Learning to be Altruistic

Abstract

Evolutionary studies account for cooperation under the shadow of the future. But how can altruism spread without direct reciprocity? Learning from punishment, including badmouth, is impossible in harsh environments, where agents do not survive rejections received. Imitation is indispensable, but what to imitate? Frequent behaviors are not necessarily socially desirable, neither fitness observable. In this paper, agents meeting infrequent but lethal food scarcity survive thanks to food sharing. Saving recipients from certain death, donations reduce altruists’ lifetime. Results show that prudent donors, helping only when above starvation point, are exploited by cheaters and soon extinguish. The same happens with agents taking reciprocity into account, and helping only when their credits are turned off. Instead, agents endowed with dynamic goals (survival vs. giving help) learn even the most unconditioned form of altruism, thus avoiding extinction. Tentative conclusions are discussed. Among others, dynamic goal-directed agents are autonomous entities learning even the most generous forms of altruism. Moreover, prudence is not necessarily more adaptive than unconditioned altruism; indeed, it may be self-defeating under given conditions.