IMAGERY: COMPUTATIONAL AND COGNITIVE PERSPECTIVES

This “Taking Issue” issue is about imagery. In particular, it is about a computational approach to imagery, an approach motivated by the laudable goal of endowing artificial systems with the capacity to exploit imagery in problem solving, much as we humans do oftentimes. This, of course, makes cognitive perspectives on imagery relevant to this issue as well. As Robert Sommer (1978) says, the processes of imagery-creating a visuospatial mental image, inspecting it, manipulating it, and making inferences from it -are mediative between the outer world (the objects being imaged) and the inner world (long-term memory representations of these objects). In other words, mental images, while being derived from the inner world, are entities that appear closer, or analogous, to the outer world. This property of imagery manifests itself in various ways. For one thing, images make explicit to 11s (and therefore make it easy to access) various visual (color and texture, for example) and spatial (e.g., relative locations) properties of objects being imaged. For another, mental images can be manipulated in ways that resemble spatial processes that objects in the outer world undergo. Both these characteristics are effectively exploited by humans when imagery is used in problem solving (for instance, Huttenlocher 1968 describes the use of imagery for solving syllogisms, and Shepard and Cooper 1982 describe experiments on mental rotation for solving object matching problems). This alone makes the question “how can processes of imagery be implemented, and used for problem solving, in a machine?” worthy of exploration to those who view the emulation of human intelligent behavior as a major goal of artificial intelligence (AI). Indeed, this is the question that Janice Glasgow addresses in her position paper, titled “The Imagery Debate Revisited: A Computational Perspective,” in this issue. The tantalizing nature of imageryit is quite widely experienced; however, its highly subjective nature stymies objective scientific analyses, at least with tools available in the early days-made it a subject of considerable interest, if not controversy, from the time of philosophers like Aristotle. More recently, the development of a variety of experimental methods (protocol analyses, reaction-time studies, tracking eye movements, and positron emission tomography scans, to mention a few) has enabled cognitive psychology to accumulate an impressive body of research on imagery. The curious reader is referred to Block ( 1 9 8 1 ~ ; 1981b), Finke (1989; 1990), Logie and Denis (1991), Pylyshyn (1978), and Tye (1991) for thorough elucidations of recent psychological research on, and philosophical issues surrounding, mental imagery. Why should A1 researchers be interested in all this, you may be inclined to ask? What should be of interest to them is, that there is a significant amount of research that analyzes imagery-related phenomena within an information processing framework, a framework in