This paper represents an attempt to bring the theory of learning into a workable relationship with the more practical application of machine learning. For both research and practice, the issues of definition and description of complex learning remain absolutely central. The primary research problem in this area, therefore, is to understand and describe learning with the precision that makes it possible to measure. Once this is accomplished, then we can decide when and if it has been achieved. Schank’s article is an attempt at defining a series of criteria that might, at the very least, help describe the problem of machine learning in more precise terms.

As part of his initial analysis, Schank derives from his current research a theoretical framework for describing the criteria for successful machine learning. He argues that there are four criteria crucial for establishing machine learning: Form, Context, Development, and Evolution. According to Schank, Form is merely knowing what it is that is supposed to be learned. Context relates to the “need for learning to occur in the course of the performance of a high level understanding task, guided by overarching memory structures.” Development refers to a system’s ability to learn or create more complex structures. Finally, Evolution denotes that a system can adapt, change, and grow as it acquires more information and knowledge. Schank then uses these criteria as the basis for identifying successful programs of learning, and considers to what extent and in what ways other programs fail to meet these criteria.

To further extend the knowledge, Schank poses several possible theoretical questions and discusses how one might conduct research to compare these positions. The questions Schank asks include: What classes of memory should we use? What are the possible indices for these structures? How should one deal with expectations, failures? How do these two items relate to structure alterations? Although these questions are formulated around Schank’s own research, they are relevant to all researchers interested in the problem of creating learning systems.

Since this is a theoretical paper, it would be inappropriate to criticize Schank’s application of the learning theory. Yet, for most researchers in the field, it is still unclear whether Schank’s own system meets the stated precise criteria he has set forth. More importantly, of course, is the question of whether any learning system can be precise, yet robust enough to handle both general and specific learning tasks. Nevertheless, Schank poses the initial questions that will serve to clarify the problem areas that exist within the learning field.