Decision-Making and Learning: The Peak Shift Behavioral Response

In a typical peak shift experiment, control subjects are trained to respond (e.g., button-press) to a positively reinforced stimulus (Sþ, for example a line of a particular orientation or a color of a particular hue). Treatment subjects are trained in a manner identical to control subjects with respect to Sþ and also to withhold response to an unreinforced or punished stimulus (S–, a line or hue similar but not identical to Sþ). Both groups of subjects are then tested without reinforcement on a continuum of stimuli comprising various line orientations or hues, including the training stimuli. Figure 1 depicts the common finding of such discrimination learning experiments: during the test, control subjects respond most frequently to the Sþ stimulus. Treatment subjects, however, respond most frequently to a stimulus they have never encountered before. The treatment subjects’ expression of a preference for an unrewarded and novel stimulus over Sþ is somewhat paradoxical. If all test stimuli are discriminable, why are the treatment subjects (Sþ/S trained) not like the control subjects (Sþ training only), responding most strongly to the stimulus they have learned is rewarding? The phenomenon that treatment subjects display in this type of experiment is known as peak shift. When the subjects’ frequency of response is plotted as a function of stimulus value, data show a peaked response gradient. The stimulus receiving the maximum, or ‘peak,’ response by the treatment subjects is said to be ‘shifted’ relative to that of the control subjects. An area shift is often noticeable, even in experiments that do not result in a significant peak shift. An area shift is characterized by an elevation of the rates of response to the novel stimuli on the side of Sþ away from S–. Area shift often co-occurs with peak shift. In addition, a shift of the most strongly avoided stimulus is also produced, off of S– in a direction away from Sþ. Peak shift is considered to be a general outcome of generalization accompanying discrimination learning. In a typical peak shift experiment, the stimuli are simple sensory perceptions, monotonically increasing in value on the stimulus domain. The experiment consists of a control group and a treatment group each undergoing training (discrimination) and testing (generalization) phases. Stimuli are presented one at a time, for instance, on a