on the screen as you deliver the presentation, also noting the feedback about your
pacing, intonation, speech clarity, and physical responses. Your computer laughs heartily at your opening joke and responds enthusiastically to your best points. You thank your computer for the help.
As we embrace the idea of personal assistants, intelligent agents, and computers that understand our needs, work styles, and preferences, we must consider instilling computers with emotional intelligence. As Rosalind Picard points out in her book
Affective Computing
, although we're conditioned to view emotions as a hindrance to clear reasoning and intelligent judgments, research supports the importance of emotions in critical thinking. Computers are little more than savants, and yet we expect them to behave like human assistants. A truly personal, interactive system must learn to associate relevance and context to the facts it holds and to make judgments based on experience and intuition.
The first part of the book covers current research on human emotion, delving into the m
ystery of what makes up emotions and how to recognize, test, and differentiate among a range of emotional states. Picard makes a convincing argument about the role of emotions in intelligence and suggests some compelling applications, such as computer interfaces, tutorials, and personal data assistants. Correctly developed and applied, emotions could enrich the computing experience and advance it to new levels of utility.
Of course, human emotion, especially in intricate areas such as creativity, is not well understood. Building emotions into computers can seem like a daunting challenge. I certainly sensed the forbidding complexity of the problem as I read the book. The most encouraging areas of research, at least in the near term, involve emotional recognition and imitation. In the aforementioned presentation scenario, a truly emotive computer would not be required. The system would need only to infer a likely emotional state by monitoring your physical responses and then react in an emotionally suppor
tive way. Sensors in a computer could track discrete signals, such as electromyograph (EMG) waves, blood-volume pressure, galvanic skin response, and respiration, to infer emotional states. The next step is to map emotional patterns to appropriate reactions by the computer.
Researchers are applying a variety of theories about human emotions to computer architectures. Picard depicts many models in the book, "deliberately not trying to establish one model or one theory of emotion." Again, it's the nebulous nature of emotion itself that constricts her; after all, there is not "sufficient understanding of human emotions to justify a comprehensive model at the level needed for computer implementation."
This book works best as a solid framework for ongoing research and development. Picard offers criteria for quantifying emotional intelligence, for verifying its effectiveness in computers, for recognizing the properties of emotional behavior and implementing appropriate computer responses, and for chara
cterizing affective patterns and recognizing them algorithmically. She also lays the important groundwork for building the models and software architectures of the future.
I might only suggest that Picard should have used fewer references to the homicidal HAL from
2001: A Space Odyssey
. Every time I launch my Web agent, I keep hearing HAL's ominous echo: "I'm afraid, I'm afraid, I'm afraid...."
Stanford Diehl is a frequent contributor to BYTE and former director of BYTE reviews. You can reach him at
Sdiehl@nebs.com
.
Book Review: A Computer to Love
