The "Fungus Eater Approach" to Emotion: A View from Artificial Intelligence

The study of emotion has always been a difficult and controversial subject. The main reason seems to be that in spite of the many years of investigation and the abundance of studies aiming at understanding human emotions, virtually no consensus could be reached. In this paper we argue that there are fundamental underlying problems and that a radically different approach is needed. We propose a new one, called the "New Fungus Eater Approach". It is illustrated by experiments with autonomous robots, the “New Fungus Eaters”, which are named after their predecessor, the “Solitary Fungus Eater” invented by Masanao Toda in 1961. This approach is outlined and it is demonstrated how it can contribute to understanding the foundations of what one might want to call emotional processes. It is also discussed how some of the basic controversies "disappear" in this way. Introduction: The “Solitary Fungus Eater” In 1961 the Japanese psychologist Masanao Toda 1 proposed to study “Fungus Eaters” as an alternative to the traditional ways of academic psychology (Toda, 1982, Chapter 7): Rather than performing ever more restricted and well-controlled experiments on isolated faculties (memory, language, learning, perception, emotion, etc.) and narrow tasks (memorizing non-sense syllables, letter perception on degraded stimuli, etc.) we should study “complete” systems, though perhaps simple ones. “Complete” in this context means that the systems are capable of behaving autonomously in an environment without a human intermediary. Such systems have to incorporate capabilities for perception, for decision making and for generating motor action. The integration of these capabilities into a system which is capable of behaving on its own, so the argument went, will yield more insights into the nature of mind or intelligence than looking at fragments of the awesomely complex human mind. The “Fungus Eater” approach can be seen as a precursor of a more ecologically minded psychology (e.g. Bateson, 1972, Neisser, 1976, 1982). The “Solitary Fungus Eater” is a creature sent to a distant planet to collect uranium ore. The more ore it collects the more reward it will get. It feeds on a certain type of fungus which grows on this planet. The “Fungus Eater” has a fungus store, means of locomotion (e.g. legs), means for decision making (a brain) and collection (e.g. arms). Any kind of activity, including “thinking”, requires energy. If the level of fungus drops to zero the “Fungus Eater” is dead. The “Fungus Eater” is also equipped with sensors, one for vision 1This is my own interpretation of his paper. I apologize for any misinterpretation. and one for detecting uranium ore (e.g. a Geiger counter). We propose to use the “Fungus Eater” to study emotions. We start from the assumption that both the Artificial Intelligence (AI) researcher and the emotion psychologist can potentially benefit from each other. First we try to delineate the concept of an emotion and point out some basic controversies. Then we discuss some fundamental problems of AI models of emotion. We then present “New Fungus Eater Approach” and discuss its merits and problems. The ideas reported in this paper have emerged from efforts over many years to develop AI models of emotion. The experiences made and the frustrations we suffered have been among the central motivating factors to work out a new and completely different approach to break loose from some of the impasses and controversies that seem inescapable. With this manuscript we hope to promote a constructive discussion which will hopefully contribute to advancing the field of emotion psychology. The concept of emotion Definition, issues, controversies In order to delineate the concept of an emotion we should try to define it. However, the large number of definitions which can be found in the literature (e.g. Kleinginna & Kleinginna, 1981) is convincing evidence for the ill-definedness of the concept. This definition problem may also be among the reasons why there are quite a few important controversial issues around. We will list some of them because this way we can get somewhat an idea of the contents of the research in emotion psychology. Examples of these controversies are: the “components” of emotion: can emotions be decomposed into “components” (like a cognitive-evaluative, a physiological, an expressive-communicative, and an action “component”)? basic emotions: is there a (relatively small) set of basic emotions (which cannot be decomposed)? primacy of cognition vs. primacy of other factors (physiological patterns, expression, action, subjective feeling): which one of these causally entails the others? general arousal vs. discrete patterning of emotions; universality vs. cultural specificity of emotions; the functions of emotion; emotion generation (i.e. how emotions come about); factors influencing the intensity of emotion There is a lot of literature about all of these issues. Discussing them would be much beyond the scope of this paper. It would in fact be against the spirit of the paper since we are arguing that many of them are the result of an inappropriate conceptualization and will disappear if a different approach is pursued. The study of the extremely rich literature yields a large number of different theories which are all highly interesting and intuitively appealing. But it somehow seems impossible to say why one should be preferred over another. For example, which one of the following should be preferred, and on what basis (the choice is 100% arbitrary): Frijda (1986), Izard (1977), Mandler (1985), Oatley & Johnson-Laird (1987), Ortony, Clore, & Collins (1988), Rolls (1990) or Scherer (1984)? Clearly, they do not all have the same focus, but there is a significant overlap in the kinds of phenomena they are aiming to explain. The large number of theories which exist in parallel is an additional indication that there are some fundamental problems with the psychology of emotion. One of the important aspects which is really lacking is a methodology which would be acceptable to all and which could be applied to evaluate the various theories. AI models of emotion A detailed review of AI models of emotion has been given in Pfeifer (1988). For the purposes of the present discussion we will only highlight a few important points. The models described in the literature (nearly) all fall into the categories of psychological modeling on the one hand and of “reasoners” about emotion on the other.