In biology, altruism is behaviour by an individual that increases the fitness of another individual while decreasing their own. Altruism in this sense is different from the philosophical concept of altruism, in which an action would only be called "altruistic" if it was done with the conscious intention of helping another. In the behavioural sense, there is no such requirement. As such, it is not evaluated in moral terms—it is the consequences of an action for reproductive fitness that determine whether the action is considered altruistic, not the intentions, if any, with which the action is performed.

The term altruism was coined by the French philosopher Auguste Comte in French, as altruisme, for an antonym of egoism. He derived it from the Italian altrui, which in turn was derived from Latin alteri, meaning "other people" or "somebody else".

Altruistic behaviours appear most obviously in kin relationships, such as in parenting, but may also be evident among wider social groups, such as in social insects. They allow an individual to increase the success of its genes by helping relatives that share those genes. Obligate altruism is the permanent loss of direct fitness (with potential for indirect fitness gain). For example, honey bee workers may forage for the colony. Facultative altruism is temporary loss of direct fitness (with potential for indirect fitness gain followed by personal reproduction). For example, a Florida scrub jay may help at the nest, then gain parental territory.

In ethology (the study of behavior), and more generally in the study of social evolution, on occasion, some animals do behave in ways that reduce their individual fitness but increase the fitness of other individuals in the population; this is a functional definition of altruism. Research in evolutionary theory has been applied to social behaviour, including altruism. Cases of animals helping individuals to whom they are closely related can be explained by kin selection, and are not considered true altruism. Beyond the physical exertions that in some species mothers and in some species fathers undertake to protect their young, extreme examples of sacrifice may occur. One example is matriphagy (the consumption of the mother by her offspring) in the spider Stegodyphus; another example is a male spider allowing a female fertilized by him to eat him. Hamilton's rule describes the benefit of such altruism in terms of Wright's coefficient of relationship to the beneficiary and the benefit granted to the beneficiary minus the cost to the sacrificer. Should this sum be greater than zero a fitness gain will result from the sacrifice.

When apparent altruism is not between kin, it may be based on reciprocity. A monkey will present its back to another monkey, who will pick out parasites; after a time the roles will be reversed. Such reciprocity will pay off, in evolutionary terms, as long as the costs of helping are less than the benefits of being helped and as long as animals will not gain in the long run by "cheating"—that is to say, by receiving favours without returning them. This is elaborated on in evolutionary game theory and specifically the prisoner's dilemma as social theory.

The existence of altruism in nature is at first sight puzzling, because altruistic behaviour reduces the likelihood that an individual will reproduce. The idea that group selection might explain the evolution of altruism was first broached by Darwin himself in The Descent of Man, and Selection in Relation to Sex, (1871). The concept of group selection has had a chequered and controversial history in evolutionary biology but the uncritical 'good of the species' tradition came to an abrupt halt in the 1960s, due largely to the work of George C. Williams, and John Maynard Smith as well as Richard Dawkins. These evolutionary theorists pointed out that natural selection acts on the individual, and that it is the individual's fitness (number of offspring and grand-offspring produced compared to the rest of the population) that drives evolution. A group advantage (e.g. hunting in a pack) that is disadvantageous to the individual (who might be harmed during the hunt, when it could avoid injury by hanging back from the pack but still share in the spoils) cannot evolve, because the selfish individual will leave, on average, more offspring than those who join the pack and suffer injuries as a result. If the selfishness is hereditary, this will ultimately result in the population consisting entirely of selfish individuals. However, in the 1960s and 1970s an alternative to the "group selection" theory emerged. This was the kin selection theory, due originally to W. D. Hamilton. Kin selection is an instance of inclusive fitness, which is based on the notion that an individual shares only half its genes with each offspring, but also with each full sibling (See footnote). From an evolutionary genetic point of view it is therefore as advantageous to help with the upbringing of full sibs as it is to produce and raise one's own offspring. The two activities are evolutionarily entirely equivalent. Co-operative breeding (i.e. helping one's parents raise sibs—provided they are full sibs) could thus evolve without the need for group-level selection. This quickly gained prominence among biologists interested in the evolution of social behaviour.

In 1971 Robert Trivers introduced his reciprocal altruism theory to explain the evolution of helping at the nest of an unrelated breeding pair of birds. He argued that an individual might act as a helper if there was a high probabilistic expectation of being helped by the recipients at some later date. If, however, the recipients did not reciprocate when it was possible to do so, the altruistic interaction with these recipients would be permanently terminated. But if the recipients did not cheat then the reciprocal altruism would continue indefinitely to both parties' advantage. This model was considered by many (e.g. West-Eberhard and Dawkins) to be evolutionarily unstable because it is prone to invasion by cheats for the same reason that cooperative hunting can be invaded and replaced by cheats. However, Trivers did make reference to the prisoner's dilemma game which, 10 years later, would restore interest in Trivers' reciprocal altruism theory, but under the title of "tit-for-tat".

In its original form the Prisoner's Dilemma Game (PDG) described two awaiting trial prisoners, A and B, each faced with the choice of betraying the other or remaining silent. The "game" has four possible outcomes: (a) they both betray each other, and are both sentenced to two years in prison; (b) A betrays B, which sets A free and B is sentenced to four years in prison; (c) B betrays A, with the same result as (b) except that it is B who is set free and the other spends four years in jail; (d) both remain silent, resulting in a six-month sentence each. Clearly (d) ("cooperation") is the best mutual strategy, but from the point of view of the individual betrayal is unbeatable (resulting in being set free, or getting only a two-year sentence). Remaining silent results in a four-year or six-month sentence. This is exemplified by a further example of the PDG: two strangers attend a restaurant together and decide to split the bill. The mutually best ploy would be for both parties to order the cheapest items on the menu (mutual cooperation). But if one member of the party exploits the situation by ordering the most expensive items, then it is best for the other member to do likewise. In fact, if the fellow diner's personality is completely unknown, and the two diners are unlikely ever to meet again, it is always in one's own best interests to eat as expensively as possible. Situations in nature that are subject to the same dynamics (rewards and penalties) as the PDG define cooperative behaviour: it is never in the individual's fitness interests to cooperate, even though mutual cooperation rewards the two contestants (together) more highly than any other strategy. Cooperation cannot evolve under these circumstances.