Clues To The Function Of Moralistic Punishment

One of the major questions I’ve spent the better part of the last few years trying to work up an answer to is the matter of why – in the adaptive sense – people punish others as third parties moralistically. That is to ask why people condemn and punish others for engaging in behavior perceived to be “wrong”. This is not the same question as why people take revenge on others for their behavior; that kind of punishment isn’t terribly mysterious. Chimps, for instance, will punish those who cheated them, but they seem fairly uninterested in punishing those who cheated others (Riedl, Jensen, Call, & Tomasello, 2012). Humans are a bit different in that regard: not only do we involve ourselves in the disputes of others, but we do so on behalf of people who aren’t kin or even social allies. This is rather strange behavior, even though we do not usually perceive it to be so. I’m not the only one who has attempted an evolutionary explanation for moralistic punishment, of course; in my hubris, however, I do feel that the explanation I ended up floating tends to fit the existing data better than other, contemporary theories. Today, I wanted to try and demonstrate that fit by considering some relative-recent empirical research.

“I don’t know, man; I think it might be a little baggy. Try a size down”

The paper in question (Kurzban, DeScioli, & O’Brien, 2007) concerns audience effects on punishment. Specifically, it addresses the following question: do people punish others more (as third parties) when they are being watched and their decisions are being made public, relative to when they’re anonymous? The question was examined across two experiments. In the first of these, 43 participants viewed the outcome of a previous trust game in which one player had the choice between two different outcomes: $39 for himself and $1 for the other player, or $20 for both. In this case, the player had selected the former, selfish option. The participants in question were provided with $7 and told they could spend any amount of that money – from $0 to $7 – to deduct from the selfish player. For each dollar they spent on deductions, the selfish player would lose $3. Importantly, the participants were either doing so in conditions under which their spending would be unknown to anyone but themselves (n = 24), or conditions where the experimenter would known how much each particular participant spent (n = 19). In the anonymous condition, 38% of participants spent at least some money on punishment, whereas in the public condition, 47% opted to punish something. Importantly, however, the amount punished varied between these conditions. Those in the public condition tended to spend more on punishment than those in the anonymous condition when they did. For instance, though precise numbers are not reported, approximately 20% of the participants in the public condition paid the full $7 to punish the selfish player, as compared with 0% in the private condition.

The second experiment examined peoples’ responses to a different kind of economic game – a prisoner’s dilemma – where one player attempted to cooperate with their partner, but was met with defection. In terms of the outcome, this means that rather than both players getting $25 (if both had cooperated), the cooperative partner got $0 while the defector got $30. The participants in this experiment were shown this outcome and given $10. As before, the participants could spend any amount of this $10 to deduct from the defecting player at the 3-to-1 rate. In this case, the punishment decisions would either be known by (a) no one else but the participant (n = 31), (b) the experimenter alone (n = 26), or (c) the entire group of participants (n = 30). Again, fewer participants opted to punish at all when punishment was anonymous (42%) relative to when punishment was known to either the experimenter (65%) or the whole group (67%); the amount spent on punishment increased as well (Ms = $1.06, $2.54, and $3.17, respectively). Further, those in the public conditions also tended to display more anger, disgust, and contempt at the cheating, suggesting that they were more upset by the transgression when other people were watching (or they were at least pretending to be).

The existence of audiences seemed to have an important impact on determining moralistic punishment: not only did the presence of other people affect the percent of third parties willing to punish at all, but it also positively influenced how much they did punish. In a sentence, we could say that the presence of observers was being used as an input by the cognitive systems determining moralistic sentiments. While this may sound like a result that could have been derived without needing to run the experiments, the simplicity and predictability of these findings by no means makes them trivial on a theoretical level when it comes to answering the question, “what is the adaptive value of punishment?” Any theory seeking to explain morality in general – and moral punishment in particular – needs to be able to present a plausible explanation for why cues to anonymity (or lack thereof) are being used as inputs by our moral systems. What benefits arise from public punishment that fail to materialize in anonymous cases?

“If you’re good at something, never do it for free…or anonymously”

The first theoretical explanation for morality that these results cut against is the idea that our moral systems evolved to deliver benefits to other per se. One of the common forms of this argument is that our moral systems evolved because they delivered benefits to the wider group (in the form of maintaining beneficial cooperation between members) even if doing so was costly in terms of individual fitness. This argument clearly doesn’t work for explaining the present data, as the potential benefits that could be delivered to others by deterring cheating or selfishness do not (seem to) change contingent on anonymity, yet moral punishment does. 

These results also cut against some aspects of mutualistic theories for morality. This class of theory suggests that, broadly speaking, our moral sense responds primarily to behavior perceived to be costly to the punisher’s personal interests. In short, third parties do not punish perpetrators because they have any interest in the welfare of the victim, but rather because punishers can enforce their own interests through that punishment, however indirectly. To place that idea into a quick example, I might want to see a thief punished not because I care about the people he harmed, but rather because I don’t want to be stolen from and punishing the thief for their behavior reduces that probability for me. Since my interests in deterring certain behaviors do not change contingent on my anonymity, the mutualistic account might feel some degree of threat from the present data. As a rebuttal to that point, the mutualistic theories could make the argument that my punishment being made public would deter others from stealing from me to a greater extent than if they did not know I was the one responsible for punishing. “Because I punished theft in a case where it didn’t effect me,” the rebuttal goes, “this is a good indication I would certainly punish theft which did affect me. Conversely, if I fail to punish transgressions against others, I might not punish them when I’m the victim.” While that argument seems plausible at face value, it’s not bulletproof either. Just because I might fail to go out of my way to punish someone else who was, say, unfaithful in their relationship, that does not necessarily mean I would tolerate infidelity in my own. This rebuttal would require an appreciable correspondence between my willingness to punish those who transgress against others and those who do so against me. As much of the data I’ve seen suggests a weak-to-absent link in both humans and non-humans on that front, that argument might not hold much empirical water.

By contrast, the present evidence is perfectly consistent with the association-management explanation posited in my theory of morality. In brief, this theory suggests that our moral sense helps us navigate the social world, identifying good and bad targets of our limited social investment, and uses punishment to build and break relationships with them. Morality, essentially, is an ingratiation mechanism; it helps us make friends (or, alternatively, not alienate others). Under this perspective, the role of anonymity makes quite a bit of sense: if no one will know how much you punished, or whether you did at all, your ability to use punishment to manage your social associations is effectively compromised. Accordingly, third-party punishment drops off in a big way. On the other hand, when people will know about their punishment, participants become more willing to invest in it in the face of better estimated social return. This social return need not necessarily reside with the actual person being harmed, either (who, in this case, was not present); it can also come from other observers of punishment. The important part is that your value as an associate can be publicly demonstrated to others.

The first step isn’t to generate value; it’s to demonstrate it

The lines between these accounts can seem a bit fuzzy at times: good associates are often ones who share your values, providing some overlap between mutualistic and association accounts. Similarly, punishment, at least from the perspective of the punisher, is altruistic: they are suffering a cost to provide someone else with a benefit. This provides some overlap between the association and altruistic accounts as well. The important point for differentiating these accounts, then, is to look beyond their overlap into domains where they make different predictions in outcomes, or predict the same outcome will obtain, but for different reasons. I feel the results of the present research not only help do that (inconsistent with group selection accounts), but also present opportunities for future research directions as well (such as the search for whether punishment as a third party appreciably predicts revenge).

References: Kurzban, R., DeScioli, P., & O’Brien, E. (2007). Audience effects on moralistic punishment. Evolution & Human Behavior, 28, 75-84.

Riedl, K., Jensen, K., Call, J., & Tomasello, M. (2012). No third-party punishment in chimpanzees. Proceedings of the National Academy of Science, 109, 14824–14829

Having Their Cake And Eating It Too

Humans are a remarkably cooperative bunch of organisms. This is a remarkable fact because cooperation can open the door wide to all manner of costly exploitation. While it can be a profitable strategy for all involved parties, cooperation requires a certain degree of vigilance and, at times, the credible threat of punishment in order to maintain its existence. Figuring out how people manage to solve these cooperative problems has provided us with no shortage of research and theorizing, some of which is altogether more plausible than the rest. Though I haven’t quite figured out the appeal yet, there are many thoughtful people who favor the group selection accounts for explaining why people cooperate. They suggest that people will often cooperate in spite of its personal fitness costs because it serves to better the overall condition of the group to which they belong. While there haven’t been any useful predictions that appear to have fallen out of such a model, there are those who are fairly certain it can at least account for some known, but ostensibly strange findings.

That is a rather strange finding you got there. Thanks, Goodwill.

One human trait purported to require a group selection explanation is altruistic punishment and cooperation, especially in one-shot anonymous economic games. The basic logic goes as follows: in a prisoner’s dilemma game, so long as that game is a non-repeated event, there is really only one strategy, and that’s defection. This is because if you defect when your partner defects, you’re better off than if you cooperated; if you partner cooperated, on the other hand, you’re still better off if you defect. Economists might thus call the strategy of “always defect” to be a “rational” one. Further, punishing a defector in such conditions is similarly considered irrational behavior, as it only results in a lower payment for the punisher than they would have otherwise had. As we know from decades of research using these games, however, people don’t always behave “rationally”: sometimes they’ll cooperate with other people they’re playing with, and sometimes they’ll give up some of their own payment in order to punish someone who has either wronged them or, more importantly, wronged stranger. This pattern of behavior – paying to be nice to people who are nice, and paying to punish those who are not – has been dubbed “strong reciprocity”. (Fehr, Fischbacher, & Gachter, 2002)

The general raison d’etre of strong reciprocity seems to be that groups of people which had lots of individuals playing that strategy managed to out-compete other groups of people without them. Even though strong reciprocity is costly on the individual level, the society at large reaps larger overall benefits, as cooperation has the highest overall payoff, relative to any kind of defection. Strong reciprocity, then, helps to force cooperation by altering the costs and benefits to cooperation and defection on the individual level. There is a certain kind of unfairness inherent in this argument, though; a conceptual hypocrisy that can be summed up by the ever-popular phrase, “having one’s cake and eating it too”. To consider why, we need to understand the reason people engage in punishment in the first place. The likely, possibly-obvious candidate explanation just advanced is that punishment serves a deterrence function: by inflicting costs on those who engage in the punished behavior, those who engage in the behavior fail to benefit from it and thus stop behaving in that manner. This function, however, rests on a seemingly innocuous assumption: actors estimate the costs and benefits to acting, and only act when the expected benefits are sufficiently large, relative to the costs.

The conceptual hypocrisy is that this kind of cost-benefit estimation is something that strong reciprocators are thought to not to engage in. Specifically, they are punishing and cooperating regardless of the personal costs involved. We might say that a strong reciprocator’s behavior is inflexible with respect to their own payments. This example is a bit like playing the game of “chicken”, where two cars face each other from a distance and start driving at one another in a straight line. The first drive to turn away loses the match. However, if both cars continue on their path, the end result is a much greater cost to both drivers than is suffered if either one turns. If a player in this game was to adopt an inflexible strategy, then, by doing something like disabling their car’s ability to steer, they can force the other player to make a certain choice. Faced with a driver who cannot turn, you really only have one choice to make: continue going straight and suffer a huge cost, or turn and suffer a smaller one. If you’re a “rational” being, then, you can be beaten by an “irrational” strategy.

Flawless victory. Fatality.

So what would be the outcome if other individuals started playing the ever-present “always defect” strategy in a similarly inflexible fashion? We’ll call those people “strong defectors” for the sake of contrast. No matter what their partner does in these interactions, the strong defectors will always play defect, regardless of the personal costs and benefits. By doing so, these strong defectors might manage to place themselves beyond the reach of punishment from strong reciprocators. Why? Well, any amount of costly punishment directed towards a strong defector would be a net fitness loss from the group’s perspective, as costly punishment is a fitness-reducing behavior: it reduces the fitness of the person engaging in it (in the form of whatever cost they suffer to deliver the punishment) and it reduces the fitness of the target of the punishment. Further, the costs to punishing the defectors could have been directed towards benefiting other people instead – which are net fitness gains for the group – so there are opportunity costs to engaging in punishment as well. These fitness costs would need to be made up for elsewhere, from the group selection perspective.

The problem is that, because the strong defectors are playing an inflexible strategy, the costs cannot be made up for elsewhere; no behavioral change can be affected. Extending this game of chicken analogy to the group level, let’s say that turning away is the “cooperative” option, and dilemmas like these were at least fairly regular. They might not have involved cars, but they did involve a similar kind of payoff matrix: there’s only one benefit available, but there are potential costs in attempting to achieve it. Keeping in line with the metaphor, it would be in the interests of the larger population if no one crashed. It follows that between-group selective pressures favor turning every time, since the costs are guaranteed to be smaller for the wider population, but the sum of the benefits don’t change; only who achieves them does. In order to force the cooperative option, a strong reciprocator might disable their ability to turn so as it alters the cost and benefits to others.

The strong reciprocators shouldn’t be expected to be unaffected by costs and benefits, however; they ought to be affected by such considerations, just on the group level, rather than the individual one. Their strategy should be just as “rational” as any others, just with regard to a different variable. Accordingly, it can be beaten by other seemingly irrational strategies – like strong defection – that can’t be affected by the threats of costs. Strong defectors which refuse to turn will either force a behavioral change in the strong reciprocators or result in many serious crashes. In either case, the strong reciprocator strategy doesn’t seem to lead to benefits in that regard.

Now perhaps this example sounds a bit flawed. Specifically, one might wonder how appreciable portions of the population might come to develop an inflexible “always defect” strategy in the first place. This is because the strategy appears to be costly to maintain at times: there are benefits to cooperation and being able to alter one’s behavior in response to costs imposed through punishment, and people would be expected to be selected to achieve and avoid them, respectively. On top of that, there is also the distinct concern that repeated attempts at defection or exploitation can result in punishment severe enough to kill the defector. In other words, it seems that there are certain contexts in which strong defectors would be at a selective disadvantage, becoming less prevalent in the population over time. Indeed, such a criticism would be very reasonable, and that’s precisely the because the always defect population behaves without regard to their personal payoff. Of course, such a criticism applies in just as much force to the strong reciprocators, and that’s the entire point: using a limited budget to affect the lives of others regardless of its effects on you isn’t the best way to make the most money.

The interest on “making it rain” doesn’t compete with an IRA.

The idea of strong defectors seems perverse precisely because they act without regard to what we might consider their own rational interests. Were we to replace “rational” with “fitness”, the evolutionary disadvantage to a strategy that functions as if behaving in such a manner seems remarkably clear. The point is that the idea of a strong reciprocator type of strategy should be just as perverse. Those who attempt to put forth a strong reciprocator type of strategy as plausible account for cooperation and punishment attempt to create a context that allows them to have their irrational-agent cake and eat it as well: strong reciprocators need not behave within their fitness interests, but all the other agents are expected to. This assumption needs to be at least implicit within the models, or else they make no sense. They don’t seem to make very much sense in general, though, so perhaps that assumption is the least of their problems.

References: Fehr, E., Fischbacher, U., & Gachter, S. (2002). Strong reciprocity, human cooperation, and the enforcement of social norms. Human Nature, 13, 1-25 DOI: 10.1007/s12110-002-1012-7

The Tension Between Theory And Reality

“In theory, theory and practice are the same. In practice, they are not.”

There is a relatively famous quote attributed to Michelangelo who was discussing his process of carving a statue: “I saw the angel in the marble and carved until I set him free”. Martin Nowak, in his book SuperCooperators (2011), uses that quote to talk about his admiration for using mathematical models to study cooperation. By stripping away the “noise” in the world, one can end up with some interesting conclusions. For instance, it was through this stripping away of the noise that led to the now-famous programming competition that showed us how successful a tit-for-tat strategy can be. There’s just one hitch, and it’s expressed in another relatively famous quote attributed to Einstein: “Whether you can observe a thing or not depends on the theory which you use. It is the theory which decides what can be observed.” Imagine instead that Michelangelo had not seen an angel in the marble, but rather a snake: he would have “released” the snake from the marble instead. That Michelangelo “saw” the angel in the first place seemed to preclude his seeing the snake – or any number of other possible images – that might have potentially been representable by the marble as well. I should probably also add that neither the snake nor the angel were actually “in” the marble in the first place…

“You see a medium for high art; I see new kitchen countertops”

The reason I bring up Nowak’s use of the Michelangelo quote is that both in his book and a recent paper (Nowak, 2012), Nowak stresses the importance of both (a) using mathematical models to reveal underlying truths by stripping away noise from the world, and (b) advocates for the readdition of that noise, or at least some of it, to make the models better at predicting real-world outcomes. The necessity of this latter point is demonstrated neatly by the finding that, as the rules of the models designed to assess cooperation shifted slightly, the tit-for-tat strategy no longer emerged as victorious. When new variables – ones previously treated as noise – are introduced to these games, new strategies can best tit-for-tat handily. Sometimes the dominant strategy won’t even remain static over time, shifting between patterns of near universal cooperation, universal defection, and almost anything in between. That new pattern of results doesn’t mean that a tit-for-tat strategy isn’t useful on some level; just that it’s usefulness is restricted to certain contexts, and those contexts may or may not be represented in any specific model.

Like Michelangelo, then, these theoretical models can “see” any number of outcomes (as determined by the initial state of the program and its governing rules); like Einstein, these models can also only “see” what they are programmed to see. Herein lies the tension: these models could be excellent for demonstrating the many things (like group selection works), but many of many those things which can be demonstrated in the theoretical realm are not applicable to the reality that we happen to live in (also like group selection). The extent to which those demonstrations are applicable to the real world relies on the extent to which the modeller happened to get things right. For example, let’s say we actually had a slab of marble with something inside it and it’s our goal to figure out what that something is: a metaphorical description of doing science. Did Michelangelo demonstrate that this something was the specific angel he had in mind by removing everything that wasn’t that angel from an entirely different slab of marble?  Not very convincingly; no. He might have been correct, but there’s no way to tell without actually examining the slab with that something inside of it directly. Because of this, mathematical models do not serve as a replacement for experimentation or theory in any sense.

On top of that concern, a further problem is that, in the realm of the theoretical, any abstract concept (like “the group”) can be granted as much substance as any other, regardless of whether those concepts can be said to exist in reality; one has a fresh slab of marble that they can “see” anything in, constrained only by their imagination and programming skills. I could, with the proper technical know-how, create a mathematical model that demonstrates that people with ESP have a fitness advantage over those without this ability. By contrast, I could create a similar model that demonstrates that people without ESP have a fitness advantage over those with the ability. Which outcome will eventually obtain depends entirely on the ways in which I game my model in favor of one conclusion or the other. Placed in that light, (“we defined some strategy as working and concluded that it worked”) the results of mathematical modeling seem profoundly less impressive. More to the point, however, the outcome of my model says nothing about whether or not people actually have these theoretical ESP abilities in the first place. If they don’t, all the creative math and programming in the world wouldn’t change that fact.

Because, eventually, Keanu Reeves will stop you.

As you can no doubt guess by this point, I don’t hold mathematical modeling in the same high esteem that Nowak seems to. While its theoretical utility is boundless, its practical utility seems extremely limited, relying on the extent to which the assumptions of the programmer approach reality. With that in mind, I’d like to suggest a few other details that have not yet seemed to have been included in these models of cooperation. That’s not to say that the inclusion of these variables would allow a model to derive some new and profound truths – as these models can only see what they are told to see and how they are told to see it – just that these variables might help, to whatever degree, the models better reflect reality.

The first of these issues is that these cooperation games seem to be played using an identical dilemma between rounds; that is to say there’s only one game in town, and the payoff matrices for cooperation and defection remain static. This, of course, is not the way reality works: cooperation is sometimes mutually beneficial, other times mutually detrimental, and still others only beneficial for one of the parties involved, and all that changes the game substantially. Yes, this means we aren’t strictly dealing with cooperative dilemmas anymore, but reality is not made up of strictly cooperative dilemmas, and that matters if we’re trying to draw conclusions about reality. Adding this consideration into the models would mean that behavioral strategies are unlikely to ever cycle between  “always cooperate” or “always defect” as Nowak (2012) found that they did in his models. Such strategies are too simple-minded and underspecified to be practically useful.

A second issue involves the relative costs and benefits to cooperation and defection even within the same game. Sometimes defecting may lead to great benefits for the defector; at others, defecting may only lead to small benefits. A similar situation holds for how much of a benefit cooperation will bring to one’s partner. A tit-for-tat strategy could be fooled, so to speak, by this change of rules (i.e. I could defect on you when the benefits for me are great and reestablish cooperation only when the costs to cooperation are low). As cooperation will not yield identical payoffs over time more generally, cooperation will also not yield identical payoffs between specific individuals. This would make some people more valuable to have as a cooperative partner than others and, given that cooperation takes some amount of limited time and energy, this means competition for those valuable partners. Similarly, this competition can also mean that cooperating with one person entails simultaneously defecting against another (cooperation here is zero-sum; there’s only so much to go around). Competition for these more valuable individuals can lead to all sorts of interesting outcomes: people being willing to suffer defection for the privilege of certain other associations; people actively defecting on or punishing others to prevent those others from gaining said associations; people avoiding even trying to compete for these high value players, as their odds of achieving such associations are vanishingly low. Basically, all sorts of politically-wise behaviors we see from the characters in Game in Thrones that don’t find themselves represented in these mathematical models yet.

We might also want to add a stipulations for in-game beheadings.

A final issue is that information that individuals in these games are exposed to: it’s all true information. In the non-theoretical realm, it’s not always clear as to whether someone you’ve been interacting with cooperated or defected, or the degree of effort they put into the venture even if they were on the cooperating side of the equation. If individuals in these games could reap the benefits of defecting while simultaneously convincing others that they had cooperated, that’s another game-changer. Modeling all of this is, no doubt, a lot of work, but potentially doable. It would lead to all sorts of new set of findings about which strategies worked and which one didn’t, and how, and when, and why. The larger point, however, is that the results of these mathematical models aren’t exactly findings; they’re restatements of our initial intuitions in mathematical form. Whether those intuitions are poorly developed and vastly simplified or thoroughly developed and conceptually rich is an entirely separate matter, as they’re all precisely as “real” in the theoretical domain.

References: Nowak, M. (2011). SuperCooperators: Altruism, evolution, and why we need each other to succeed. New York: Free Press

Nowak, M. (2012). Evolving cooperation Journal of Theoretical Biology, 299, 1-8 DOI: 10.1016/j.jtbi.2012.01.014

No, Really; Group Selection Still Doesn’t Work

Back in May, I posed a question concerning why an organism would want to be a member of group: on the one hand, an organism might want to join a group because, ultimately, that organism calculates that joining a group would likely lead to benefits for itself that the organism would not otherwise obtain; in other words, organisms would want to join a group for selfish reasons. On the other hand, an organism might want to join a group in order to deliver benefits to the entire group, not just themselves. In this latter case, the organism would be joining the group for, more or less, altruistic reasons. For reasons that escape my current understanding, there are people who continue to endorse the second reason for group-joining as plausible, despite it being anathema to everything we currently know about how evolution works.

The debate over whether adaptations for cooperation and punishment were primarily forged by selection pressures at the individual or group level has gone on for so long because, in part, much of the evidence that was brought to bear on the matter could have been viewed as being consistent with either theory – if one was creative enough in their interpretation of the results, anyway. The results of a new study by Krasnow et al (2012) should do one of two things to the group selectionists: either make them reconsider their position or make them get far more creative in their interpreting.

Though I think I have a good guess which route they’ll end up taking.

The study by Krasnow et al (2012) took the sensible route towards resolving the debate: they created contexts where the two theories make opposing predictions. If adaptations for social exchange (cooperation, defection, punishment, reputation, etc) were driven primarily by self-regarding interests (as it is the social exchange model), information about how your partner behaved towards you should be more relevant than information about how your partner behaved towards others when you’re deciding how to behave towards them. In stark contrast, a group selection model would predict that those two types of information should be of similar value when deciding how to treat others, since the function of these adaptations should be to provide group-wide gains; not selfish ones.

These contexts were created across two experiments. The first experiment was designed in order to demonstrate that people do, in fact, make use of what the authors called “third-party reputation”, defined as a partner’s reputation for behaving a certain way towards others. Subjects were brought into the lab to play a trust game with a partner who, unbeknownst to the subjects, were computer programs and not real people. In a trust game, a player can either not trust their partner, resulting in an identical mid-range payoff for both (in this case, $1.20 for both), or trust their partner. If the first player trusts, their partner can either cooperate – leading to an identical payoff for both players that’s higher than the mid-range payoff ($1.50 for both) – or defect – leading to an asymmetrical payoff favoring the defector ($1.80 and $0.90). In the event that the player trusted and their partner defected, the player was given an option to pay to punish their partner, resulting in both their payoffs sitting at a low level ($0.60 for both).

Before the subjects played this trust game, they were presented with information about their partner’s third-party reputation. This information came in the form of questions that their partner had ostensibly filled out earlier, which assessed the willingness of that partner to cheat given freedom from detection. Perhaps unsurprisingly, subjects were less willing to trust a partner who indicated they would be more likely to cheat, given a good opportunity. What this result tells us, then, is that people are perfectly capable of making use of third-party reputation information when they know nothing else about their partner. These results do not help us distinguish between group and individual-level accounts, however, as both models predict that people should act this way; that’s where the second study came in.

“Methods: We took 20 steps, turned, and fired”

The second study added in the crucial variable: first-party reputation, or your partner’s past behavior towards you. This information was provided through the results of two prisoner’s dilemma games that were visible to the subject, one which was played between a subject and their partner and the other played between the partner and a third party. This led to subjects encountering four kinds of partners: one who defected both on the subject and a third party, one who cooperated with both, and one who defected on one (either the subject or the third party) but cooperated with the other. Following this initial game, subjects again played a two-round trust game with their partners. This allowed the following question to be answered: when subjects have first-party reputation available, do they still make use of third-party reputation?

The answer could not have been a more resounding, “no”. When deciding whether they were going to trust their partner or not, the third-party reputation did not predict the outcome at all, whereas first-party reputation did, and, unsurprisingly, subjects were less willing to trust a partner who had previously defected on them. Further, a third-party reputation for cheating did not make subjects any more likely to punish their partner, though first-party reputation didn’t have much value in those predictions either. That said, the social exchange model does not predict that punishment should be enacted strictly on the grounds of being wronged; since punishment is costly it should only be used when subjects hope to recoup the costs of that punishment in subsequent exchanges. If subjects do not wish to renegotiate the terms of cooperation via punishment, they should simply opt to refrain from interacting with their partner altogether.

That precise pattern of results was borne out: when a subject were defected on and the subject then punished the defector, that same subject was also likely to cooperate in subsequent rounds with their partner. In fact, they were just as likely to cooperate with their partner as they were cases where the partner did not initially defect. It’s worth repeating that subjects did this while, apparently, ignoring how their partner had behaved towards anyone else. Subjects only seemed to punish the partner in order to persuade their partner to treat them better; they did not punish because their partner had hurt anyone else. Finally, first-party reputation, unlike third-party reputation, had an effect on whether subjects were willing to cooperate with their partner on their first move in the trust game. People were more likely to cooperate with a partner who had cooperated with them, irrespective of how that partner behaved towards anyone else.

Let’s see you work that into your group selection theory.

To sum up, despite group selection models predicting that subjects should make use of first- and third-party information equally, or at least jointly, they did not. Subjects only appeared to be interested in information about how their partner behaved towards others to the extent that such information might predict how their partner would behave towards them. However, since information about how their partner had behaved towards them is a superior cue, subjects made use of that first-party information when it was available to the exclusion of third-party reputation.

Now, one could make the argument that you shouldn’t expect to see subjects making use of information about how their partners behaved towards other parties because there is no guarantee that those other parties were members of the subject’s group. After all, according to group selection theories, altruism should only be directed at members of one’s own group specifically, so maybe these results don’t do any damage to the group selectionist camp. I would be sympathetic to that argument, but there are two big problems to be dealt with before I extend that sympathy: first, it would require that group selectionists give up all the previously ambiguous evidence they have said is consistent with their theory, since almost all of that research does not explicitly deal with a subject’s in-group either; they don’t get to recognize evidence only in cases where it’s convenient for their theory and ignore it when it’s not. The second issue is the one I raised back in May: “the group” is a concept that tends to lack distinct boundaries. Without nailing down this concept more concretely, it would be difficult to build any kind of stable theory around it. Once that concept had been developed more completely, then it would need to be shown that subjects will act altruistically towards their group (and not others) irrespective of the personal payoff for doing so; demonstrating that people act altruistically with the hopes that they will be benefited down the road from doing so is not enough.

Will this study be the final word on group selection? Sadly, probably not. On the bright side, it’s at least a step in the right direction.

References: Krasnow, M.M., Cosmides, L., Pederson, E.J., & Tooby, J. (2012). What are punishment and reputation for? PLOS ONE, 7

Group Selectionists Make Basic Errors (Again)

In my last post, I wrote about a basic error most people seem to make when thinking about evolutionary psychology: they confuse the ultimate adaptive function of a psychological module with the proximate functioning of said module. Put briefly, the outputs of an adapted module will not always be adaptive. Organisms are not designed to respond perfectly to each and every context they find themselves in. This is especially the case regarding novel environmental contexts. These are things that most everyone should agree on, at least in the abstract. Behind those various nods of agreement, however, we find that applying this principle and recognizing maladaptive or nonfunctional outputs is often difficult for people in practice, laymen and professional alike. Some of these professionals, like Gintis et al (2003), even see fit to publish their basic errors.

Thankfully for the authors, the paper was peer reviewed by people who didn’t know what they were talking about either

There are two main points to discuss about this paper. The first point is to consider why the authors feel current theories are unable to account for certain behaviors, and the second is to consider the strength of the alternative explanations put forth. I don’t think I’m spoiling anything by saying the authors profoundly err on both accounts.

On the first point, the behavior in question – as it was in the initial post – is altruism. Gintis et al (2003) discuss the results of various economic games showing that people sometimes act nicely (or punitively) when niceness (or punishment) doesn’t end up ultimately benefiting them. From these maladaptive (or what economists might call “irrational”) outcomes, the authors conclude, therefore, that cognitive adaptations designed for reciprocal altruism or kin selection can’t account for the results. So right out of the gate they’re making the very error the undergraduates were making. While such findings would certainly be a problem for any theory that purports humans will always be nice when it pays more, and will never be nice when it pays less, and are always able to correctly calculate which situation is which, neither theory presumes any of those things. Unfortunately for Gintis et al, their paper does make some extremely problematic assumptions, but I’ll return to that point later.

The entirety of the argument that Gintis et al (2003) put forth rests on the maladaptive outcomes that are obtained in these games cutting against the adaptive hypothesis. As I covered previously, this is bad reasoning; brakes on cars sometimes fail to stop the car because of contextual variables – like ice – but that doesn’t mean that brakes aren’t designed to stop cars. One big issue with the maladaptive outcomes Gintis et al (2003) consider is that they are largely due to issues of novel environmental contexts. Now, unlike the undergraduate tests I just graded, Gintis et al (2003) have the distinct benefit of being handed the answer by their critics, which are laid out, in text, as such:

Since the anonymous, nonrepeated interactions characteristic of experimental games were not a significant part of our evolutionary history, we could not expect subjects in experimental games to behave in a fitness-maximizing manner. Rather, we would expect subjects to confuse the experimental environment in more evolutionarily familiar terms as a nonanonymous, repeated interaction, and to maximize fitness with respect to this reinterpreted environment.

My only critique of that section is the “fitness maximizing” terminology. We’re adaptation executioners, not fitness maximizers. The extent that adaptions maximize fitness in the current environment is an entirely separate questions to how we’re designed to process information. That said, the authors reply to the critique thusly:

But we do not believe that this critique is correct. In fact, humans are well capable of distinguishing individuals with whom they are likely to have many future interactions, from others, with whom future interactions are less likely

Like the last post, I’m going to rephrase the response in terms of arousal to pornography instead of altruism to make the failings of that argument clearer: “In fact, humans are well capable of distinguishing [real] individuals with whom they are likely to have [sex with], from [pornography], with [which] future [intercourse is] less likely.”

I suppose I should add a caveat about the probability of conception from intercourse…

Humans are well capable of distinguishing porn from reality. “A person” “knows” the difference between the two, so arousal to pornography should make as little sense as sexual arousal to any other inanimate object, like a chair or a wall. Yet people are routinely aroused by pornography. Are we to conclude from this, as Gintis et al might, that, therefore sexual arousal to pornography is itself functional? The proposition seems doubtful. Likewise, when people take birth control, if “they” “know” that they can’t get pregnant, why do they persist in having sex?

A better explanation is that “a person” is really not a solitary unit at all, but a conglomeration of different modules, and not every module is going to “know” the same thing. A module generating arousal to visual depictions of intercourse might not “know” the visual depiction is just a simulation, as it was never designed to tell the difference, since there never was a difference. The same goes for sex and birth control. That the module that happens to be talking to other people can clearly articulate that it “knows” the sex on the screen isn’t real, or that it “knows” it can’t increase its fitness by having sex while birth control is involved, other modules, could they speak, would give a very different answer. It seems Gintis et al (2003) fail to properly understand, or at least account for, modularity.

Maybe people can reliably tell the difference between those with whom they’ll have future contact and those with whom they likely won’t. Of course, there are always risks that module will miscalculate given the uncertainty of the future, but that task might have been something that a module could plausibly have been designed to do. What modules were unlikely to be designed to do, however, is interact with people anonymously, much less interact anonymously under the specific set of rules put forth in these experimental conditions. Gintis et al (2003) completely avoid this point in their response. They are talking about novel environmental contexts, and are somehow surprised when the mind doesn’t function perfectly in them. Not only do they fail to make use of modularity properly, they fail to account for novel environments as well.

So the problem that Gintis et al see is not actually a problem. People don’t universally behave as Gintis et al (2003) think other models predict they should. Of course, the other models don’t make those predictions, but there’s an even larger issue looming: the solution to this non-problem that Gintis et al favor introduces a greater, actual issue. This is the big issue I alluded to earlier: the “strong reciprocity” trait that Gintis et al (2003) put forth does make some very problematic assumptions. A little juxtaposition will let one stand out, like something a good peer reviewer should have noted:

One such trait, which we call strong reciprocity (Gintis, 2000b; Henrich et al., 2001), is a predisposition to cooperate with others and to punish those who violate the norms of cooperation, at personal cost, even when it is implausible to expect that these costs will be repaid either by others or at a later date…This is not because there are a few ‘‘bad apples’’ among the set of employees, but because only 26% of employees delivered the level of effort they promised! We conclude that strong reciprocators are inclined to compromise their morality to some extent, just as we might expect from daily experience. [emphasis mine]

So the trait being posited by the authors allows for cooperation even when cooperating doesn’t pay off. Leaving aside whether such a trait is plausibly something that could have evolved, indifference to cost is supposed to be part of the design. It is thus rather strange that the authors themselves note people tend to modify their behavior in ways that are sensitive to those costs. Indeed, only 1 in 4 of the people in the experiment they mention could even potentially fit the definition of a strong altruist, even (and only) if the byproducts of reciprocal altruism modules counted for absolutely nothing.

25% of the time, it works 100% of the time

It’s worth noticing the trick that Gintis et al (2003) are trying to use here as well: they’re counting the hits and not the misses. Even though only a quarter of the people could even potentially (and I do stress potentially) be considered strong reciprocators that are indifferent to the costs and benefits, they go ahead a label the employees strong reciprocators anyway (just strong reciprocators that do things strong reciprocators aren’t supposed to do, like be sensitive to costs and benefits). Of course, they could more parsimoniously be labeled reciprocal altruists who happen to be behaving maladaptively in a novel circumstance, but that’s apparently beyond consideration.

References: Gintis, H., Bowles, S., Boyd, R., & Fehr, E. (2003). Explaining altruistic behavior in humans Evolution and Human Behavior, 24 (3), 153-172 DOI: 10.1016/S1090-5138(02)00157-5

No, Really; Group Selection Doesn’t Work.

Group selection is kind of like the horror genre of movies: even if the movie was terrible, and even if the main villain gets killed off, you can bet there will be still be a dozen sequels. Like the last surviving virgin among a group of teenage campers, it’s now on my agenda to kill this idea once again, because it seems the first couple dozen times it was killed, the killing just didn’t stick. Now I have written about this group selection issue before, but only briefly. Since people seem to continue and actually take the idea seriously, it’s time to explicitly go after the fundamental assumptions made by group selection models. Hopefully, this will put the metaphorical stake through the heart of this vampire, saving me time in future discussions as I can just link people here instead of rehashing the same points over and over again. It probably won’t, as some people seem to like group selection for some currently unknown reason, but fingers crossed anyway.

Friday the 13th, part 23: At this point, you might as well just watch the first movie again, because it’s the same thing.

Recently, Jon Gotschall wrote an article for Psychology Today about how E.O. Wilson thinks the selfish gene metaphor is a giant mistake. As he didn’t explicitly say this idea is nonsense – the proper response – I can only assume he is partially sympathetic to group selection. Et tu, Jon? There’s one point from that article I’d like to tackle first, before moving onto other, larger matters. Jon writes the following:

In effect, this defined altruism-real and authentic selflessness–out of existence. On a planet ruled by selfish genes, “altruism” was just masked selfishness.

The first point is that I have no idea what Jon means when he’s talking about “real” altruism. His comments there conflate proximate and ultimate explanations, which is a mistake frequently cautioned against in your typical introductory level evolutionary psychology course. No one is saying that other-regarding feelings don’t exist at a proximate level; they clearly do. The goal is explain what the ultimate function of such feelings are. Parents genuinely tend to feel selfless and act altruistically towards their children. That feeling is quite genuine, and it happens to exist in no small part because that child carries half of that parent’s genes. By acting altruistically towards their children, parents are helping their own genes reproduce; genes are benefiting copies of themselves that are found in other bodies. The ultimate explanation is not privileged over the proximate one in terms of which is “real”. It makes no more sense to say what Jon did than for me to suggest that my desire to eat chocolate cake is really a reproductive desire, because, eventually, that desire had to be the result of an adaptation designed to increase my genetic fitness. Selfish genes really can create altruistic behavior, they just only do so when the benefits of being altruistic tend to outweigh the costs in the long run.

Speaking of benefits outweighing the costs, it might be helpful to take a theoretical step back and consider why an organism would have any interest in joining a group in the first place. Here are two possible answers: (1) An organism can benefit in some way by entering into a coalition with other organisms, achieving goals it otherwise could not, or (2) an organism joins a group in order to benefit that group, with no regard for its own interests. The former option seems rather plausible, representing cases like reciprocal altruism and mutualism, whereas the latter option does not appear very reasonable. Self-interest wins the day over selflessness when it comes to explaining why an organism would bother to join a group in the first place. Glad we’ve established that. However, to then go on to say that, once it has joined a coalition, an organism converts its selfish interests to selfless ones is to now, basically, endorse the second explanation. It doesn’t matter to what extent you think an organism is designed to do that, by the way. Any extent is equally as problematic.

You will need a final answer at some point: Selfish, or Selfless?

But organisms do sometimes seem to sometimes put their own interests aside to benefit members of their group, right? Well, that’s going to depend on how you’re conceptualizing their interests. Let’s say I’m a member of a group that demands a monthly membership fee, and, for the sake of argument, this group totally isn’t a pornography website. I would be better off if I could keep that monthly membership fee to myself, so I must be acting selflessly by giving it to the group. There’s only one catch: if I opt to not pay that membership fee, there’s a good chance I’ll lose some or all of the benefits that the group provides, whatever form those benefits come in. Similarly, whether through withdrawal of social support or active punishment, groups can make leaving or not contributing costlier than staying and helping. Lacking some sort of punishment mechanism, cooperation tend to fall apart. The larger point there here is that if by not paying a cost, you end up paying an even larger cost, that’s not exactly selfless behavior requiring some special explanation.

Maybe that example isn’t fair though; what about cases like when a soldier jumps on a grenade to save his fellow soldiers? Well, there are a couple of points to make about the grenade-like examples: first, grenades are obviously an environmental novelty. Humans just aren’t adapted to an environment containing grenades and, I’m told, most of us don’t make a habit of jumping into dangerous situations to help others, blind to the probably of injury to death. That said, if you had a population of soldiers, some of which had a heritable tendency to jump on grenades to save others, while other soldiers had no such tendency, if grenades kept getting thrown at them, you could imagine which type would tend to out-reproduce the other, all else being equal. A second vital point to make is that every single output of an cognitive adaptation need not be adaptive; so long as whatever module led to such a decision tended to be beneficial overall, it would still spread and be maintained throughout the population, despite occasional maladaptive outcomes. Sometimes a peacock’s large tail spells doom for the bird who carries it as it is unable to escape from a predator, but that does mean, on the whole, any one bird would be better suited to just not bother growing their tail; it’s vital for attracting a mate, and surviving means nothing absent reproduction.

Now, onto the two major theoretical issues with group selection itself. The first is displayed by Jon in his article here:

Let’s run a quick thought experiment to see how biologists reached this conclusion. Imagine that long before people spread out of Africa there was a tribe called The Selfless People who lived on an isolated island off the African coast. The Selfless People were instinctive altruists, and their world was an Eden. 

The thought experiment is already getting ahead of itself in a big way. In this story, it’s already assumed that a group of people exist with these kind of altruistic tendencies. Little mind is paid to how the members of this group came to have these tendencies in the first place, which is a rather major detail, especially because, as many note, within groups selfishness wins. Consider the following: in order to demonstrate group selection, you would need a trait that conferred group-level fitness benefits at individual-level fitness costs. If the trait benefited the individual bearer in any way, then it would spread through standard selection and there would be no need to invoke group-level selection. So, given that we’re, by definition, talking about a trait that actively hinders itself getting spread in order to benefit others, how does that trait spread throughout the population resulting in a population of ‘selfless people’? How do you manage to get from 1 to 2 by way of subtraction?

Perhaps it’s from all that good Karma you build up?

No model of group selection I’ve come across yet seems to deal with this very basic problem. Maybe there are accounts out there I haven’t read that contain the answer to my question; maybe the accounts I have seen have an answer that I’ve just failed to understand. Maybe. Then again, maybe none of the accounts I read have actually provided a satisfying answer because they start with the assumption that the traits they’re seeking to prove exists already exists in some substantial way. That kind of strikes me as cheating. Jon’s thought experiment certainly makes that assumption. The frequently cited paper by Boyd and Richardson (1990) seems to make that assumption as well; people who act in favor of their group selflessly just kind of exist. That trait needs an explanation; simply assuming it into existence and figuring out the benefits from that point is not good enough. There’s a chance that the trait could spread by drift, but drift has, to the best of my knowledge, never been successfully invoked to explain the existence of any complex adaption. Further, drift only really works when a trait is, more or less, reproductively neutral. A trait that is actively harmful would have a further hurdle to overcome.

Now positing an adaptation designed to deliver fitness benefits to others at fitness costs to oneself might seen anathema to natural selection, because it is, but the problems don’t stop there. There’s still another big issue looming: how we are to define the group itself; you know, the thing that’s supposed to be receiving these benefits. Like many other concepts, what counts as a group – or a benefit to a group – can be fuzzy and is often arbitrary. Depending on what context I currently find myself in, I could be said to belong to an almost incalculably large number of potential groups, and throughout the course of my life I will enter and leave many explicitly and implicitly. Some classic experiments in psychology demonstrate just how readily group memberships can be created and defined. I would imagine that for group selection to be feasible, at the very least, group membership needs to be relatively stable; people should know who their “real” group is and act altruistically towards it, and not other groups. Accordingly, I’d imagine group membership should be a bit more difficult to just make up on the spot. People shouldn’t just start classifying themselves into groups on the basis of being told, “you are now in this group” anymore than they should start thinking about a random woman as their mother because someone says, “this woman is now your mother” (nor would we expect this designated mother to start investing in this new person over her own child). That group membership is relatively easy to generate demonstrates, in my mind, the reality that group membership is a fuzzy and fluid concept, and, subsequently, not the kind of thing that can be subject to selection.

Now perhaps, as Jon suggested, the selfless people will always win against the selfish people. It’s an possible state of affairs, sure, but it’s important to realize that it’s an assumption being made, not a prediction being demonstrated. Such conditions can be artificially created in the lab, but whether they exist in the world, and, if they do, how frequently they appear, is another matter entirely. The more general point here is that group selection can work well in the world of theory, but that’s because assumptions are made there that define it as working well. Using slightly tweaked sets of assumptions, selfless groups will always lose. They win when they are defined as winning, and lose when they are defined as losing. Using another set of assumptions, groups of people with psychic abilities win against groups without them. The key then, is to see how these states of affairs hold up in real life. If people don’t have psychic abilities, or if psychic abilities are impossible for one reason or another, no number of assumptions will change that reality.

Finally, the results of thought experiments like the foot-bridge dilemma seem to cut against the group selection hypothesis: purposely sacrificing one person’s life to save the lives of five others is, in terms of the group, the better choice, yet people consistently reject this course of action (there, B=5, C=1). When someone jumps on a grenade, we praise them for it; when someone throws another person on a grenade, we condemn them, despite this outcome being better from the group perspective (worst case, you’ve kill a non-altruist who wouldn’t jump on it anyway, best case, you helped an altruist act). Those outcomes conflict with group selection predictions, which, I’d think, should tend to favor more utilitarian calculations – the ones that are actually better for a group. I would think it should also predict Communism would work out better than it tends to, or that people would really love to pay their taxes. Then again, group selection doesn’t seem to be plausible in the first place, so perhaps result like these shouldn’t be terribly surprising.

References: Boyd, R., & Richerson, P.J. (1990). Group selection among alternative evolutionary stable strategies. Journal of Theoretical Biology, 145, 331-342.

Somebody Else’s Problem

Let’s say you’re a recent high-school graduate at the bank, trying to take out a loan to attend a fancy college so you can enjoy the privileges of complaining about how reading is, like, work, and living the rest of life in debt from student loans. A lone gunman rushes into the bank, intent on robbing the place. You notice the gunman is holding a revolver, meaning he only has six bullets. This is good news, as there happen to be 20 people in the bank; if you all rush him at the same time, he wouldn’t be able to kill more than six people, max; realistically, he’d only be able to get off three or four shots before he was taken down, and there’s no guarantee those shots will even kill the people they hit. The only practical solution here should be to work together to stop the robbery, right?

Look on the bright side: if you pull through, this will look great on your admissions essay.

The idea that evolutionary pressures would have selected for such self-sacrificing tendencies is known as “group selection”, and is rightly considered nonsense by most people who understand evolutionary theory. Why doesn’t it work? Here’s one reason: let’s go back to the bank. The benefits of stopping the robbery will be shared by everyone at the abstract level of the society, but the costs of stopping the robbery will be disproportionately shouldered by those who intervene. While everyone is charging the robber, if you decide that you’re quite comfortable hiding in the back, thank you very much, your chances of getting shot decline dramatically and you still get the benefit; just let it be somebody else’s problem. Of course, most other people should realize this as well, leaving everyone pretty uninclined to try and stop the robbery. Indeed, there are good reasons to suspect that free-riding is the best strategy (Dreber et al., 2008).

There are, unfortunately, some people who think group selection works and actually selects for tendencies to incur costs at no benefit. Fehr, Fischbacher, and Gachter (2002) called their bad idea “strong reciprocity”:

“A person is a strong reciprocator if she is willing to sacrifice resources (a) to be kind to those who are being kind… and (b) to punish those who are being unkind…even if this is costly and provides neither present nor future material rewards for the reciprocator” (p.3, emphasis theirs)

So the gist of the idea would seem to be (to use an economic example) that if you give away your money to people you think are nice – and burn your money to ensure that mean people’s money also gets burned – with complete disregard for your own interests, you’re going to somehow end up with more money. Got it? Me neither…

“I’m telling you, this giving away cash thing is going to catch on big time”

So what would drive Fehr, Fischbacher, and Gachter (2002) to put forth such a silly idea? They don’t seem to think existing theories – like reciprocal altruism, kin selection, costly signaling theory, etc – can account for the way people behave in laboratory settings. That, and existing theories are based around selfishness, which isn’t nice, and the world should be a nicer place. The authors seems to believe that those previous theories lead to predictions like: people should “…always defect in a sequential, one-shot [prisoner's dilemma]” when playing anonymously.That one sentence contains two major mistakes: the first mistake is that those theories most definitely do not say that. The second mistake is part of the first: they assume that people’s proximate psychological functioning will automatically fall in-line with the conditions they attempt to create in the lab, which it does not (as I’ve mentioned recently). While it might be adaptive, in those conditions, to always defect at the ultimate level, it does not mean that the proximate level will behave that way. For instance, it’s a popular theory that sex evolved for the purposes of reproduction. That people have sex with birth control does not mean the reproduction theory is unable to account for that behavior.

As it turns out, people’s psychology did not evolve for life in a laboratory setting, nor is the functioning of our psychology going to be adaptive in each and every context we’re in. Were this the case, returning to our birth control example, simply telling someone that having sex when the pill is involved removes the possibility of pregnancy would lead to people to immediately lose all interest in the act (either having sex or using the pill). Likewise, oral sex, anal sex, hand-jobs, gay sex, condom use, and masturbation should all disappear too, as none are particularly helpful in terms of reproduction.

Little known fact: this parade is actually a celebration of a firm understanding of the proximate/ultimate distinction. A very firm understanding.

Nevertheless, people do cooperate in experimental settings, even when cooperating is costly, the game is one-shot, there’s no possibility of being punished, and everyone’s ostensibly anonymous. This poses another problem for Fehr and his colleagues: their own theory predicts this shouldn’t happen either. Let’s consider an anonymous one-shot prisoner’s dilemma with a strong reciprocator as one of the players. If they’re playing against another strong reciprocator, they’ll want to cooperate; if they’re playing against a selfish individual, they’ll want to defect. However, they don’t know ahead of time who they’re playing against, and once they make their decision it can’t be adjusted. In this case, they run the risk of defecting on a strong reciprocator or benefiting a selfish individual while hurting themselves. The same goes for a dictator game; if they don’t know the character of the person they’re giving money to, how much should they give?

The implications of this extend even further: in a dictator game where the dictator decides to keep the entire pot, third-party strong reciprocators should not really be inclined to punish. Why? Because they don’t know a thing about who the receiver is. Both the receiver and dictator could be selfish, so punishing wouldn’t make much sense. The dictator could be a strong reciprocator and the receiver could be selfish, in which case punishment would make even less sense. Both could be strong reciprocators, unsure of the others’ intentions. It would only make sense if the dictator was selfish and the receiver was a strong reciprocator, but a third-party has no way of knowing whether or not that’s the case. (It also means if strong reciprocators and selfish individuals are about equal in the population, punishment in these cases would be a waste three-forths of the time – maybe half at best, if they want to punish selfish people no matter who they’re playing against – meaning strong reciprocator third-parties should never punish).

There was some chance he might of been being a dick…I think.

The main question for Fehr and his colleagues would then not be, “why do people reciprocate cooperation in the lab” – as reciprocal altruism and the proximate/ultimate distinction can already explain that without resorting to group selection – but rather, “why is there any cooperation in the first place?” The simplest answer to the question might seem to be that some people are prone to give the opposing player the benefit of the doubt and cooperate on the first move, and then adjust their behavior accordingly (even if they are not going to be playing sequential rounds). The problem here is that this is what a tit-for-tat player already does, and it doesn’t require group selection.

It also doesn’t look good for the theory of social preferences invoked by Fehr et al. (2002) when the vast majority of people don’t seem to have preferences for fairness and honesty when they don’t have to, as evidenced by 31 of 33 people strategically using an unequal distribution of information to their advantage in ultimatum games (Pillutla and Murnighan, 1995). In every case Fehr et al. (2002) looks at, outcomes have concrete values that everyone knows about and can observe. What happens when intentions can be obscured, or values misrepresented, as they often can be in real life? Behavior changes and being a strong reciprocator would be even harder. What might happen when the cost/punishment ratio changes from a universal static value, as it often does in real life (not everyone can punish others at the same rate)? Behavior will probably change again.

Simply assuming these behaviors are the result of group selection isn’t enough.The odds are better that the results are only confusing when their interpreter has an incorrect sense of how things should have turned out.

References: Dreber, A., Rand, D.G., Fudenberg, D, & Nowak, M.A. (2008). Winners don’t punish. Nature, 452,  348-351

Fehr, E., Fischbacher, U., & Gachter, S. (2002). Strong reciprocity, human cooperation, and the enforcement of social norms. Human Nature, 13, 1-25.

Pillutla, M.M. & Murnighan, J.K. (1995). Being fair or appearing fair: Strategic behavior in ultimatum bargaining. Academy of Management Journal, 38, 1408-1426.