issue 162 - August 1986
Illustration: Hector Cattolica
The end of intelligence
Some of the most intensive research in the computer industry is
directed towards artificial intelligence - the creation of the Ultra-Intelligent
Machine as the 'next stage along the evolutionary path'. Now, says
Michael Shallis, is the time to blow the whistle on this project
and reassert human rather than mechanical values.
IMAGINE a television game show in which blindfolded contestants have to guess whether they are talking to a person or a computer. Any questions can be asked and any response given; answers do not have to be truthful. If the contestant cannot distinguish between person and computer they have to conclude that the machine they are talking to is 'intelligent'.
An imaginary game of this type - the Turing Test - was devised in the 1940s by Alan Turing, regarded as one of the founders of computer science. If (or when) a computer wins it will be, by the rules of the game, as intelligent as a human being.
The aim of the science called Artificial Intelligence, or Al for short, is to build such a machine. But neither the idea nor the intention stops there. The aim of Al research is to develop thinking machines that outstrip all human intellectual capabilities; leaving mankind, in the words of Donald Michie, Britain's leading Al researcher, 'living in the interstices of uncomprehended, incredibly intelligent electronic organisms, like fleas on the backs of dogs'.
The desirability of human creation superseding humanity is not questioned in Al circles; Michie has said that such developments cannot be classified as 'anything but good'. Such an enterprise is justified both on the grounds of evolutionary progress and in terms of sheer practical necessity.
The world's problems, it seems, have become so complex that human intelligence is no longer adequate. The only hope is that powerful rational machines will move us all towards a more secure and richer state. Each new generation of machines assists in the design of the subsequent, superior generation until 'the leapfrogging of machine intelligence evolves into a higher realm than human thought,' as Turing put it.
The stated intentions of the 'artificial intelligentsia' may sound more like science-fiction than anything real. So how far have they got along this road? Certainly when confronted by a word processor or desk-top computer we might have our doubts: the limitations of the machine are all too apparent. We don't seem to be supplanting ourselves with a supermind. So what is the reality?
Over the past thirty years Al research has been directed toward three main areas; pattern recognition, machine learning and language development. Success in all three areas has been surprisingly rapid, as a few examples will illustrate.
As far as pattern recognition is concerned there are already commercial (albeit very expensive) systems that can analyze telephone conversations, picking out over 500 key, pre-programmed words and identifying speakers from their voice prints.
This was no easy problem to solve. Human speech requires a context from which to identify the beginnings and endings of words as you will discover if you say aloud to yourself: 'It's not easy to wreck a nice beach'. Nevertheless conversation can to a limited degree be sufficiently well analyzed to provide a powerful system.
Visual images can in a similar way be recognized by the well-trained computer - even to the extent of extracting information about vehicles, buildings and people from electronic images. Police forces, for example, use computers linked to television cameras for checking the number plates on cars travelling at speed. The computer can be instructed to disentangle from a fuzzy image an object that it has been taught to look for - particularly useful for analyzing pictures taken by spy satellites of military vehicles and weaponry.
The second general area of Al is machine learning. It was argued in the early days of Al that machines would only be able to do what they were instructed to do (which is more or less true) and hence they would never be able to learn for themselves. The challenge to Al scientists was to demonstrate the falsity of this premise. If a machine could be instructed how to learn it would thereby transcend its instructions.
Computers even today can be taught by example. A paint-spraying robot can be guided through a process by a skilled human technician and the robot then remembers the skill and can repeat it ad infinitum.
But in a more sophisticated version, the computer can be shown objects and then interrogate its teacher. If it has learned about pencils, for example, and is then shown a pencil without a pointed end it may ask why this object is a pencil (and be told it has not yet been sharpened). The learning is guided, concept-building and experiential. Remembering that computers can be given senses and mobility (wheels, television sight, touch sensors, electro-chemical taste organs, microphone hearing) then one can see that the ultra-intelligent machine may not be too far along the road of research.
A third area of intensive research has been computer languages. The Japanese are developing the 'Fifth Generation' machines whose aim is not only to recognize but understand spoken and written language as well as 'image' (or imagine) the concepts they receive. By combining methods of language analysis, problem-solving systems and a massive knowledge base, the Fifth Generation machines are intended to be at least like a limited but highly clever specialized colleague.
But the forerunners of the Fifth Generation computers already exist in the form of 'expert systems'. These are programs designed to solve particular problems such as how to put out a fire on an oil-rig; they are designed to encapsulate the knowledge of an acknowledged expert in the field. Typically the user will respond to a series of questions from the computer about a particular problem and then be offered the most appropriate solution.
In practice the science fiction is rapidly becoming science fact. After all, even the cheapest electronic calculator can outstrip human ability at simple arithmetic. And the vastly complex computer systems used, for example, in commercial airline bookings or defence systems far outstrip large teams of qualified personnel. The question today is not whether intelligent machines can be built but whether they should be built.
Professor John McCarthey of Stanford University once asked 'what do judges know that we cannot tell a computer?' The answer at the level of pure reason is 'nothing', but in terms of the human condition the answer is overwhelmingly 'everything'. The intelligent machine is not, and never can be, human. Its intelligence is machine intelligence, alien intelligence. Even if machines became self-conscious, feeling organisms they would still be machines. When you cut them they would not bleed, but would short-circuit. Their awareness' would be machine awareness and the more sophisticated they become the more alien they become. To be judged by a computer might be fine in terms of its thoroughness in analysing the law but it would be inhuman and to that extent it would be unethical. But the implications for this new and undoubtedly powerful technology do not stop there.
Computer technology, with Al at its cutting edge, is a product of the linear, mechanistic and reductionist thinking, which has characterized Western society for around three hundred years. It is born out of a particular techno-economic and political philosophy where efficiency, quantity and imperialistic power are the idols.
We, as computer users (for that is what we all are, even if indirectly), cannot win. In a computer game, however skillful the player becomes, the machine continues unrelentingly, making the game harder until it inevitably wins. So in the wider world we, the players, lose our skills, have our actions confined to pushing buttons and watching a screen, hence becoming actually less human, our abilities atrophied. This is true whether the skills are those of games, skills in the workplace or even skills in handling our lives. As the technology intrudes so we become reduced.
The intellectual imperialism of the artificial intelligentsia needs to be seen through and transcended. We, as human beings, need to reassert our worth, our values, our culture, instead of succumbing to the subtle, persuasive and hypnotic forces of the market place and its 'thinking' technology.
Michael Shallis is a lecturer in Physical Sciences at the University of Oxford, UK and the author of The Silicon Idol (OUP).