Artificial intelligence is the science of making machines that think like men.
As a science, it has been around almost as long as digital computers have been. But most of the research in artificial intelligence has been in the basic nature of intelligence, or in fascinating but intractable fields such as language translation.
By the late 1960's, artificial intelligence concepts were being applied for the first time to solving problems in a strictly limited domain of knowledge. As a contributor to a recent book on expert systems (Hayes-Roth et al., 1983) put it, " a new methodology become viable: building systems in limited domains of expertise with knowledge elicited from human experts." These were the first expert systems.
An expert system is a computer program which copies the reasoning and the knowledge of an expert. in a limited field of expertise.
At first, research concentrated on areas which were too difficult for human experts. But gradually researchers realized that an expert system was a tool which could be used to "clone" experts in fields where there was a high demand for, and short supply of, specialists. Expert systems in such fields were found to have a big advantage over previous expert system: they were much more modest in their hardware demands.
In exploration geophysics we have had such problems for years: we always seem to have a shortage of experts, both in good times and in bad, especially if it involves working in a remote area. Any field of expertise, from survey planning to interpretation, is a candidate for an expert system to help less experienced or broadly trained people perform as well as a specialist. Expert systems could be used in a standard mode simply to advise a person in decision making, or they could be integrated into a data collection or processing system so that they can use data directly for guiding the decision, and put the decision into effect automatically.
A few years ago, the study of expert systems, known as "knowledge engineering", was an arcane science almost entirely in the hands of a small community of academics at institutions such as the Mass achusets Institute of Technology and Stanford University. As Barr and Feigenbaum state in the preface of The Handbook of Artificial Intelligence (1981), "those who understand symbolic computation have been speaking largely to themselves for the first 25yearf of AI's history". (By the way, they still do!). Ever if someone in industry knew what an expert system was, the cost of using one was hardly economic. An expert might cost $25,000 per year while just to run an existing expert system required a dedicated mini computer costing $100,000.
Until recently, even to get hands-on experience with developing an expert system was quite expensive. In the September, 1984, issue of BYTE magazine there was a review of M. 1, a "knowledge engineering tool" written in the Prolog language, and designed to run on an IBM Personal computer with 128 kb of random access memory1. This tool is intended as an exploratory tool for deciding whether knowledge engineering can be- applied to a specific problem. Jerold Kaplan, of Teknowledge (the supplier of the M.1 system) is quoted as saying "Before M.1 it was a couple 'of hood red thousand dollars. You had to go out and buy an $80,000 dollar LISP machine, take two guys and set them to work on it for six months, in addition to buying some $60,000 software package."