Metatheoretical Issues in Cognitive Science

information may be more difficult to use without some concrete example. Abelson's script theory can not only cope with a lot of the nonformal vagaries of human reasoning but provides a handle on dealing with attitudes and cognitively mediated social behavior. Attitudes are simply the ensemble of scripts concerning some object, person or activity; cognitively mediated social behavior involves the selection of a script for a certain situation and taking a participant role in that script. Apparently, computer simulation may be able to handle many cognitive aspects of social interaction, as well as the proceses of judgment, decision making, and attitude formation. A problem, however, according to Miller (1974), is that though the computer metaphor is relevant to cognitive theories, it has little place for the affective components of a person's acts. Certainly, emotion plays an important role in social interaction, and even in solitary activities like reading and writing we must take account of emotional factors. In activities which are primarily cognitive there are emotional components like interest or boredom. To what extent could a computer system have any understanding of emotion? T o what extent can we learn about human emotions using computer simulation? Seemingly, it makes little sense to attribute "feeling" t o an inorganic system; however, we must also point out that there is a great deal more to emotions than affect. The seminal work of Shachter and Singer (1962) demonstrated that emotions are a function of situational cues plus the labelling of feelings, and subsequent work has carried these ideas much further. Emotion probably has a lot to do with interpreting situations via schemata with varying cognitive content (shame, pride, vanity, etc.). The bodily manifestation of an emotion may have more to d o with strength of emotion than with the particular form an emotion takes. It is true that bodily manifestations might not be readily simulatable. Nevertheless, the theoretical goal is to represent the origins and effects of emotions rather than mimic their actual bodily form. The core structure of cognitive choices, which even as phenomenologically oriented a writer as deRivera (1977) finds essential to each emotion, are far more accessable to implementation on an artificial symbol sytem. T o understand emotion it becomes important to articulate the background knowledge, the concepts, desires, abilities, plans, and so forth, in terms of which the emotion plays a role in a person's psychological life. It is via scriptal frames covering themes like comfort, rescue, violation, and so forth that a person's emotional life might be understood. Only through the rich interconnection between emotion, situation, activity, and other phenomena do we respond appropriately to friends, understand works of art , engage in productive work, and d o any number of other essentially human actions and projects. Clearly, since virtually all emotion has some sort of cognitive content, and since we are able to speak of emotions in a shared language expressing their qualitative content, it is quite probable that a computer system could "comprehend" the "cognitive content," and "speak" of emotions, responses, and so forth with some facility and appropriateness. "...To say that a computer could have no real underMETATHEORETICAL ISSUES 171 standing of emotions-no matter how plausibly it used emotional language-on the ground that it supposedly cannot experience feelings, is to make a highly dubious claim" (Boden, 1977, p. 442). Given a computer implementation that can make appropriate emotional interpretations of situations and other people's actions (whether directly or indirectly exposed to these), a computer could be designed to take into account the emotions of its human interactants. If a computer is going to simulate emotions, however, it has to be capable of placing itself as participant in emotion-laden scripts which have procedural implications for its own activities, symbolic and otherwise. Some emotions, and their corresponding (or identifying) procedural effects, may be crucial to the central controlling operations of the computer as a system. These sorts of effects may be quite important to any sort of intelligent system which must function in a world of such complexity. In a sense, many emotions can provide information about how to proceed: When confident we proceed with efficiency, things are working well, we do not have to be constantly checking out possible alternatives and dangers; at other times anxious checking of multiple possiblities may be in order. Sussman's (1975) HACKER simulation employs a distinction along these lines, working in CAREFUL mode when first attempting a new kind of problem, becoming more "confident" with greater experience. Emotional states like agitation or obsession might be represented by different sorts of search processesbreadth first versus depth first, for example. Other sorts of emotions, like humiliation and guilt, might be represented by quite culturally dependent sorts of themes and scripts-directing reoccurrence of certain mental operations, creating certain imperatives for action, and so forth. Many emotions provide warnings about present or future dangers to physical, psychological, social, or reflexively, emotional well being of the system. Such warnings may initiate searches for avoidance strategies, or provide suggestions of things to watch out for or keep in mind (particulars deserving attention or actions needing to be executed). These suggestions are admittedly speculative, but they do indicate directions for research and provide at least intuitive evidence for the usefulness of the computer metaphor in exploring various aspects of human emotional life. The essential cognitive or information processing component to emotion is also emphasized by some empirical work in psychology. For example, Isen, Shalker, Clark and Karp (1978) indicate that mood may be an effective retrieval cue, i.e., for a subject in a certain mood, certain sorts of cognitions are more accessable than others. The authors also suggest that mood may be appropriately conceptualized as a cognitive state. The claim is that the cognitive processes or changes which were originally thought to mediate the relationship between mood and behavior, actually produce, or are the same as, the mood itself. A number of problems may be solvable by reference to alternative modes of processing, different means of accessing or constructing various kinds of information, and different procedures for dealing with or operating on information in various circumstances. One of these problems concerns the role of fringe consciousness in attentional focus: since the operation of an attention focusing device must depend on the whole field, it must include unattended parts (Posner, 1978). Some beginnings to solving this problem have been attempted making a "local context" more accessible to ongoing processing (e.g., Winograd, 1972). A capacity for taking different perspectives may be programmed into a system via the operation of different retrieval strategies operating on memory, allowing the kind of changes in recall following a shift in perspective found by Anderson and Pichert (1978). Similarly, different "states" of consciousness may not be states at all but rather different organizations of encoding, processing, and retrieval operations of a single "state of consciousness." As Kaplan (1971) points out, it is unlikely that a general theory of cognition can be established without considering all forms of mentation: that of infants, children, schizophrenics, archaic and preliterate people, the senile, and of oneiric, hypnogogic and hypnopompic "states." Work in this direction is really only now beginning.