The "architecture" of human cognition. The influence of computer scientists who strive to develop "artificial intelligence" has focussed more attention on the role of knowledge in human cognition. It has also lead to the concept of an "architecture" of a human cognitive system that is based on the metaphor of the mind as a computer^{10, 11} (see figure 5.1 in Chapter 5). In this approach, the mind is considered to have a long term memory that stores knowledge. This long term memory is essentially infinite in capacity.

Additionally, there is a "working" or "short term" memory that contains our thoughts of the moment. The working memory calls on, or "addresses" knowledge in our long term memory, or what is sometimes called our "knowledge" or "data" base, and uses that information in the comprehending, learning, communicating, reasoning and so forth that it is involved in at the moment. But, unlike the long term memory, the capacity of the working memory is severely limited. We cannot keep too many things in mind at one time because of the limited capacity of our working memories. For instance, if we want to dial a telephone number, we may repeat the number over and over again to keep it in working memory long enough to dial it, then we can forget it.

Among the very important findings from studies of the limited capacity of working memory is that the capacity can be expanded if some of the mental processes involved are automated. For instance, in reading, it has been found that students who must occupy their limited working memory in decoding print to speech, as in phonics, cannot comprehend well what they are reading. Comprehension requires additional processing "space" in working memory, particularly in regard to addressing knowledge in long term memory and merging it with the new information picked-up from the book. In order to efficiently read and comprehend, it is necessary that the decoding aspect of reading become automatic, that is, performed without conscious attention. This can only be accomplished by hours and hours of practice in reading. This is one of the reasons why "quick-fix" reading programs cannot make much of an improvement in reading comprehension of those low in decoding skill. A second reason, of course, is that to improve reading comprehension much, one must develop a large body of knowledge in long term memory relevant to what is being read. Like skill, the development of large bodies of knowledge takes a long time.

The limitations of working memory are also a major factor in learning and applying mathematics¹². For instance, in trying to calculate the cost of a meal in a restaurant, the working memory must deal with the locating of information and comprehending the written description of what each item was and what it cost. If working memory space must also be given to recalling how to add, subtract, multiply and divide simple two or three digit numbers, while searching the check, there is much opportunity for mistakes. However, if one has automatized a large number of calculations, such as in learning the multiplication tables up to 12 or so, then calculations can be made while working memory space is left for searching and comprehending the check.

This concept of a human cognitive system with a knowledge base in long term memory and information processing in a limited capacity working memory is one that has emerged over the last quarter century and was not understood in this sense when most of the intervention programs of the War on Poverty were designed and implemented. For this reason the important implications of studies of these system components, their contents, and how they interact with one another and with the information in the external world "outside the head" was not incorporated into the design of these programs. And, for the most part, the knowledge that cognitive scientists have developed about this system has still not influenced education programs (mainstream or intervention) to the degree that it should.