NOTES ON ARTIFICIAL INTELLIGENCE (Al) Introduction, Study notes of Artificial Intelligence

Download NOTES ON ARTIFICIAL INTELLIGENCE (Al) Introduction and more Study notes Artificial Intelligence in PDF only on Docsity! NOTES ON ARTIFICIAL INTELLIGENCE (Al) Alien Newell Carnegie-Mellon University 6 March 1981 Introduction The purpose of these brief notes is simply to introduce some basic terms and statements. They will all be discussed in detail and illustrated concretely in the lecture. Aims of Al Artificial Intelligence is a subpart of Computer Science, whose essential focus can be described by several different types of questions: What is the fundamental nature of intelligent action? What is the fundamental nature of mind? How are computers to be made more sophisticated, i.e., to carry out new intellectual and perceptual functions? What is the nature of the methods, algorithms and representations useful for solving tasks where little knowledge and analysis exists or where variability is so great that little can be known for sure? Core Concepts of Al A basic set of core concepts shows up repeatedly in all of the work in artificial intelligence, though in many different guises. Not all these concepts distinguish Al from the rest of computer science; indeed, some of them are the reason the study of intelligence is part of computer science. Symbol systems are systems that have symbols and symbol structures, which can designate external entities and which can be created, interpreted, changed, modified, etc. Only symbol systems can be organized so as to exhibit general intelligent action. Digital computers are symbol systems, though they are not usually organized to exhibit intelligence. Many other varieties of symbol systems exist besides digital computers. i Representations are just another term for structures that have symbolic function, hence can refer to some external situation or object. All processing for intelligent action occurs on representations of the task situation, and the difficulty or ease of a task depends radically on the representation used. Knowledge is a way of referring to all the guidance for action that can be extracted from a representation with indefinite amounts of processing. It is the concept of capacity or competence, when dealing with intelligence: An intelligent agent cannot do more than its knowledge permits; it often does less - failing to apply its knowledge. Goals are symbol structures that hold the knowledge of the situation or object desired and how to access additional knowledge useful in attaining it. Goals are the control structures through which symbol systems become purposeful. Intelligence is the ability to bring knowledge to bear to select actions to attain goals. This is the most useful simple definition of intelligent action. Search is the fundamental process that occurs in all intelligent behavior. It comes in two forms: combinatorial search in a problem space, which arises from uncertainty in how to construct or find the solution; and finite search in a pre-organized memory, which arises in finding the knowledge that will guide and limit the search in the problem space. Weak methods are schemes of action that are useful under conditions of little knowledge (the pervasive situation in problem solving). The well known methods are generate and test, hill climbing, heuristic search, means-ends analysis and simple abstraction planning. Weak methods provide the frames on which additional knowledge is accreted to form more intelligent methods. All artificial intelligence programs are composed of weak methods augmented by extra knowledge. Areas of Application of Al Al is only beginning to be applied. A harbinger of its potential can be seen in the diversity of ways application has started or is nascent (as indicated below). As in all technology transfer, the applications occur in small and particular ways, often utilizing the less advanced but more robust aspects of the field. Each of these areas draws on the core ideas of the field (in agreement with the assertion at the beginning). But in addition each draws on some quite distinct aspects, thus reinforcing this image of .extreme diversity in the application potential of Al. Knowledge-based systems. Draws on the development of architectures for large numbers of parallel conditiorvaction rules (called production systems). Works in domains with little problem solving and large amounts of expertise. The major impact will be on all manner of human expert consultation and decision situations. Heuristic search. Draws on combinatorial heuristic search schemes in spaces with well- defined symbolic representations. The major impact will be on engineering and science and, ultimately, on programming itself. Natural language. Draws on detailed study of natural language and to a lesser extent on knowledge-based architectures. The major impact will be in communicating with computers. Vision. Draws on detailed studies of the knowledge in the optic array and also on advances in digital image processing. The major impact will be in robotics. Speech. Draws on detailed studies of the knowledge in the acoustic wave and also on advances in digital signal procesing. The major impact will be in communication with computers.