On the Assignment-of-Credit Problem in Operant Learning
Operant conditioning is the selection of particular activities by the activity-dependent occurrence of pleasant or unpleasant events (reinforcers). Reinforcers are most effective if they closely follow the target activity (contiguity), but it is also important that the reinforcer and activity be correlated (contingency). No formal model for the action of response-reinforcer contiguity exists and several anomalies—cases where the behavior contiguous with the reinforcer is not strengthened, or an activity is strengthened by response-independent reinforcement—indicate flaws in the informal contiguity account. We show that an exceedingly simple, nonassociative (context- free) process provides a qualitative account for both anomalous and nonanomalous properties of operant conditioning. The process can be extended to permit associative effects, including classical conditioning and stimulus control; it may therefore represent the initial processing stage for all conditioning in higher vertebrates.
