Not Science Fiction
In Our Own Image: Building an Artificial Person.
by Maureen Caudill.
Oxford University Press. 242 pp. $23.00.
Artificial Life: The Quest for a New Creation.
by Steven Levy.
Pantheon. 390 pp. $25.00.
Why can’t a machine be more like a man? After her rapid but thorough discussion of current research in a wide variety of computer-related fields, Maureen Caudill feels able to answer the question. Essentially her answer is: it can be—and maybe better.
This reply should come as a surprise to anyone who has had any contact with today’s computers, which, though extremely competent at performing some tasks that challenge even intellectually gifted adults, are yet totally incapable of others that young children accomplish without effort. For example, computers have been programmed to play chess at the grandmaster level, but it has so far proved impossible to build a robot that can walk across a room as well as the average two-year-old.
Caudill suggests that this situation will change fairly soon. Computer scientists, she writes, now recognize that many human skills, since they do not rely solely on logical reasoning, cannot be duplicated by traditional computers, which are suited only for well-defined tasks that can be performed in a systematic and logical fashion. Therefore, instead of relying solely on logic, researchers have refocused their attention on the way human and animal bodies and brains actually operate, and how human skills develop.
_____________
Using insights gained in this fashion, scientists have found a variety of ingenious methods—summed up in the catchall term, “artificial intelligence”—to attack commonsense problems of the kind in which computer logic must be supplemented with factual knowledge of the real world. And they have achieved some success in building simple models of a variety of human skills by abandoning digital computers in favor of neural networks: arrays of simple electronic elements (neurodes) in which each element is connected to all the others in a way patterned on the arrangement of elementary brain cells (neurons) in humans and animals.
Neural networks, for example, can “learn” to recognize patterns. Much as animal behavior can be conditioned by methods long known to animal trainers and intensively studied by psychologists, these electronic systems can be trained to produce a desired output signal when presented with a particular input. Initially, the network will respond by producing something quite different from what is desired. The training process then consists of repeatedly presenting the network with the same input while systematically adjusting the connections among the neurons until the actual output is identical with the desired one.
As an expert on neural networks, Caudill believes that growing sophistication in their use, together with advances in understanding how intelligent behavior can be modeled, will help machines acquire the qualities required by an android—an intelligent robot with near-human skills. Her ambitious list includes: mobility; dexterity and tactile skills; memory; vision; hearing; learning ability; problem-solving ability; speech-generation and -comprehension; and a basic knowledge of common-sense facts about the worlds of nature and society. Preferably, the whole device would be contained in a package bearing a general resemblance to the human body.
Caudill reviews, clearly if sometimes in so much detail as to daunt the nontechnical reader, the progress made so far toward this ultimate goal by researchers working in artificial intelligence, neural networks, and robotics. While admitting frankly that current systems are drastically limited in their ability to perform even the less ambitious items on her list, she is optimistic about future progress, assuming that rapidly developing computer technology will be able to overcome all deficiencies.
_____________
In the concluding part of her book, Caudill ponders a number of nontechnical issues. Some have been familiar since the industrial revolution, when the replacement of human labor by machines that could surpass men in brute force or in specialized repetitive skills produced major social changes. As she writes, far greater changes—and far deeper psychological reactions—would surely follow upon the introduction of machines that could compete with people on a near-equal level.
One troubling issue is what would then happen to the relationship between humans and the machines they have created. Caudill argues that a true android with all the capabilities she specifies—particularly the ability to communicate intelligently with humans—would of necessity also possess the consciousness and self-awareness that have until now been the signs of human uniqueness. The main question this raises for her is rather surprising. Under such circumstances, she proposes, the use of androids merely as servants to carry out human purposes would amount to slavery, and would therefore be morally unacceptable. (Caudill also believes that it is morally wrong for people to enslave animals.)
She takes this line of thought even further. Androids, once beyond a certain stage of development, would be able to replicate themselves and even to produce continuously improving models. Taken together with their other capabilities, this means that androids would eventually possess all the qualities needed to meet the biologists’ definition of life itself, albeit with a different chemical basis from life as we know it today. This new form of life, Caudill believes, would fairly quickly prove superior to humanity, with all its well-known flaws. In light of her concern about slavery, she therefore concludes that we would be well-advised to treat the first primitive androids nicely, in the hope that their descendants will not at some time in the future take an appropriate revenge on us.
_____________
With sentiments like these, one might be tempted to think we have passed into the realm of science fiction. After all, it may never be possible for computers to develop the skills Caudill describes, and even if clever engineers were able to program machines to mimic human behavior, they would still be machines, not living creatures—no matter how biologists or positivist philosophers care to define life.
Not so, according to the scientists whose work is described in Steven Levy’s Artificial Life: The Quest for a New Creation. If anything, researchers modeling living systems of various types have gone beyond Caudill’s conclusion that robots may one day evolve into a form of life to the more radical belief that computer programs can already not only model life but even embody it.
While Caudill concentrates on specific technologies, Levy devotes much of his book to theoretical explanations of computer science and to limning the often colorful personalities who inhabit the small world of artificial-life research. He begins with the hypothesis, put forward 50-odd years ago by the polymathic genius John von Neumann, that it would be theoretically possible to design what von Neumann called a self-reproducing automaton. This would be a device that could not only make copies of itself but also serve as the universal computer imagined by Alan Turing, a computer which could model any machine in the world—including, Turing asserted, a living creature.
_____________
In the 1980’s, mathematicians intrigued by von Neumann’s suggestion began to play with systems known as cellular automata, whose behavior is determined by a few simple rules. One, the Englishman John Conway, invented a game he called Life. The game began as a checkerboard populated by a few black dots which were born, continued living, or died, depending on what their neighboring “cells” were doing. It soon turned out that this simple system could give rise to unexpectedly complex behavior, as the dots formed different geometric configurations that were able to move in different ways. Soon it proved too time-consuming to follow the system’s moves with pencil and paper (or, to be more accurate, with shells on the floor of the Oxford common room where the game started), and so a computer was enlisted which could rapidly track the system’s evolution through thousands of generations.
Later researchers found that they could use different cellular automata to model the behavior of real living systems surprisingly well. One, for example, put on his computer screen a large number of creatures he called “boids,” and, assuming that each one would fly in a direction influenced by what its closest neighbors were doing, discovered that the whole population behaved in a way that described flocking birds better than ornithologists had previously been able to do.
Cellular automata provide only one example of how computers can model living processes. There is, for example, a close analogy between the way DNA molecules control the development and reproduction of living creatures and the way programs both embody and direct a computer’s flow of information. As Levy shows, this analogy has been used fruitfully as a tool to investigate in detail the development and workings of biological systems—and, in reverse, to improve computer programs and systems, including the kinds needed to solve the problems discussed in Caudill’s book.
But where Caudill is content to suggest that solving those problems will in time enable man to create new kinds of living creatures, some of Levy’s researchers believe that man has already created at least one new form of life—namely, computer viruses. According to these scientists, such viruses demonstrate all the essential features of life: organization, self-reproduction, and an ability to evolve. Other scientists, although waiting for something a little more ambitious and less troublesome to come along, agree that it has already become possible to conceive of life within a computer. None, in the light of these developments, would be willing to concede that the human species can expect to retain forever its sovereign position in the world.
_____________
Since the Darwinian revolution, many biologists have taken it as a fundamental truth that man is simply one species among others, a species with no claim to a special place in the universe but whose current supremacy is simply the lucky and quite likely transient result of a series of random changes in his genes. Therefore, as Caudill says explicitly, if androids come to replace man, so be it: that is just the way evolution works. Or, in the words of the physicist Norman Packard, quoted by Levy,
There’s a harmony to the evolutionary process . . . completely divorced from the existence or the maintenance of a particular species.
But if this is so, why should Caudill be so concerned about the morality of slavery, or about man’s mistreatment of other species? And why does Packard go on to say that “the only thing that worries” him is “that somehow we would be so inept as to introduce an element of cruelty in our successors”? A cursory glance around the world of nature confirms that such issues are of no concern whatsoever to any species other than man.
Is man different after all? The most impressive phenomenon discovered over and over again by Levy’s artificial-life researchers is that of “emergence”: the behavior of large systems of individual entities repeatedly following simple rules over long periods of time produces collective results that are qualitatively new, unpredictable, and often surprisingly complex. The human mind (unaided by computers) cannot grasp the emergent properties of even the simple rules it imposes on simple systems. But it is clear that those rules determine those emergent properties.
For artificial-life systems, the rules are made by humans. Thus, to anyone who believes that man does have a special role in the universe, the development of computers at this moment in history may serve as a particularly timely reminder that the existence of rules is a fundamental property of the universe. Machines may one day learn to mimic much of man’s ability to manipulate nature, but the uniqueness of man remains: he is the only creature who has been given both the ability to understand that there are rules of nature and the freedom—but only at his own risk—to disregard them.
