The Ontology of Emotions

What, in the end, are emotions? What do they ultimately consist in? A variety of possible answers to this "ontological" question suggest themselves in the light of the above account. They might be physiological processes, or perceptions of physiological processes, or neuropsychological states, or adaptive dispositions, or evaluative judgments, or computational states, or even social facts or dynamical processes. In fact most philosophers would assent to most of these descriptions while regarding all as partial. In view of the acknowledged complexity of emotional functions, it seems wise to rephrase the question not in terms of ontology, but in terms of levels of explanation. The trichotomy first introduced by David Marr (Marr, 1982) remains an excellent starting point. At the computational level (which most would now call the functional level), we need to identify the emotions' basic teleology: what they are for. This will be appropriate even if one believes, as some traditionally have, that emotions actually represent the breakdown of smoothly adaptive functions such as thought, perception and rational planning. For in that case the emotions will be understood precisely in terms of their failure to promote the smooth working of the cognitive and conative functions. Such a failure will trigger a descent to a lower level of explanation, adverting to the counterproductive exercise of mechanisms at the algorithmic and implementational levels. The first—-more or less equivalent to the design level of (Dennett, 1971)—refers to the sub-functions that natural selection has set up to perform the functions said to be disrupted by emotion. The second designates the actual neurophysiological processes whereby, in animals built on a specific plan such as mammals or humans such as we, these sub-functions are normally carried out.

This trichotomy has been reinterpreted in various ways, but it still serves. It is generally agreed that the simpler emotions, those whose expression and recognition Ekman (1972, 1989) has shown to be universal, are driven by the basic needs of organisms such as mating, defense or avoidance of predators and social affiliation. All complex mammals require swift, relatively stereotyped responses to these challenges. These are the "affect programs" favored by Ekman (1972, 1989), DeLancey (2001) and particularly Griffiths (1997), to be "what emotions really are." Opinions divide as to whether the same sort of functional analysis can be applied to a wider range of what Griffiths has called the "cognitively penetrable" emotions. Placing severe constraints on what is to count as a "natural kind", Griffiths argued that Ekman's six basic affect programs and only they, form natural kinds: the others, he claimed, are for the moment beyond the reach of useful scientific investigation. Each affect program comprises a coordinated syndrome of responses (which we attribute to the algorithmic level) implemented at the physiological (hormonal and neurological), muscular-skeletal and expressive levels in ways that owe their uniformity to homology, that is to say their common ancestral origin. Other emotions, however, bear only relations of analogy with these and don't count as natural kinds either singly or as a class.

Against this Charland (2002) has argued that a sufficient level of homology can be found to unite at least the basic emotions as a class and that we should regard emoters and hence their emotions, as a natural kind. Relying on Panksepp (1998, 2000), Charland argues that the integrated mechanism of seven basic emotions (Panksepp's list differs slightly from Ekman's) are implemented by distinct circuits forming natural kinds not only in the human but more widely in the mammalian brain. Emoters form a distinct kind in view of their ancestral organization in terms of certain basic functions, the specific algorithms that contribute to those functions and their implementation in terms of physiological, expressive, hormonal and motivational processes. This is sufficient not only to justify treating the specific emotions as natural kinds, but to treat emotion in general as a natural kind. (Charland 1995, 1997). This view seems to require that we regard emotions as a set of processes distinguished at all three levels of explanation. Emotions in general should then be viewed as a genus of processes typically involving five different component aspects or components, comprising subjective feeling, cognition, motor expression, action tendencies or desire and neurological processes (Scherer, 2005). On this view, individual emotions would owe their specific identity both to the subfunctions they are designed to serve and to their characteristic physiological implementation.

Another way of organizing the various approaches might appeal to the dominant theoretical models on which they rest. It has often been said that in the history of the philosophy of mind, every epoch has tended to redefine its subject matter in terms of the most fashionable technological metaphor. The notion of emotions as "springs of action" alludes to the once fashionable model of clockwork. The dominant metaphor in Freud's early work was hydraulic (Freud, 1895). What does this observation lead us to expect for emotions?

Modern conceptions of emotions, as we have seen, have been frequently couched in terms of other mental terms. In these cases, there is nothing sui generis that emotions are: any "ontological" question about their nature belongs derivatively to the ontology of desire and belief. (At a different, more remote level of explanation, theories favoured by cognitive science are likely also to appeal to evolutionary ideas.) This leaves three other dominant contemporary models which one could expect to lay claims on emotion theory: physiology, computation and dynamical systems.

Physiological processes are conceded by all philosophers to be involved in clearly prototypical cases of emotion. But no philosopher, for fear perhaps of defining themselves out of relevant competence, has been willing to concede that emotions just are physiological processes. Instead they are held to be complexes in which physiology plays a part at the level of implementation of some higher-level process. The higher-level process in which an emotion consists owes its overall structure to functional needs and typically comprises, in addition to physiological aspects, behavioural, expressive and phenomenological, components.

Computational theories of emotion seem to have been particularly attractive to psychiatrists and psychoanalysts. They were broached early by a couple of psychoanalysts turned hackers (Peterfreund, 1971), (Shank and Colby, 1973) and played an important role in the theoretical elaborations of John Bowlby's work on the mechanisms and psychological consequences of early separation and loss. (Bowlby, 1969-1980). These works attempted to model Freudian concepts of the dynamics of conscious and unconscious mental life in computational terms. Colby even constructed a simulation of a paranoid patient, "Parry", which famously fooled some psychiatrists. The key idea was to set up second-order parameters that acted on the first-order modules of perception, belief and desire, thus regulating or disrupting the operation of perceptual and action programs. From the sidelines, de Sousa (1987) suggested that connectionist systems or analog models stand a better chance of modeling emotion than those based on classical von Neumantype digital computation, but that suggestion hasn't gone anywhere. From the point of view of computational theory, the prevailing wind, backed by both evolutionary speculation and neurological findings on control systems and relatively independent affect-programs, has tended to favour modular conceptions of emotion rather than holistic ones. (Charland, 1995, Robinson, 2003).

Still, some philosophers and computer scientists have continued to be interested in integrating computing theory with emotions. Aaron Sloman has elaborated the sort of ideas that were embryonic in Shank and Colby into a more sophisticated computational theory of the mind in which emotions are virtual machines, playing a crucial role in a complex hierarchic architecture in which they control, monitor, schedule and sometimes disrupt other control modules. (Wright, Sloman and Beaudoin 1996). The notion of architecture here adverts to the complex hierarchy of control of component modular mechanisms. In line with the three-level schema I have cited from Marr, (cf. also (Dennett, 1971)) we should understand the approach elaborated in this work as pertaining both to the functional and to the algorithmic level. It explicitly eschews hypotheses about implementation. Joining the growing consensus that emotion phenomena reflect distinct, successively evolved behavioral control systems, Soloman distinguishes between a primitive or primary stream rooted in relatively fixed neurophysiological response syndromes, a more elaborate control system bringing in cortical control, as well as a third level, probably exclusive to humans, which most closely corresponds to the layer of emotions that we are most concerned with when we think of the emotional charge of art and literature or of the complexity of social intercourse. Rosalind Picard (1997) lays out the evidence for the view that computers will need emotions to be truly intelligent and in particular to interact intelligently with humans. She also adverts to the role of emotions in evaluation and the pruning of search spaces. But she is as much concerned to provide an emotional theory of computation as to elaborate a computational theory of emotions. Marvin Minsky (2006) explores the many-faceted nature of mental life, including emotions, from a computer modeling point of view. Paul Thagard (2005) has elaborated computer models in which emotional valence interacts with evidential strength to determine a mode of emotional coherence.

Dynamical systems theories have been relatively slow to emerge, despite their increasingly fashionable status in more central areas of cognitive science and contrary to the prediction made by de Sousa (1987) that connectionist systems or analog models would be more successful in modeling emotion than those based on classical von Neumantype digital computation. One remarkable attempt to integrate the perspective of dynamical systems into understanding of emotional life is that of (Magai and Haviland-Jones, 2002), who draw on dynamical systems theory to model the elusive combination of unpredictability and patterned coherence found in the life-long evolution of individuality. Like predecessors such as Bowlby (1969-1980), they are motivated by a goal of understanding at the level of conscious experience as well as of underlying mechanisms: dynamical systems theory is only one of their tools. It is therefore particularly pertinent to the preoccupations of those who are interested in the normative dimensions of emotions: their rationality and their irrationality, their capacity for enhancing or inhibiting self-knowledge and their moral implications. I address these questions in the next three sections.

< Prev   CONTENTS   Next >