Artificial Intelligence and Globalization 2980

I don't think the analogy to chess is apt. Chess is a very closed world with very precise rules. I remember being at the first ACM meeting where chess was being played between computers (1964 I think) and recall that one of the players could be rolled in (a little bigger than a refrigerator) but most had to be remote because of their size. Now chess playing computers can beat grandmasters - but are still larger than them. :) And that line will not be re-crossed.

But I also remember being at the New York World's Fair (1968, I think). In the IBM pavilion there was a computer translating Russian into English and typing it out on a Selectric typewriter. A major point of the exhibit was that we are this far now and thus perfect translation must be just around the corner. I remember thinking back then: Don't you have to be able to understand context and therefore be able to think like a human to do a perfect translation.

Well we are in the same place today - perfect translation must be just around the corner, though some of the translation programs are very, very good. (Perhaps spurred on by the DOD's need for rapid translation of large amounts of material.) It may be one of those problems where you cut the distance to the finish line by half every couple of years.

The reason computers are not as good as humans at translation is that you need to have roughly the same kind of intelligence as humans do in order to be able to complete a good translation. You need understanding of the subject matter in order to know whether a specific word, in this context, means this or that. AI is not up to that yet.

But it is making progress. There is a race out in Calif.-Nv. where they race SUVs. The SUVs must go through a 150 mile, or so, course marked by GPS checkpoints over the desert. They have no knowledge of particular hazards and can not receive human instruction along the way. They do have cameras, radar, computers, etc. In the first year the race was run, none got further than 25 miles. They all hung up on something, overturned, etc. In the second year 10 or so finished.The main reason for the improvement was that in the first year they used strict rules to decide things. In the second year they moved on to fuzzy logic.

So the point is that AI is improving within limits. How fast this is happening and the logical consequences of this are discussed in a book "The Singularity Is Near: When Humans Transcend Biology" by Ray Kurzweil.

He suggests that technology will continue to develop at the same rapid rate as today and somewhere between 2025 and 2040 machines will be as intelligent as humans - both because of increased speed and because of software improvement. (I don't agree with the timetable. I don't see how Moore's law can continue to hold. We are beginning to have to manipulate objects at the atomic level and heat buildup is a significant problem. But it has been given a reprieve by recent developments.) (I also think this will require some kind of self generating software.)

Actually, as the breastle implies, Kurzweil suggests that what evolves will be some kind of human-machine hybrid. I don't think of that as a human. The "Singularity" in the breastle refers to the singularity around a black hole which has an event horizon beyond which no information can flow out. (Except via Hawking Radiation which Hawking himself recently conceded.)

So Kurzweil is suggesting that you really can't see beyond the point when machines (or the human-machine hybrid) become more intelligent than humans. And that makes sense to me.

Getting back to the original issue. Yes, AI will affect Globalization. But at the point AI evolves so it can do tasks like language translation as well as humans, it will be about as smart as humans. And whether humans will exist much beyond that point is a question.

Bill