A drive-reinforcement neural network model of simple instrumental conditioning

A network of classically conditionable drive-reinforcement neurons learned to choose an appropriate instrumental response to cues in a T maze when the cues could be utilized to anticipate the presentation of a positive or negative reinforcer or the absence of a reinforcer. To prove generality, it was shown that a network trained to respond to one configuration of reinforcers in the maze could learn to respond appropriately when the configuration is reversed. When the learning system turned toward the one arm of the maze it was negatively reinforced. When this happened the opposing intentional neuron corresponding to the active effector became active. The opposing intentional neuron corresponding to the inactive effector remained inactive because its high threshold precluded activity unless both the negative reinforcement center and its corresponding effector were active. When either opposing intentional neuron became active, it inhibited the currently active effector neuron and excited the other. A negatively reinforcing event therefore had the effect of inhibiting the responses that led to it. causing the connection strength from the T sensor to that effector neuron to decrease and also excited alternative actions