The Turing Option

Brian had to smile at this. “Sorry. You’re right, I better go back to the beginning. Renal refers to kidney functions. And Expert Systems, as you know, are knowledge-based programs for computers. What we call computer hardware is the machinery that just sits there. Turn off the electricity switch and all you have are a lot of expensive paperweights. Turn it on and the computer has just enough built-in programming to test itself to see if it is working all right, then it prepares to load in its instructions. These computer instructions are called software. These are the programs that you put in to tell the hardware what to do and how to do it. If you load in a word processing program you can then use the computer to write a book. Or if you load a bookkeeping program the same computer will do high-speed accountancy.”

J.J. nodded. “I’m with you so far.”

“The old, first-generation programs for Expert Systems could each do only one sort of thing, and one thing only — such as to play chess, or diagnose kidney disease, or design a computer circuit. But each of those programs would do the same thing over and over again, even if the results of doing so were unsatisfactory. Expert Programs were the first step along the road to AI, artificial intelligence, because they do think — in a very simple and stereotyped manner. The self-learning programs were the next step. And I think my new learning-learning type of program will be the next big step, because it can do so much more without breaking down and getting confused.”

“Give me an example.”

“Do you have a languaphone and a voxfax in your office?”

“Of course.”

“Then there are two perfect examples of what I am talking about. Do you take calls from many foreign countries?”

“Yes, a good number. I talked with Japan quite recently.”

“Did the person you were talking to hesitate at any time?”

“I think so, yes. His face sort of froze for an instant.”

“That was because the languaphone was working in real time. Sometimes there is no way to instantly translate a word’s meaning, because you can’t tell what the word means until you have seen the next word — like the words ‘to,’ ‘too,’ and ‘two.’ It’s the same with an adjective like ‘bright,’ which might mean shining or might mean intelligent. Sometimes you may have to wait for the end of a sentence — or even the next sentence. So the languaphone, which animates the face, may have to wait for a complete expression before it can translate the Japanese speaker’s words into English — and animate the image to synchronize lip movements to the English words. The translator program works incredibly fast, but still it sometimes must freeze the image while it analyzes the sounds and the word order in your incoming call. Then it has to translate, again, into English. Only then can the voxfax start to transcribe and print out the translated version of the conversation. An ordinary fax machine just makes a print of whatever is fed into a fax machine at the other end of the connection. It takes the electronic signals that it receives from the other fax and reconstructs a copy of the original. But your voxfax is a different kind of bird. It is not intelligent — but it uses an analytical program to listen to the translated or English words of your incoming telephone calls. It analyzes them, then compares them with words in its memory and discovers what words they make up. Then it prints out the words.”

“Sounds simple enough.”

Brian laughed. “It is one of the most complex things that we have ever taught computers to do. The system has to take each Japanese element of speech and compare it to stored networks of information about how each English word, phrase or expression is used. Thousands of man-hours of programming have been done to duplicate what our brains do in an instant of time. When I say ‘dog’ you know instantly what I mean, right?”

“Of course.”

“Do you know how you did it?”

“No. I just did it.”

“That I just did it is the first problem faced in the study of artificial intelligence. Now let’s look at what the computer does when it hears ‘dog.’ Think of regional and foreign accents. The sound may be closer to dawg, or daw-ug, or any other countless variations. The computer breaks down the word into composite phonemes or sounds, then looks at other words you have recently said. It compares with sounds, relationships, and meanings it holds in memory, then uses a circuitry to see if its first guesses make sense; if not it starts over again. It remembers its successes and refers back to them when it confronts new problems. Luckily it works very, very fast. It may have to do thousands of millions of computations before it types out ‘dog.’ ”

“I’m with you so far. But I don’t see what is expert about this voxfax system. It doesn’t seem to be any different from a word processing system.”