Could a computer think like a human?
THE development of artificial intelligence (Al) is being intensively pursued in several locations in the world
THE development of artificial intelligence (Al) is being intensively pursued in several locations in the world. Al means a computer capable of rational thought of the same quality as human rational thought.
Many computer experts believe the development of Al is possible, even inevitable humanists disagree saying there are fundamental reasons why a machine can never think like a human.
How would we know if a computer could think like a human? It is commonly accepted in Al circles that the goals of artificial intelligence would be achieved if the computer could pass the Turing Test. This test is named after Alan M. Turing, a British mathematician and computer expert, who died in 1954.
Turing, one of the fathers of the modern computer, was a genius. He not only made contributions of fundamental importance in mathematics, but an enormous contribution to the Allied victory in the second World War when, working as a cryptographer for the British Foreign Office, he cracked the German naval Enigma code. He had a sad and lonely personal life and died by poisoning, aged 41, probably by his own hand.
The Turing Test involves a human interrogator who questions two interviewees, using only a keyboard and printer for communication. One of the interviewees is a computer and the other is humane Both are unknown to the interrogator and both are located in a separate room, out of his or her sight.
The interrogator asks questions designed to find out which of the interviewees is human. If the computer can fool the interrogator, then it has passed the Turing Test.
Humanists cite three reasons for claiming that it will not be possible to develop a computer capable of thinking like a human. The first is the human capacity for intuition. They say computers will never be able to think intuitively because they rely exclusively on rules, whereas humans, in addition, employ a subtle and sophisticated form of inference from experience.
For example, humanists claim a computer could never be a good physician. However this argument performed poorly in a practical test. In the US a medical computer programme called Internist I was given laboratory test data from real case histories and missed 18 out of 43 diagnoses.
The same data were given to a team of clinicians at Massachusetts General Hospital, who missed 15, while a committees of medical experts missed eight diagnoses. So, although the computer programme was significantly worse than the team of medical experts, it performed almost as well as the team of clinicians from one of America's best teaching hospitals.
It can be confidently predicted, therefore, that the computer programme would outperform a significant minority of physicians in general practice, even though the development of such computer programmes is only in its infancy. So, the humanists' first argument is unconvincing.
Their second questions the validity of the Turing Test itself. According to this argument, even if computers can one day perform the same intellectual tasks as humans, the concept of thought cannot be applied to the type of information processing employed by the computer. The analogy of the "Chinese Room" is quoted on this point.
The Chinese Room contains no doors or windows and only a small post box slit. Through this slit two sheets of paper are passed into the room. A story in Chinese is written on one of the sheets, and a number of questions about the story, also written in Chinese, are on the second sheet. After a period, a sheet emerges from the room through the slit. The writing on the sheet is in Chinese and correctly answers the questions.
It would seem obvious from this experiment that the room contains a knowledge of Chinese. However, the humanists argue that this deduction is not necessarily valid.
For example, the room could contain a person who knows no Chinese but merely follows rules given to him for converting some kinds of foreign squiggles into other kinds. In this case the room would contain no true understanding of Chinese, although it behaved as if it did.
This second argument is probably too clever for its own good. Scientific definitions are stated in behavioural terms. If something looks like a duck waddles like a duck, and quacks like a duck, then it is a duck.
If we cannot accept this, then we must be very sceptical of things in general, including human minds. For example, how could we be sure that anyone else on the planet is truly conscious except ourselves? I, think most people will agree that this question is not worth agonising over.
The third reason put forward by the humanists is that whether or not a computer could pass the Turing Test would depend on the mind of the interrogator.
According to this argument, if the interrogator had well developed human faculties he/she would not be fooled into thinking any artificial intelligence was human. To take an extreme example, it would be easy for a computer to pass the `Turing Test if the interrogator was a child.
It is not uncommon for a four year old child, when handed a speak and spell toy, to initially believe the toy is alive. Likewise it is easy to " see how an emotionally cold interrogator of only moderate intelligence and little experience of life could be fooled by a smart computer in the Turing Test.
But the more intelligent, experienced and human the interrogator, the more difficult it would be for the computer to fool him. The third argument claims that interrogators of a sufficiently high calibre would always diagnose artificial intelligence as alien.
I find this convincing. It is very difficult to see how any intelligence that was not born of a mother, did not face the various trials and challenges encountered in human development, had no sexual identity, was I never tortured by self doubt or exhilarated by success, had no emotional experience of family ties or romantic love, and had no experience of the fear of death, could think in a truly human way.
No doubt, in the future, computers will be able to simulate some or many of the characteristics of the human mind (readers may remember the haunting image of the computer HAL, in the film 2001 A Space Odyssey, begging for his "life" as the astronaut gradually unplugged his electronic brain circuitry), and even exceed some of them in capacity, but I do not foresee the development of a computer that would pass a properly administered Turing Test.
The further back in history one goes, the closer to uniqueness became the position assigned to humans in the natural order of things. However, with the development of biology, and more recently sociobiology, many of the characteristics previously thought to be solely human e.g. maternal love, self sacrifice, co-operation and altruism, were seen to be characteristics fairly widely shared with animals.
The one thing that survived as being uniquely human was the capacity for rational thinking. Humans are thinking animals. And now this last attribute common only to humans faces the threat of having to share its pedestal with the computer. Personally, I don't think the computer will succeed in mounting this pedestal.