to give more importance to a recent grooming interaction. On the other hand, intermediate ranked subjects seem to give more weight to the opportunity of strengthening their hierarchy rank by aggressing their box partner in the presence of an audience that outranks this partner. Such a difference seems to highlight the complexity of living in a social group with a stable hierarchy. Finally, we did not find any evidence that a dominant audience compels subjects to be aggressive towards partners through punishment. Instead, receiving coalitionary support from an audience seems to be the key temptation to become less tolerant. The degree to which the benefits of being more aggressive are simply due to easy victories or in addition due to reputational gains (sensu [54,55]) needs further evaluation. In this context, fighting Siamese male fish have been shown to be more aggressive in the presence of a male audience [56], especially when there are nests within their territory. Future studies in a natural feeding context should also evaluate whether provisioning has influenced the results of the current study.

In conclusion, we provide evidence for complex abilities in vervet monkeys to flexibly adjust social behaviour to the specifics of a social interaction, as postulated already by early proponents of the social intelligence hypothesis [1,36]. Optimal solutions often depend on both an individual's and the partner's choice of action, along with a variety of variables like recent interactions with the partner, current needs, cues given by the partner, rank relationship, potential stakes, etc. This introduces high degrees of freedom and hence uncertainty. Uncertainty increases even further when interactions take place within a communication network, where bystanders eavesdrop in order to obtain valuable information [4] and may interfere as a function of their own current state, recent social experiences, or relationships with the interaction partners. The vervet females' ability to consider multiple variables for their behavioural choices between reciprocation and rank reinforcement shows a high degree of social competence [57,58]. The fine-tuning of the choice to tolerate or aggress a subordinate in a foraging context corresponds to the classic definition of Machiavellian intelligence sensu de Waal [37] with its emphasis on the struggle for power. The major challenge for future research will be to find ways to investigate the correlation between such strategic Machiavellian decision-making and the evolution of brain structures, as postulated by the social brain hypothesis [3].

Ethics

The study was approved by the relevant local authority, Ezemvelo KZN Wildlife and by the University of Cape Town, South Africa. The subjects were three groups of wild vervet monkeys habituated to human presence.

Data accessibility

Dataset has been deposited on Figshare and will be available upon submission under https://figshare.com/s/dc16dfe6cffbba74684c (doi:10.6084/m9.figshare.4569793).

Authors' contributions

C.B. collected the data, ran the statistical analyses, and wrote the first draft. R.B. designed the experiment, wrote the second draft, and was granted the funding for data collection. A.S. ran the genetic analyses and commented the manuscript. M.K. supervised and financed the genetic analyses and commented on the manuscript.

Competing interests

We have no competing interests.

Funding

This study was funded by the Swiss National Science Foundation (Sinergia: CRSI33_133040)