Both AI and the DS fields have grown up in roughly the same time period - roughly the last 40 years. While some of their bases can be traced back centuries (e.g. Bayesian Analysis - see Chapter 5), these fields have both matured with the digital computer. This is likely because the computational burden of their techniques otherwise limited their growth. The birth of the term "Artificial Intelligence" is attributed to Professor John McCarthy of Stanford. It came into common usage with a conference held at Dartmouth in the summer of 1954. In a sense the term is self descriptive. Artificial is anything man made. And intelligence is .... well maybe not so easily defined. Because there is a general lack of agreement on a definition for intelligence there remains considerable debate about the term artificial intelligence. Some have gone as far as to say that AI is simply not possible. For as soon as AI researchers break a barrier in what could be termed "intelligent" behavior by the computer, that behavior is no longer considered to possess the attributes of intelligence. Refer to Saunders [PS9] for a more in depth discussion of this paradox.
Perhaps the most interesting definition for intelligence has been provided by Professor Douglas Hofstadter in his Pulitzer Prize winning book Godel, Escher, & Bach[AI6]. He defines intelligence as the ability to;
respond to situations very flexibly,
take advantage of fortuitous circumstances,
make sense out of ambiguous or contradictory messages,
recognize the relative importance of different elements of a situation,
find similarities between situations despite differences which may separate them,
draw distinctions between situations despite similarities which may link them,
synthesize new concepts by taking old concepts and putting them together in new ways, and
come up with ideas which are novel.
This list may well be the crux of what AI scientists are attempting to embed in software. But at this time, we are far from this goal. On a much more practical note, Patrick Henry Winston[AI11] has phrased a benign definition for AI by focusing upon its goals. They are; 1) to make computers more useful, and 2) to understand the principles that make intelligence possible.
We will have considerable latitude in working with AI methods within the context of this definition and this paper, but will not include any methods which have no place in one of the more renowned works in the field. These works include but are not limited to Winston[AI11], Giarrantano[ES2], Feigenbaum Barr & Cohen[AI2-5], and Shapiro[AI1]. The origin of the decision sciences is much less clear. However early research and the eventual publishing of von Neumann and Morgenstern's book The Theory of Games and Economic Behavior [DM7] in 1944 is a distinct watershed. Simon's The New Science of Management Decision [DM6] is also considered an early classic work.
Complete areas that have grown, capitalizing in part upon this early work, include decision analysis, mathematical programming, production & operations management, forecasting, organizational behavior & development, and parts of economics and statistics. Since the Decision Sciences now encompass such a large, diverse area, single reference sources of a renowned nature do not exist. Yet some areas do continue to stand out as focusing upon the broader decision science issues and concepts. These include decision making, decision analysis, and decision support systems.
An excellent treatment of decision making is available through Harrison[DM3]. Decision analysis techniques and principles are covered well by Watson and Buede[DA3] and Gregory[DA1]. Finally Turban[DS3] and Hopple[DA3] both address the broad topic of decision support systems. Both AI and DS continue to engage debate in the efficacy of their right to be called distinct disciplines. It is frequently stated that we are now in an AI "winter", i.e., a "dark" period for AI. And the foundations and practicality of the decision sciences are frequently challenged by students and practitioners alike. But debate is a fundamental tenet in establishing any discipline.
Fortunately both AI and the decision sciences are based upon theory fairly well established in other areas including statistics, psychology, computer science, economics and engineering. In a similar manner, both AI and the DSs have as their central goal the solving of general problem classes such as resource allocation with constraints, planning of all types, and prediction. A more thorough list of these general problem areas follows in the next chapter. If the focus of AI and the DSs is problem solving, then before we can examine each of these disciplines and their confluence, we should first explore the foundation areas problem solving, decision making, and modeling.