Emotion Research: Cognitive Science / Artificial Intelligence

Framework for Categorizing Computational Models

The primary research methodology used by AI and Cognitive Science researchers is the computational modeling approach. Researchers using this approach are exploring emotion at varying levels of abstraction (e.g., agent architectures vs. single phenomena), using different computational methods (e.g., connectionist vs. symbolic), addressing different emotional phenomena (e.g., emotions as heuristics used in planning vs. psychopathology), and basing their models on different theories of affect (e.g., cognitive appraisal theory and primacy of cognition vs. tight coupling between cognition and affect with no process having a dominant status). Several dimensions can be used to characterize the space of computational model of emotion.

• Level of Abstraction at which Emotions are Addressed
  
  The end points of this spectrum are represented by models of individual circuits or simple psychological phenomena on the one hand, and entire architectures integrating affective processing on the other. At the higher level of abstraction are architecture-level models which embody emotional processing. At an intermediate level of abstraction are task-level models of emotion, which focus on addressing a single task, such as natural language understanding or specific problem solving. At lower levels of abstraction are mechanism-level models, which attempt to emulate some specific aspect of affective processing.
  
• Ultimate Goal of the Research
  
  At one end of the spectrum models of emotion represent a means to an end and goals of these research efforts are to build more robust agents and systems, ones appearing to be more like humans, better able to interact with human users, exhibit better performance on specific tasks such as natural language understanding or specific problem solving. Following the nomenclature from the human performance modeling literature we term these models output models. At the other end of this spectrum, emotional processing itself is the object of study. The aim of these models is to better understand the nature of affective processing, interaction of the affective processing systems with other subsystems (e.g., cognition, neurophysiology), and the role of emotion in human behavior. Again, following the nomenclature developed in human performance modeling we term these approaches the process models.
  
• Basic Underlying Assumptions about Emotional Processing
  
  One end of this spectrum represents the view that emotional processing is secondary to cognitive processing, perhaps arising as an emergent property of cognition. The other end of this spectrum represents the position that emotional processing is a distinct form of computation in the brain, taking place alongside of what we generally understood as cognition (e.g., memory, learning, planning, etc.) and influencing it in subtle and pervasive ways.
  
• The Degree to which the Model or Theory has been Implemented
  
  The final test of a theory can only occur once the details have been adequately defined to allow thorough testing. In considering computational models it is important to specify to what extent the model has been implemented and tested.
  

Editor: Eva Hudlicka [psychometrixassociates.com]

Please send us your comments.

...Navigate The Emotion Site