Artificial Intelligence—Is It Intelligent?
THE competition was intense. Move by countermove, the opponents were battling it out over the chessboard. But they were not ordinary chess players. On one side was the world computer chess champion Cray Blitz. On the other side was the challenger Hitech. Both of them were specially programmed supercomputers, running on different programs. And both of them were powerful enough to outplay all but the top-ranking human chess players. They were engaged in the world computer championship title match.
By the final round, Hitech gained the upper hand, as everyone had expected. It needed only a draw to win. But to everyone’s surprise, Hitech failed to notice a subtle move Cray Blitz was developing. Suddenly, Cray Blitz came from behind and defeated Hitech, retaining its world championship!
Stories like this tend to leave some people feeling uneasy. It is somehow unnerving to learn that computers have become better than most humans at playing chess, solving puzzles, or proving mathematical theorems; that there are robots that can hear, see, and even talk; or that doctors consult computers for opinions on treatment or diagnosis. Is science fiction coming true? Have computers become so smart that they will soon be masters of the world?
Such concerns are justifiable because we normally associate activities like problem solving and use of language with intelligence. We do not expect machines to do these things—not even computers—because ordinary computers are no more than high-speed information processors that respond to commands. But computers like Hitech and Cray Blitz are far from ordinary. To describe what they are trying to make such computer systems do, scientists have coined the term “artificial intelligence,” or AI. And the claims and predictions they have made for these machines have not helped to calm the uneasiness.
In 1957 Herbert Simon, now a Nobel laureate, predicted: “Within 10 years a digital computer will be the world’s chess champion.” More recently, computer scientist Harvey Silverman of Brown University in Providence, Rhode Island, claimed that “in a few years we expect to develop [a computer] that will have a vocabulary of 5,000 words and will understand most conversations in plain English.” Really, is the human mind becoming obsolete?
What Is Artificial Intelligence?
To reason, to understand, to discover meaning, to deal with unfamiliar circumstances, and to make decisions—these are things usually associated with the human mind. The ability to do these and many other tasks is what intelligence is all about. Ever since the 17th century, scientists have been dreaming of a “thinking machine” that could solve mathematical and logical problems. However, it was not until the advent of the electronic computer in the mid-1950’s that the dream began to take on substance.
Most of us are familiar with the computer’s ability to store, retrieve, and process large amounts of information at great speed. Because of this, computers are used in accounting and bookkeeping; handling files, catalogs, indexes, and so forth. In all these operations, raw data is fed into the computer’s memory, and the computer is given a set of instructions, or program, on what to do with the data. In a computer used for accounting, for example, the machine may be programmed to process all the information at the end of the month to produce bills and statements for all the accounts.
Of course, it takes a certain kind of intelligence to do the kind of work described. Essentially, however, such systems merely follow a predetermined set of steps, specified by humans, until the job is accomplished. If something is missing or has gone wrong during the course of operation, the machine stops and waits for further instructions from the human operator. Such machines can be said to be efficient but hardly intelligent. Computers with artificial intelligence, though, are to be a different breed.
Basically, artificial intelligence is a set of instructions, or programs, that attempts to enable a computer to solve problems on its own—the way a human does. In one approach, rather than follow a spelled-out, step-by-step procedure that leads to the solution, the computer tackles the problem by trial and error. The result of each trial is analyzed and used as the basis for working out the next trial.
This principle may sound simple in itself, but when it is applied to real-life situations, things can become very complicated. Why? Because few things in real life are as simple as yes or no, black or white. Rather, everything is full of shades of meanings and subtle implications. For example, if a certain medical procedure is to be recommended only for patients over six years of age, what about a child five years and ten months old? Such decisions are far beyond what today’s computers can handle. However, if the field of application is restricted, AI can be successfully applied.
For example, armed with artificial intelligence, Hitech was able to defeat some of the better chess players all on its own, without any external, human direction or intervention. But how does it do this? The computer carefully examines the opponent’s move, then searches through the thousands of positions in its memory to come up with the countermove that would minimize the opponent’s potential for gain. To do this, it checks through 175,000 chess positions every second, or over 30 million positions in the three minutes it usually takes to come up with the right move.
AI at Work
Good as Hitech is at chess, it is totally helpless at other games or tasks. This is because Hitech is only programmed to play chess. Its memory has been stocked with a vast amount of information about chess moves and step-by-step instructions that enable it to “think” in a logical fashion. In other words, as far as chess playing is concerned, Hitech is an expert. And that is precisely what computer scientists call devices such as Hitech—expert systems.
An expert system is basically a computer stocked with an extensive collection of information in a particular field. Along with this, it is programmed in such a way that it can guide a user to the precise information he needs with a minimum of time and effort. It often does this by means of a set of if-then rules: If a certain condition is true, then a certain action should be followed. The user “communicates” with the expert system through a keyboard and video screen or some other device. The store of information and the if-then process give such expert systems the appearance of intelligence—artificial intelligence.
Today, expert systems are being used in various aspects of medicine, computer design, mineral prospecting, accounting, investment management, space flight, and so on. Computer scientists are working on expert systems that can process not just one if-then situation at a time but many such operations simultaneously, as does the human mind. Also under development are systems that can “see,” “hear,” and “speak,” albeit in a limited way. All of this has caused concern in some circles. Will computers become as smart as, or even smarter than, man?
Is There Any Limit?
What scientists have been able to do with expert computer systems is truly impressive. There remains, however, the crucial question: Are these systems really intelligent? What would we say, for example, of a person who can play powerful chess but can do or learn hardly anything else? Would we really consider him intelligent? Obviously not. “An intelligent person learns something in one area and applies it to problems in other areas,” explains William J. Cromie, executive director of the Council for the Advancement of Science Writing. Here then is the crux of the matter: Can computers be made to approach the level of intelligence found in humans? In other words, can intelligence really be artificially made?
So far, no scientists or computer engineers have been able to reach that goal. In spite of the prediction about chess-playing computers, made over 30 years ago now, the world champion is still a human. And in spite of the claim that computers will be able to understand conversations in English or other natural languages, this still remains at a rudimentary level. Yes, no one has learned how to build the quality of generality into a computer.
Take language, for instance. Even in simple speech, thousands of words are strung together in millions of combinations. For a computer to understand a sentence, it must be capable of checking all the possible combinations of every word in the sentence simultaneously, and it must have an enormous number of rules and definitions stored in its memory. This is far beyond what present-day computers can do. Yet, even a child can manage all of this, plus perceive the nuances beyond the spoken words. He can discern whether the speaker can be trusted or is being devious, whether a statement is to be taken literally or as a joke. The computer is not up to these challenges.
The same can be said about expert systems with the ability to “see,” like the robots used in automotive manufacturing. One advanced system with three-dimensional vision takes 15 seconds to recognize an object. It takes the human eye and brain only one ten-thousandth of a second to do the same. The human eye has the innate ability to see what is important and filter out nonessentials. The computer is simply inundated by the mass of details it “sees.”
Thus, in spite of the advances and promises of the state of the art in AI, “most scientists believe that computer systems will never have the broad range of intelligence, motivation, skills, and creativity possessed by human beings,” says Cromie. Likewise, renowned science writer Isaac Asimov states: “I doubt the computer will ever match the intuition and creative powers of the remarkable human mind.”
A fundamental obstacle in achieving true intelligence artificially is the fact that no scientist or computer engineer fully understands how the human mind really works. No one knows the precise relationship between the brain and the mind or how the mind uses the information stored in the brain to make a decision or to solve a problem. “Because I don’t know how I do [certain things with my mind], I cannot possibly program a computer to reproduce what I do,” confesses Asimov. Putting it another way, if no one knows what intelligence really is, how can it be built into a computer?
g88 7/8 pp. 12-15