Abstract This paper gives an overview of a major integrated oil company's experience with expert systems over the last five years. First artificial intelligence, AI, and expert systems are defined. Then the development of an artificial intelligence group is chronicled, including details on development tool selection, project selection strategies, and potential pitfalls. Three potential pitfalls in the areas of technology, marketing and social behavior that the knowledge engineer should be aware of when developing a new AI application are identified. Next other artificial intelligence technologies are summarized. Finally eight completed expert systems are described.
Introduction Artificial Intelligence, AI, has been around since the mid-fifties and has evolved slowly over the last 40 years. Most objective observers would agree that AI has not come close to fulfilling its early promise of allowing computer systems to act in a manner normally associated with human thought processes. Hence, to the business person, AI is not currently popular. However, there is a change on the horizon. Several expert system success stories and a few neural network applications are slowly changing people's minds. Often thought of as the stuff for the new millennium, AI is becoming a reality for many enterprises in the 1990's. Expert systems, rather than being an independent technology, are becoming part of an integrated strategy.
At ARCO we have had some success over the past five years. Since 1989, when we formed the AI Development Group, we have completed 15 production systems, several prototypes and have several more systems under development. We have completed software trades with other oil industry companies, and have signed licensing agreements with third party software vendors. We also have a small, but healthy research effort in the group. A patent has been awarded to one of the authors and another has been applied for. We have also had a few failures, but we have learned from these mistakes. In addition to AI systems we have also developed graphical user interfaces, GUI'S, and several "smart" systems especially databases.
DEFINITION OF AI Webster's dictionary defines AI as: "The capacity of a machine to imitate intelligent human behavior." We believe this definition can be improved if the additional clause "in a limited domain" is added. The scientific definition of AI is controversial and is derived from the question: . . . "Can machines think?" … an interesting subject that has been the topic of intense argument.
For practical applications, we propose an engineering definition of AI derived from a different question: "Which of the AI technologies can we use, and how can we use these technologies to make a computer more useful, productive, and smart?"
P. 209^