| | | | | |       Listen to Me       | | | | | |

The Kurzweil Applied Intelligence Alumni Newsletter


Go to: Welcome Table of Contents What's New Registration Database

[From the Cover]
Stupid Computer Tricks: How Virtual Reality, Speech Recognition and Other Good Ideas Can Hurt Business

Trends in Computing

Taking Computers to Task

Scientific American, July 1997, Volume 277 Number 1, page 82
by W. Wayt Gibbs

[Excerpts]

Will new 3-D interfaces, speech recognition and other highly touted computer technologies do anything to make workers more productive? A no-nonsense look at the value of new computer features, from the overhyped to the overlooked.

Coming generations of computers will be more fun and engaging to use. But will they earn their keep in the workplace?

WORKING HARD OR HARDLY WORKING? Computers can now tune into Internet "channels," make video calls, recognize speech and navigate 3-D models. But few of these innovations demonstrably boost productivity.

Researchs presented at least 83 novel interfaces at the CHI '97 (Computer-Human Interaction) conference. Many do truly nifty things. One program takes dictations with 97 percent accuracy from radiologists. ... But only nine of those 83 projects compared workers' performance on real tasks using the new interface with their current way of doing things. Four offered no gains at all. Radiologists completed their reports faster without the computer.

Beyond the Desktop

Speech Recognition and Understanding

Researchers at M.I.T. have built a state-of-the-art system called GALAXY that can understand and answer spoken questions about weather, airline flights and the city of Boston. Even on such relatively simple, focused tasks, however, the system often makes mistakes. Like most other speech-driven programs, it is not yet clearly faster than old-fashioned methods, such as consulting travel agents or newspapers.

"People rarely speak in grammatically correct sentences," observes Tom Landauer of the University of Colorado. "So when you put the best speaker-independent systems in real- world situations, at most they can recognize 10,000 words-- about a sixth of what we use in day-to-day life-- and they get at least one quarter of the words wrong."

Despite more than 25 years of research on the subject, says cognitive scientist Donald Norman, "language understanding by machines remains decades and decades away. And on top of that," he adds, "anyone who has ever struggled to communicate with an assistant knows that this isn't the answer."

Copyright ©Scientific American


| | | | | | Kurzweil AI Alumni News | | | | | |
Go to: Welcome Table of Contents What's New Registration Database

Questions or Problems? Send e-mail
June 15, 1997