Roger Schank wants to set the record straight. Despite what "Star Wars" and "2001: A Space Odyssey" have led us to believe, computers will never become "steel persons," capable of such human attributes as anger and greed.
Nevertheless, they may one day be smart enough to run the government, says the Yale professor who has for the last 17 years been in the vanguard of American research on the creation of truly intelligent computers.
In his new book, The Cognitive Computer (Addison-Wesley, $17.95), Schank makes what he calls a "half-serious" proposal: that future citizens elect computerized "belief systems" instead of politicians.
Computers, he says, could be programmed as either liberal or conservative and people could vote for the candidate that most closely shares their beliefs.
The great advantage mechanical leaders would have over human ones, muses Schank, is consistency in their policies and an absence of lies. Voters would no longer have to worry, for example, that the peace candidate would change his mind and get them involved in an unpopular war.
It's hard to tell how much the bearded, 38-year-old director of the Yale Artificial Intelligence Project actually believes in the futuristic scenario he has painted, and how much of it is an attempt to get a rise out of critics who fear that computers will be running everything.
"I want to point out," he says, "that there are advantages to computers running things."
But don't ask him if they're going to take over the world.
"It's the wrong question. It assumes that what we're after is the creation of this steel person. That's not really what's going on. We're creating an entity which is going to think, but that entity isn't also going to want to eat, for example."
Schank is concerned that the general public doesn't understand what he and other researchers at such universities as Yale, MIT and Stanford are attempting to do as they try to construct programs to mimic the complexities of the human brain.
"Artificial intelligence research has been terribly oversold," he says, with many magazine and newspaper articles declaring -- erroneously -- that the intelligent computer is here at last.
The truth is, admits Schank, that researchers are only at the Model T stage of development. Some computers can now read electronic versions of newspapers, summarize what they have read, and come to a new insight based on their reading. But those insights won't necessarily be correct, at least not at this point in their evolution, says Schank.
Calling the American hysteria over computer literacy "absolute hype," Schank advises those who want to be prepared for the world of tomorrow to simply take a typing course instead of trying to learn programming or a confusing computer language.
"Computers will change. The things you learn today will not be of any use in five years. With AI artificial intelligence systems where you can type English, probably you'll only have to know how to type."
Once the general public can "talk" to computers in plain English, says Schank, the true democratization of society will begin.
"I grew up in Brooklyn, not a particularly rich section of Brooklyn. I saw a lot of what happened amongst poorer people was really lack of information." They didn't know how to invest their money, what college to send their children to, what kind of car to buy. They didn't know how to find experts to advise them, so they would ask "the smart relative" for advice, which may or may not be sound.
One day, Schank predicts, people will be able to sit down at the smart computer that will be available in nearly every house, and simply type in such questions as "What college should I attend?" or "I'm serving chicken for dinner and I don't want to spend more than $5, so what kind of wine should I buy?"
The computer would ask the user questions -- either a few or many, depending on the complexity of the query -- to clarify the person's preferences. And out will come the answer.
Physicians, business people and other professionals are already beginning to get advice from so-called "expert systems," some of which understand English. Those systems, which can do everything from help a doctor make a diagnosis to suggest where geologists might mine, were designed by feeding the computer a great deal of technical information and basic rules on how professionals make decisions.
One system designed by Schank's company, Cognitive Systems Inc., allows Belgian bank customers to get computerized advice about the stock market.
"It follows," says Schank, "some very dull rules. You can teach those rules to a stockbroker when he first starts out and he'll be as good as those rules. Our machines will be as good as those rules, too. But that stockbroker five years later should be better. The machine will be the same -- as of now."
Therein lies the AI researchers' chief hurdle: teaching the computer to learn from experience, as people do.
Before a truly intelligent computer can be created, researchers must unlock the secrets of the human mind, which allow people to think of the right thing at the right time, to remember what is important, to forget what is extraneous and to arrive at intelligent insights.
In one program, called Cyrus, Yale researchers convinced the computer that it was Cyrus Vance, secretary of state during the Carter administration, by feeding it a lot of personal and professional information about Vance.
Cyrus' crowning achievement was that it was able to come to sensible conclusions about things it had no direct information on. When asked, says Schank, if Vance's wife, Grace, had ever met Aliza Begin, the late wife of the former Israeli prime minister, the computer responded, "Probably on such and such a day."
The computer came to that conclusion by sorting through its data banks and discovering that both Grace Vance and Aliza Begin had attended a state dinner on a particular date. From that information, it assumed the two must have met.
Such inferences would be a snap for most people. But to Schank, it was one more step forward in an often frustrating attempt to teach computers to do what humrns do so effortlessly -- think.