We have tested this last implication in a spatial learning experiment with hummingbirds ( Cole, Hainsworth, Kamil, Mercier, & Wolf, 1982). The procedure involved a two part trial. In the first part of the trial, the information stage, the bird was presented with a single artificial flower located on either the right or left side of a styrofoam feeding board. This flower contained a small amount of 0.5M sucrose solution, and its location was random from trial to trial. The animal was allowed to take this nectar. The feeding board was then removed, and the second, choice stage of the trial began 10-12 sec later. During the choice stage, the hummingbird was presented with two artificial flowers on the feeding board, only one of which contained sucrose solution. The bird could choose either to return to the location visited during the information stage, or to avoid that location.
During stay learning, return visits were rewarded; during shift learning, only visits to the position not visited during the information stage were rewarded. All 8 hummingbirds received both stay and shift learning, with order of training counterbalanced. Training continued until a criterion of 3 consecutive days above 80% correct was achieved. The results were quite dramatic. Shift learning was always much more rapid than stay learning, regardless of whether it came first or second. The fastest stay learning took 282 trials and 130 errors, the slowest shift learning took 180 trials and 96 errors. There were two major differences between performance during shift learning and performance during stay learning. At the outset of the experiment, all birds showed a pro-experimental basis towards shifting. This may have been innate, or a result of the previous experience of these wild-caught birds. However, the rate of shift learning (in percentage improvement per day) was also significantly higher than the rate of stay learning. This hold even when each task was being learned after the other, and initial rates of performance were equal. These results suggest that nectar feeding birds have an evolved tendency to shift locations of flower visits, which is manifested in both starting performance and in differential rates of learning.
To be sure that there is no misunderstanding, I would like to make the reasoning behind this experiment very explicit The logical structure of the stay and shift tasks was very similar. In each case, the only cue predicting the location of food during the choice stage was the location visited during the information stage. In each case, the correct response was a visit to a specific location. Therefore, the reason for the difference in learning rates can be conceived of as residing in the structure of the animal being tested rather than in the tasks themselves.
For most nectarivorous birds, the major source of energy is floral nectar located in small, easily depleted, slowly renewing specific locations. Return visits to recently emptied flowers will never be rewarded; visits to now flowers will frequently be rewarded. Over evolutionary time, natural selection will favor those birds who learn to shift flower locations, but not those who repeatedly return to the same location. This process could be responsible for the results of our spatial learning experiment, and also is consistent with the ecological view.
Several points need to be made about an ecological view of learning and cognition. First, it needs to be contrasted with the idea of "biological constraints on learning" (e.g., Seligman & Hager, 1972). The basic idea of