Article Preview
TopIntroduction
Over the last fifteen years, researchers have started to investigate different forms and roles of affect in virtual agents and robots. Their efforts are, at least in part, based on the recognition that affective states like pleasure, happiness, elation, admiration, anxiety, remorse, disgust, anger, and many others are involved in many cognitive processes in humans and animals, and that, as a result, affective processes might have important functions in cognitive architectures which might benefit artificial agents. There is ample evidence from psychology (Frijda, 1986; Izard, 1993; Scherer, Schorr, & Johnstone, 2001), neuroscience (Damasio, 1994; LeDoux, & Fellous, 1995; Panksepp, 2000; Hamm, Schupp, & Weike, 2003), and ethology (Lorenz & Leyhausen, 1973; Lorenz, 1981; McFarland, 1981) that affective processes are, among other functions, involved in (1) the initiation, selection, regulation and coordination of behavior, (2) the management of motivation and goals, (3) the formation of memories and memory recall, (4) attentional control, (5) different forms of associative and reinforcement learning, (6) social signaling and reacting to signals of other animals.
In simple organisms with limited cognitive and representational capacities, affect seems to control behaviors mainly by providing an internal measure of “what is good and bad” for the organism (e.g., Humphrey, 1992). The basic evaluation in terms of hedonic values causing the organism to be attracted to what it likes and to avoid what it does not like (e.g., Gray, 1990) forms the basis of the organism’s behaviors. If another organism poses a perceivable threat, a fear-anger system (Berkowitz, 2003) may generate fight-or-flight behavior. And while emotional states such as fear and anger control immediate actions (LeDoux, 1996), other affective states may operate on longer term behavioral dispositions (e.g., anxiety caused by repeated triggering of fear leads to increased alertness without the presence of any immediate threat).
In humans, affect is deeply ingrained in the cognitive architecture, biasing and influencing attentional mechanisms (such as interrupting and distracting the current processing; see Derryberry & Tucker, 1994, or broadening the attentional focus, see Fredrickson, 1998). Positive and negative affect often influence problem solving strategies, with negative affect causing local, bottom-up processing, while positive affect tends to cause global, top-down approaches in many cases (Bless, Schwarz, & Wieland, 1996; Schwarz, 1990). Humans also seem too often rely on affective memory (e.g., Blaney, 1986; Bower & Cohen, 1982) to evaluate a situation quickly instead of performing a longer, more complex cognitive evaluation (Kahneman, Wakker, & Sarin, 1997), which suggests that affective evaluations might encode implicit knowledge about the likelihood of occurrence of positive or negative future events (e.g., Damasio, 1994; Clore, Gasper, & Conway, 2001). Finally, affect is crucially involved in social coordination (Frijda, 2000; Cosmides & Tooby, 2000) ranging from signaling emotional states (e.g., pain) through facial expressions and gestures (Ekman, 1993) to perceptions of affective states that cause approval or disapproval of one’s own or another agents’ actions (relative to given norms), which can trigger corrective responses (e.g., feeling guilty).