Following training on a variable-interval food-reinforcement schedule, rats were exposed to three unsignaled shocks during each SO-min session. Although leverpressing was initially suppressed, responding was significantly accelerated following offset of the third shock, regardless of when in the session it occurred. Control sessions in which only two shocks were programmed, one early and one late, did not yield baseline acceleration. Evidence of "counting to three" was less obvious in subjects simultaneously exposed to a temporal autocontingency, that is, for which each shock also predicted a minimum S·min safety period. The addition of a signal prior to each shock eliminated evidence of counting behavior altogether. We conclude that rats may be taught to count, but such behavior is highly unnatural and may be blocked or overshadowed by more salient sources of information.Although Clever Hans, the counting horse, appears to be a permanent part of our psychological folklore (Watson, 1914), the formal possibility that infrahuman animals can count has received little serious attention in the modern experimental literature.In this paper, we will present evidence that rats can count to three to predict safety from shock. In evaluating our results, we have considered definitions of counting adapted from human performance. However, even the mechanisms that underlie counting behavior in humans are far from understood (cf. Gelman & Gallistel, 1978;Piaget, 1952). We have therefore adopted the view of counting that is typically held in infrahuman research, viz, an animal is assumed to be counting if behavioral sensitivity to number-related stimuli can be demonstrated. This view requires that all alternative, and presumably simpler, sources of control, such as changes in exteroceptive stimuli or temporal cues, be precluded.We have recently reviewed a variety of evidence that suggests that infrahuman animals are, in fact, capable of counting (Davis & Memmott, 1982). For example, Davis and Mcintire (1969) reported that when three unsignaled shocks were regularly superimposed upon a baseline of food-reinforced leverpressing, responding did not remain totally suppressed, but, rather, recovered immediately following delivery of the third shock. Seligman and Meyer confirmed these findings and concluded that "rats act as if they are able to 'count to three' and useThe preparation of this paper was supported in part by Grant A06973 from the Natural Science and Engineerin. Research Council of Canada to Hank Davis, and by an NSERC postdoctoral fellowship award to John Memmott. Reprints may be obtained from Hank Davis, Department of Psychology, University of Guelph, Guelph, Ontario Nl G 2WI, Canada.
95the information conveyed by the third shock-no further shock-as a safety signal" (1970, p, 2(6).In order to explore this source of control, Davis, Memmott, and Hurwitz (1975) introduced the concept of autocontingencies (ACs), subtle relationships among stimulus events that often go undetected by the experimenter. Our initial work descri...