The general session was held in Hall "A" in the new convention center annex. There wasn't any carpet, and the acoustics weren't all that great but there was a lot of room. Cecelia Cowles (president of SHARE) welcomed the group and announced record attendance for a southern SHARE - 5,000 people were registered, with a high number of walk-ins.
Our speaker was Ray Kurzweil. He is an MIT veteran and is well known in the music industry; he manufactures and markets music synthesizers that are widely acknowledged to be the finest available. Kurzweil's latest love is speech recognition; he is in competition with IBM to produce the first practical device. I don't honestly know who will win this one; IBM has all the resources, but Kurzweil has a track record for getting things done.
Kurzweil is a bright boy, but not a particularly dynamic speaker; I heard many complaints throughout the week about his lecture. He droned on for an hour in an irritating monotone and kept fiddling with his microphone. Distracting.
We got the standard lecture on the explosive growth in power of computing systems - you know, the one where the speaker says "If the automobile industry had kept pace with the computer industry, cars would cost two dollars apiece". Unlike most speakers, Ray Kurzweil has been on the leading edge of technology long enough to appreciate the growth curve firsthand. At MIT he had access to the University's IBM 7094 - a behemoth with 147 K bytes, executing 250,000 instructions per second. Nowadays he deals with microprocessors that execute hundreds of millions of instructions per second.
All this computing capability will be needed. Today it is possible to speak to a computer and be understood. The computer's vocabulary is large, but you must enunciate each word separately and clearly - "connected" or "continuous" speech is still hard to parse. Kurzweil sees this mainly as a hardware problem - throw enough hardware at the problem and it will go away. By the mid-1990s processors will be fast enough and cheap enough to solve the connected-speech problem. This will have high value to handicapped persons.
In an offhand remark he clarified for me the difference between "handicapped" and "disabled" persons:
"...{speech recognition systems can} eliminate the handicaps associated with ... disabilities."The first industrial revolution displaced physical effort. Nobody complained much about this, because it was widely held that humans are fairly mediocre machines. Horses run faster than humans. Oxen can haul more timber than humans. Birds can certainly fly much better than humans. The evidence is overwhelming that humans come in second place in any physical activity.
We humans have an ego problem when it comes to matters of the mind. We have always believed that we are the universe's great thinkers. We are unique, we are at the apex of evolution! The second industrial revolution is upon us, and it displaces mental effort; many people are frightened by this idea.
Artificial intelligence ("brain within a harness") applications are relatively narrow. People should not be too concerned about AI efforts that take away some of the computational drudgery that we deal with on a daily basis. Kurzweil draws an analogy: Music is the only common human cultural expression. It has always used the highest technology available - new advances in materials sciences have been incorporated quickly into instruments. With modern synthesizers and sequencers, sound is being divorced from playing technique; you don't have to learn to use a violin in order to produce a piece of music that sounds like a violin. The artist still has the creative outlet, but no longer has to be a technician.
Computing technologies of today are very good at information retrieval, and we have some superior calculating engines. We have systems that "understand" the calculus, that can prove theorems, and that could in essence write the Principia Mathematica. We have automated chess masters such as CMU's "Deep Thought", which analyze one million board positions per second and play at near grandmaster level.
But our best efforts at automated pattern matching are dismal failures in comparison. No computer can tell the difference between two pictures of a dog and a cat (a two-year-old-human can do this). Vision systems that merely do edge identification require somewhere on the order of 100 trillion floating-point operations per second. Pattern matching is expensive! Kurzweil sees a trend away from digital devices back to analog components; they are useful in applications not requiring much in the way of accuracy. A few transistors can do a fast analog multiplication, and may be just what the vision systems people need.
The English mathematician Alan M. Turing proposed a test to determine if a artificial construct had acquired intelligence. Put a human judge on a teletype and have the judge ask questions of the entity at the other end of the telephone wire. If the judge cannot tell the difference between a computer and a human, then the computer is intelligent. The Turing Test has been widely accepted as a valid test for intelligence in the years since, and Turing himself expected that a machine would pass the test by the year 2000. Alas, the best guess these days is that no machine will pass the Turing Test for another half century beyond.
Because Kurzweil's business deals with a number of "leading-edge" issues, he is concerned with fostering creativity among his employees. One way he does this is to get people from different disciplines to work each others' problems. He "has gotten the signal processing experts to work on a linguistics problem {for example}; they tend to approach the problem differently. We've had some success doing this..."