The Turing Test
This test was proposed by British
mathematician Alan M. Turing in his 1950 article (now available
Machinery and Intelligence (Mind, Vol. 59, No. 236, pp. 433-460).
The test was intended to be used to determine if a computer program
could display a human level of intelligence - i.e. to show that a
computer could think (a radical concept in those days).
Originally it was to work as
follows: The test would consist of two people and a computer
program - all online and communicating with each other by teletype but
none could actually see each other. One of the people was the
judge and his job was to ask questions of the other two and to try to
determine which was the person and which was the computer
program. If the computer program could fool the judge into
thinking it was the human 50% of the time, then that program was said
to display a human level of intelligence.
Conceptually it sounds very easy, but for
computer programmers is has been extremely difficult to create a
program which could answer arbitrary questions like a human
does. Indeed to date there are few if any programs which can
fool a human inquisitor. There is in fact a whole field of computer
programs called chatterbots which attempt to do just that.
Many people have probably already realized
that a superintelligence already exists - it is called the internet.
Clearly, all we need to do is to interface the androids to the
internet and they will rapidly become superintelligent too. We at
Android World plan to introduce such an interface for our Valerie
android within a year. Perhaps some readers are familiar with the Loebner
prize which is a real prize for a computer program which can pass
the Turing test. That site gives links to their past
winners which you can go to and review many of the conversations
between the programs and the inquisitors. I think you will be
surprised at how poorly the programs seem to perform. That will
Although knowledge, which is merely
the recitation of known facts, is NOT intelligence, it can
certainly be very impressive and can appear to be intelligence.
Imagine that we had a computer program which could access the internet
and answer virtually any question to which the answer is known and
posted on the internet. This would vastly exceed the performance
of any human or in fact any group of humans which you could assemble.
Therefore, I can imagine an inquisitor
simply asking large numbers of difficult (but known) questions and
noting the responses. The human would be revealed by his
inability to answer more than a small fraction of the questions
correctly. The program on the other hand would be able to answer
all the questions correctly. Thus the inquisitor would be
able to identify the computer as the one who got all the questions
correct and the human as the one who didn't know the
answers. Thus in order to fool the inquisitor, the
computer program would need to purposely answer questions incorrectly
or to state that it didn't know the answer. Now
we have arrived at exactly the opposite of the intention of Alan
Turing - the program must now pretend to be stupid in order to
convince the inquisitor that it is the human and to pass the test
showing that it is intelligent.
Comments? Email me at firstname.lastname@example.org