PENULTIMATE DRAFT, FINAL VERSION FORTHCOMING IN THE EVOLUTION OF COOPERATION VOL.2: SIGNALLING, COMMITMENT & EMOTION (MIT PRESS) False advertising in biological markets: Partner choice & the problem of reliability Ben Fraser 9/25/2011 This is the penultimate draft of my chapter in a forthcoming book on the evolution of cooperation, edited by Kim Sterelny, Richard Joyce, Brett Calcott and myself. The partner choice approach to understanding the evolution of cooperation builds on approaches that focus on partner control by considering processes that occur prior to pair or group formation. Proponents of the partner choice approach rightly note that competition to be chosen as a partner can help solve the puzzle of cooperation (Noe 2006; Miller 2007; Nesse 2007). I aim to build on the partner choice approach by considering the role of signalling in partner choice. Partnership formation often requires reliable information. Signalling is thus important in the context of partner choice. However, the issue of signal reliability has been understudied in the partner choice literature. The issue deserves attention because – despite what proponents of the partner choice approach sometimes claim – that approach does face a cheater problem, which we might call the problem of false advertising in biological markets. Both theoretical and empirical work is needed to address this problem. I will draw on signalling theory (Maynard-Smith & Harper 2005; Searcy & Nowicki 2005) to provide a theoretical framework within which to organise the scattered discussions of the false advertising problem extant in the partner choice literature. I will end by discussing some empirical work on cooperation, partner choice, and punishment among humans (Barclay 2006; Nelissen 2008; Horita 2010). The Problem of Cooperation Numerous definitions of ‘cooperation’ have been offered in the biological literature. Some researchers use ‘cooperation’ very generally to cover all acts by one individual that benefit one or more other individuals (Sachs et al. 2004, 137). Others consider such usage too liberal, since it counts as cooperative behaviours that benefit others only incidentally, and restrict ‘cooperation’ to behaviours that have been selected because they benefit others (West et al. 2007, 416). Thus, while some researchers would count an elephant that defecates and thereby feeds a dung beetle as cooperating with the beetle, others would not. Minimally, though, the various definitions on offer agree that cooperation involves one organism A benefiting another organism B, where that means A’s behaviour increases B’s fitness. There is also general agreement about when such behaviour is, from an evolutionary perspective, prima facie puzzling: it is when the increase in B’s fitness comes at an apparent cost to the fitness of A. This is the problem of cooperation upon which I wish to focus. One way to solve the puzzle of cooperation is to show how, appearances aside, A is not actually sacrificing its own fitness to benefit B.1 Hamilton (1964) put together a good deal of the puzzle when he conceived of the pieces not as individual organisms but as far smaller units: genes. Hamilton’s theory of kin selection showed how A helping B to survive and/or reproduce could increase A's “inclusive” fitness (even if doing so decreased A's “personal” fitness) so long as A and B were sufficiently close relatives. Even after Hamilton’s elegant insight, however, much of the puzzle of cooperation remained fragmentary. Partner Control Models of Cooperation Cooperation among unrelated individuals has predominantly been viewed through the lens of reciprocity (Trivers 1971). The key question here is how an organism A should behave toward a given partner B over a series of encounters in order to maximise its total pay-off over that series. Should A cooperate with B, or refrain from cooperating, or sometimes do the one and sometimes the other? The important insight grounding the reciprocity-based approach is that, under certain conditions, the immediate cost A pays to benefit B can be recouped (and more) over time if B repays A’s help (whether the repayment is in kind, or in a different currency). For example, the Tit-for-Tat strategy in Axelrod's (1984) iterated Prisoner’s Dilemma tournament enjoyed success because it conditionalised its own cooperative behaviour on the cooperativeness of the strategies with which it found itself paired. Studying cooperation from the perspective of reciprocity focuses attention on what has been called “partner fidelity” (Bull & Rice 1991), “partner verification” (Noe & Hammerstein 1994), and “partner control” (Noe 2001). I will use the term ‘partner control approach’ here to mean approaches to understanding the evolution of cooperation that focus on how an individual should manage its interactions with a given partner, and in particular on how individuals engineer the incentive structure of a given interaction partner for their own benefit. 1 I set aside here group selectionist explanations (see Sober & Wilson 1998), which show how, under certain conditions, genuinely fitness-sacrificing behaviour can be maintained in populations under selection. One reason for dissatisfaction with the partner control approach is that it is surprisingly under-supported by empirical data. Uncontroversial examples of reciprocity-based cooperation among non-human organisms have proven elusive.2 A more important reason – and the focus here – is that the partner control approach neglects important aspects of cooperation. It models only one aspect of a complex process. As noted above, the partner control approach focuses on how an individual manages its interactions with a given partner. However, interactions between organisms can be imagined as consisting of three stages: a pair/group formation stage, a decision stage (e.g. cooperate or not?) and, finally, a division stage, in which the yield of the interaction is apportioned among those involved (Noe 1990, 79; Dugatkin 1995, 4).3 The partner control approach considers only the second of these stages. Importantly for current purposes, that approach typically assumes that individual agents have no control over partnership or group formation, whereas in nature it is likely to often be the case that organisms exert some degree of partner choice (Dugatkin 1995). The partner control approach is thus limited in scope and makes unrealistic assumptions. 2 One reason that has been suggested for this is that the cognitive demands of managing reciprocal interactions are rarely met (Stevens et al. 2004; Stevens & Hauser 2005). But, the cognitive demands of reciprocity need not be high – even if most reciprocating organisms are in fact quite cognitively sophisticated – since even plants and paramecia can trade costs and benefits in a way that counts as reciprocity as defined above. Another reason for the dearth of clear empirical examples of reciprocity is that excluding alternative hypotheses is difficult. The classic case of (supposed) reciprocity-based cooperation is blood-sharing in vampire bats (Wilkinson 1984). Subsequent work has challenged this as an example of reciprocity, since the data fail to rule out the possibility that blood-sharing results from kin selection plus the occasional misidentification of kin, for instance, or is a form of tolerated theft (Clutton-Brock 2009). Even demonstrating Tit-for-Tat reciprocity in action in the laboratory is problematic, given the difficulty of designing experiments that faithfully replicate an iterated Prisoner’s Dilemma (Noe 2006, 11). 3 There is yet another stage that must also be considered, namely, the generation of benefit stage, at which the cooperating individuals actually produced whatever good is then distributed. Both the partner control and the partner choice model have tended to neglect this stage, and I must do so here simply for reasons of space For detailed discussion, see Calcott (2008; this volume). Partner Choice Models of Cooperation The shortcomings of the partner control approach to the problem of cooperation have left some researchers inclined to explore an alternative: the partner choice approach. The emphasis here is not on how best to deal with a given partner but on “the option of choosing and switching partners” (Noe 2006, 5). It is important to note that partner control and partner choice models are not competing, mutually exclusive alternatives. Rather, the partner choice approach is meant to complement the partner control approach. I will discuss in a later section precisely how the two approaches relate to each other, For now, I am concerned to define some key terms, before critically discussing the partner choice approach. (The following material is drawn from an pair of influential early papers by Noe & Hammerstein 1994; 1995.) A ‘biological market’ exists whenever organisms engage in mutually beneficial exchanges of resources and, crucially, when at least one of the trading individuals can exercise choice in selecting a trading partner. Put negatively, this second requirement says that the market metaphor does not apply when desired resources can simply be taken (i.e. theft) or when individuals can force others to partner with them (i.e. coercion). In many biological markets, organisms can be divided into ‘trader classes,’ according to the kind of resource they offer. For example, numerous ant species can be classed together as ‘protection’ traders, and this class forms a market with various species of aphids, butterflies, and plants, which together comprise a (rather heterogeneous) trader class dealing in nutrients. Notice, in some biological markets, there may be only one commodity on offer. For example, in some forms of hunting there may be a marketplace for hunting skill, in which would-be hunters compete to be chosen as participants by hunt-leaders. Even if a hunting group comprises several specialised roles, each requiring a distinct skill-set, there may nevertheless be competition to fill each role. For example, consider turtle hunting among the Meriam Islanders of the Torres Strait (see Smith & Bliege-Bird 2000): hunt leaders assemble groups comprising a boat driver, a harpooner, and several ‘jumpers’, and competition to fill each role may result in ‘sub-markets’ for specific hunting skills. So, it is not necessary to set up a biological market that traders fall into two classes, although this is certainly where the emphasis falls in the partner choice literature. Given the existence of a biological market, competition among individuals to be chosen as a trading partner and pressure on individuals to make good partner choices are both to be expected. There will accordingly be ‘market selection,’ defined as “selection of traits that maximize fitness in biological markets, such as the ability to compete with members of the same trading class, the ability to attract trading partners, and the ability to sample alternative partners efficiently” (Noe & Hammerstein 1995, 336). The partner choice approach to understanding cooperation, then, attempts to explain cooperation as the result of competition to be picked as a partner in profitable exchanges. An early statement of this approach was given by Bull & Rice (1991, 68), who defined partner choice models as those in which cooperation is evolutionarily stable because “an individual of species A is paired with several members of species B for a single interaction, but A chooses to reward only the most co-operative members of B.” Recently, Nesse (2007, 151) has discussed the approach under the heading of “the social selection perspective,” which “shifts the focus of attention away from decisions to cooperate or defect and abilities to detect cheating, and toward the quite different tasks of selecting carefully among a variety of potential partners, trying to discern what they want, and trying to provide it, so one is more likely to be chosen and kept as a partner.” According to the partner choice approach, puzzling cooperation – A benefiting B at some cost to itself – becomes intelligible when it becomes clear that benefiting B allows A to either establish or maintain a mutually beneficial interaction with B: it is an entry fee. Although the partner choice approach is presented as complementary to, rather than in competition with, the partner control approach, proponents of the former do claim that it enjoys several advantages over the latter. For one thing, the partner choice approach does not rely on unrealistic assumptions about the nature of cooperative interactions (such as that the interacting individuals have no control over partnership formation). For another, empirical support for partner choice models is comparatively abundant (for example, see work by Bshary and colleagues on cleaner fish, and a review in Sachs et al. 2004). Most importantly for current purposes, it is claimed that the partner choice approach, unlike the partner control approach, does not face the so-called ‘cheater problem’ (or at least faces that problem to a far lesser extent). The Relationship Between Partner Choice and Partner Control Models As noted above, the partner choice approach is typically presented as complementary to the partner control approach, rather than in competition with it. Both are supposed to be important to fully understanding cooperation. However, the precise way in which the approaches complement each other requires clarification. There are two possibilities here. The two approaches may (1) model different stages of one complex interaction, or (2) model different kinds of interaction.4 I discuss these possibilities in turn. One way in which the partner choice approach may complement the partner control approach is by modelling different stages of the complex process that is cooperation. Partner choice focuses on the “formation” stage of the cooperative process, at which partnerships or groups come together. Partner control approaches assume that partnership or group formation is, if not random, at least not under the control of the interacting individuals. Partner choice models supplement partner control models by considering the formation stage of the cooperative process in more realistic detail. Noe seems to see the relationship between the partner control and partner choice approaches as complementary in the sense above. There is, though, another way to understand ‘complementary’. Partner choice and partner control models may be relevant – not to different stages of the one extended cooperative interaction – but to importantly different kinds of cooperative interactions. In some cooperative interactions, the benefits of cooperation are distributed to the partners in sequence, serially rather than simultaneously. This is the cases in so-called ‘reciprocal altruism’, such as blood sharing among vampire bats (the classic study being Wilkinson 1984). In such cases, control mechanisms for preventing or punishing defection are important. (Interestingly, one such mechanism may be the threat of partner switching in response to defection, and here, the distinction between mechanisms of partner control and those of partner choice begins to blur). So, we may see partner control models as particularly relevant when cooperative interactions involve the sequential distribution of benefits. 4 Thanks to one of the editors – Kim Sterelny – for pointing this out. In other cooperative interactions, by contrast, the benefits of cooperation are not generated and distributed sequentially. Rather, partners reap the rewards of cooperation with each other simultaneously. Crucially, in some (perhaps many) such cooperative interactions, the magnitude of the benefit generated by cooperation may depend on the nature of the partner(s) involved in the interaction. When this is so – when the size of one’s reward depends on the quality of one’s partners – then partner choice mechanisms will be especially vital, provided of course that there is scope for choice in the first place. To sum up, one way to understand the claim that partner choice and partner control models of cooperation are complementary is to see each as modelling a different stage of the same extended interaction. In some cases, this may well be the correct way to understand the claim. But, there is another way in which the two approaches can complement each other, namely, by each illuminating a qualitatively different kind of cooperative interaction. In this case, whether a partner choice or a partner control model is likely to be most illuminating will depend (among other things, of course) on the way in which benefits are produced and distributed. We should expect partner choice, rather than partner control, models to be particularly helpful in understanding cooperative interactions that generate benefits simultaneously for the cooperators, where the extent of those benefits depends on the quality of the interacting agents. Cooperation and the Cheater Problem In the context of the partner control approach, the problem of cheating is as follows. If B receives a benefit from a cooperative individual A but does not repay that benefit, or repays less than was received, then B-type individuals may eventually replace cooperators like A in the population of interacting individuals, since B-type individuals enjoy the benefits of cooperation without paying any of the associated costs. Showing how such cheating (defection, exploitation, free-riding) can be prevented from undermining cooperation is a central concern for the partner control approach. Cheating is supposedly not a problem – or is at least less of a problem – for the partner choice approach (see e.g. Noe & Hammerstein 1994, 2; Nesse 2007, 145). To assess this claim, it is necessary to specify just what the advocates of partner choice mean by ‘cheating’. Noe & Hammerstein identify one kind of cheating in the context of partner choice as reneging on a proposed trade. They write: “cheating, that is changing the value of the commodity offered after the pair has been formed…” (1994, 6 [emphasis added]). The claim is then that such post-choice changes of offer are often impossible, and hence that cheating is often not a problem for the partner choice approach to cooperation. Noe & Hammerstein write: To our minds the cheating option can safely be ignored in the large number of cases in which either the commodity cannot be withdrawn or changed in quality or quantity once it is offered on the market (1994, 2) For example, Noe & Hammerstein observe that the “food bodies of myrmecophilous plants are examples of such irretrievable offers [since] once the plant has ‘decided’ to provide a quantity x of food bodies, these remain available to the ants” (1994, 3). In this case, the plant traders have a certain amount of the commodity of interest to the ant traders, but cannot withdraw that commodity once it is offered (or at least cannot easily do so). One problem here is that cheating (in this sense) does not seem precluded for the ants. Moreover, there is some evidence that plants in such partnerships can and in fact do withdraw the commodities on offer. For example, Edwards et al. (2006) studied a particular ant-plant mutualism and found that ‘ant shelters’ (domatia) on stems that lose leaves – a sign that protector ants may not be patrolling enough – tend to wither away. So, it may be that the particular example chosen by Noe & Hammerstein to make the case that cheating is precluded in biological market interactions was poorly chosen. In any case, even if post-choice changes of offer are impossible in many biological marketplaces, it has become apparent that the term ‘cheating’ is used to pick out different things in the contexts of partner control and partner choice. The partner control approach faces something aptly called a “cheater problem” – free-riding – while there is another thing – reneging – which is also aptly described as a cheater problem and which the partner choice approach avoids. Cheater problems come in many varieties. Even if reneging is impossible in many biological marketplaces, there is a different kind of cheater problem that can arise in the context of partner choice. To appreciate this problem, it will help to first note the importance that proponents of the partner choice approach assign to signalling in the context of partner choice. Signalling and the Problem of Reliability Noe & Hammerstein write that “trading may take place on the basis of an honest signal that is correlated with access to a commodity, instead of being based on the commodity itself” (1995, 336). Noe also notes that “choosing partners implies a number of mechanisms [including] judging the partner’s quality, a memory for the partner’s quality and location, searching strategies, judging the honesty of signals and so on.” (2006, 5 [emphasis added]). Indeed, Noe thinks it is important to distinguish between markets “in which commodity values are measured directly and those in which signals play an intermediating role” (2001, 108). Bull & Rice (1991, 69) and Sachs et al. (2004, 141) both point out that explanations for cooperation in terms of partner choice depend on there being some way in which individuals can assess and discriminate between potential partners. Partner assessment need not necessarily involve signals (as will be discussed below), but via signalling is one way that it can be done. Once it has been acknowledged that signals play an important role in at least some partner choice scenarios, the problem of reliability becomes unavoidable. In brief, the problem of reliability is as follows. Organisms are often interested in unobservable qualities of other organisms: a sexual rival’s fighting ability, an offspring’s hunger level, the evasive ability of potential prey, or the quality of a potential mate. Making adaptive decisions depends crucially on estimating these unobvious qualities. As it turns out, potential partners, predators and prey often provide the relevant information: they roar, beg, stot, sing or dance, for example, or signal in some other way. The problem of reliability arises when we ask why, given the often strong incentive to mislead signal receivers, signal senders do not do so more often. Why do signallers not exaggerate the relevant quality to their own advantage? Why do signalling systems not collapse as a result of receivers eventually ignoring a cacophony of dishonest signals? ‘False Advertising’ in Biological Markets It is now possible to state a cheater problem that faces the partner choice approach, one that is (I claim) currently underappreciated and insufficiently addressed. The problem is that of false advertising: a trader of one class may present itself as a better partner than it actually is, in order to increase its chances of being chosen as a partner by members of the other trading class. The problem of false advertising in biological markets is a specific case of the more general problem of reliability in biological signalling systems. The problem of false advertising differs from reneging as described above. False advertising is not a matter of a trader in a biological market genuinely having the relevant commodity or property but not delivering it, but rather of the trader lacking that commodity or property while convincing others otherwise. There is some recognition of the false advertising problem in the extant literature on partner choice. For example, Noe (2001, 94) noted that in biological markets the “commodities on offer can be advertised [and] as in commercial advertisements there is a potential for false information”. Noe also mentions the possibility in some markets for “subtle cheating” where “signals associated with the future transfer of resources are occasionally dishonest” (1995, 338; see also Noe 2001, 94). Thus, I do not take myself to be pointing out something overlooked by proponents of the partner choice approach. Rather, I am suggesting that this kind of cheater problem is more pressing than has yet been acknowledged and, further, that the treatments of the problem offered to date are unsatisfactory. Current Treatments of ‘False Advertising’ Attempts to address the false advertising problem for the partner choice approach have been too sanguine in dismissing the problem, or have made misdirected efforts to address it, or have been disunified and in need of clarification. Dismissing the Problem Nesse (2007, 145) in his discussion of partner choice and cooperation claims that social selection “will select for displays of resources and selective altruism that reflect an individual’s potential value as a partner.” Nesse is too quick to assume selection will favour honest signalling of partner quality. One should wonder why displays that reflect partner value would be selectively favoured, rather than those that flatteringly exaggerate it to the displayer’s advantage. Speaking specifically about cooperation among humans, Nesse says that “deception and cheating have been major themes in reciprocity research, and they apply in social selection models, but their effects are limited by inexpensive gossip about reputations and by the difficulty of faking expensive resource displays” (2007, 145). Nesse’s claims about the limited scope for deception in human partner choice are questionable. For one thing, not all signals of partner value require the expenditure of large quantities of resources. Displaying qualities like kindness, honesty, and patience – all plausibly valuable qualities in many kinds of cooperative partnerships – may be quite cheap in terms of energy, risk, and material resources. In addition, gossip may not be so cheap. The risk of making enemies is a hard-to-quantify but nevertheless real cost of gossiping, which ought not to be ignored. While these brief remarks do not settle matters any more than do Nesse’s own, but they serve to show that Nesse is too sanguine regarding the problem of false advertising in biological markets as minor. Missing the Point Sachs et al. (2004) note that partner choice models of cooperation must incorporate effective partner assessment systems. They write: “the [partner] assessment system is the biological arena in which one or more potential partners are observed for their cooperative tendencies, such that their level of cooperation in further interactions can be predicted… [It] allows an individual to gain information about which partners are cooperative and how cooperative they are” (2004, 141-142). Sachs et al. identify “parcelling” and “distribution” as two ways in which partner assessment may be conducted. Parcelling and distribution both involve splitting up a resource to be invested. In the former case, the resource is divided temporally, while in the latter case, the resource is divided spatially. There are two problems with taking parceling and distribution to be partner assessment systems that allow effective partner choice. One problem is conceptual, the other empirical. I will discuss them in turn. The conceptual problem is that parceling and distribution occur after members of two trading classes have partnered up. Two impala grooming each other in brief bursts have already formed a grooming pair. A yucca plant selectively aborting those of its many flowers that have been over-exploited by selfish yucca moths is already interacting with its many partners. This is not to say that parceling and distribution are unimportant in the context of the partner choice approach. On the contrary, they are good ways of deciding when to do some partner “switching” (Noe 1995, 337). There is a difference, though, between partner choice, which occurs prior to the formation of a trading pair or group, and partner switching, which is a matter of strategically leaving one’s current partner for greener fields elsewhere. The difference is one that proponents of the partner choice approach are themselves at pains to mark. Treating parceling and distribution as ways of making effective partner choices is thus conceptually confused and potentially misleading. These are clearly ways of engineering the incentives of partners – and are important as such – but they are not mechanisms of partner choice. The empirical problem with parceling and distribution as ways of making good partner choices is that neither will be an option in biological markets where indivisible resources are at stake. For example, in mating markets where one trader class consists of monandrous females (those who mate with only one male), the commodity on offer is exclusive reproductive access, which cannot be parceled or distributed. In such cases, the timing of partner assessment matters crucially. Partner switching after dabbling a toe, so to speak, will not be an option. Traders offering indivisible resources must identify who is genuinely a high-value partner and who is not prior to committing to a trade. This problem is not limited to cases in which the resource being traded is indivisible, either. The problem may also arise in cases where the costs of partner switching are high. For instance, if searching for a new partner is very costly in terms of time, energy and/or risk, then a choice once made may be effectively fixed. Here again, traders must be able to identify who is genuinely a high-value partner and who is not prior to committing to a trade. A third means of partner assessment mentioned by Sachs et al. is “image scoring” (2004, 142; see also Nowak & Sigmund 1998). For example, potential clients of cleaner fish, while waiting for service, observe the cleaner’s interaction with its current client and are much more likely to interact with a cleaner if its current interaction ends peacefully instead of in conflict. This way of making partner assessments can be used prior to pair formation, and is hence potentially a mechanism for genuine partner choice. Waiting clients that see the current cleaning interaction end in conflict can simply swim away. It is worth drawing a distinction at this point between cues and signals. Signals are behavioural or morphological traits that alter the behaviour of other organisms, have evolved because of that effect, and are effective because the response of receivers has also evolved (Maynard-Smith & Harper 2003, 15). A cue, by contrast, is any “feature of the world, animate or inanimate, that can be used by an animal as a guide to future action” (Maynard-Smith & Harper 2003, 15). Showing that a behavioural or morphological trait is a signal is far more demanding than showing that trait to be a cue. In the former case, much must be established about the evolutionary history of the trait. To show something is a cue, though, we need only show that other organisms attend to it when deciding how to act. Returning now to the case of the cleaner-client fish interaction, it seems that image scoring is better described in terms of cues than of signals. The relevant observation made by potential clients is not of any specific behavioural display by the cleaner that is designed to entice clients. It is rather the observation of a certain state of affairs: an amicable end to the cleaner’s current interaction. This seems more akin to predators choosing prey via cues than it is to, say, peahens choosing mates based on signals like the peacock’s extravagant tail. A predator choosing which of a herd of prey animals to chase wants to pick one it is likely to catch, and watching to see which ones limp is a good way to find out which ones will be most easily caught: limping here is a cue. A client fish wants to interact with cooperative cleaners, and watching to see whether a cleaner’s current interaction ends peacefully – rather than in a cheating-precipitated chase – seems like a good way to obtain at least some information about the cooperativeness of the cleaner. The fact that partner assessment is sometimes done using cues is not in itself any kind of problem for the partner choice approach. Indeed, one might think that assessing partners using cues is less problematic than relying on signals from potential partners. Cues can be more or less accurate predictors, but at least they don’t provide scope for false advertising. However, such deception in the context of partner assessment is possible even when assessments are cue-based, as become evident when we pay closer attention to the cleaner-client fish case. Cleaner fish prefer to eat clients’ mucus rather than parasites (Bshary & Grutter 2003). Large fish have more defection-tempting mucus than do smaller ones. Nonpredatory clients cannot eat a cheating cleaner. Mobile clients – those whose home range encompasses more than one cleaning station – tend not to bother with punitive chases, instead simply swimming away from cheating cleaners. These facts together make large, non-predatory, mobile clients the perfect ‘marks’ for cheating cleaners. It turns out that the image scoring system in the cleaner-client market is exploited by a certain class of cleaners, dubbed “biting cleaners” (Bshary 2002, 2088). A biting cleaner servicing a small client and being observed by a large, nonpredatory, mobile client, will often rub its pelvic fins around the small client’s dorsal fin area (Bshary & Wurth 2001, 1495). This behaviour has been termed “host stabilization” and apparently renders the current client quiescent, ensuring that the cleaning interaction ends peacefully (Potts 1973, 274).5 The biting cleaner thus sets up the score by ensuring that its mark observes the reassuring cue and approaches for service. The cleaner then defects, ignoring the large client’s parasites and plundering its abundant mucus. It is unclear whether a dorsal rub provides any benefit to the small client. If it does not – if it merely wasted time and thus inflicts a net loss – then biting cleaners manage an impressive deception indeed, subtly cheating one client while appearing cooperative to another. Biting cleaners should perhaps be dubbed ‘Machiavellian masseurs.’ It is important to note that biting cleaners do not somehow fake the relevant cue: their interaction with the small client really does end peacefully. Their deception consists in exploiting the cue-based image scoring system of partner assessment operative in this particular biological market. They make sure they are perceived as cooperative under precisely those conditions when being so perceived will open up the most profitable defection opportunities. There is an interesting question to be asked here regarding the classification of the Machiavellian masseur’s behaviour: should such strategic massaging be counted as a cue, or instead, does it qualify as a signal insofar as it has evolved in part in order to influence others’ behaviour? Even if the behaviour is counted as a signal rather than a cue, it should be stressed that the 5 Host stabilization is typically used to soothe clients after a conflict, or to induce waiting fish to remain in the area when the cleaning station is crowded and busy. signal is parasitic upon the cue-based system of partner assessment. (There is perhaps a parallel of sorts to be drawn here with cases of Batesian mimicry.) Thus, false advertising is an issue even in biological markets where partner assessment and choice is conducted on the basis of cues. Piecemeal Solutions Noe’s discussions of partner choice emphasise the importance of signalling in biological markets and acknowledge the problem of reliability (what I am calling ‘false advertising’). The problem with Noe’s treatment of the issue is certainly not a lack of ideas. It is rather a lack of unity and detail. Image scoring (as just discussed) is one of many proposals Noe offers about how the problem might be solved (Bshary & Noe 2003). There are several others. Noe has sometimes appealed to costly signalling theory in his discussions of partner choice. He claims that: the handicap principle predicts that in the context of mate choice, agonistic competition or predation, receivers of signals only pay attention to those signals that are costly to produce… because only individuals that are fit enough to back-up the signal will produce it at high intensity (2001, 109). Noe is here claiming that advertisements in biological marketplaces where the interests of different trader classes conflict must be costly if they are to be believable. Elsewhere, though, he has said things in conflict with this. For example, Noe has suggested that partner choice is facilitated when traders of one class can signal their inability to pursue courses of action detrimental to the interests of traders of the other class. As an example, Noe describes cooperative breeding among purple martins (Noe & Hammerstein 1994, 7). A dominant male will allow ‘tenant’ couples to breed on his territory, in exchange for sexual access to the female tenants. The deal goes sour for the ‘landlord’ if his male tenants sneakily mate with many females on his territory. Noe writes: [W]e expect the ‘choosing’ class, i.e. the dominants, to prefer partners with an ‘honest signal’ of inferiority: an easily perceptible character that cannot change overnight, and that constrains its bearer to keep to its role (Noe & Hammerstein 1994, 6). As it turns out, landlords prefer male tenants with juvenile plumage. Males bearing juvenile plumage are ‘sexually handicapped’ when it comes to attracting females. By foreclosing his option of mating with numerous tenant females, then, a male with juvenile plumage makes himself a non-threatening and thus appealing tenant. It is worth pointing out, though, that displaying juvenile plumage is not costly in the way Noe envisions when talking about the handicap principle. In yet other places, Noe appeals to yet another kind of barrier to false advertising (Noe & Voelkl, this volume). He notes that during “outbidding competition” – in which members of one trader class vie to be chosen as a partner by a member of the other class – the competitors: may be forced to produce their commodity at the maximum possible level. Thus, while their output from this competition cannot be taken as a proxy for how much they will invest later on, it provides – at least – reliable information about their potential. The field of signalling theory has identified several mechanisms that can ensure the honesty of signals (even when sender and receiver interests conflict). Noe’s discussions of signalling in partner choice mention many of these, but in a haphazard way, often in the context of specific empirical examples, and without an eye to the bigger picture. The partner choice approach would benefit from having in place a unified theoretical framework for thinking about kinds of solutions to the problem of false advertising in biological markets. Such a framework would help guide investigation of specific market interactions. A Theoretical Framework for Addressing the False Advertising Problem In this section, I will draw on work in signalling theory to provide a framework within which to organise and clarify Noe’s various discussions of the issue of signalling and reliability in biological markets. The mechanisms that can underwrite the reliability of a signal can be divided into three broad classes: costs, constraints, and commitments. Below, I will discuss each class of mechanism, and will show how the scattered discussions of signal reliability in the partner choice literature can be fitted into the framework this three-way distinction provides. In each case, I suggest research questions that can usefully inform future work on partner choice and cooperation. Cost and Honest Advertisement Handicaps are signals kept honest by costs. Amotz Zahavi’s (1975; 1977) solution to the problem of reliability in signalling was to point out that signals can be relied upon to be honest if it is prohibitively expensive to send a dishonest signal, that is, if the costs of sending such a signal outweigh whatever benefits might be gained by doing so. Zahavi called his solution to the problem of reliability the “handicap principle”. Noe rightly latches on to the handicap principle as a means of solving the false advertising problem. Many cases of signalling to potential partners in mating markets are amenable to this kind of explanation; for example, the peacock’s tail (see e.g. Petrie & Halliday 1994). However, Noe gives a rather simplistic presentation of the costly signalling idea that is based on Zahavi’s initial formulation of the handicap principle. Zahavi’s idea has gone through several incarnations since it was first suggested. He initially claimed that high signal costs impose a test on signallers – a test that only high-quality individuals can pass – and that signalling and surviving is thus an effective way to advertise one’s quality (1975). Zahavi later suggested that “the phenotypic manifestation of the handicap is adjusted to correlate to the phenotypic quality of the individual” (1977, 603). In the initial formulation of the handicap principle, both high quality and low quality individuals were assumed to pay the costs of signalling. In this later version, high signalling costs are paid only by those who can afford those costs (i.e. the genuinely high quality individuals), while those who cannot afford high signalling costs either do not signal at all, or, signal at a lower intensity that is affordable given their quality. In light of these three variations on the costly signalling idea, Searcy & Nowicki (2005, 10) distinguish between “Zahavi” handicaps, “conditional” handicaps, and “revealing” handicaps. Future work on signalling to potential partners in biological markets should take account of the advances in discussions of costly signalling theory. In particular, the issue of signal costs should not be treated in too cavalier a fashion. Careful accounting of the costs involved in a behavioural or morphological display is needed to substantiate the claim that the display is a costly signal. A detailed discussion of the challenges posed by such accounting is given in Kotiaho (2001), and a schema for classifying signal costs is given by Searcy & Nowicki (2005). Too narrow a focus on Zahavi’s initial – and relatively primitive – statement of the costly signalling idea can only handicap attempts to use this idea to understand signalling in the context of partner choice and cooperation. Constraints and Honest Advertisement An index is a signal that is kept honest by constraints against faking. Whereas faking a handicap is possible but unprofitable, faking an index is simply not possible. Indices are signals “whose intensity is causally related to the quality being signalled, and which cannot be faked” (Maynard-Smith & Harper 2003, 15). This is the most likely place to fit Noe’s example of members of a trading class engaging in outbidding competition by signalling at maximum output. Actual empirical work investigating this possibility is rather thin, though. For a specific example, consider the production of nectar by caterpillars in order to attract protector ants. Noe mentions studies of this interaction that report that a lone caterpillar’s nectar production initially increases with increasing numbers of attending ants, but soon hits a ‘ceiling’ (additional ants don’t prompt greater nectar production). This ‘ceiling’ effect may well show that it is not possible for a caterpillar to produce nectar above some particular level, that is, nectar production is constrained. But this is not yet to show that nectar-production is an index used in out-bidding competition. The crucial experiment for determining whether nectar-producing caterpillars really are engaging in out-bidding competitions for the services of protector ants, has not yet been done. That experiment would keep the number of ants fixed, but vary the number of caterpillars. If nectar production is a form of out-bidding competition, then an increase in caterpillar numbers (i.e. more competing bidders) should generate an increase in each individual caterpillar’s level of nectar production (up to some ceiling level that will doubtless vary across individuals). Outbidding competition conducted via indices is an intriguing theoretical possibility, but is currently empirically under-supported. Future empirical work investigating this possibility must take into account at least two issues. Obviously, one is the relation between the signal and the quality signalled: establishing that the possession of the quality constrains the production of the signal is needed in order to show that the signal is an index. The other is the relation between the signal and the context in which it is sent. In cases of outbidding competition, signal intensity should rise as marketplaces become more crowded, that is, as more bidders enter the competition. Commitment Devices and Cooperation A commitment device provides reliable information about one’s likely future actions in virtue of restricting the space of actions one is able to take, or strongly biasing one toward certain of the available options. The work of economist Robert Frank (1988; cf. Fessler & Quintelier, this volume) provides a good example of this kind of approach. Frank’s starting point is the idea of a “commitment problem” (1988, 4). A commitment problem arises whenever an agent can best serve his own interests only by credibly committing himself to act in a way that may later be contrary to his selfinterest. An agent might need to make a credible promise of honesty in order to reassure and secure would-be partners in cooperative endeavours where cheating would be profitable and undetectable. An agent might need to make credible threats of revenge in order to deter would-be exploiters in situations where avenging a wrong would be more costly than not doing so. Commitment problems are common and solving them is important. Frank suggests that evolution has endowed humans with the means to solve commitment problems, namely, “moral sentiments”: anger, contempt, disgust, envy, greed, shame and guilt (1988, 46, 53). Moral sentiments help us solve commitment problems because “being known to experience certain emotions enables us to make commitments that would otherwise not be credible” (1988, 5). The promises of an agent known to be prone to guilt will be for that reason more trusted, Frank suggests, and threats from agents known to be prone to anger will be for that reason taken more seriously. Moral sentiments alone may suffice for solving personal commitment problems, where the goal is for an agent to act in their own longer-term interest despite shorterterm temptation. For interpersonal commitment problems, though, more is needed (as Frank recognised). If the commitment device is internal to the agent (“subjective”, as it is put by Fessler & Quintelier, this volume), then there must be some way for other agents to tell – and tell reliably – that one is committed to being honest, or punitive, or cooperative, as the case may be. For Frank, it is unfakeable expressions of emotions associated with moral sentiments that allow other agents to tell this (here, the index and commitment accounts intersect). Of course, commitment devices need not be internal to agents and signalled in some way to influence the partner choices of others. Commitment devices themselves may be discernable to others. Noe’s example of purple martin landlords preferring male tenants who are ‘sexually handicapped’ by juvenile plumage might fit here. Commitments bind agents; they foreclose some future option(s). Assuming that a male displaying juvenile plumage at the start of a breeding season cannot change his appearance rapidly enough to seduce tenant females that same season, the male has bound himself (at least in the short term) to being a relatively sexually nonthreatening tenant for the dominant landlord male. Then again, we may want to reserve the term ‘commitment device’ for factors that foreclose options indefinitely. And in any case, it is unclear whether males displaying juvenile plumage really are juveniles (making plumage a cue), or whether they are mature birds that have retained juvenile plumage as a breeding strategy (making plumage, potentially at least, a signal). A clear case of commitment playing a role in partner choice-mediated cooperation comes from the case of ritual scarification, tattooing, and other forms of highly visibly body modification (Fessler & Quintelier, this volume). Such modifications can serve to mark the modified individual as a member of a particular group. Depending on the wider social dynamics, such marking may strongly prejudice an individual’s partner choices, even to the extent of precluding some choices, such as the choice to defect to a different group. If so, then marked individuals may well be more attractive than unmarked ones as partners in group-beneficial cooperative endeavours, precisely because such individuals’ fates are tied to the fate of the group. To sum up, there are a plurality of ways in which signalling might work in the context of partner choice. Nothing in this section has been ground-breakingly new. Even so, it is worthwhile to organise the scattered discussions of false advertising extant in the partner choice literature, and to explicitly bring together work in signalling theory with the market perspective on cooperation. Punishment and Partner Choice I want to turn now to cooperation and partner choice in humans, and consider the role of punishment. I think punishment is a neglected option in the partner choice literature. Noe, for instance, is sceptical about the possibility that punishment of false advertising might serve to maintain signal reliability in the context of partner choice (see e.g. Noe & Hammerstein, 1995, 337). This is partly because he underestimates the ways in which punishment can be efficacious. He says that: Punishment as revenge for past behaviour without future fitness advantages cannot be ‘evolutionarily stable’. ‘Punishment’ can only work in long-lasting relationships in which the aggression of the punisher moulds the behaviour of the punished individual in a manner beneficial to the punisher. (2001, 101, 105) Noe assumes here that the only way that punishment could benefit the punisher is if it rehabilitates the punishee. While that is one way in which punishment might benefit the punisher, it is not the only way. Importantly (and importantly by Noe’s own lights as a proponent of the partner choice approach), punishment might benefit the punisher by influencing the partner choices of observers in the punisher’s favour. Experimental economists have studied effect of costly punishment on partner choice in humans. Rob Nelissen investigated “how the costs invested in an altruistic act influence its interpersonal consequences” (2008, 243). By ‘altruism,’ Nelissen meant moralistic punishment, specifically, the paying of a cost to punish unfairness. He predicted that “people [would] confer social benefits (both in terms of enhanced preference and financial rewards) on altruistic punishers proportionally to the cost they incurred in punishing” (2008, 243-244). Nelissen’s subjects were given a sum of money with which to play a “trust game”, which worked as follows: each player was given a sum of money and the option of sending some or all of that money to another player. Any money sent to a trustee would be tripled, and the trustee would then have the option of returning some, all or none of that amount to the sender. Subjects had to choose a partner for the trust game from among the participants of a previous experiment. Subjects were told that their three potential partners – labelled A, B, and C – had observed a “dictator game” in which the dictator split $10 unevenly with the receiver, keeping $8 and giving only $2. Subjects were also told that A, B, and C had had the opportunity to spend some of their own money to take money away from the dictator: giving up $1 would reduce the dictator’s total by $2. Finally, subjects were told that A chose to spend $0 out of $5 on punishment, B chose to spend $1.50 out of $5 on punishment, and C chose to spend $1.50 out of $10 on punishment. In one condition, subjects were randomly matched with A, B, or C and were then asked how much they would entrust to that partner. In the other condition, subjects were asked which of A, B, and C they wanted to play the trust game with and were then asked how much they would entrust to that partner. Nelissen found that subjects chose B over A and C (2008, 244). He also found that, when pairing was random, subjects paired with B sent the most money in the trust game (2008: 246). As Nelissen interprets the findings: [T]he costs incurred in altruistic punishment were perceived as signalling the extent to which punishers value fairness… [P]eople prefer punishers more [as trust-game partners] if they invest more to punish unfairness but only if the invested amount can be perceived as a reliable index of fairness concerns (2008, 244, 246). The difference between B and C lies in the relative cost each paid in order to punish unfairness. Punishment was, relatively speaking, twice as costly for B as for C. Subjects thus seem to take the cost of punishment into account when deciding who to interact with or how to behave toward partners that are forced upon them. While this does not show that more costly acts of punishment are more reliable signals that the punisher values fairness, it at least suggests that observers judge them to be such.6 The evidence for a signalling role for punishment in the context of human cooperative partner choice is admittedly rather thin at this stage. The influence of costly moralistic punishment on partner choice is yet to be fully described. Very recent work by Horita (2010) indicates that being a punisher influences others’ partner choice to one’s own advantage in some cases, but works to one’s disadvantage in others. Specifically, punishing unfairness is good for one’s prospects of being chosen as a partner when one will play the role of provider of resources (i.e. when punisher 6 Nelissen’s work on moralistic punishment dovetails with Barclay’s (2006) work on the topic. Barclay ran an experiment in which accepting a cost to punish free-riding during a Public Goods Game benefitted other players. He found that individuals who paid to punish were subsequently rated as more trustworthy and more worthy of respect than non-punishers and were chosen over non-punishers as partners in subsequent trust games (2006, 330). will have control over distribution of resources in later interaction). However, punishers seem to be chosen less frequently than non-punishers when, in the coming interaction, the chooser has control over the distribution of resources. At least in the case of humans, punishment may play an important role in partner choice and cooperation, both by imposing costs on non-cooperation and by influencing decisions about whom to pair with for mutually beneficial interactions. Proponents of the partner choice approach should therefore not be too quick to dismiss punishment as a potential means of solving the false advertising problem, at least in the context of human cooperation and partner choice. Conclusion There is a cheater problem for the partner choice approach: the problem of false advertising. That problem is currently under-studied. Addressing it will require importing some signalling theory into the partner choice approach. I have started that process here, but much remains to be done. Hopefully, my contribution will help inform future work on partner choice and cooperation. Works Cited Axelrod, R. 1984. The Evolution of Cooperation. New York: Basic Books. Barclay, P. 2006. “Reputational benefits for altruistic punishment.” Evolution and Human Behaviour 27: 325-344. Bshary, R. 2002. “Biting cleaner fish use altruism to deceive image-scoring client reef fish.” Proceeding of the Royal Society of London B 269: 2087-2093. Bshary, R & Grutter, A. 2003. “Cleaner wrasse prefer client mucus.” Proceeding of the Royal Society of London B 270: S242-S244. Bshary, R. & Noe, R. 2003. “Biological markets: the ubiquitous influence of partner choice on the dynamics of cleaner fish – client reef fish interactions.” In Genetic and Cultural Evolution of Cooperation, ed. P. Hammerstein, 167-184. Cambridge, MA: MIT Press. Bshary, R. & Wurth, M. 2001. “Cleaner fish Labroides dimidiatus manipulate client reef fish by providing tactile stimulation.” Proceeding of the Royal Society of London B 268: 1495-1501. Bull, J. & Rice, W. 1991. “Distinguishing mechanisms for the evolution of cooperation.” Journal of Theoretical Biology 149: 63-74. Calcott, B. 2008. “The other cooperation problem: generating benefit.” Biology and Philosophy 23: 179-203. Calcott, B. This volume. “The evolution of complex cooperation.” Clutton-Brock, T. 2009. “Cooperation between non-kin in animal societies.” Nature 462: 51-57. Dugatkin, L. 1995. “Partner choice, game theory, and social behaviour.” Journal of Quantitative Anthropology 5: 3-14. Dugatkin, L. 1997. Cooperation Among Animals. Oxford: Oxford University Press. Edwards, D., Hassal, M., Sutherland, W. & Yu, D. 2006. “Selection for protection in an ant-plant mutualism: host sanctions, host modularity, and the principal-agent game.” Proceeding of the Royal Society of London B 273: 595-602. Fessler, D. & Quintelier, K. This volume. “Suicide bombers, weddings, and prison tattoos.” Frank, R. 1988. Passions Within Reason. New York: W. W. Norton & Co. Hamilton, W. 1964. “The genetical evolution of social behavior (I and II).” Journal of Theoretical Biology 7: 1-52. Horita, Y. 2010. “Punishers may be chosen as providers but not as recipients.” Letters on Evolutionary Behavioural Science 1: 6-9. Kotiaho, J. 2001. “Costs of sexual traits.” Biological Reviews 76: 365-376. Maynard-Smith, J. & Harper, D. 2003. Animal Signals. Oxford: Oxford University Press. Miller, G. 2007. “Sexual selection for moral virtues.” The Quarterly Review of Biology 82: 97-121. Nelissen, R. 2008. “The price you pay: cost-dependent reputation effects of altruistic punishment.” Evolution and Human Behavior 29: 242-248. Nesse, R. 2007. “Runaway social selection for displays of partner value and altruism.” Biological Theory 2: 143-155. Noe, R. & Hammerstein, P. 1994. “Biological markets: supply and demand determine the effect of partner choice in cooperation, mutualism and mating.” Behavioural Ecology and Sociobiology 35: 1-11. Noe, R. & Hammerstein, P. 1995. “Biological markets.” Trends in Ecology and Evolution 10: 336-339. Noe, R. & Voelkl, B. This volume. “Cooperation and biological markets: the power of partner choice.” Noe, R. 1990. “A Veto game played by baboons: a challenge to the use of the Prisoner’s Dilemma as a paradigm for reciprocity and cooperation.” Animal Behaviour 39: 78-90. Noe, R. 2001. “Biological markets: partner choice as the driving force behind the evolution of mutualisms.” In Economics in Nature: Social Dilemmas, Mate Choice, and Biological Markets, eds. R. Noe, J. van Hoof & P. Hammerstein, 93-118. Cambridge: Cambridge University Press. Noe, R. 2006. “Cooperation experiments: coordination through communication versus acting apart together.” Animal Behaviour 71: 1-18. Nowak, M. & Sigmund, K. 1998. “Evolution of indirect reciprocity by image scoring.” Nature 393: 673-677. Petrie, M. & Halliday, T. 1994. “Experimental and natural changes in the peacock’s (Pavo cristatus) train can affect mating success.” Behavioral Ecology and Sociobiology 35: 213-217. Potts, G. 1973. “The ethology of Labroides dimidiatus on Aldabra.” Animal Behaviour 21: 250-291. Sachs, J., Mueller, U., Wilcox, T. & Bull, J. 2004. “The evolution of cooperation.” The Quarterly Review of Biology 79: 135-160. Searcy, W. & Nowicki, S. 2005. The Evolution of Animal Communication. Princeton, NJ: Princeton University Press. Smith, E. & Bliege-Bird, R. 2000. “Turtle hunting and tombstone opening.” Evolution and Human Behavior 21: 245-261. Sober, E. & Wilson, D. 1998. Unto Others. Cambridge, MA: Harvard University Press. Stevens, J. & Hauser, M. 2004. “Why be nice? Psychological constraints on the evolution of cooperation.” Trends in Cognitive Sciences 8: 60-65. Stevens, J., Cushman, F. & Hauser, M. 2005. “Evolving the psychological mechanisms for cooperation.” Annual Review of Ecology, Evolution and Systematics 36: 499-518. Trivers, R. 1971. “The evolution of reciprocal altruism.” Quarterly Review of Biology 46: 35-57. West, S., Griffin, A. & Gardner, A. 2007. “Social semantics.” Journal of Evolutionary Biology 20: 415432. Wilkinson, G. 1984. “Reciprocal food sharing in the vampire bat.” Nature 308: 181-184. Zahavi, A. 1975. “Mate selection: a selection for handicap.” Journal of Theoretical Biology 53: 205214. Zahavi, A. 1977. “The cost of honesty: further remarks on the handicap principle.” Journal of Theoretical Biology 67: 603-605.