Picard, Affective Computing, Chapter 1

In this chapter Picard frames her discussion with what she—and most others—see as the two primary aspects of human emotion—the physical an the cognitive. Within—and across (as the distinctions between “physical” and “cognitive” are far from clear-cut)—these two aspects Picard explores particular features of human emotion (or “emotion theory”) she feels are important to consider in order to create genuinely affective computing, to give  computers “the ability to recognize, express, and ‘have’ emotions” (p. 44). These features are “social display rules, universal vs. person-specific responses, primary vs. secondary emotions, the role of emotions in creativity and general memory processes, the existence of multiple paths for emotion expression in humans, and emotion inducement” (p. 44).

Picard also spends some time highlighting ways in which “emotion can be expressed through sentic modulation—including facial expression, vocal intonation, gesture, posture, and other bodily changes” (p. 44). “Sentic modulation,” in turn, are expressions “such as voice inflection, facial expression, and posture,” are “the physical means by which an emotional state is typically expressed, and is the primary means of communicating human emotion” (p. 25).

One of the things I found particularly interesting about this chapter is Picard’s consideration of computers’ “recognition problem”—wherein they have (or at least had, at the time of the book’s publishing) difficulty recognizing, or focusing-in on and interpreting, human speech acts and meaning under certain circumstances. She argues for a

“universal” recognizer… [that] would first ask “Which category is this person most similar to?” In speech, this might be likened to asking “who sounds like this—both accent-wise and voice-quality wise?” Subsequently, a recognizer can be used that was trained on the prototype person for that category. A benefit of this approach is that it is also likely to reveal categories of affective expression that theorists have not yet identified. (p. 34)

I’d never thought of such a problem before, and it leaves me wondering what kind of recognition systems currently exist, and how much they do or don’t resemble such a kind as Picard describes.

Advertisements

Picard, Affective Computing, Introduction

In the Introduction to Affective Computing (1997), Picard overviews and sets up her argument for the book, stating that it “proposes that we give computers the ability to recognize, express, and in some cases, ‘have’ emotions” (p. 1). She proposes this because, she contends (after raising and critiquing various points that might be made against such a proposal), “[a]fter nearly a half century of research … computer scientists have not succeeded in constructing a machine that can reason intelligently about difficult problems or that can interact intelligently with people” (p. 1). Picard further sketches the outlines of the book as “lay[ing] a foundation and construct[ing] a framework for what I call ‘affective computing,’ computing that relates to, arises from, or deliberately influences emotions” (p. 3).

Much of the rest of the Introduction is devoted to Picard making the case for (simply put) emotional intelligence—both as a vital element in human cognitive functioning (including decision-making, a subject she spends some time on here) and as a form of intelligence we should strive to create for and in computers. “[C]omputers,” she argues, “if they are to be truly effective at decision making, will have to have emotions or emotion-like mechanisms working in concert with their rule-based systems” (p. 12).

One of the things that struck me about Picard’s argument in the Introduction is this phrase, “emotion-like mechanisms.” At this—early—point in her book, I can’t decide whether her offering it is a compromised position (from computers having “real” emotions or emotional capabilities) or not, whether it seems to come somewhat too soon or somewhat too late in her argument, even though it’s only a dozen pages into the book.