[Sigia-l] "Gait code"

[Listera]

Continuing Eric's education on contextual analysis by machines: ;-)


One afternoon I drove across the Microsoft campus to visit a man who is
trying to achieve precisely that: a computer that can read your mind. His
name is Eric Horvitz, and he is one of Czerwinski's closest colleagues in
the lab. For the last eight years, he has been building networks equipped
with artificial intelligence (A.I.) that carefully observes a computer
user's behavior and then tries to predict that sweet spot - the moment when
the user will be mentally free and ready to be interrupted.

Horvitz booted the system up to show me how it works. He pointed to a series
of bubbles on his screen, each representing one way the machine observes
Horvitz's behavior. For example, it measures how long he's been typing or
reading e-mail messages; it notices how long he spends in one program before
shifting to another. Even more creepily, Horvitz told me, the A.I. program
will - a little like HAL from "2001: A Space Odyssey" - eavesdrop on him
with a microphone and spy on him using a Webcam, to try and determine how
busy he is, and whether he has company in his office. Sure enough, at one
point I peeked into the corner of Horvitz's computer screen and there was a
little red indicator glowing.

"It's listening to us," Horvitz said with a grin. "The microphone's on."

It is no simple matter for a computer to recognize a user's "busy state," as
it turns out, because everyone is busy in his own way. One programmer who
works for Horvitz is busiest when he's silent and typing for extended
periods, since that means he's furiously coding. But for a manager or
executive, sitting quietly might actually be an indication of time being
wasted; managers are more likely to be busy when they are talking or if
PowerPoint is running.

In the early days of training Horvitz's A.I., you must clarify when you're
most and least interruptible, so the machine can begin to pick up your
personal patterns. But after a few days, the fun begins - because the
machine takes over and, using what you've taught it, tries to predict your
future behavior. Horvitz clicked an onscreen icon for "Paul," an employee
working on a laptop in a meeting room down the hall. A little chart popped
up. Paul, the A.I. program reported, was currently in between tasks - but it
predicted that he would begin checking his e-mail within five minutes. Thus,
Horvitz explained, right now would be a great time to e-mail him; you'd be
likely to get a quick reply. If you wanted to pay him a visit, the program
also predicted that - based on his previous patterns - Paul would be back in
his office in 30 minutes.

With these sorts of artificial smarts, computer designers could re-engineer
our e-mail programs, our messaging and even our phones so that each tool
would work like a personal butler - tiptoeing around us when things are
hectic and barging in only when our crises have passed. Horvitz's early
prototypes offer an impressive glimpse of what's possible. An e-mail program
he produced seven years ago, code-named Priorities, analyzes the content of
your incoming e-mail messages and ranks them based on the urgency of the
message and your relationship with the sender, then weighs that against how
busy you are. Superurgent mail is delivered right away; everything else
waits in a queue until you're no longer busy. When Czerwinski first tried
the program, it gave her as much as three hours of solid work time before
nagging her with a message. The software also determined, to the surprise of
at least one Microsoft employee, that e-mail missives from Bill Gates were
not necessarily urgent, since Gates tends to write long, discursive notes
for employees to meditate on.

New York Times
<http://www.nytimes.com/2005/10/16/magazine/16guru.html?pagewanted=5>

---- 
Ziya

Best Practices,
For when you've run out of your own ideas and context.