Information Technology Reference
In-Depth Information
not a legitimate step toward passing the Turing test. I know
of no case in the history of science where such “naysayers”
actually helped with a new discovery.**
With the constant evolution of technology, we can be reason
ably confident that computers in the future will have significantly
greater capabilities than we expect or imagine today. The nature of
such capabilities, however, requires speculation. Over the past sev
eral decades, the most successful computer applications typically
have focused on the solution of specific, narrow problems. When
problems are contained, techniques can take advantage of details in
the problem, such as limited vocabulary, constrained choices, iden
tifiable algorithms, a wealth or history of examples (perhaps stored
in a database), and detailed mathematical models. With this track
record, it seems reasonably safe to predict that future computers
will be able to solve a wide range of complex problems, well beyond
current solutions and applications.
At the same time, past history suggests that applications tend to
be less successful when they try to tackle a range of general prob
lems. From this perspective, we might expect computers in a decade
or two to show blackbox intelligence in significantly more areas,
but we would not expect computers to demonstrate the general
knowledge needed to handle questions on various subjects as might
be given by the human investigator in a general Turing Test. Some
current research aims to find methods for general problem solving
or conversation, but progress in this area has been slow.
Further, although considerable research is exploring how people
function, how the brain works, what electrochemical mechanisms
produce what actions, and the like, commerciallevel applications
seem to be most successful when they do not try to mimic human
functioning. Although we can expect many new and spectacular suc
cesses from computers, we might be skeptical about whether those
successes will come from better understandings of human thought
and reasoning. You may or may not believe whitebox testing for in
telligence is appropriate, but it seems unlikely that computers will
satisfy any whitebox intelligence tests any time soon.
All of this, however, leaves open the question of whether com
puters might, someday, be able to think. We may not expect comput
ers to think in the next 10 or 20 years, but is computer intelligence
**Bledsoe, W.W. “ I Had a Dream: AAAI Presidential Address” 19 August
1985
AI Magazine
7(1):58.
