Page 42 - Robots Androids and Animatrons : 12 Incredible Projects You Can Build

Neural networks can be trained to perform visual recognition.
They can learn to read or perform quality control by visual analy-
sis of parts. One such example is Papnet, discussed in Chap. 1.
Other networks can be taught to respond to verbal commands
(speech recognition) and generate speech. Statistical nets can
predict the future behavior or probability of complex nonlinear
systems based on historical examples. These networks have been
used to predict oil prices, monitor aircraft electronics, and forecast
the weather. Networks have also successfully been employed to eval-
uate the stock market, mortgage loan applicants, and life insurance
contracts better than standard rule-based expert-system programs.

What is artificial intelligence?

This is a legitimate question. We most certainly will develop neural
networks that are intelligent before we develop nets that become
conscious. So in attempting to create neural networks that are intel-
ligent or demonstrate intelligence, what criteria should one use to
determine if this goal has been achieved?
Alan Turing, a British mathematician, devised an interesting proce-
dural test that is generally accepted as a valid way to determine if a
machine has intelligence. The test is conducted as follows: A per- 21
son and the machine hold a conversation by typing messages to one
another via a teletype. If the machine can carry on a conversation
without the person being able to determine whether a machine or
person exists at the other teletype, the machine can be classified as
intelligent. This is called the Turing test and is one criteria used to
determine AI.
Although the Turing test is well accepted, it isn’t a definitive test
for AI. There are a number of “completely dumb” language pro-
cessing programs that come close to passing the Turing test. The
most famous program is named ELIZA, developed by Joseph
Weizenbaum at the Massachusetts Institute of Technology (MIT).
ELIZA simulates a psychologist, and you are able to conduct a
conversation with ELIZA. For instance, if you typed to ELIZA that
you missed your father, ELIZA might respond with “Why do you miss
your father?” or “Tell me more about your father.” These responses
may lead you to believe that ELIZA understands what you have said.
It doesn’t. The responses are clever programming tricks constructed
from your statements.

Therefore, if we like, we could do away with the Turing test and
consider a different criterion. Perhaps consciousness or self-
awareness would be a better signpost of intelligence. A self-aware
Team LRN Artificial life and artificial intelligence