IS A PICTURE REALLY WORTH A THOUSAND WORDS?

If the author wishes to bring the imagery debate to the artificial intelligence (AI) community, I believe it is important that we begin with a better definition of what one means by “imagery.” Of course, almost all humans have an intuition about what it is like to imagine something; but it is doubtful that such experiences inform us very much about the format or processes underlying imagery. However, this does not lead me to dismiss empirical methods as the primary means for understanding imagery. Cognitive psychologists are constantly facing this challenge-our goal is to develop paradigms that circumvent or otherwise control for subjects’ intuitions, but result in measurable behaviors that reveal properties of the underlying mechanisms and representations of cognition. Thus, at least for a first pass, I wish to retain the notion of imagery as a uniquely human phenomenon and suggest that it should be studied as such. In some ways this may seem to be a highly unsatisfactory assumption, for we are left with a definition of imagery that is a tautology: imagery is simply those instances in which humans imagine. However, there is more to this definition than meets the eye. First, because this definition necessitates a conscious being to experience the image, it immediately narrows the meaning to encompass only the human (and possibly other animals’) experiences of imagery, while eliminating instances in which a so-called image is processed by a machine (at least until we develop conscious machines). Second, because this definition is modality neutral, it accommodates any experience in which we encounter images, including imagery in all of our senses, as well as nonsensory thoughts and desires. This issue is raised, but is rightly tempered by the fact that most studies of imagery, as well as the debate itself, have centered on visual imagery. However, while the author takes this as a basis for quickly shifting from imagery to spatial representations,’ this emphasis obscures the essential point of the imagery debate-whether or not human information processing beyond early perceptual input utilizes modality-specific representations. Thus, the imagery debate was not about whether humans have visual representations (of course they do), but whether such representations are retained and play some functional role in our cognition. This point is crucial to current theories of imagery-rather than simply stating that we have representations that “are like” structures in the world, it has been argued that we retain perceptual input in a form that allows us to reinstantiate the prior perceptual processing of this information by activating some of the same structures used for the original perception. Thus, it is not the fact that images are derived from percepts that makes them images, but rather that they can function as percepts at some later time.