Why do humans cooperate? Why do we behave "altruistically"? These are the sort of "big" questions which the human sciences explore. From the vantage point of evolutionary biology there has been a long history of exploring, and attempting to explain, altruistic behavior. And yet such questions weren't always considered of great value, the great W.D. Hamilton was discouraged from his exploration of this topic by his department head. At one point he took up carpentry to furnish for himself an income while he pursued his science in his spare time, so pessimistic was he about receiving academic backing for the questions which interested him. Of course, his two papers, The Genetical Evolution of Social Behavior, became widely cited classics (perhaps attested by the fact that Wikipedia has an entry devoted just to these two papers!). Hamilton explored inclusive fitness using the model of haplodiploid eusocial insects, and showed how Hamilton's Rule could predict the emergence of altruistic behavior amongst genetically related individuals (subsequent work tends to call into question the importance of haplodiploidy in the generation of eusociality via inclusive fitness, but if that is so, Hamilton was brilliantly wrong). These ideas triggered a revolution in the minds of many biologists, and a great deal of the exposition in Richard Dawkins' books are distillations and extensions of fundamentally Hamiltonian concepts. But in this post, I would like to explore the ideas of another individual who has contributed to our understanding of the evolution of sociality, Robert Trivers. Just as Hamilton's 1964 papers were seminal in our understanding of social evolution and its relation to genetics, so Trivers' 1971 paper The Evolution of Reciprocal Altruism was an important watershed in the understanding of the intersection between ethology and evolution. Just as Hamilton's work was like a hammer in the intellectual landscape of influential popularizers like Richard Dawkins, so Trivers' paper looms large in Steven Pinker's development. Like it or not, Trivers' paper foreshadows many elements of the evolutionary psychological toolkit, from the extrapolation from animal models to the usage of the prehistoric thought experiment. In Defenders of the Truth sociologist Ullica Segerstrale makes the case that Trivers was also a major engine behind various shifts in the thinking of E.O. Wilson during the early 1970s when the ideas which went into Sociobiology were crystallized. My own reading of Trivers' 1971 paper will be framed by his presentation, with a new biographical introduction and postscript, in his book Natural Selection and Social Theory. You may read the whole chapter online (PDF). I won't cover Trivers' biographical material much, except to repeat five points he offers in "How to Write a Classic Paper": 1) Pick an important topic. 2) Try to do a little sustained thinking on the topic, always keeping close to the task at hand. 3) Generalize outward from your chosen topic. 4) Write in the language of your discipline but, of course, try to do so simply and clearly. 5) If at all possible, reorganize existing evidence around your theory. With that in mind, on to Trivers' reciprocal altruism. What is it? And why did Trivers enter into this territory? As noted in the introduction clearly Trivers saw a gap in Hamilton's papers on the origins of altruism: why do unrelated individuals help each other? One of Trivers' empirical examples is between different species, who by definition are unrelated, at least to the degree of genetic affinity in which kin selection might be salient. The paper begins with a human example, which I think is influenced by Trivers' own biases in starting with humans and expanding outward, rather than using a "simple" animal as an initial model. The scenario is as follows: A) A man has a 1/2 chance of drowning as an unrelated man watches B) If the unrelated man intervenes the man who was in need has a 1/20 chance of drowning, as does the man who intervenes The issue is this: as humans I suspect the scenario recounted in A is one we find repugnant, and yet the cost vs. benefit mathematics seems to imply that that is exactly what we should do on first glance. Consider an allele, h for "help," and another, nh for "no help." If you assume additivity of effect model the phenotypes like so: hh ⇒ strong tendency to help (i.e., always help) hnh ⇒ mild tendency to help (i.e., help half the time, don't help the other half) nhnh ⇒ no tendency to help (i.e., never help) Clearly you have a situation where the fitness increases for every substitution of nh, and decreases for every substitution of h. Altruistic genes can't get a break! Of course, in the inclusive fitness models one might imagine that the drowner also carried h, and so altruistic alleles could increase their fitness because the cost is outweighed by the aggregate gain. But remember that the situation is contrived so that these two individuals are nonrelatives. Certainly aiding a brother or sister, or a child, seems to be a "no brainer." And yet our human intuition tells us that we also aid friends, or neighbors, in times of need. Why? The key is to think beyond one situation to repeated iterations. That is, one must step outside the frame of one circumstance where an individual is in need and another is capable of aiding, and realize that these situations occur commonly throughout the lifetime of an individual. Importantly, the one in need and the one capable of aiding are not fixed but shift between individuals, so an individual looking on might in the future be drowning himself! Now with this iterated scenario in mind let's reformulate our model. In one instance the cost to the helper is a 1/20 chance of death, while the benefit to the helped is nearly a doubled chance of survival. Now, imagine that the next instance is one where the one who was aided now can aid, and the one who helped is now at risk. The chances are inverted. Explicitly Iteration 1: Situation 1: Individual A is drowning, B watches on without intervening, A as 1/2 chance of survival, B has a 100% chance of survival Situation 2: Individual A is drowning, B intervenes, A & B have a 95% chance of survival. Iteration 2: Situation 1: Individual B is drowning, A watches on without intervening. B has 1/2 chance of survival, A has a 100% chance of survival Situation 2: Individual B is drowning, A intervenes, A & B have a 95% chance of survival. If the intervention of A upon B's behalf in iteration 2 is contingent upon B's intervention on behalf of A, assuming A survives, the "payoff" is clear insofar as the small risk of death for B in iteration 1 - situation 2 is offset by an enormous gain in iteration 2 - situation 2. Of course, there is a problem with this scenario: one can cheat. Consider individual A, who is aided by B in iteration 1, but simply watches on in iteration 2! This is obviously optimal in terms of survival likelihood, for those two situations!. Again, repeated iterations mean that uncertainty is introduced into the equation so that the payoff of cheating is balanced by the possibility that the individual who manages to survive in iteration 2 despite not being aided remembers the lack of altruism and no longer intervenes in a possible iteration 3. To further explore this issue Trivers presents three scenarios: 1) Random dispensation of altruism 2) Nonrandom dispensation by reference to kin 3) Nonrandom dispensation by reference to the altruistic tendencies of the recipient The second scenario is Hamiltonian, and can be ignored. The first scenario is clearly one in which altruistic tendencies simply cannot spread because non-altruists can "free ride" upon the good will of altruists, who make no attempt to combat this behavior. Finally, the last situation is one where altruists can spread because they may favor each other, and ignore non-altruists who attempt to free ride. The last situation is obviously one that implies important constraining parameters. Trivers suggests that length of lifetime, dispersal rate, degree of mutual dependence, parental care, dominance hierarchies and aid in combat are all issues that can be viewed through the reciprocal altruism lens, and weight the die in regards to how likely such behavior is. Organisms who are long lived and interact many times with a finite number of conspecifics who they are familiar with and engage in social behavior are clearly prime candidates for instances of reciprocal altruism. In contrast, ephemeral creatures which do not interact with conspecifics are clearly not good candidates. Trivers also points out interesting angles such as the fact that older individuals whose reproductive value potential is far lower might have less incentive to behave altruistically toward non-relatives as their own future payoffs are diminished by their shorter time window. The necessity for symmetry of behavior, that is, one who is aided may aid in kind to a similar degree, makes Trivers cautious about possibilities for altruism between individuals who are embedded in a strict and vertical hierarchy where dominants are in a position to take with impunity resources while subordinates are in no position to reciprocate. At the time of the writing of the paper J.M. Smith was just beginning to formulate his "Theory of Games" in the context of evolutionary biology, but the ideas were transmitted to Trivers via Hamilton. Hamilton suggested to Trivers that one could gauge the likelihood of the origination and spread of reciprocal altruism in proportion to the benefits and costs traded and the number of likely encounters. Clearly greater benefits and more encounters would tilt the playing field in favor of the spread of alleles which induced discriminatory altruism toward other altruists. Conversely, the invasion of altruist alleles would be less likely in a scenario where the number of encounters was minimal or the benefits to cheaters was great (and so the implied cost to altruists also great). One interesting point though is that at low frequencies altruists may have a difficult time increasing in frequency if their selective advantage is proportional to the encounters with other altruists, because at low frequencies such encounters would be very rare. Clearly a large number of total encounters would increase the likelihood of altruists at low frequencies increasing each other's fitness vis-a-vis the non-altruists. Trivers offers up three empirical illustrations of the model that he presents above. These are symbioses between fish who clean and those who are cleaned, bird warning calls, and finally human social cooperation. Since the paper can be read in full I will be brief, in particular since Trivers suggests that his illustrations were not particularly powerful and the nature of the altruism was not what he had originally understood. In regards to the cleaning symbioses, the idea is that small fish which clean the gills of larger, normally predatory fish, increase the fitness of the latter by consuming ectoparasites while feeding. Trivers attempts to show not just relationships between specific fish species, but relationships between specific fish! That is, cleaners who haunt particular locales and seek out other specific predator fish. The example is interesting because we do not generally consider fish to be "intelligent" organisms capable of behavorial plasticity, and yet here some level of interspecific recognition must be going on for particular dyads to emerge. The relative complexity of these behavorial relations imply that even "simple" organisms can engage in rational altruism. The second example Trivers uses is that of warning calls for birds. I won't cover this in detail because Trivers seems to suggest that this example is not a case of direct reciprocal altruism, where the bird giving a warning is also warned later by the birds which it aids, but a case of "return effect altruism." In this case the benefits to giving a warning are more generalized and do not depend on future aid, as the bird giving the warning may benefit from the disorientation caused to the predator by the scattering, or the discouragement of the predator forming a "search image" on which to fix in regards to its species. And finally Trivers moves on to his human example. Here the empirical example is one where Trivers mines the social psychological literature and attempts to slot in the behavior into his model. Much of this section prefigures later behavorial ecological and evolutionary psychological literature. Trivers' allusions to a psychological arms race between various individual humans as they attempt to detect cheaters and cheat, as well as maintain good relations within their group, receives fuller treatment in subsequent books such as Grooming, Gossip and the Evolution of Language, or The Mating Mind. The importance of Trivers paper is clear insofar as it sketched the outlines of whole disciplines. But it also can be traced back to the principles which he elucidated earlier in regards to how to write a classical paper, he reduced a rather expansive issue, the evolution of social behavior, down to an atomic unit from which larger constructs and dynamics could be generated. Trivers' verbal model (he admits that the mathematics is illustrative rather than the root of any original insight) was later expanded upon by Robert Axelrod and W.D. Hamilton in their 1981 paper The evolution of cooperation in Science, where they showed the power of the tit for tat "strategy." They showed more formally the basic idea which Trivers offered, that simple altruistic behaviors could lead to great complexity in social organization as they spread because of their advantages vis-a-vis naive selfishness. Of course I think it is safe to say that in regards to human sociality a bit more complexity must be added to the model to achieve the sort of explanatory power which we would want, but it is upon the shoulders of such simple and elegant ideas like reciprocal altruism that such things may rise.