Three Types of Artificial Intelligence

Lighthill begins by dividing artificial intelligence into three areas which he calls A, B, and C. A stands for advanced automation, the objective being to replace human beings by machines for specific purposes, for example, industrial assembly, military reconnaissance or scientific analysis. A large body of work in category A is concerned with pattern recognition, with the programming of computers to read documents or to recognize spoken words. C stands for computer-based central-nervous-system research. The objective here is to understand the functioning of brains, either human or animal, using the computer as a tool to complement and interpret the facts of experimental neurophysiology. A more remote aim is to understand the architecture of the brain so completely that we {279} can borrow the brain's architecture in building a new generation of computers. Finally, B stands for bridge, an area of work which aims to make contact between A and C, to make use of neurophysiological models in designing machines to perform practical tasks. The main activity in area B has been the building of robots. Lighthill's main conclusion is that while work in areas A and C is promising and worthy of support, area B is largely illusory. Both advanced automation and neurophysi-ology are real sciences with concrete achievements, but the bridge linking them together is nonexistent. Insofar as artificial intelligence claims to be the unifying bridge, artificial intelligence has no real existence.

In the United Kingdom, Lighthill's sweeping condemnation of area B had the effect of a self-fulfilling prophecy. Funding of efforts in area B was withdrawn, and areas A and C continued their independent development. But in other parts of the world, and in particular in the United States, the same decline of area B occurred without Lighthill's intervention. I conclude that Lighthill's diagnosis was accurate, that his harsh words about area B were well founded.

Here is Lighthill's famous caricature of area B:

Most robots are designed to operate in a world as like as possible to the conventional child's world as seen by a man: they play games, they do puzzles, they build towers of bricks, they recognize pictures in drawing-books, "bear on rug with ball," although the rich emotional character of the child's world is totally absent. Builders of Robots can justly reply that while robots are still in their infancy they can mimic only pre-adult functions and a limited range of those at most, and that these will lead on to higher things. Nevertheless, the view to which this author has tentatively but perhaps quite wrongly come is that a relationship which may be called pseudomaternal comes into play between a Robot and its Builder.

As Lighthill predicted, the fruitful development of artificial intelligence during the last ten years has occurred in area {280} A and not in area B. The successful programs are utilitarian tools designed to perform specific tasks without any pretensions of intelligence. They are not supposed to understand what they are doing, nor to mimic the operations of a human intelligence. Their software incorporates large quantities of human knowledge, but this knowledge is supplied to them from the outside, not generated on the inside by any process of internal ratiocination. Artificial intelligence has been practically useful only when it abandoned the illusion of being intelligent.

What of the future beyond the year 2000? I agree with Lighthill in expecting advanced automation and neurophysiology to continue to develop as separate sciences. They still must grow within their separate domains before bridge building will be possible. But sooner or later the two areas are bound to come into contact. The time will come when brain-architecture in area C begins to be understood in detail and program-architecture in area A begins to acquire some of the sophistication of natural human language. At that stage the time will be ripe for building bridges, and the further progress of the two areas will be merged rather than separate. Machine builders will be able to incorporate the structures of neuro-physiology into their designs, and neurophysiologists will be able to monitor neural processes with properly matched connections between brains and computers. When progress has reached this point, the grand claims of artificial intelligence, so prematurely made and so justly ridiculed, will at last be close to fulfillment. The building of truly intelligent machines will then be possible. The artificial intelligence revolution will be upon us in full force.

How long will this take to happen? My guess is about fifty years from now, some time between the years 2000 and 2050. I am old enough so that I do not need to worry about seeing my guess proved wrong. What will be the human consequences of artificial intelligence? To guess the consequences is even more hazardous than to guess the date of the revolution. I will say only that my view of the consequences is not {281} apocalyptic. I do not see any real danger that human intelligence will be supplanted by artificial intelligence. Artificial intelligence will remain a tool under human control. To conclude my assessment of the future of artificial intelligence, I quote again from Lighthill:

The intelligent problem-solving and eye-hand co-ordination and scene analysis capabilities that are much studied in category B represent only a small part of the features of the human central nervous system that give the human race its uniqueness. It is a truism that human beings who are very strong intellectually but weak in emotional drives and emotional relationships are singularly ineffective in the world at large. Valuable results flow from the integration of intellectual activity with the capacity to feel and to relate to other people. Until this integration happens, problem-solving is no good, because there is no way of seeing which are the right problems. The over-optimistic category-B-dominated view of artificial intelligence not only fails to take the first fence but ignores the rest of the steeplechase altogether.

My verdict agrees with Lighthill's. I believe that artificial intelligence will succeed in jumping the first fence before the year 2050, but that human intelligence is far ahead and will remain far ahead in the rest of the steeplechase, as far into the future as I can imagine. Man does not live by problem solving alone. Artificial intelligence will not only help us with solving problems, but will also give us freedom and leisure for exercising those human qualities which computers cannot touch.