emotions "affective computing" 1
emotion and thought Globe & Mail, March 25, p. A15 "Emotions are an intrinsically human process..", [Dr. Mayberry] said. "One cannot separate emotions from thinking." Are there reasons for AI to incorporate "emotional state" into programs? 2
reasons for studying affective computation synthetic personalities games & interactive entertainment synthetic actors are examples of autonomous agents use "emotional state" in implementing autonomy (partial?) solution to the agent's "what to do next?" 3
forms of autonomy independence from external control untethered robots functionally independent softbots agents which treat computers as locations in which to act move from computer to computer distributed AI : benign "viruses"? agent chooses among multiple competing goals based on its own history and state implicit optimization? of what? 4
"what should I do now?" why should an agent do anything? very high level "meta-goals"? survival? => control over outcomes? equilibrium? ability to recover from equlibrium also implies control need very general evaluations of current and possible states emotions are one way of structuring general evaluations of events, objects, relationships 5
emotions as a "logic" for actions use emotions to enrich the bases for selecting actions and responses: contrast differential evaluation of the field of possibilities if one is happy or one is angry lots of complexity and subtlety in structure of emotions: contrast jealousy and resentment 6
sources for structures? cognitive science psychological modelling from a computational perspective concepts overlap with AI cognitive scientists want interesting models that "work" AI researchers want interesting implementatiion problems relating to human intelligence and cognition 7
research examples Ortony, Clore, Collins, The cognitive structure of emotions, 1988 Frijda and Swagerman, "Can computers feel? Theory and design of an emotional system" Cognition and Emotion 1987 Bates, J. et al., "An architecture for action, emotion, and social behaviour", 1992 8
ingredients what are computational theories made of? memory structures - what else? categorizations variables rules for assigning values 9
categorizations (Ortony) Emotions about things: liking, disliking. Emotions about persons: approving, disapproving about self: pride, shame about others: admiration, reproach. 10
events Emotions about events for self: pleasing, displeasing for other: gloating, pity, resentment, happy-for. about events in the future: hope, fear. realized events (positive): satisfaction, relief. (negative): disappointment, fears-confirmed. Emotions about another person's role in events: gratitude, anger. about self's role: 'gratification', remorse. 11
a "frame" type: e. g. hope valence: e. g. positive subject: self time: future etc. 12
a typical theory (Frijda) "emotions" are structures created by an evaluation process primary: matching input events to concerns matches are + or - or 0 (summation = "affect": + or - feelings) secondary: evaluate attributes of significant event in context of current situation is it fair? is it expected? is it controllable? 13
outcome of appraisal control precedence value = interrupt priority changed action readiness appraisal activates action tendencies action tendencies approach or avoid attend or ignore activation intense, low, neutral 14
modulation/regulation of behaviour "experience of the valenced relevance signal, interruptability of the control precedence, and the impulse corresponding to the action tendencies, is the subjective feeling of an emotion." a model of the theory models the "feeling" by modelling the structures and values and relations given by the theory 15
testing a theory what tests should a cognitive theory of emotion pass? what tests should an AI model of emotion pass? 16
the dark side cartoon in the Star, March 23, 2000 the salesman offers the lady a "Very User Friendly" computer "and every 15 minutes it announces that it loves working with you very much." the "media equation" Reeves, B., and C. Nass, The media equation: how people treat computers, television, and new media like real people and places, Cambridge University Press: 1996. 17
it's all pointers "The basic data structures in AI might be the rule, the frame, and the script... but from a programmer's point-of-view these are really interchangeable-- just arrangements of pointers in memory-space. So the real challenge is to discover which interpretations of these various arrangements of pointers best match the way human thinking works." Jorn Barg 18
labels aren't meanings DON'T take the labels and names in your program literally to find out what a computational structure really "means": consider the terms and statements as pure graphs, with all symbolic labels stripped off. example: member(x, [X]). member(x, [_ Y]) :- member(x, Y). does member mean here?. 19
member(x, [X]). member(x, [_ Y]) :- member(x, Y). All black circles represent distinct variables. Diamonds represent functors (same functor indicated by same colour.) The colour of a line corresponds to the position of the argument in the term. 20