access to the computer system. But proficiency in communication between human beings, and between machines and people, requires (1) mutual intimate familiarity with contextual understanding, (2) a very large base of data, (3) linguistic inferential capability, and (4) broad utilization of jointly accepted models and symbols. The process, very complex and detailed, demands expensive computer hardware and software to achieve accurate and efficient translation between machine and human languages. Extensive research is now in progress in the AI field to better understand the fundamentals of human language and to improve the quality of communication between man and machine.
1.1.4 Expert Systems
Scientific expertise typically develops in human beings over many years of trial and error in some chosen field. So-called "expert systems" permit such individual expertise to be stored in a computer and made available to others who have not had equivalent experience. Successful programs have been developed in fields as diverse as mineral exploration, mathematical problem solving, and medical diagnosis. To generate such a system, a scientific expert consults with software specialists who ask many questions in the chosen field. Gradually, over a period of many months, the team builds a computer-based interactive dialogue system which, to some extent, makes the expert's experience available to eventual users. The system not only stores scientific expertise but also permits ready access to the knowledge base through a programmed capacity for logic and inference.
Typically, a user interrogates the expert system via a computer terminal, typing in, for example, statements about apparent symptoms in a medical case. The system may then inquire about other conditions or symptoms, request that specific tests be performed, or suggest some preliminary diagnosis, thus attaching a probability or level of confidence to its conclusion and supplying an explanation upon demand. Therefore, user and system interact and gradually approach an answer to some question, whether on a diagnosis of an illness, the location of a mineral deposit, or the solution to a problem in mathematics.
1.1.5 Automation, Teleoperation, and Robotics
Automatic devices are those that operate without direct human control. NASA has used many such machines for years for diverse purposes including antenna deployment, midflight course changes, and re-entry parachute release.
Teleoperation implies a human operator in control of a mechanical system remotely. Executive signals are transmitted from controller to device over hard wires if the distance is small, as in the case of a set of master-slave arms in an isolation room (e.g., "P4" biohazard facility, radioisotope handling, etc.). Or, control signals may travel millions of kilometers over a radio wave link to a planet light-hours away.
Robotic devices have the capacity to manipulate or control other devices. They may be mobile, able to move to some distant physical location where an action must be taken. Robots can be either automatic or teleoperated.
1.1.6 Distributed Data Management
Large amounts of data are involved in the operation of automatic and robotic devices. This may include control information that specifies the next action to be taken in some sequence of operations, archived data that are being transmitted from one memory bank to another, or sensed or measured data that give the status of a geographical area, the position of an actuator, or the speed of a spacecraft. The field of distributed data management is concerned with ways of organizing such data transmission and distribution so that it is accomplished rapidly, efficiently, and in a manner which best supports overall system operation, and with ways of optimizing cooperation among independent but mutually interacting databases.
1.1.7 Cognition and Learning
In this study, cognition and learning refer to the development of a machine intelligence capable of dealing with new facts, unexpected events, and contradictory information in novel situations. Many potential applications of advanced automation require a level of adaptive flexibility that is unavailable with present technology. Today's automatic computer-controlled machines handle new data by a method or approach programmed into them when they were developed. Tomorrow's more sophisticated tools may need the ability to learn, even understand, in the sense of changing their mode of operation as they encounter new situations.
1.1.8 Research and Development in Artificial Intelligence
At present, there is a great deal of AI theoretical research (and in some cases practical development) in progress at several institutions in the United States and throughout the world. Much of the early work in the field was accomplished at five major centers: Carnegie-Mellon University, Edinburgh University, MIT, SRI International, and Stanford University. Today, however, the list of active sites is much longer and includes, in the United States alone, such schools as the University of Illinois, the University of Massachusetts, Yale University, the University of Southern California, Texas University, the University of California at Berkeley, etc. Corporations with ongoing work include Bolt Beranek and Newman, General Motors, IBM,
