Payback without bookkeeping: The origins of revenge and retaliation

ABSTRACT

Current evolutionary models of revenge focus on its complex deterrent functions. Nevertheless, there are some retaliatory behaviors in nonhuman animals that do not appear to have a deterrent function and that are notably similar to revenge in their motivational structure. I propose a novel explanation for the origin of revenge motives that does not rely on deterrence, and I argue that revenge motives as understood on this model could easily be co-opted to serve subsequent functions of deterrence. This explanation appeals to a well-understood family of game-theoretic models known as the ‘war of attrition.’ One theoretical implication concerns the starting point for evolutionary psychology. Starting with an understanding of how human adaptations might have arisen from more ancient mechanisms and selection pressures provides helpful constraints and correctives.

KEYWORDS:

• Revenge
• evolution of punishment
• evolutionary psychology
• war of attrition
• anger
• emotion

Acknowledgments

This paper would not be what it is without the guidance, encouragement, and feedback from the following: Justin Bruner, John Doris, Robert Fischer, Robert Kurzban, Edouard Machery, Ron Mallon, Joseph McCaffrey, Adam Morris, Burkay Ozturk, Felipe Romero, Rory Smead, and a generous reviewer for Philosophical Psychology.

Disclosure statement

No potential conflict of interest was reported by the author.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

1. I make no assumptions about the relations between these motives. They may be different names for one and the same trait, or they may be derivatives of a single, ancestral trait. They may also have functions aside from payback, but this depends on how widely one understands payback. For example, I am inclined to categorize “vending machine rage” as part of the phenomenon of payback.

2. Many cast revenge as an explicitly social and highly intellectual phenomenon. For example, Frijda suggests that it includes appraisals regarding social comparison (Frijda, 1994, p. 274). While I agree that this is an interesting and important phenomenon, it is not my focus here. Rather, my interest is in payback motives conceived more broadly, of which revenge is but one species. I also doubt that payback is strongly linked to social comparisons (see Footnote 4).

3. This is perfectly consistent with saying that the function of revenge is to secure some future benefit for the agent. The distinction between proactive and reactive aggression concerns the agent’s motive rather than the evolved function of that motive.

4. This is not to say that aggression against others cannot be learned. For example, the phenomenon of shadenfreude consists in pleasure at the pain of another (usually a member of an outgroup). Experiencing such pleasure passively may very well reinforce actions that cause suffering in another (Cikara, 2018; Cikara & Fiske, 2013; Leach & Spears, 2008). In any case, I am uncertain whether this phenomenon should be lumped together with revenge, though the two certainly have common elements. If revenge is an instinctual motivation, then this kind of “payback” almost certainly has a distinct motivational structure.

5. See Daly and Wilson (1988, p. 226). See also McCullough et al. (McCullough et al., 2012). A related consideration is that norms prohibiting or restricting revenge are very widespread (cf. Daly & Wilson, 1988, Chapter 10). Frank (1988) draws out the implications of this fact with characteristic elegance (Frank, 1988, p. 39).

6. Nevertheless, learning and culture clearly do explain much about the current function of payback motives and their functions across different cultures (e.g., Nisbett & Cohen, 1996).

7. For ease of expression, I will use “revenge” interchangeably with “payback motives.”

8. In other words, payback motives are not necessarily negative social motives which focus on satisfying negative other-regarding preferences, for example, the preference that another person suffers (cf. Jensen, 2010, p. 2643).

9. Another caveat is that a spiteful motive may be triggered by imperfect indicators of far-off benefits without an agent having the goal of bringing about those benefits. On my view, spite is about the motivation that guides ongoing behavior, rather than the situation that triggers the motivation. Accordingly, my talk of “calculation” has to do with the ongoing motivation to aggress, rather than that which triggers aggression.

10. For example, Frank (1988, pp. 36–37) criticizes Trivers’s notion of moralistic aggression on the grounds that in many situations, this motive appears to be less optimal (and no simpler) than a tit-for-tat strategy, which is directly in line with self-interest. That is, if material benefits were one’s only motivation, then one would only punish a defector to achieve a higher payout, to cut losses, or to coerce a partner to cooperate (in accordance with a tit-for-tat policy).

11. Hence, I leave aside a great deal of important work on punishment in evolutionary game theory, such as games in which the immediate benefits of punishment do outweigh its costs (cf. the discussion of “loss-cutting” strategies in Nakao & Machery, 2012). I leave aside other work because I do not find it adequate for explaining the origins of payback motives as opposed to their subsequent modification (e.g., Clutton-Brock & Parker, 1995; Nowak & Sigmund, 2005).

12. I do not intend “individualistic” to rule out increases in inclusive fitness that sometimes translate to increased reproductive fitness in an individual’s kin. Similarly, I do not mean to rule out retaliation or revenge at the level of groups. First, retaliation of a group against another may exist because of a pre-existing tendency of individuals to retaliatiate. Second, group retaliation may exist because of the fitness benefits it confers on individuals.

13. Corvids are an obvious exception. Nevertheless, the mindreading capacities of corvids are probably best explained by parallel evolution rather than common descent. Thus, the evolution of complex bookkeeping capacities in corvids is irrelevant to its origination in mammalian species.

14. Dawkins, however, expresses skepticism at the idea of a population-level function. He would also likely balk at calling anything a population-level adaptation, as I do below. Regardless, we can agree that the aggression of owners or residents is not an adaptation that exists to benefit individuals.

15. See Lloyd (2007) for an overview of the “units of selection” debate that sheds light on this distinction.

16. This is not at all clear for polymorphic ESSs, where there is a stable equilibrium with two or more strategies. In that case, the benefits accrue to more than one trait, and no single, heritable trait is the source of the benefit. Thus, neither trait appears to be an adaptation.

17. Aaron Sell (2005) has proposed an evolutionary model of anger that also appeals to the war of attrition model. Nevertheless, Sell’s model falls squarely within the family of recalibrational models discussed above and thus falls prey to the criticisms offered in Section 4. Gintis (2006) and DeScioli and Wilson (2011) have also referred to this model to explain patterns of territorial behavior in humans. Nevertheless, their efforts have been directed at explaining peoples’ desires to keep what they have (i.e., endowment effects) or the fact that they are usually successful in doing so, rather than the motivational states that lead one to defend what one has. My contention here is that revenge is one such motivational state.

18. For instance, Archer and Huntingford (1994) discuss the application of the sequential assessment model to escalated aggressive encounters (e.g., Enquist & Leimar, 1983). I consider this model to be an extension of the generalized war of attrition discussed below in that it assumes that costs build up over time and that assessment of asymmetries in resource-holding power are decisive in determining contest outcomes. The main difference is that the sequential assessment model assumes that the costs of fighting (e.g., the risk of injury) increase for both contestants, sometimes in stages, as the contest proceeds and that assessment strategies use information obtained during the competition. Something like a reserve strategy (discussed below) remains intact, and that is why “bluff” strategies (which occur beyond the stage at which a contestant assesses its resource-holding power to be lower than that of its competitor) are unstable as a rule.

19. It may be that payback motives also implement strategies in discrete games like hawk-or-dove. Nevertheless, these games seem to me too idealized to capture important dimensions of animal conflict (cf. fn. 22). Moreover, such models make it more difficult to isolate and explain critical strategic elements (i.e., the reserve strategy).

20. See also Skyrms (1996, Chapter, p. 4) for a helpful discussion of correlated conventions that break symmetry.

21. As Skyrms (1996) puts it, once the correlated costs are figured in, “the basin of attraction of the bourgeois equilibrium will now be larger than that of the paradoxical strategy” (p. 78). See also Dawkins (2006, p. 81).

22. It will be obvious to some that the war of attrition model is not the only one in which symmetry can be broken by correlated convention, and perhaps not the only model in which the convention must be enforced. By focusing on the war of attrition model as opposed to, for instance, hawk-or-dove, I have suggested that the origin of revenge derives from temporally extended contests; however, does the argument generalize to discrete or iterated games in which conventions break symmetry? It may. However, I have a misgiving about explaining the origins of revenge in terms of the iterated hawk-or-dove game, which is that iterated games simply do not apply to a large class of animal conflicts. The hawk strategy is usually conceptualized as a discrete decision to “escalate and continue until either opponent retreats or until injured” (Maynard Smith & Parker, 1976, p. 161). However, almost all animal conflicts have the possibility of temporal extension. Insofar as injury is unlikely to occur in the initial moment of a contest, the space of strategies expands to include decisions about not just whether to escalate but also how long to persist, and thus, in these conditions, the interaction can be reduced to a war of attrition after all. The same is true of various iterated games, such as the retaliation game (cf. Maynard Smith & Parker, 1976, p. 173).

23. In reality, I am not claiming that one cannot learn to follow a bourgeois convention (see, e.g., Skyrms, 1996, Chapters 71–75). However, I do claim that the bourgeois reserve strategy as evaluated in war of attrition models cannot be learned. That is, learning from individual experience would not tend to converge on the bourgeois reserve strategy (as it is ordinarily defined) across all of the conditions required for its stability. Since learning would not reliably produce the relevant phenotype, the bourgeois reserve strategy cannot develop by this mechanism. If one asks why these models should not include reserve-learning in their strategy space, there are two reasons. First, the symmetric war of attrition seems to me the most plausible model for resource competition at the outset. Moreover, it is difficult to see how the mixed ESS (to pick a duration interval from a certain probability distribution) could be learned in this game. Moreover, the mixed ESS in the symmetric game is identical to the reserve strategy in ESSs in the asymmetric war of attrition. Thus, the most plausible evolutionary trajectory from the symmetric game to the asymmetric game is a reserve strategy that is not learned. Second, recent work also suggests that punishment is unlikely to be learned when it is costly, and strategies that always punish without learning are more likely to evolve if stealing resources (similar to violating a convention) is rewarding (Morris, Macglashan, Littman, & Cushman, 2017).

24. Another asymmetry involves the value of a resource for an individual (Grafen, 1987; e.g., Parker & Rubenstein, 1981). If an individual has a greater need for food, for instance, the value of a given food item will be greater for that individual than for an individual who is less hungry.

25. For a formal description of these models, see Parker and Rubenstein (1981, pp. 223–225).

26. This model addresses a worry about using the ESS methodology in the war of attrition (cf. Huttegger, 2010; Huttegger & Zollman, 2012): The worry is that the bourgeois reserve strategy may not be an ESS against a simple bourgeois strategy, which respects the convention but does not play reserve. That is, at many possible population states, there will be no behavioral differences between the strategies that play reserve and those that do not. If mutant strategies do not invade and role assessment is perfect, then these strategies will look behaviorally identical. If so, this introduces the concern that the convention would not be stable against drift in the persistence-time of owners or intruders. Indeed, it is not stable under these conditions (Hammerstein & Parker, 1982). Haccou and Glaizot’s (2002) model resolves this worry for the generalized war of attrition by showing that when role perception is not perfect, something like the reserve strategy remains intact, even when a wider range of strategies are at play, at least under certain plausible assumptions (e.g., assumptions about the likelihood of mistakes).

27. Anger at loss of control develops quite early and only represents an implicit grasp of the first-possession convention. However, children develop a more explicit grasp of this convention as early as three or four years of age, when they rely heavily on the convention to make judgments about ownership that concern third parties (see Friedman & Neary, 2008).

28. The frustration–aggression link may very well have been established by a more ancient and general evolutionary problem of overcoming obstacles to one’s goals. This evolutionary problem is not essentially a social one, since the obstacle to one’s goal could just as easily be a rock as a conspecific. Nevertheless, revenge is a distinctively social interchange, so I would argue that a motive for aggression does not become a payback motive until it begins to be shaped for a distinctively social purpose. A related point concerns the nature of anger qua payback motive: It seems quite possible to me that anger has other functions aside from revenge. By calling anger a payback motive, I mean only that it has been shaped by selection to implement revenge, in addition to whatever function it already had or subsequently acquired.

29. One of the most prominent accounts of the nature of social norms makes central appeal to social expectations (Bicchieri, 2006).

30. I have focused here on how payback phenomena are shaped by modifications to bookkeeping capacities that serve as inputs to payback motives. However, there is much else to explain about the outputs of payback motives, including the many ways that payback is channeled or directed. For example, the aim of revenge can be to restore “karmic balance,” to adjust relative status, or even to balance one’s pain with the pain of a transgressor. While these are important facts to explain, it is reasonable to leave them for another time. By comparison, if I were explaining the ancient origins of hunger, it would be reasonable to leave aside the question of why hunger leads to differentiated food cravings (e.g., for double-chocolate-fudge ice-cream or ricotta-spinach gnocchis). Thanks to an anonymous referee for posing this problem.