Лекция: The nature and objectives of fifth generation computers
Fifth generation computers aim to be able to solve highly complex problems, ones which require reasoning, intelligence and expertise when solved by people. They are intended to be able to cope with large subsets of natural languages, and draw on very large knowledge bases. In spite of their complexity, fifth generation computers are being designed to be used by people who are not necessarily computer experts.
In order to achieve these very ambitious aims, fifth generation computers will not have a single processor, or a small number of tightly coupled processors as computers do today. They are being designed to contain a large number of processors, grouped into three major subsystems: a knowledge base system, an inference mechanism and an intelligent user interface.
The knowledge base system has a very large store of knowledge, structured in one of the ways described above, with a set of processors which access and update the knowledge. It is likely that knowledge bases will evolve from current work in relational databases. Operations on knowledge bases require the manipulation of large numbers of individual elements: this manipulaton will be done in parallel by the arrays of knowledge processing elements.
The inference mechanism draws reasoned conclusions from the knowledge base. Much of its processing will be drawing logical inferences of the:
if <condition> then <action>
variety. Accordingly, the processing power of a fifth generation computer is expressed in logical inferences per second (lips). The target is in the range 50 to 1000 million lips (compared with a current performance of 10 to 100 thousand lips). Most of this improved performance is planned to be achieved via highly parallel architectures, such as the dataflow and graph reduction architectures discussed below.
The intelligent user interface is the point of contact between a fifth generation computer and its user. Many of these will be based on communication in a large subset of a natural language. Others will make extensive use of advanced graphics, including image processing. The intention is to build a user interface which is close to the natural way of thinking of the user, rather than close to the way of working of the computer, as is the case with contemporary user interfaces. The intelligent interface will contain its own set of processing elements — image processing systems may have an array of processors, one per pixel of the display.