Computers and Language

So-called artificial intelligence programs are still a long way from tackling language. And by language I mean languages which are spoken by humans, like English or Japanese. "Computer languages," sets of instructions followed by computers, are not the same as human languages. It has not been demonstrated yet that the languages which we speak can be reduced to such sets of instructions. If they can, we're still a long way away from doing it. If they can't, then, in my opinion, that would be one of the reasons not to worry about the machines eventually rising up and killing us all. If a computer was capable of having a conversation with me which was indistinguishable from a conversation with a human, then I'd be startled. And possibly a little spooked as well. I would include written conversations like those in chat rooms.

Math is exact and language is not. Often times letters, the symbols used to record some languages, have been used as mathematical symbols. Roman numerals are one well-known example of that. But while the symbols used in language and math may be the same, what they refer to is not. X + IV = XIV means exactly the same as 10 + 4 = 14, and any number of different systems of notation can be used to express exactly the same thing as 10 + 4 = 14. However, many times the simplest sentences are untranslatable from one language to another. And very many, perhaps most sentences cannot be exactly translated. Furthermore, in many cases, perhaps most, the best translation is a matter of opinion. Highly-qualified experts in linguistics and literature routinely disagree about whether this translation of a poem or novel is better than that one.

As long as we're talking about translation made by humans, that is. With all of the astounding advances made in computing, the best computer translation programs still routinely produce results which are comically bad and far inferior to any work done by any competent professional human translator. The same goes for original written compositions by computers compared with those written by ordinary lit students.

Computers are far beyond humans now when it comes to playing chess. (And if recent headlines have not misled me, computers are about to pass us as Go players as well.) But computers play chess differently than humans, by crunching enormous amounts of data. How do we humans do it? Well, we don't know yet.

Perhaps human intelligence would be less mysterious if the possibility were more often considered that it involves things which aren't reducible to math. Perhaps researchers sometimes resist considering that, because one of the things it would mean is that we're not, in fact, on the brink of developing artificial intelligence. Well, actually, many people think that we're already well past the brink, and that artificial intelligence has already existed for some time.

No doubt, information technology has produced many amazing things, and continues to do so with no end in sight. Maybe there's no reason for me to object to calling some of those things artificial intelligence. I don't go around angrily telling IT people to stop using the term "computer language."

But as I've said, a computer language is a fundamentally different thing than a language spoken by humans. And artificial intelligence, if we want to use that term for things which already exist, and why not -- I mean, people are using that term for things which exist, whether we approve or not -- is still fundamentally different from what human brains do. Playing chess, computers win. Writing poems, computers still have not given us any competition. Perhaps the things needed in order to write great poems are quantifiable. But perhaps they're not.

And perhaps the latter possibility is too often overlooked.