No one can deny that computers have enabled humans to do amazing things, like watch movies of the insides of our bodies, golf on the moon, and shrink the planet with a world-wide communication Net. And computers do keep getting faster and roomier. For about twenty years now, “Moore’s law” has held true, as every two or three years we’ve seen a tenfold improvement in computer capacity and speed. But factual reports of quantitative progress can easily slip into science fiction speculation, as has happened with the next three myths.

“Computers can accurately forecast the weather”

This myth is easily dismissed, and not just by personal experience. In the last half-century we’ve learned that the accurate prediction of the future state of a “chaotic” system like the world’s weather is not a time-consuming computation problem; it’s theoretically impossible. Dynamical systems (the technical term) are inherently unpredictable, because they exhibit what is called “sensitive dependence on initial conditions”. So computers forecasting next year’s weather are no more believable than computers squaring the circle, or flying faster than the speed of light.

“Computers can understand natural languages”

A “natural” language is what people use to communicate with each other, e.g. English, Japanese, Mandarin, Swahili. A “formal” language is what we use to pass on a set of instructions to a computer, e.g. Java, COBOL, C. Natural language understanding is what I hope you’re experiencing right now: making sense of this article. And natural language understanding is, and has been for some time, an unachieved goal of computer science.

In the 1960s, computer scientist and educator Joseph Weisenbaum wrote and released a computer program that dispensed pre-packaged advice modeled after that from a Rogerian psychoanalyst. The program purported to carry on a learned dialogue by simply scanning input text for trigger words like “mother” and “love” and regurgitating canned questions like: “Did you have a happy childhood?”

Weisenbaum knew full well that his computer program was hopelessly unprepared to master the complexity of human language. He said later that he wrote the program, called Eliza, in part to expose the myth of computer understanding of natural language.

“Computers have artificial intelligence”

Almost from their start, computers have far exceeded the human ability to calculate, to store and retrieve information, and to quickly and reliably respond to stimulae. Computers have also been programmed to solve complex, “knowledge-based” problems in “limited domains” such as mineralogy, drug interactions, and chess. There’s an annual contest at MIT of computer programs seeking to pass a limited version of the “Turing test” for computer intelligence, where contestant programs exchange email messages with human examiners in an attempt to fool them into judging that the computer correspondents are human.

Those dialogues at MIT have all been restricted in content to very limited domains, e.g. World Series records, the chemistry of free radicals, the use of metaphor in 19th Century Turkish Fiction. That’s not what Alan Turing had in mind when he invented his test in 1950. Turing wanted to test for “natural language understanding”, which can’t be exhibited in a limited domain of discourse.

That’s because in few human dialogues are there predefined domains. We solve without apparent effort what cognitive scientists call the “frame problem”, determining the best frame of reference to bracket a discussion. That is something that no computer has ever done. There are some useful computerized language tools, such as programs to assist human experts in the translation of technical documents, or perform single-speaker voice-to-text transcription. But there are no computerized tools that have demonstrated language understanding; no computer has come close to matching the language performance of the average three-year-old child.

You can buy a cheap, computerized toy that will beat you at checkers, or scrabble, or chess (If you’re a grandmaster, you’ll have to pay a bit more), but for no amount of money can you buy a computerized assistant who can follow instructions like these:

“Find me the … you know, thingy, that I’ll need for my meeting with that fat guy I met at Bob’s Grey Cup party.”

Most executive assistants could easily handle that request, and probably wouldn’t consider it to be much of an intellectual challenge. But there is no computer anywhere on this planet that can do that; humans have not yet built a computer that can match of the intelligence of the average two-year-old child, never mind that of a mature adult. And there are many things even a one-year-old child can do that a computer can’t, including recognizing his mother’s voice, and knowing when he’s being ignored.

Intelligent androids have been a science fiction staple for many years, from Stanislaus Lem’s Robots to Star Trek’s all-knowing but unfeeling Data. But will real computers ever be as smart as people? I don’t think they will. “A big mistake, Stan!” any student of the history of technological progress would doubtless warn me, recounting the embarrassing predictions of some famous naysayers of human progress. The day may indeed come when my opinion is condemned as misinformed, or even as bigoted “human race-ism”. But that day has not yet arrived; though the pro- and anti-AI disputants (e.g. Marvin Minsky and Roger Penrose) all make use of sophisticated computer applications to disseminate their opinions, there are as yet no computer intelligences contributing to the debate.

But surely, you say, I must at least concede that …

“Computers improve our lives”

I admit that I may be crossing the line calling this last proposition a “myth”. Consider just computerized advances in telecommunications, where many of you no doubt believe your lives are improved by products such as cell phones, CallerID, and Nortel’s Meridian Mail. Without computers, you couldn’t get cash at 3 am, or pay bills with a plastic card. Without computers, doctors couldn’t diagnose and repair many maladies; and some Alzheimer’s patients wouldn’t have their confidence restored, and their lives improved, by using Palm PDAs to compensate for their no longer reliable, biological memories.

But many other, far less desirable things would not even be conceivable without computers, like telephone voice jails that block customer access to service personnel, or Net porn and email Spam. And for we regular computer users there’s the background stress of the anticipation and experience of crashes of self-confidence every time confusion with a new, “improved” version of an essential computer tool makes us feel like a child who can’t even walk upright. The designer of the C++ computer programming language tells the story about once wishing that his computer were as easy to use as his telephone. Bjarne Stroustrup says that his wish has finally come true … since he now has difficulties using his telephone.

Computers have certainly improved my own life in many ways, by providing me employment as a softsmith and an online instructional designer, and by providing me with anecdotes for articles like this one. But my years of experience with computers, and my professional focus on the quality of user interfaces, have inclined me to be alert to, and to be stressed by encounters with computer systems that treat me badly, and by blatant misrepresentations of the capabilities of those systems. Computer systems will certainly not be a positive factor in my life, or yours, if hardware and software manufacturers neglect to spend sufficient time and money designing and maintaining quality user interfaces, and performing thorough testing of new and changed systems.

So what remedies am I suggesting? That members of design, or development, or testing teams stand up to their bosses and say things like “My responsibility to the larger community compels me to reject your unrealistic project plan”, or “We should tell customers the truth about what our product can and can’t do”? Well, sure, that would help. But not all of us work as developers or distributors of computer systems (at least not yet); and not many of us the courage of a Gandhi to change the world. And of course a single individual is almost always powerless. But as Margaret Mead and others remind us, power can be and always has been exercised by individuals working in concert. I for one believe that the time has come to act, and I promise to shame and debunk promoters falsely claiming human qualities in their computer system products.

In a world with so many important causes demanding your attention, it’s possible that you don’t share my commitment to this obscure cause. You may even judge my arguments about computer abuse of human beings to be sophomoric, or fanciful, or at best premature. But in time your opinion may change; so if and when you do conclude that some threshold of threat has been crossed, I ask that you remember these reflections and exhortations. If the time comes when you decide to take action, I hope you recall these warnings: Be wary of the motives of the powerful and self-interested scorning your opinion; and remember that it’s rarely true that complex problems have simple solutions.

© 2004 Stan Yack
Instructional Designer and Softsmith
www.softsmith.ca