Artificial intelligence - metaphor or oxymoron?

The field of artificial intelligence is interesting to a student of metaphor, because it was explicitly founded upon a metaphor - several of them, in fact.

In the 1950s, a group of scientists decided to try to provide the computer with intelligence. Their goal seemed attainable due to a common metaphorical identification of the computer with a brain. From their efforts emerged the field of artificial intelligence, or AI. As I thought about the basic, or root metaphors of AI, I realized that they took a form resembling a classical syllogism:

* Major Premise: The computer is a brain.

* Minor Premise: Thinking is computing.

* Conclusion: If we provide the computer with sophisticated programs, it will develop a mind similar to human minds.

Note how both premises of the syllogism are metaphors. Both have received widespread acceptance in popular and technical discourse. This underlying syllogism of metaphors generated a very persuasive discourse, which defined the field of AI. Once you assumed that the computer was a brain, and what it did was thinking, then it made perfect sense to expect your computer programs to generate a mind for the computer.

This persuasive syllogism can be seen at work in the predictions made by MIT's Marvin Minsky, quoted in Life magazine in 1970:

In from three to eight years we will have a machine with the general intelligence of an average human being. I mean a machine that will be able to read Shakespeare, grease a car, play office politics, tell a joke, have a fight. At that point the machine will begin to educate itself with fantastic speed. In a few months it will be at genius level and a few months after that its powers will be incalculable. (Darrach, 1970.)

In the more scholarly confines of Scientific American, Minsky also made some predictions:

Once we have devised programs with a genuine capacity for self-improvement, a rapid evolutionary process will begin. As the machine improves both itself and its model of itself, we shall begin to see all the phenomena associated with the terms "consciousness," "intuition," and "intelligence" itself. It is hard to say how close we are to this threshold, but once it is crossed the world will not be the same. (Minsky, 1966.)

Such assumptions have become the staple of science fiction movies and television shows, with cute little robots making clever quips to their human buddies. It is easy to write in a character and call it a robot. It has proved considerably harder, however, to actually build one.

Many of the optimistic predictions of early AI researchers have not been fulfilled. There seems to be a pattern of initial progress, leading to great expectations, followed by a leveling off, with little further advance. (For a discussion, see Dreyfus and Dreyfus, 1986.)

Many of the stumbling blocks appear when computers are set at doing commonsense tasks which humans perform routinely, but which necessitate programming huge amounts of background information into the computer. This background information, or "tacit knowledge," plays a crucial role in orienting humans to their world, and has proved almost impossible to duplicate in computers. (See Polanyi, 1958.)

Tacit knowledge is required to understand human language. Since computers don't have it, they have still not been adequately programmed to "understand" and produce natural human language. Computers are idiot savants when it comes to semantics - they can look up words in dictionaries for us, they can scan words into their memory files, they can even translate words into different languages. But when the task goes beyond relatively simple matching and rule-applying, computers do not function well.

After years of optimistic predictions, many researchers now feel the goal of complete computer comprehension of natural language is far off. "It's not in sight," Stanford professor Terry Winograd told Atlantic Monthly in 1988. "I'm not saying it will never happen, but it's not something that can be done by improving and tuning up existing systems." (Wallraff, 1988.)

I think what we are seeing here is the expiration of the valid entailments of the founding metaphors of AI. Early on, the field needed definition, it needed a sense of direction. The syllogism of metaphors identifying the computer with a brain, and computing with thinking, provided that guidance. Researchers thought they were describing future computer programs and their potentialities, when they were actually just tracing out the entailments of their metaphors.

For a time, the metaphors proved useful. But then, as all metaphors must, they diverged from the realities they purported to describe. The persuasive conclusion of the syllogism of metaphors - that giving the computer sufficiently sophisticated programs would endow it with a conscious mind - turned out not to be accurate.

During these years, however, a professional field grew up around the founding metaphors - complete with professorships, expensive laboratories, grants, degree-granting programs, etc. Many people in this field are so committed to the founding metaphorical syllogism that they attack any questioners as heretics or traitors. (See West & Travis, 1991.)

And yet it is possible that the founding metaphors will not turn out to be valid metaphors at all - they may be oxymorons instead.

This is an interesting situation to the student of metaphor, because we tend to assume a metaphor is a metaphor, and that's that. But that may not be that.

We may have a series of oxymorons masquerading as metaphors. Oxymorons are composed of terms which don't belong together, or which are contradictory in some way. Examples of humorous oxymorons are "jumbo shrimp," "graduate student," "standard deviation." (See Blumenfeld, 1989.)

The computer may be nothing like a brain. Identifying the computer with a brain may be putting together things that don't belong: creating an oxymoron.

Computing may be just a poor imitation of one small part of the vast and little-understood process we call "thinking." Thinking is computing may be another oxymoron, joining opposites together.

When you define oxymoron, and define a metaphor, the definitions don't fall too far apart. A metaphor also puts terms together which are usually thought of as separate - although usually not opposites, as in the case of the oxymoron. Still, the family resemblance is there, which makes it understandable how an oxymoron could, indeed, masquerade as a metaphor.

The early AI researchers were pioneers, exploring uncharted territory. They needed maps, and metaphors provided them with those maps. The powerful metaphors equating human intelligence with computer computation paid dividends in conceptual clarity and fertility. It is understandable that these researchers should follow where the bright implications of their founding metaphors led them, as they ventured further into the dark confines of the digital labyrinth.

Indeed, the founding metaphorical syllogism still lives, as we hear predictions for speech recognition systems which will be used by intelligent agents, perhaps powered by neural network programming. The old dreams die hard.

The computer - and its potentials - force us to examine our assumptions about humanity, mind, and language. Metaphors are an integral part of this process. We should realize their limitations, and not follow blindly where they lead. Computers are not brains. Computing is not thinking. Confusing the two domains will just make us seem dumber and won't make computers any smarter.

REFERENCES

Warren Blumenfeld, Pretty Ugly. (New York: Perigee Books, 1989.)

Brad Darrach, "Meet Shaky, The First Electronic Person." (Life, November 20, 1970, pp.58B-68.)

Hubert Dreyfus & Stuart Dreyfus, Mind Over Machine. (New York: Free Press, 1986.)

Marvin Minsky, "Artificial Intelligence." (Scientific American, September, 1966, pp.246-260.)

Michael Polanyi, Personal Knowledge. (Chicago: University of Chicago Press, 1958.)

Barbara Wallraff, "The Literate Computer." (Atlantic Monthly, January, 1988, pp. 64-71.)

D. West & L. Travis, "The Computational Metaphor and Artificial Intelligence." (AI Magazine, 12, (1), 1991, pp.64-79.)

Dr. Raymond Gozzi, Jr., is Associate Professor in the Television-Radio Department at Ithaca College, Ithaca, New York.