An artificial fruit is an experimental apparatus which functionally mimics natural food items, where an outer shell or casing must be removed via manipulation in order to access an edible core or kernel.
Artificial Fruits and the Two-Action Method
The term “artificial fruit” was first used by Whiten et al. (1996) to describe an apparatus provided to both human children and chimpanzees (Pan troglodytes). This apparatus built upon the “two-action method,” first used by Dawson and Foss (1965) in their study of observational learning in budgerigars (Melopsittacus undulates). Two-action tasks expose naïve “observer” animals to “demonstrator” or “model” animals which operate on a task designed to be dealt with in either of two (or potentially more) alternative ways. The observer animals are then given access to the task, and the ways in which they interact with it are recorded. If observer animals tend to manipulate the task using the action modelled by the demonstrator more frequently than the alternative action, this indicates that social learning occurred. This method allows researchers to distinguish low-fidelity forms of social transmission, such as stimulus enhancement (in which animals learn to orient their behavior towards a particular object) from higher-fidelity forms of social transmission, such as imitation, in which an animal learns the form of a behavior through observation. Low-fidelity social transmission within a two-action artificial fruit task would lead to observers being attracted to the apparatus more than control individuals who saw no model, but then solving the task by trial-and-error, so this would not result in animals matching the demonstrator’s solution more frequently than the alternative action.
Although two-action artificial fruit tasks identify relatively high-fidelity social learning, they do not necessarily differentiate imitation (copying the form of actions, like sliding versus lifting a door) from emulation (copying the results of actions, such as “the door slides” versus “the door rises,” which could be achieved using actions different to those observed). Alternative methods, such as ghost controls, can be used to distinguish between these relatively high-fidelity forms of social learning.
Artificial fruits can incorporate multiple barriers, each of which may be removed in different ways in order to access a reward, thus exemplifying the two-action method. For example, Whiten et al.’s (1996) task featured bolts which could be removed using either a poking or twisting action, a barrel latch incorporating a pin that could be removed using either a turning or spinning action and a handle that could be turned or pulled upwards. However, artificial fruits do not necessarily require high levels of manipulative complexity; some designs have incorporated barriers as simple as a door which could be either lifted up or slid to one side to access a food reward within a box (e.g., the “doorian,” Horner et al. 2006).
While retrieving the food from within all artificial fruits requires physical manipulation of the task, the first artificial fruits did not generally require the use of tools. More recently, the term has also been used to describe tasks requiring tool use. Though some of these tool-use tasks (e.g., the “Pan-pipes,” Whiten et al. 2005) require forms of behavior that are perhaps more analogous to wild primate behaviors such as termite-fishing or honey-dipping, rather than fruit processing (and so perhaps might be better termed “artificial foraging tasks”), they fulfil the same goal of providing a manipulative task that can be solved using one of at least two distinct behaviors, and so can be broadly described as “artificial fruit approaches.”
Artificial fruits have primarily been used to examine social learning processes in both humans and nonhuman animals but have also been incorporated in experimental paradigms examining cognitive abilities such as behavioral flexibility and planning (e.g., Miyata et al. 2011). Within the field of social learning, artificial fruits have been used in dyadic settings to investigate the capacity of multiple species, including great apes (chimpanzees: Whiten et al. 1996, gorillas: Stoinski et al. 2001, orangutans: Custance et al. 2001), and other nonhuman primates (marmosets: Caldwell and Whiten 2004, capuchin monkeys: Custance et al. 1999), birds (keas: Huber et al. 2001), and human children (e.g., Whiten et al. 1996), to engage in high-fidelity social learning. They have also been used to study social learning biases (such as from which models animals are most likely to learn, and the impact of context upon such biases) and the social dynamics which affect the transmission of behaviors within groups.