Considerate Computing
From: Scientific American - 01/2005 - Vol. 292, No. 1, P. 54
By: W. Wayt Gibbs

Digital gadgetry's propensity to interrupt users with alerts is not only a
source of social embarrassment, but of declined productivity: Studies support
the idea that interruptions in normal routines slow people down and make them
more likely to commit errors. "If we could just give our computers and phones
some understanding of the limits of human attention and memory, it would make
them seem a lot more thoughtful and courteous," notes Microsoft's Eric
Horvitz; he is part of a small but expanding group of researchers trying to
develop "attentive" systems capable of inferring their owners' whereabouts
and activities, weighing the value of the messages they wish to send against
the consequences of interruption, and selecting the best time and manner of
interjection. A study of human "interruptibility" conducted by Carnegie
Mellon University and IBM Research found that truly useful attentive systems
must be over 65 percent accurate in detecting when users are close to their
cognitive thresholds. The researchers determined that adding microphones to
pick up conversations within earshot increased that accuracy to 76 percent,
while the detection of mouse movement, keyboard activity, and computer
application status raised accuracy to 87 percent. Carnegie Mellon's Scott
Hudson recommends that the attentive system analyze but not record input data
streams in order to address privacy concerns. Roel Vertegaal with Queen's
University in Ontario has made everyday appliances capable of responding to
users' vocal commands and shutting themselves off when the user's gaze is no
longer fixed on them through a combination of speech recognition and infrared
scanning. Another approach to attentive system design is Bayesian networks,
which are employed in spam filters and network firewalls to statistically
learn a user's preferences in terms of wanted and unwanted messages. However,
the University of Maryland's Ben Schneiderman reports that more attentive
systems are less predictable, and notes that the scientific community has a
history of building "smart" technologies that go unused because their
operational principles are not easily understood.  

Links:
Eric Horvitz homepage
http://research.microsoft.com/~horvitz/

Polite computers win users' hearts and minds
http://www.newscientist.com/article.ns?id=dn6156
