Here's an example of how the accuracy of speech recognition has improved
so much that it is being seen more in commercial applications. At the
same time, the customer service career path for blind persons is in
jepordy with widespread deployment of this technology.
kelly
from the New York times
June 21, 1998
The Voice on the Phone Is Not Human, but It's Helpful
By JOHN MARKOFF
L ater this year, many callers wanting flight information from
United Airlines will speak not to a person but to a computer that
acts like one. It will ask when and where the caller wants to
travel, look up flight schedules and converse using synthesized
speech.
A caller can say, "I want to fly from New York to San Jose next
Wednesday morning." The system recognizes the names of 1,200
airports around the world and understands concepts like "next
Wednesday." Employing advanced forms of artificial intelligence, it
interacts with a caller just as an agent might, even checking to
confirm that it has heard correctly -- "Did you say Austin or
Boston?"
Beep, 'How May I Help You?' Beep
United Airlines has created a fully automated telephone reservation
system for its employees using speech recognition technology. Here is
part of a recent dialogue involving an employee using the system to
make plane reservations.
RESERVATION COMPUTER Welcome to the new speech-activated United
Airlines employee travel service. If it's your first time calling or
if you havent yet received a personalized letter describing this
service, say: Instruction. Please enter or say your your employee file
number and check digit.
EMPLOYEE
101967-5.
COMPUTER
Next, enter or say your six-digit PIN number.
EMPLOYEE ENTERS NUMBERS ON A KEYPAD.
COMPUTER
Please hold while I check your file number and PIN. (PAUSE) Please say
your departure city.
EMPLOYEE
Chicago.
COMPUTER
Please say your arrival city.
EMPLOYEE
New York.
COMPUTER
United Serves more than one airport in New York. They are J.F.K., New
York; La Guardia, New York; White Plains, New York and Newark, New
Jersey. Please say the airport code or name you want.
EMPLOYEE
La Guardia.
COMPUTER
What day would you like to depart? For example, March 5th.
EMPLOYEE
Next Tuesday.
COMPUTER
Sorry, I didnt understand. Please say just the month and day you will
be departing.
EMPLOYEE
Next Tuesday.
COMPUTER
Got it. Do you have a particular flight number in mind?
EMPLOYEE
No.
COMPUTER
Approximately what time of day would you like to leave?
EMPLOYEE
Late afternoon.
COMPUTER
Before I search for flights, let me confirm your itinerary. You want
to fly from Chicago OHare, Illinois, to La Guardia, New York, at 3
P.M. on Tuesday, June 23d. Is that correct?
EMPLOYEE
Yes.
COMPUTER
Please hold while I look for flights that match your itinerary.
(15-SECOND PAUSE) I found eight nonstop flights that match your
itinerary and eight with stops or connections. I'll give you the
nonstop flights first. If you need instruction, say: Help.
_________________________________________________________________
United's system is only one example of a wave of new computer
technologies that understand spoken language and are poised to
sweep through the American economy. In areas as diverse as airline
reservations, retailing, directory assistance, banking, medical
transcription, computer help desks and secretarial services,
machines that can recognize thousands of words and phrases -- and
in accents as diverse as Brooklyn and Pakistani -- are rapidly
becoming commonplace.
"Speech recognition has passed a threshold," said Raymond Kurzweil,
a leading researcher in the field of artificial intelligence. In
the next few years, he predicted, "The bulk of business
transactions will take place between a person and an automated
personality."
There could be problems at first. Some people will have to speak
more slowly or clearly than they normally do. And the computers
still get tripped up by jargon, accents and questions they are not
programmed to answer. For example, an airline reservations system
might find itself stumped by a question about ground
transportation.
But the systems promise customers new ease in performing every type
of transaction. Executives at companies developing the technology
say that speech recognition systems are being eagerly embraced by
consumers who have grown weary of waiting for customer service
representatives and of using a keypad to navigate the seemingly
endless mazes of automated menus.
And the relatively few consumers who have found themselves talking
to a computer are generally positive about the experience.
For users of United Airline's speech recognition system, there is
often a reaction of wonder when they realize that a computer
understands their words.
"Sometimes it feels like it's smarter than I am," said Tony
Molinaro, a United manager. He began using the company's
reservation system, which is now available only to employees,
several months ago and routinely uses it to book flights.
At first he said his sessions would take slightly longer than with
a human being because the computer would ask him additional
questions about which airport he wanted to travel to in the Los
Angeles area. But he has since become an expert user.
The system has enough sophistication to know about the members of
his family who are eligible to fly. For example, when he refers to
his father, the computer asks, "Do you mean Ben?"
"It's very neat that it understands me," Molinaro said.
Nor does the user's age appear to have any bearing on the ease of
adapting to speech recognition. Arthur Edwards of Fort Myers, Fla.,
82, is a satisfied user of the voice service offered by Charles
Schwab & Co., the discount stockbroker.
"I like to get stock quotes and don't like to have to wait,"
Edwards said. The system almost always recognizes without error the
stocks he mentions, he said, although he did recall that it once
had trouble with some stocks traded on a Hong Kong exchange.
Edwards said he was quite comfortable talking to computers. "I've
seen lots of changes in my life," he said, "and these days I spend
my time trying to keep up with the grandchildren."
At the same time, the new systems will have a big effect on the
American work force. The number of jobs created or destroyed by
such systems is a matter of much debate. But many labor experts
agree that the new technologies will contribute to the growing
polarization of the job market into high- and low-skilled jobs and
a corresponding disparity in wealth.
On one hand, they will create high-paying jobs for computer
programmers and for the many marketing people who sell their work.
But in the process, they will destroy semiskilled jobs in customer
service, and many people now earning moderate incomes will be
forced to move into low-skill jobs at the bottom of the employment
ladder.
"This is the scariest economic problem of our time," said Timothy
Bresnahan, an economist at Stanford University who has studied the
effect of new technologies on income distribution. "The most
important technological changes tend to be related to big changes
in the distribution in wages."
Indeed, jobs that have been among the fastest growing, like travel
reservations, telephone sales and customer service, are in danger
of being replaced by inexpensive and increasingly flexible speech
recognition systems.
Last month, Sears, Roebuck & Co. became the nation's first retailer
to install a computer system that answers all phone calls at the
company's 833 stores, responding to queries and automatically
routing calls to the right department.
Sears executives say that while 3,000 jobs were affected by the new
system, no workers were laid off. Because of the strength of the
economy, the company was able to reassign its telephone operators
to new jobs as sales and stocking clerks, but those jobs might not
be possible in times of slower economic growth.
The same bullish economy is also fostering widespread investment in
speech recognition technologies.
Last fall, Charles Schwab began allowing customers to buy and sell
mutual funds over the phone using software developed by a Silicon
Valley start-up, Nuance Communications. The system understands the
names of more than 1,300 funds and can respond to requests for
price quotes for more than 13,000 stocks.
In March, American Express Co., the nation's largest travel agency,
began testing a service that will allow its corporate customers to
call computers for flight information or to make airline, hotel and
car reservations.
Silenced Operators
As speech recognition technology has become more prevalent, the number
of directory assistance and corporate telephone operators has
declined.
[INLINE]
1996 figures are latest available.
Source: Bureau of Labor Statistics
_________________________________________________________________
Automation has already had a profound effect on telephone
operators. According to the Bureau of Labor Statistics, the number
of jobs for telephone, directory assistance and in-house operators
fell to 164,000 from 400,000 between 1970 and 1996, a decline of
236,000 jobs.
More than a fifth of those -- 51,000 jobs -- have been lost in the
six years since the first rudimentary speech recognition was
introduced, according to labor economists and AT&T executives. That
system, which can recognize just five terms -- "collect,"
"operator," "third party," "credit card" and "person to person" --
saves "several hundred million" dollars a year, most of it in labor
costs, AT&T says.
Since 1907, when AT&T and Western Electric Co. combined their
engineering operations and created the Bell Telephone Laboratory,
the growth of the telephone industry has been marked by cycles of
huge expansions in employment followed by periods of automation.
The era of operators who manually connected calls at large plug
boards began to end in 1919, when mechanical switching was
introduced, ushering in the era of dial telephones. The first
automated call processing went into service in Brooklyn in 1938,
but it was not until an entirely electronic switching system was
introduced in 1963 that operators began to disappear entirely from
the process of completing calls.
More recently, the creation of "megacenters" that handle calls for
large regions has significantly reduced phone company employment.
Among the workers affected by the concentration of jobs in such
centers was Addie Brinkley, who after 40 years of working as an
operator in eight different cities quit her job in August 1996
after AT&T closed its call center in Modesto, Calif., and asked her
to move to Reno.
Ms. Brinkley's job had fallen victim to a new computerized system
that performed many of the functions traditionally handled by
operators. AT&T executives say they accomplished the cutbacks with
almost no layoffs, but union representatives and operators like Ms.
Brinkley tell a more unsettling story.
Her decision to quit rather than move to a directory assistance
megacenter in Reno, she said, mirrored the experience of thousands
of other phone company workers who have been eased out in the last
six years.
"I saw thousands of people lose their jobs," she said, "and I saw
the devastated lives."
While few economists assert that speech recognition will be a job
killer overall, many predict that like other information-based
technologies introduced in the last decade, it will bring profound
changes, positive and negative, for workers and consumers.
While today's speech recognition systems are not perfect -- United
Airlines officials said their system had a call completion rate of
98 percent -- they are adequate when used with scripts that reduce
the range of expected human answers. And they will improve rapidly
if the development of current speech recognition technology is any
indicator.
"It's a confluence of small things, from better software design to
dramatically cheaper and faster computers," said Ronald Croen,
chief executive of Nuance. "There was no single breakthrough."
Speech recognition began in the 1960s when computer scientists at
the Massachusetts Institute of Technology, Carnegie Mellon
University, Stanford University and other centers began to research
the idea with financing from the Pentagon's Advanced Research
Projects Agency, which created the original Internet.
Despite the optimism of researchers and prototypes built in the
1980s, commercial applications remained elusive. It was not until
the early 1990s that companies ranging from giants like IBM to
start-ups like Dragon Systems began offering the first commercial
systems. These systems, designed for personal computers, initially
recognized only individual words and forced users to speak slowly
and unnaturally.
The commercial turning point came in 1992, when AT&T introduced the
five-term speech recognition technology into its nationwide
long-distance network.
One indicator of the effect that speech recognition is likely to
have in the workplace emerged soon after the United Parcel Service
lost its battle with the teamsters' union last August. Scrambling
for new ways to save labor costs, UPS quietly deployed a speech
recognition system developed by Nuance that gives package tracking
information in response to a caller's spoken commands. Last
Christmas Eve, the system handled 193,000 calls, double the daily
average for UPS. Company officials declined to say what the system
cost but said it had paid for itself in less than three months.
Rand Wilson, a teamsters spokesman, said that the system had
resulted in "a tremendous loss of workers," adding, "This has been
a contentious issue."
For Croen of Nuance, the changes are justified because the jobs
that are being replaced are so tedious. He cites annual employee
turnover rates of more than 60 percent for call centers at which
operators answer repetitive questions.
However unrewarding the work, some labor experts worry about the
number of jobs at risk.
"One of the main sources for new jobs for the middle class over the
past 15 years have been telephone sales and information-related,"
said Robert Reich, the former secretary of labor, who is a
professor of economic and social policy at Brandeis University.
"Now all those jobs are on the line."
Copyright 1998 The New York Times Company
|
