Trade-offs in the adaptive use of social and asocial learning.

A common assumption by ethologists, behavioral ecologists, and anthropologists, albeit rarely made explicit, is that the acquisition of learned information from others hhenceforth ‘‘social information’’) is inherently adaptive. Individuals are deemed to gain fitness benefits by copying others on the assumption that they acquire adaptive information while avoiding some of the costs associated with learning for themselves (the costs of ‘‘personal information’’). Social learning is known to enable naı¨ve animals to acquire information relevant to many life skills, including when, where, what, and how to eat (Galef and Giraldeau, 2001), with whom to mate (White, 2004), or fight (Peake and McGregor, 2004), as well as which predators to avoid and how (Griffin, 2004). The unspoken supposition is that the acquisition and exploitation of such information will inevitably confer fitness benefits on the learner, since individuals will save themselves the costs, for instance, of searching their entire home range, sampling all potential foods, or learning to escape predators for themselves. In fact, the use of social information does not guarantee success (Boyd and Richerson, 1985; Laland, 2004). Individuals face evolutionary tradeoffs between the acquisition of costly but accurate information and the use of cheap but potentially less reliable information (Boyd and Richerson, 1985). Theoretical models investigating the adaptive advantages of different forms of learning conclude that social earning cannot be employed in a blanket or indiscriminate manner, and that individuals should adopt flexible strategies that dictate the circumstances under which they copy others (Laland, 2004). Such theoretical analyses reveal that social learners would have higher fitness than asocial learners only when copying is rare, when most potential demonstrators would be asocial learners who have acquired and display accurate information about the environment (Boyd and Richerson, 1985, 1995; Giraldeau et al., 2002; Rogers, 1988). As the frequency of social learners increases, however, the value of using social information would decline, because the proportion of individuals demonstrating accurate personal information would decrease. At the extreme, with all individuals copying, the population would have to rely upon unreliable and possibly inaccurate information1 as no one would have acquired accurate personal information by sampling the environment. In order for the use of social learning to be adaptive, individuals must use social learning selectively and engage in the collection of accurate personal information some of the time (Galef, 1995; Laland, 2004). The circumstances under which individuals might switch between reliance on different sources of information remain relatively unexplored. What context‐dependent rules have evolved in animals dictating how they exploit both personal and social information? Do animals copy the behavior of others when they are uncertain how to solve a problem? Do they copy others when it is easy to do so and only learn asocially when copying is not an option? Or is social learning a last resort when asocial learning has failed? Following Laland (2004), the term ‘‘strategies’’ is used here to equate such learning heuristics with those strategies commonly analyzed using evolutionary game theory (Maynard‐Smith, 1982). Of course, animals need not be aware that they are following a strategy, nor need they understand why such strategies may work. Until relatively recently, the existence and characteristics of social learning strategies have not been allotted a great deal of empirical attention, despite assumptions and predictions pertaining to such strategies in theoretical models. However, experimental support is now emerging for the existence of two broad classes of strategies, dictating both when animals will use social information and from whom they will learn (Laland, 2004). In this article, we will focus on when strategies, reviewing the current empirical support for the putative strategies of copy others when asocial learning is costly and copy others when uncertain. We hope that by emphasizing consistent findings in a range of species, including fishes, birds, and mammals, the prevalence of trade‐offs in the use of social and asocial learning will become apparent, and will be taken into account in future studies of social learning. A further aim is to encourage the integration of theoretical and empirical work in animal social learning, where there is considerable potential for combining laboratory experiments and game theoretical analyses (Laland and Kendal, 2003; Laland, 2004)