TESTING FOR INTELLIGENCE
HOW CAN WE TEST TO SEE IF A COMPUTER IS REALLY THINKING?
We
are fortunate that such a test was devised in the 1940s by the brilliant English
mathematician Alan Turing. He invented the Turing Test for Thinking Machines
and like all great ideas, it is a perfectly simple concept. Turing suggested
isloating someone in a room with two computer terminals, through which conversations
could be conducted for any length of time and on any subject. One terminal
would be operated directly by a humanl the other by a computer. Both would
have to answer questions and converse with the subject of the test. To pass
the test, the computer system's conversation would have to be indistinguishable
from the human being's.
To most person's relief, no machine has passed the test yet. But there have been a few examples of unpredictability in machines which can only serve as smoke signals for what may soon become reality.
One often quoted example is that of the scientist at the Massachusetts Institute of Technology who was talking to a friend via a computerized telecommunication network. This is called teleconferencing and enables groups of people without actually having to come together physically. On this ocassion, the scientist's friend left the terminal without giving any prior warning, but it was some time before the other scientist noticed any difference, so sophisiticated was the computer system's conversation.
There are also ocassions in a chess program where a program written to play chess end-games had apparently mystified its opponents by breaking what were considered cardinal rules of strategy and going for a completely different approach. It had, in fact, discovered a new way of playing end-games.
This sort of thing
is possible not through any advances in hardware. but through the development
of much more sophisticated software. Most programs consist of a defined set
of instructions and tests which are executed sequentially, except when a certain
condition arises and the program 'branches' to other instruction. In artifical
Intelligence, a lot of work has been done on programs which can adapt or modify
themselves, according to experience. A simple example might be that of a program
which tested the colour of bozes coming down a production line. Its instructions
might be :
Is it red?
Is it blue?
Is it yellow?
If after a day's operation, the program had only encountered yellow boxes, it might decide to put the test for 'yellow' first rather than wasting time looking for colours which rarely, if ever, occured. The principle in this sort of programming is to show the system how to learn, rather than specifying in advance what they are to learn. The end-game program described above is one example of this, where the system, given the rules of the game and the objectives, has come up with something that even grand masters of chess had never thought of before!
......................................................................................
