Artificial Intelligence (A.I.) is a multidisciplinary field whose goal is to automate activities that presently require human intelligence. Recent successes in A. I. include computerized medical diagnosticians and systems that automatically customize hardware to user requirements. The major problem areas addressed in A.I. can be summarized as Perception, Manipulation, Reasoning, Communication, and Learning. Perception is concerned with building models of the physical world from sensory input (visual, audio, etc.). Manipulation is concerned with articulating appendages (e.g., mechanical arms, locomotion devices) in order to affect a desired state in the physical world. Reasoning is concerned with higher level cognitive functions such as planning, drawing inferential conclusions from a world model, diagnosing, designing, etc. Communication treats the problem understanding and conveying information using language. Finally, Learning treats the problem of automatically improving system performance over time based on the system’s experience.
Many important technical concepts have arisen from A.I. that unify these diverse problem areas and that form the foundation of the scientific discipline. Generally, A.I. systems function based on a Knowledge-Base of facts and rules that characterize the system’s domain of proficiency. The elements of a Knowledge Base consist of independently valid (or at least plausible) chunks of information. The system must automatically organize and utilize this information to solve ‘&e specific problems that it encounters. This organization process can be generally characterized as a Search directed toward specific goals. The search is made complex because of the need to determine the relevance of information and because of the frequent occurrence of uncertain and ambiguous data. Heuristics provide the A.I. system with a mechanism for focusing its attention and controlling its searching processes.
The necessarily adaptive organization of A.I. systems yields the requirement for A.I. computational Architectures. All knowledge utilized by the system must be represented within such an architecture. The acquisition and encoding of real-world knowledge into A.I. architecture comprises the subfield of Knowledge Engineering.
This article provides a basic Introduction to Artificial Intelligence (A.1.). No prior experience with or knowledge about A.I. is assumed. The following sections describe the fundamental goals of A.I., the major sub-disciplines of A.I., other fields with which A.I. is intimately involved, the central issues in A.I. today, some more technical details that elaborate the general discussion, important general approaches to development of A.I. applications, and finally an assessment of the current accomplishments and position of the field.
The Field of A.I.
The ultimate scientific goal of Artificial Intelligence is to construct a formal model of Mind that fully explains the activity of “thinking” as it is performed by humans. The methodology of A.I. research dictates that such a model can only be validated if it is made operational. Thus, A.I. is centrally concerned with building operational systems that exhibit behaviors that illustrate the systems’ ability to “think”.
From a practical perspective, automated thinking systems promise to be quite useful; e.g., as consultants to human experts, as autonomous entities that continuously monitor and control complex industrial processes, as systems that rigorously explore the possibilities in a design space, as robots that intelligently explore physical environments that are too harsh for reasonable exploration by humans, etc. Based on this perception of utility, the scientific endeavor of A.I. has spawned an engineering discipline whose primary goal is to build useful systems that incorporate the techniques for implementing intelligent rational behavior that have emerged from A.I. research. At this point in time, A.I. has developed techniques that render the most primitive of such systems feasible.
Humans exhibit intelligent behavior in a plethora of activities. To pursue the goal of automating intelligent reasoning, A.I. researchers have explored many of these different activities and attempted to build systems that successfully implement the activities. This has led to a subdivision of the scientific field of A.I. into specialized areas that are concerned with the “types of thinking” that characterize each fundamentally different type of human intelligent behavior. From the engineering perspective, each different type of intelligent activity can be used as the basis for different useful applications that require specialized implementation techniques. Some of the important intelligent activities that A.I. has explored, and is continuing to explore, are summarized below.
Perception can be defined as the ability to identify objects and relationships between objects in the world based on sensory input data. There are many examples of the requirement for intelligence in the performance of this task. For example, recognizing that a visual image is an instance of the concept of “chair” requires a deep understanding of the function of a chair, its likely orientations and configurations, etc. Similarly, the ability to deduce that a table you are sitting at has several legs when only a subset of those legs is visible might require reasoning about structural soundness and likely symmetry. One of the primary subfields of A. I. is examining the general problem of building an accurate model. of a physical environment based on visual sensory data. Less-developed subfields are attacking the analogous problem for other types of sensory data; e.g., radar and sonar. These research endeavors have identified as a common property the necessity for the A. I. systems to have knowledge about the types of objects that are likely to be found in their respective environments, the physically valid configurations of these objects, the intended functions of the objects as a basis for identifying possible forms, the expected shapes of the objects, etc. All this knowledge must be organized to generate, constrain, and evaluate possible interpretations of the raw sensory data.
Manipulation can be defined as the ability to articulate oneself or one’s environment in order to establish some desired physical situation. Robotics is a subfield of A. I. that is researching this form of intelligent activity. As an example of the requirement for intelligence in successful manipulation, consider the simple problem of moving oneself from one location to another. This requires planning to determine a path that is likely to lead to the desired destination, knowledge about how to alter this path if unexpected obstacles arise, knowledge about what constitutes a reasonable place to step and how to identify possible problems (e.g., slippery spots, unstable spots), etc. A s in the case of perception, successful manipulation requires a substantial amount of knowledge about the domain in which the system is operating together with the ability to apply this knowledge to an analysis of specific goals and to the exigencies that arise.
Communication can be defined as the ability to transfer information from one entity to another. Research in A. I. on this form of intelligent behavior has concentrated on the problem of communication between humans using language. To understand a specific linguistic communication, it is necessary to have a strong model of the domain about which that communication is concerned. As a simple example, consider the sentence, “John saw the girl with the telescope.” Ir. this case, the girl could possess the telescope, or John could have used the telescope as an instrument to accomplish his viewing of the girl. This ambiguity could be resolved with further information; e.g., “John had a telescope and a pair of binoculars. He saw the girl with the telescope.”, or “There were two girls gazing at stars on the hill. One of them had a telescope. John saw the girl with the telescope.” Recognizing the possibility of an ambiguity at all in the original sentence requires knowledge about the properties of objects in the 8ornain; e.g., “John saw the girl with the orange” does not have the same ambiguity (at least in “rational” contexts). Thus, properly interpreting the meaning of language requires deep knowledge about the subject matter of that language and the ability to reason about the situation that the language describes. Similarly, producing good linguistic descriptions requires the ability to anticipate the understanding powers of the listener, so that the language one generates does not seem too redundant nor too terse.
which of A.I. is concerned with more abstract forms of Reasoning that do not involve a direct coupling with a physical environment. For example, there has been work in the automated design of devices of a given type (e. g . , V L S I ) to meet specific goals, in the automatic construction of computer programs that accomplish a specified computational task, in the automated diagnosis of disorders in humans (medicine) or in machines, in game playing (e.g., Chess), in the intelligent scheduling of complex activities, in the automated solution of various mathematical problems etc. A very important subfield of these abstract reasoning A. I. applications is that of Expert Systems. An Expert System can be defined as the computer embodiment of the knowledge and general problem-solving expertise that is used by a human exert problem solver in a specific domain (e.g., a diagnostician or a financial planner). A common property of all these reasoning applications is the fundamental requirement of the system to have both extensor-e knowledge about its area of expertise and the ability to accomplish analytic reasoning by applying this general knowledge to the peculiarities of situations.
An area of A. I. that is beginning to receive serious research attention is that of Learning. Several fundamentally different types of learning have been identified. For example: a) tuning, where a system’s performance is gradually improved through its experiences in a particular domain without any fundamental alteration of the system’s basic approach to its problem-solving) concept formation, where a system identifies and utilizes new fundamental concepts that arise from its experiences, and c) discovery, where a system explicitly undertakes the process of exploring the properties of its domain in order to learn interesting or useful attributes. Learning, especially in its more general forms, is an extremely difficult activity. However, there are those who argue that people are always learning, and that learning is a fundamental component of any intelligent behavior.
In order to accomplish its fundamental goals, A.I. both draws upon and makes contributions to other fields of study. For example, Computer Science is the primal reservoir of knowledge about the engineering of software systems. Examples of spin-offs from A. I. into general Computer Science are video games, time sharing, and word processing. A.I. has consistently been a fruitful area for the development of new areas of software application and of new software engineering techniques. Cognitive Psychology is a field that has some of the same higher-level goals as A.I., viz., the development of successful models of Mind. However, the methodology of Cognitive Psychology is based more on observation of and experimentation with humans as opposed to the construction of operational implementations required by A.I. Philosophy has historically been concerned with the principles of valid reasoning. Many of the techniques for automating reasoning developed in A.I. have their roots in Philosophy (primarily in Logic). Linguistics is concerned with the study of language. A.I. research in language has drawn upon linguistic models and has developed new concepts about the important features of language that derive from A.I.’s insistence on operational functionality (e.g., A.I. has been concerned more with the meaning thatis conveyed by language while linguistics has been primarily concerned with the structure of language). Finally, physical properties of the world have fundamental importance in the areas of perception and manipulation.
Another view of the system engineering side of A.I. is that it is concerned with pushing the frontiers of applied Computer Science in the direction of systems that are flexible, adaptive, able to cope with the unexpected, and able to formulate their own analyses and solution procedures by applying general knowledge to specific situations.