Abstract

This dissertation reports analyses of performance for rewarding brain stimu- lation in a three-part sequential task. A session of self-stimulation was composed of three trial types, during which the strength and opportunity cost of electrical stim- ulation were kept constant. Within a trial, a lever allowed animals to harvest brain stimulation rewards as soon as the lever had been held for a set, cumulative amount of time. When the time spent holding reached this criterion, the lever retracted and a burst of rewarding electrical stimulation was delivered. A flashing house light and 10s inter-trial interval signalled the start of a new trial. Rats were presented with strong/inexpensive/certain stimulation on one trial, a randomly selected strength, cost and risk on the next trial, and weak/inexpensive/certain stimulation on the third trial of a sequence. The sequence then repeated. Rewards during the second trial of this sequence were delivered with cued probabilities ranging from 0.5 to 1.0. The current thesis evaluates the ability of a previously published (molar) model of performance during a trial to accurately detect the effect of risk on payoff but not reward intensity. Although animals were less willing to work for stimulation trains that may not be delivered than those delivered with certainty, risk did not change the relative reward produced by stimulation. We also present evidence on a fine time scale that self-stimulating rats develop a model of their world. The first pause made as a trial began was a function of the payoff the animal had yet to receive, indicat- ing that rats had a model of the triad sequence. Analysis of the conditions under which pauses were uncharacteristic also provides evidence of what this model might be. Analysis of the fine scale of performance provides evidence that animals had a model of the stability of trial conditions. Finally, we present a (molecular) model of performance for brain stimulation rewards in real-time. Our results demonstrate that rats develop a model of the testing paradigm and can adjust to changes in reward contingencies with as few as one exemplar.