Learning from others' mistakes: a paradox revisited.

Some researchers have reported the paradoxical finding of enhanced social learning when naive observers learn from unskilled rather than skilled demonstrators, particularly in discrimination tasks. In two experiments with starlings, Sturnus vulgaris, I considered whether this enhanced learning is because the observer (1) sees incorrect responses only, (2) sees both correct and incorrect responses or (3) sees an increase in the proportion of correct responses over trials. In experiment 1, individual starlings observed a demonstrator bird perform multiple simultaneous discrimination tasks. In one group, the demonstrator always picked the correct stimulus; in another group, the demonstrator always picked the incorrect stimulus; in a third group, the demonstrator consistently picked the correct stimulus 50% of the time. Those subjects that observed only incorrect choices performed significantly better than the other two groups, but none of the birds achieved the 90% correct performance criterion. Experiment 2 involved a single discrimination task; thus, a fourth group was added to control for individual learning. Again, subjects that observed only incorrect responses learned the discrimination significantly more quickly than the other three groups. Subjects that observed the demonstrator make both correct and incorrect responses were equally likely to select the same (correct) or opposite (incorrect) stimulus when the demonstrator picked the correct stimulus. When the demonstrator picked the incorrect stimulus, however, these subjects were significantly more likely to pick the opposite (correct) stimulus. These findings suggest that when learning a discrimination problem, observing a foraging companion's lack of success is more informative than observing its success. Copyright 1998 The Association for the Study of Animal Behaviour.