Computers

We are blessed to live next door to a bright and charming young lady named Lauren. Lauren just turned three years old; and already she is the living embodiment of the reason computers cannot now nor will they ever be able to think or act like humans.

Just the other day Lauren wanted me to take her next door to Ruby's house. We were standing outside in the front yard and her mother was chatting with my wife over near the front door, while Lauren and I were standing around the corner of the house near the driveway. I told Lauren I couldn't take her over to Ruby's unless her mother said it was o.k. She thought about that for a moment, and I could see the cogs turning behind those innocent eyes. She then said, "O.K., you wait right here and I'll go ask her." She then went around the corner, and as I sneaked up to listen, told her mother that I wanted to take her to Ruby's, and was this all right. Her mother said she didn't think so since Ruby was probably busy. A conversation which Lauren then returned to me to report as her mother saying it was o.k. for her to go. And this is a perfect example of why computers will never be capable of thinking like people.

You see, human beings, even very young ones, are devious creatures. Our world is only partially logical and linear. Very few matters are straightforward, and lies and deceit are commonplace human traits. Human beings are even capable of vast feats of self-deception. Computers, by contrast, are very straightforward creatures. At the most fundamental level of a computer's being, its universe is very honest and logical. Everything is either "yes" or "no" to a computer. Very few things are "yes" or "no" to human beings. And while one can program a computer to provide a fair approximation of some of these traits; at its core, the computer is entirely rule-governed, even when the rules are not obvious on the surface.

Computer designers know this of course, although they don't like to admit it. There is even a hot new area in computer design that uses a so-called "fuzzy logic." The idea is that instead of a computer relay being either "on" or "off" ("yes" or "no") the circuit is open or closed according a probability equation. Thus, in this naive way, the designers think they can make a computer that more closely mimics human thought. But even "fuzzy logic" is logical and rule-governed. Even probability equations are equations and are based on the confident assumption that we want devices that follow the rules.

Sometimes, when confronting the real unpredictability of human behavior, the engineers even try to mimic this by introducing some degree of randomness into their equations; within careful limits of course, because no one wants a truly random machine. We want machines that do useful and predictable work for us, but we also want them to be "creative" in some way so we think that by making them unpredictable to a limited degree that this makes them "creative." Which is pretty silly. But the engineers' substitute for the unpredictability of human beings (randomness) misses the point by a wide mark. Human behavior is not random. It is deceitful and devious and self-serving and oblique and confused and insane and all sorts of other things, none of which have anything to do with randomness.

So computers can either be rigid ("yes" or "no"), highly probable, or have a dash of eccentric randomness. None of which, individually or in concert, is remotely close to how human beings operate.

Human beings, it hardly needs saying, are not nearly so neat and orderly as computers. Some of our behavior may be rule-governed, and much of it may be vaguely predictable. But we often act and think in sloppy, undisciplined ways. This is anathema to rule-based machines like computers. When humans create machines like computers they deliberately design them to be much more orderly and well-behaved than we are ourselves--that is, in a sense, the whole point of the machine. The last thing anyone wants in a machine is one that behaves like a human! But for computers to really think like humans, we would have to design unmanageable, unpredictable, ill-behaved, capricious, lying, deceiving, double-talking machines. And nobody wants to do anything like that, even if we could.

There are other reasons why computers can't think like people. Two simple examples:

A few weeks after Lauren learned the utility of cutting corners, her mother, Gail, was putting up new wallpaper in their house. These things are a matter of taste of course, but their new wallpaper looked to my eye like an overstuffed Victorian sofa had exploded all over the wall, leaving the imprint of some floral phantasmagoria from floor to ceiling. Gail asked me what I thought of the new wallpaper. Seeing that the job was three-quarters done, and that Gail was obviously pleased, I knew perfectly well what the right answer was: "It looks great," said I. "Thanks," said she.

As routine a human interaction as one could find. Why would I be so casually "deceitful?" For the simple reason that Gail was not really asking me what I thought about her wallpaper; she was asking me a question the hidden form of which was: "I am very pleased with my new wallpaper. Please tell me you like it too so I can be reassured that it is as nice as I think it is." To which the only polite and caring response is to say: "It looks great."

Then there is my wife Gaby, who regularly says odd things to me, but has one charming idiosyncrasy, which I will now reveal. Whenever I start out the door, car keys in hand, my wife always says the same thing: "Drive carefully." Now is this what it appears to be: a simple imperative sentence about driving? Nope. If it were, then the proper response might be to strike myself on the forehead with the palm of my hand and say, "Gee, I'm glad you said that. I was planning to go out there and drive like a banshee from Hell. Now I will most certainly drive with great care." Do I say this? Not on your life! I always have the same reply: "Yes dear, I will."

Why do I respond in such an "inappropriate" way? Simple, the real message behind my wife's words is: "I love you very much, and whenever you go out I worry that something bad might happen to you, and so please reassure me that you will be o.k." To which message the appropriate response is: "Yes dear, I will." By which I am really saying: "I know you worry dear, but I will be fine, there is nothing for you to worry about."

The point in both these examples is simple: human beings often do not say what they mean and do not mean what they say. And yet we all understand each other perfectly well and even children catch on to this "meta-level" conversation very early in life. Computers, however, choke to death on a simple two-word sentence that appears to be about driving, but is really about something else. All efforts to program computers to understand natural language involves teaching them the "meanings" of various words and the contexts in which these words are used. What cannot be taught is the simple fact that humans routinely communicate in words whose meanings are irrelevant to the message.

When I was in college a visiting philosophy professor from Berkeley came to deliver a lecture on "Minds and Machines and Why Computers Cannot Think." He had even written a whole book on the subject. He told us that humans have minds and computers don't. That humans can feel love and grief and joy and fear and computers feel nothing. He told us that people are creative while computers are merely captives of rote. He told us that humans are noble and philosophical and religious and thoughtful and that computers are just fast calculators. He talked of human souls and ridiculed their lack in computers. He spoke of many grand and glorious attributes of human beings which, he assured us, could never be attained or even mimicked by mere machines. Perhaps not.

What all this determined philosophizing was aimed at was to parry a perceived threat from computers. It is somehow dimly thought that to say a machine can think and act like a human being demeans humans, debases them and implies that our nobility and majesty is being denied. Thus, the instinct is to say that computers cannot be like people because people have noble emotions and subtle perceptions and deep understandings and refined sensibilities and that a mere machine could never have any of this.

It pains us to realize that in a few years a chess-playing computer will finally beat Gary Kasparov (probably one using a program marketed by Kasparov himself, no doubt). Or to suppose that a computer might someday spin out the same boorish nonsense as a Sartre or a Wittgenstein or, horror of horrors, your average Berkeley professor. Could a computer write poetry? Sure. Could it compose inspiring music? Maybe. Could it make scientific discoveries? Yes. Could it replace human beings in lots of jobs and responsibilities? Count on it.

So does this mean that someday computers will equal or even surpass human beings? Nope. There is a whole huge domain in which human beings are unequaled. But it is not in our nobility or grandeur. It is not in our intellects or our accomplishments. No, it is rather in the guile of a three-year old child who has figured out how to manipulate the adults in her life to get what she wants. It is in all those conversations that appear to be about one thing but are really about another. It is in our readiness to lie and deceive each other and ourselves. It is in our perverseness, our unpredictableness, our contrariness, our fine-feathered madnesses. It is the sloppy, less-than-admirable aspects of our character in which human beings are, and will remain forever, unsurpassed. In these areas, no machine will ever be our equal. We ought to take comfort in this.