Wave of the Future?
The following piece was originally written as part of an on-line discussion and is fairly informal in style. It presents another approach to the whole question of computers and Machine Translation, this one zeroing in from AI (Artificial Intelligence) in general and also from a whole other AI division, namely MV, or Machine Vision.
It was originally composed as a reply to John McCarthy's assertion on the newsgroup alt.fan.unabomber that our highways will soon be dominated by "electronic chauffeurs." John McCarthy is of course the granddaddy of AI and actually invented the phrase back in '59 or so. He also was one of the prime creators of the LISP computer language.
Anyone who wants to hear John's contribution to this discussion in detail can look up the messages from that newsgroup for the spring and summer of 1998 on Deja News or go directly to John's own website using the URL provided in the Links Related to This Site section at the bottom of the home page. You can also find out a fair amount about John by picking up a copy of Steven Levy's Hackers and finding him in the index.
I should start off by saying that I have a great deal of personal respect for John in every way. He's obviously accomplished a great deal in his lifetime, and one has to respect that. He's also a very pleasant person on line and fun to talk to, though he doesn't suffer fools gladly (come to think of it neither do most of us). But i want to stress how real and human this man is.
If you visit his website, you'll find, not surprisingly, that John believes in the possibility of virtually infinite development, just of people on this planet, before you even get to AI. He's absolutely sure that human beings will be around here for at least the next billion years, maybe even 4 billion. And he's totally certain that there will always be sufficient resources for everyone, even if the population continues to rise to 15 billions (where for some reason he foresees it as leveling off), and even if that means lots of people will have to eat artificial food (perhaps AF?) like synthetic proteins and carbohydrates.
But the totality of John's optimism doesn't stop therego look at his website and see for yourself. Not surprisingly, he also describes himself as a "neo-conservative" and is convinced that anyone who disagrees with him must be a wimpy liberal under the influence of other silly liberals who just can't think anything through.
There's a tag line he puts at the end of all his messages that sums up his attitude fairly well:
He who refuses to do arithmetic is doomed to talk nonsense.
In his book Hackers, Levy says that even back at MIT in the 50's, quite a few people talked about John's "boneheaded optimism." I hope the preceding provides a fair picture of this remarkable man who has accomplished so much in so many different ways.
John took a lot of criticism and a lot of put-downs during his time on our newsgroup, but he seemed to weather them all quite tranquilly. Talking about the Unabomber seemed equally important to both of us, though for quite opposite reasons. For John, I think Kaczynski was the ultimate image of flight from technology for the wrong reasons, for me the man represents a kind of sanity even approaching sanctity (at least in the Dostoyevskian sense) in recognizing that the most certain danger facing humans is our failure to perceive that the planet's resources are far more limited than anyone really wants to believe, that even the discovery of fabulous new technologies tomorrow morning will not balance out all the forces threatening us that we've already set into motion. You can find more about my own view of the Unabomber in the Theatre section of this program. (And no, of course I don't defend the Unabomber's crimes).
I should perhaps add that I'm certainly not a LudditeI love my computers, my central heating, and even, within limits, cars and airplanes, just as you do, so I'm every bit as caught up in the planet-destroying process as anyone else. but I don't feel I can ignore what is clearly happening all around us.
I don't merely like technology, I revel in it, just as most of us on-line do. But I also recognize that there might be a price to pay for it. And that price might yet turn out to be a lot higher than most of us are willing to pay.
And now we come to my treatment of John McCarthy's assertion that all our cars will soon come equipped with...
from alt.fan.unabomber, June 10, 1998
Okay, I'm back in NYC for a while, let's see if I can start where I left off.
I recently showed how three areas of so-called AImachine translation, voicewriters, and robotsare far less formidable than either their suporters claim or their enemies fear.
Now let's take up John's latest little cutesie-pie-in-the-sky gadget, the so-called "automatic chauffeur." This one is a REAL LOSER and makes even the mischief-making potential for more conventional robots look almost innocent by comparison.
First of all, let's ground ourselves in one important statistic. Ross tells us that 45,000 Americans die each year in auto accidents. I was just upstate, where i have a somewhat older Almanac putting the number closer to 50,000. Either way, we're headed for the near-holocaust figure of almost 5 million dead every hundred years, simply because we can't figure out a safer way of moving people around. And that's just for deaths in the USA.
So either 45 or 50 thousandno big deal. Either number is undoubtedly peanuts by comparison if the Stanford AI Delinquent All-Stars ever get their hands on our autos.
In a piece I recently posted, I asked how many people were willing to let a Machine Vision system make their next left turn for them, much less drive them off into the sunset. So let's start by concentrating just on left turnsas we all know, these are intrinsically more dangerous than right turns because they can make us a target from three different directions, and they frequently depend on making sense of a not always perfectly readable traffic light. Let's assume that Americans by and large take a mere 100 million trips in their cars each day, each requiring an average of six left turns (with three right turns on the outward drive which turn into left turns driving home).
So far our figures could not possibly be more conservative: 100 million trips per day, six left turns each.
That becomes 600 million left turns per day.
Let's assume our combination of Machine Vision plus Global Positioning Satellite (or WHATEVER technology they end up using) malfunctions only once for every 100,000 left turns it regulates. That gives you the potential for 6,000 accidents each day, of which one out of ten, or 600, may prove to be fatal. Multiply that by 365 and you come up with 219,000 fatal auto accidents per year, over four times as many as we now have.
And that's the total ONLY for accidents DUE TO LEFT TURNS.
But what if I've been too generous here, and the actual number of misfunctions turns out to be one for every 10,000? That means we are now dealing with 60,000 accidents per day, of which 6,000 may prove fatal. Which adds up to 2,190,000 fatal accidents per year. Not to mention all the permanently impairing ones and other injuries.
Or to come a lot closer to my every-day experience of computers, and probably yours as well, let's assume there's some programming malfunction on one out of 1,000 of those left turns. After all, our very best Optical Character Readers make about one mistake in a thousand, and those are based on a far more simplified form of Machine Vision than we'll need in our real-time, real-world robot chauffeurs. Such a percentage of errors (or a considerably WORSE one) may be intrinsic and endemic to any MV system required to make sensitive choices in a complex environment amid changes in light, shadow, color, visibility, temperature, humidity, rain/snow/sleet,and relative motion among multiple objects. And these sources for error are entirely separate from other soon-to-be-added extrinsic sources for error arising from unwanted but unavoidable electrical discharges.
That means we're now up to 600,000 accidents per day, with 60,000 fatalities. Our total annual fatalities now stand (or rather lie buried) at 21,900,000.
That also means we would now lose about 8% of the U.S. population each year, thanks to this wonderfully life-enhancing gizmo.
So we could also do away with all Americans in just over 12 years. This would make a lot of other people around the world very happy. (yes, I know, my math is off herethere's an equation for a convergent rate of a diminishing amount, but I forget it).
In any case, WOWwhat a great idea! To be fair to John, he didn't necessarily invent this one at all. I remember reading about it in blazing "colour" in the London Sunday supplements during the 'Sixties, along with other future features on how by 1980 vast numbers of people would live full-time under the oceans, except when they took off for their cabins on the moon.
Of course none of the above figures takes into account the various kinds of accidents the system could still create with right turns, on the straightway, lane changes, or backing up. To say nothing of errors our MV, GPS, or other system might make in estimating the speed of all these vehicles relative to each other.
Nor do they take into account the probability that Americans make a great many more than 100 million car trips per day, and that each of them may require far more than six left turns.
Whether springing from a base of 1,000, 10,000, or 100,000, the causes for yet further extrinsic electric malfunctions are common, ubiquitous, and unavoidable, ranging from larger external electrical energy bursts (meteorological, solar, even cosmic events), smaller external disturbances (discharges from nearby buildings, power lines, other vehicles, tools used by road repair crews, or nearby electrical devices including the car's own radio and battery system), or internal discharges arising in the chauffeur system itself.
It's bad enough having all these discharges leak into our word processors, DBs, and netbrowsers every day, as we all know they do. But out on the highway, in our own cars in real time, depending on such a system will be nothing less than an unmitigated unending disaster.
Also, let's not forget the problems involved in persuading people to switch to such a system. Ideally, all vehicles would need to be converted at once. But since this is clearly impossible, there would have to be an intro period, when some vehicles had the system but others didn't. This could lead to yet further chaos and accidents, since the vehicles without MV & GPS could not sense those with it, and to some extent vice versa.
When you put a plane on auto-pilot, you still have a lot of air space around it before it can do something really wrong, and you also have a pilot and co-pilot keeping careful watch. But cars don't have a lot of air space around them on the highway, besides which it sounds like the whole point of this invention is to let everybody forget about driving and cram for exams, play PONG on the web, or make love in the back seat while the car is on the road.
[Let me interject one other detail here: no, I'm not some kind of car buff, I dislike them as much as many of you on this newsgroup [a.f.u] do. I don't drive in the city at all, put in about 300 miles a year on a rusted 1974 Plymouth Valiant upstate during the summer, since the nearest towns are 5 and 9 miles away.]
Oh, I forget one other little detail. The price. How much is this jolly system going to cost per car, per car-owner? Not to mention insurance rateswill GEICO & the others readily accept a totally unproven system without hiking the rates?
Then of course there MIGHT be the slight problem of getting the Teamsters' Union to go along with the plan, since it would mean that every single one of their 1.6 million members would be Luddified out of a job. Hey, maybe there really was no reason at all for the Unabomber to send those little packages to computer scientists. All we have to do is let the Teamsters know what these AI guys are planning, and... :-)
I could go on and on ripping this little fantasy to shreds, but i think you get the point and can come up with the rest on your own. In other words, send this idea back to the drawing board, along with that other non-starter, a personal helicopter for every citizen.
One thing would be interesting to know, however. Did the Stanford crew or Minsky and his minions ever make a Social and Environmental Impact Report on their plans for automatic chauffeurs? Or for that matter, on their ideas for machine translation, voice-writers, or robots? Was it, and is it, a good idea to put all truck, bus, and taxi drivers, all household and manual workers, all translators and interpreters, and all stenographers and dictation-taking secretaries out of work? Or did these blessed ones blithely and blandly assume that everything AI has to offer would be A GOOD THING FOR EVERYBODY and look no further? Wouldn't it be a good idea if they did go to work making such a report? Even now, at this rather late stage? The answer to this question might just be what this newsgroup is really about. How do you feel about this, John?
He who refuses to do arithmetic is doomed to talk nonsense.
Ontology recapitulates philology.
James Grier Miller
(cited by W.V.O. Quine)
Okay, I don't think I will comment very much after all, since I think the piece speaks for itself. All I want to say is that each time I tell you there are things computers may not be able to do, all "some of you" seem able to come up with is the claim that there are sure to be new technologies none of us can even dream of that will take care of all our problems, including the ones I have been mentioning here about MV and earlier about MT. And that I must be some dumb Luddite idiot not to realize that those new technologies and new solutions are certain to arise JUST WHEN WE NEED THEM.
Sorry. among the various skills I've gained over the years, I'm also a trained journalist. I'm not sure that's all that radically different from being a trained scientist. What this means is that I've been trained how to ask really hard questions and how to listen carefully and patiently for real and certain answers to them.
But what I'm hearing here from at least some of you are not real and certain answers at all.
They are much closer to a religion. To pie in the sky. To "don't worry, be happy."
I don't have the Bible quotation in front of me, but somewhere it says something like:
Yea, verily, how shall I ask you to believe in those things which you _do_ see? This is not true faith. Only in so far as ye shall believe in those things you can _not_ see, only in this way shall ye possess true faith and enter the kingdom of heaven.
Sorry. I just don't buy it. Or at least I don't buy it where the future of the planet is concerned. And I know that some of the people I most respect in the history of this planetpeople like Swift, Voltaire, Mark Twain, H.L. Menckenwouldn't buy it either. And the more I learn about Alan Turing, the more I'm sure he would think it's a crock too.
It's not even good science. And it shouldn't be considered good technology either.
very best, alex
This article is Copyright © 1999
by Alexander Gross. It may be
reproduced for individuals and for
educational purposes only. It may
not be used for any commercial (i.e.,
money-making) purpose without
written permission from the author.
to linguistics menu