Information Technology Reference
In-Depth Information
history and also review all possible moves it might make, con
sider possible responses by its opponent, and so forth, look
ing a dozen or more moves into the future to determine which
current move was most promising.
In 2000, SpeechWorks was able to extract meanings from
speech in certain contexts, in order to allow computer re
sponses to various types of requests. For example, Amtrak
uses an automated telephone system by SpeechWorks to allow
callers to get train schedules, book reservations, and charge
tickets to their credit cards—all based on deciphering a caller's
spoken words. After identifying likely words through an
analysis of phonemes, it combines the words into phrases and
sentences based on both grammatical rules and a vast data
base of common expressions and sentence fragments.
In each of these cases, computers have performed a task that we
might associate with intelligence: medical diagnosis, transcribing
speech, winning at chess, and responding to spoken requests regard
ing trains and reservations. Although each of these applications is a
wonderful technical accomplishment, the application has a limited
focus, and the computer system does not attempt to respond to
more general questions or problems. In each case, the computer
might make a credible respondent in some type of Turing Test for
the specific area of expertise involved; however, it would have no
capacity to work more generally, and the investigator in the Turing
Test would have no trouble distinguishing quickly between the com
puter and the human respondent. The computer would fail the sim
plest blackbox testing of the Turing Test or other test variations.
These applications also would fail whitebox testing, as they fol
low rather different processes than we might expect from people
working in similar situations. For example, although Deep Blue re
views roughly 1 billion positions a second, a chess master, such as
Garry Kasparov, normally reviews just three positions per second—
focusing on which alternatives seem to have the greatest potential.
Rather than utilize insight or understanding, Deep Blue proceeds by
remarkable computational power. Deep Blue follows a vast array of
rules, but it does not comprehend what it is doing.
Interestingly, many popular computer applications have been
able to become successful
because
they do not try to seem intelli
