Information Technology Reference
In-Depth Information
Although many such tests might be contemplated, two general
approaches involve
black-box testing
and
white-box testing
. In
blackbox testing, we give the test subject various problems or in
puts, and we examine the results. We do not worry about how the
subject determined the answers—we just review the conclusions. A
simple example of blackbox testing might be a multiplechoice test;
questions are given, and the person being tested responds by identi
fying which answer is correct. In grading the test, we do not con
sider how the subject came up with the answer; we just mark each
answer as right or wrong.
In whitebox testing, we examine the results, but we also review
the process used for reaching the conclusion. For example, in a
longanswer test involving mathematics, a grader may check how a
solution was first set up, what approaches were used to solve vari
ous equations, what intermediate results were obtained, and how
the various pieces led to the conclusion. In whitebox testing, the fi
nal answer is only one factor among many in determining whether
the subject passes the test.
Perhaps the best known proposal for the blackbox testing of in
telligence is called the
Turing Test
, proposed by Alan Turing in
1950 in the early days of electrical computers. Turing proposed that
a (human) investigator be placed in one room, a computer in a sec
ond room, and a human respondent in a third room. The investiga
tor would have a teletype (similar to an old typewriter) attached to
the computer and another for communication with the human re
spondent. (See Figure 16.1.) Although the investigator would know
that one teletype communicated with a computer and the other cor
responded with a human, the investigator would not know which
was which. The investigator was charged with asking questions to
each party and analyzing the responses in whatever way seemed ap
propriate. The goal was for the investigator to identify which tele
type went to the computer and which to the human. Although the
human was to answer questions honestly, the computer was to re
spond as it wished. If the investigator was able to pick the human
only half the time (the same fraction as would be obtained by ran
dom guess), then the computer was deemed to have passed this in
telligence test.
The rationale for the Turing Test depended on the premise that
the human respondent would be intelligent in her or his behavior. If
the computer seemed to consistently behave in a similar way, then
