ABSTRACT
Extensions of basic learning go in two directions: (a) secondary learning, in which the link between the primary reinforcer and the behavior becomes remote; and (b) unfamiliar learning, in which the system must learn new responses or highly unfamiliar material. Both directions present a variation of the same problem. The learner is currently at skill level A. To perform the terminal activities implied by the new learning, the learner must have skill level N. If the learner is simply presented with the terminal tasks (those that require skill level N) until finally succeeding, the chances of the learner performing successfully within a reasonable amount of time is highly unlikely. On each trial, the learner receives feedback that its attempt leads to failure. However, the feedback does not provide the learner with information about what to do—only what not to do.
