Customer Reviews

Affective Computing

5 star:		(1)
4 star:		(2)
3 star:		(1)
2 star:		(1)
1 star:		(1)

 

 

Most of this book is a primer for non-clinicians on what is meant by 'human emotions', and how a computer in physical contact with someone could identify that person's mood and respond appropriately to it. Picard makes her case that 'emotional intelligence' would be a useful attribute for software. A human who loses the ability to feel emotions becomes, not admirably logical like Mr. Spock, but unable to make quick, simple, arbitrary decisions and prone to repeat mistakes. Just like most software today. Picard relates the use of affective computing primarily to the 'wearable computers' that researchers at MIT have been playing with for over 10 years to do mostly trivial functions like take photographs and generate muzak. There wasn't much here for those of us who have to interact through keyboards/mice and monitors, and surprisingly no attempt to connect affective computing with related techniques such as fuzzy logic. There is an excellent source reference list at the back.

This is an interesting book, and I strongly agree with Picard's assertion that computers ought to be able to "recognize" and respond to human emotions. She does an excellent job of making and supporting this point. The other part of her thesis, that computers themselves should have "emotions" is much less clear. She never seemed to adequately make the case that a computer with its own emotions would be of any significant value for anything, and frankly I can't think of any useful applications for such an ability. Some sort of emotional component may be needed to fully support and achieve AI (and she makes this point) but in terms of sort of the standard user interface types of applications it's hard to imagine how such a capability could be useful.

Anyway, good book on a very interesting topic.

A fascinating book with many implications for the fields of artifical intelligence and human-computer interaction. Picard provides a rich background on modern research in emotion and puts forth compelling arguments for the need to incorporate affective abilities in computers as, perhaps, the only way to allow them to respond intelligently to their environment and make rational decisions. An entertaining and mind-opening read.

In pop-culture there is the usual dichotomy between someone who "thinks with their head" and one who "thinks with their heart". In fact, the more enlightened thinkers realize that this dichotomy is at least half false: people who are emotionally-impaired are not more rational than the rest of us, but are rather quite crippled and incapable of facing everyday life. (Though if you are not already convinced of this, this book will do little to persuade you.)

At the time when I first read this book, nearly a year ago now upon the recommendation of a friend, I was already convinced of the usefulness of emotions in AI, and was hoping to find some real concrete and useful results here regarding AI-emotion, which I could then apply to the design and construction of an AI which would presumably be of use to someone in the real world. Much to my dismay, there are NO such applications listed; Picard suggests that AIs should be given emotions, but doesn't bother to give any real applications in which these emotions would be useful, or what kind of emotions they should be given, or even what an "emotion" is for an AI!

The book discusses almost exclusively the problem of AIs, not having emotions, but understanding the emotions of humans. Sounds great, how about some applications? Picard then proceeds to suggest the most absurd applications imaginable; here are a few of them:

-Emotive Markup Language: Modify the hardware of a keyboard such that the computer can tell how much pressure was applied on each keystroke. Then have the machine interpret these pressure levels as "happy typing", "angry typing", etc., and then mark each portion of text appropriately, with say, big red bold letters for "angrily typed" word, and so on.

-The understanding user interface: The user interface receives occasional feedback from the user, (blood pressure levels, questionnaire, whatever) from which it is to judge the user's mood, such as anger or frustration, and then try to help the user out somehow if the user is becoming frustrated. Little does Picard realize that most users find a clairvoyant-wannabe computer more annoying than helpful.

-Intelligent Answering Machines: Our answering machine receives a phone call, and presumably by talking to the individual on the other line, gathers some information as to the phone-call's content. Meanwhile the answering machine is monitoring the emotional-state of its master, and if it infers that its master is in a mood that can be interrupted, and that the phone-call is of interest to its master, then the answering machine will tell its master that there is a call waiting, otherwise it will just take a message.

If those are the most important problems facing an AI-researcher today, then the problem of AI must already be quite solved! In fact, in the past year I have been further enlightened, and have realized that AIs in fact don't need emotions: just because humans need them is no argument at all that AIs need them! It is foolhardy to simply give AIs emotions without understanding WHY emotions evolved: we would just be copying superficial similarity; feathers aren't the key to flight! It turns out that emotions are evolution's own peculiar way of implementing probabilistic reasoning and goal-systems: every emotion can be translated into purely decision-theoretic terminology. For example, "curiosity" is a heuristic which can be replaced with a system which sets up experiments so as to maximize its expected information gain on each experiment.

Of course, Picard could not simply say as much: she hints several times throughout her work that she believes in god, and that she intends her AIs to appreciate god as well. For example, we have the following quote:

"A system that truly operates in a complex and unpredictable environment will need more than laws; it will essentially need values and principles, a moral compass for guidance, and perhaps even religion." (page 134)

Funny, I seem to be doing quite fine without religion!

Overall and like most works of pure-philosophy, this book is intellectually quite sparse: Picard says more or less everything she has to say in the first 50 pages, but then somehow manages to drag her book out for another 200 pages by mentioning various things only tangentially related to the topic under discussion and rephrasing what she has already said. This short review alone contains a good deal more content than do a dozen pages from this book.

This book is meant for a pop audience. Its completely out of date and does not take into account the vast research at major computer-human interaction conferences such as ACM SIG-CHI

Rosalind Picard's book shouldn't have broken new ground, but it did. The ignoring of the role of emotion in computing is both appalling and typical. Picard begins to rectify this "oversight" ("Whoops, I forgot humans have feelings!") in a fascinating and useful book.