Alan Turing: can machines think?

An Aged Question

Can machines think? This question wasn't born with the first computers. We find it a lot of time ago. In fact, since the XVII century philosophers discuss about this. The first one who treated the argument giving a clear answer was René Descartes in the first half of 1600. The French philosopher, founder of Rationalism, was impressed by the automatons that the improved mechanical technique built in his period. He found that human behaviours can be mechanical and rational. The first ones are the movements performed by the body, the others the activities of mind. Descartes was convinced that these last ones couldn't be realized with a machine. This convincement came from the fact that in his period automatons were built to imitate movements and machines performing calculations were not so common (see Charles Babbage: a man beyond his time). The French philosopher thought that a clever behaviour is a complex behaviour, while machines were able to realize only simple movements. Now this argument can be refuted considering that, if we build small sectional pieces, we can assemble them in several ways getting a behaviour complex as we want. A harder Descartes' argument was the following: a machine can't change its behaviour when a different situation arises. Machines like that were built with pieces that cannot be changed by the machine itself, they were made of fixed hardware specifically studied for a particular purpose. This kind of machines were the only available until the first decades of the XX century.

The Turing's Machine

Apart from the forgotten attempt of Charles Babbage, a step towards an universal computing machine was made by the English Alan M. Turing. In a work of 1936, at the age of 24, he pointed out the concept of mechanical procedure. That work was about a mathematical topic regarding the Computability and the concept of Process that can be precisely described and realized. He found that something (i.e. a number, a theorem, etc.) is computable if it is computable by a simple machine that we now call the Turing's Machine.

The reason for its architecture was the finiteness of human memory: we can perform a lot of calculations mentally, but we usually use a paper to write down the intermediate results. So the Turing's Machine has two main units:

a tape divided in ordered squares in which an input/output unit can write or read a finite set of simbols s₀, s₁, ... s_N (the symbol s₀ is the blank)
a memory that can assume a finite set of states q₀, q₁, ... q_M (q₀ is the stop state)

The running is the following: the tape has some symbols written on it (the initial data) and the machine starts with the memory in a state q_I and the I/O unit on a square with a symbol s_J. Then the machine can write a new symbol, move the I/O unit to the right or to the left, change its memory state. These three operations are a function only of the current memory state and of the read symbol. When the q₀ is reached the machine stops and the symbols written on the tape are the results of the calculation.

For example, the program below realizes the logic OR function. Symbols are blank, 0, 1, while memory states range from 0 to 5. The I/O unit starts on the first symbol and the result is written in the square after the second symbol. On the left there are two examples 1 OR 0 = 1 and 0 OR 1 = 1.

An OR Implementation
NOW		NEXT
Read	Memory	Write	Memory	Move
*	q₀	-	-	stop
0	q₁	-	q₂	right
0	q₂	-	q₃	right
*	q₃	0	q₀	stay
1	q₂	-	q₄	right
*	q₄	1	q₀	stay
1	q₁	-	q₅	right
*	q₅	-	q₄	right

Examples

1 OR 0 = 1
State	Squares
q₁	1	0	_
q₅	1	0	_
q₄	1	0	_
q₀	1	0	1

1 OR 0 = 1
State	Squares
q₁	0	1	_
q₂	0	1	_
q₄	0	1	_
q₀	0	1	1

For a Java version of the Turing Machine click here.

The Turing's Machine is the first example, even if abstract, of universal calculator before the advent of computers. With the Turing's Machine we have the (second) born of the concept of universal computing: we don't need a machine for every task but a single one is sufficient and the realization of automatisms changes from hardware to software. Moreover we have a method for defining realizable algorithms because this machine is intrinsecally machanical and, finally, with such a theoric construction it's possible to demostrate some powerful theorem about computability and its limits.

The Thinking Machine

Now, what's about our starting question? The final Descartes' argument against the possibility of machines to think was their incapacity in conforming the behaviour to unexpected events. With an universal calculator all this changes. In fact, the behaviour complexity is related to the available memory and now the memory is no longer static and implemented in hardware but it can be written and erased by the machine itself: it is a tabula rasa that can be filled with any kind of information and it's not devoted to a particular purpose.

Alan Turing proposed a test, now known as the Turing's Test, that can be used to verify how much a computer is a thinking machine. The starting point is the Imitation Game: there is an examiner chatting at two terminals, one of these is connected to a computer, while the other one is connected to a man. With this test a computer is seen as an information elaborator which simulates the human brain to a functional level. How many possibilities has an average examiner to understand what terminal is connected to the computer in a short time? This time is related to the human cleverness that the machine has reached and, in a work of 1950, Turing wrote that according to him before 2000 an examiner will have only 70 per cent probability to find the computer in 5 minutes. He was convinced that before the end of the century the concept of thinking machines would be accepted even by common people.

In 2001: Space Odissey Arthur C. Clarke wrote that the famous HAL 9000 was built in 1997. HAL, where are you?

In the next Issue Science History will be

Galileo Ferraris: the rotating magnetic field