Artificial intelligence and self-organizing systems: Teaching a digital computer to assist in making decisions

DESPITE THE ADVANCES we have made, techniques of processing information so as to satisfy specific objectives of management are still needed in business. The resources available for meeting these needs are three types of services: the services of machines, e.g., high-speed digital computers; the services of materials, e.g., punched cards and magnetic tapes; and the services of humans. Most of us are not satisfied, however, when we ask ourselves “How well are we using these resources to design and develop 'business systems'?” Computers, for example, are employed too often for counting rather than computing. We talk about, but make little practical use of computer capabilities for performing the tasks ordinarily delegated to humans. In short, we frequently fail to utilize fully the machine and human services available.

Within this context, the technique to be described is one in which the human services include teaching a digital computer to make decisions that are satisfying to management. In addition, the machine services are not limited to processing information so as to help man make decisions. Instead, the computer captures the essentials of human information-processing in a form such that it can make adequate decisions, when provided the information available to human decision-makers. Given adequate instruction and input information, the machine simulates management's decision-making activities. With this technique, the computer can take account of the human factors—even company politics—that preclude operating a business on purely logical bases. This approach also resolves problems often met in specifying management's objective, or policy, and in translating it into action. The computer learns a policy and implements it in the form of decisions. This learning takes place when the machine is given as sample of the results of several satisfying decisions and the information used by management in reaching these decisions. The computer treats this sample of decisions as a variable to be predicted, the information known prior to the decisions as predictor variables. The following examples illustrate this technique of using a digital computer to assist in decision-making.