Are computers faster than humans?

[Tor]

Here is one I just came across that is rather in keeping with another thread going now: http://honoraryunsubscribe.com/shakuntala_devi.html. Carl, I know you will have a rather notably better understanding of the machines referenced in this one than I do, but I remain very impressed by the arithmatic.

[crfriend]

Every so often this sort of thing happens and the general public is astounded that a human can "beat" a computer at anything. However, not all is what it may seem.

First and foremost, there is a certain subset of the population -- savants -- who can do seemingly asounding things, and this is one of those cases. Secondly, in this particular exercise, the computer in question was a model introduced in 1964 -- not exactly the "stone age" but certainly not what we see in today's machines. What we see here is a savant pitted against an old machine, and we do not know how the machine was programmed to handle the task at hand. I'm certainly not saying that this was a one-off, but the challenge was not representative of humankind as a species, nor of the modern state-of-the-art for computers.

That having been said, there are still things that humans are splendidly suited to that computers have yet to master (or be taught to master), human-language translation being the most vexing example. A human who is fully fluent in two languages will be able to translate between the two with better speed and far greater accuracy than even the best computer algorithm going, and this does not seem to be changing as quickly as the computational boffins would have predicted (or lots of folks would have liked). This is because human language is an associative enterprise where the form of a word in one language inherently "means" something to someone who is fluent in that language: a computer must go through the process of "looking up" each word it encounters. Going further, human languages use a number of different grammars which any attempt at computer-translation must take into account whereas the human will make those changes instinctively. The human wins, if not every time, then most every time.

Savants are an interesting phenomenon, and one that really hasn't gotten the study it might. It's highly probable that savants have "different wiring" that allows them their "abilities" (e.g. the ability to see or otherwise perceive numbers, and "translate" them into "results" in much the same way that a human translator transposes one language into another). It is worth noting that it's not all that uncommon for individuals to exhibit some of these traits: the wife of a good friend "feels" time and starts feeling claustrophobic as deadlines approach. Another guy that I used to work with many moons ago was a savant when it came to sports-statistics; he could answer any sports-related question with 100% accuracy, but once off that topic was, well, slightly dimmer than most -- a good guy, but not the sharpest tool in the shed.

So, in short, that was an interesting exercise, but not particularly relevant to humanity or computing. For math, a computer -- even a very slow one -- will kick my arse 100% of the time, no questions asked. (Not that I'm perhaps representative of humanity as a whole.)

[Tor]

While I knew well that anything of today would be rather faster, I can't say I'm terribly surprised at the age of the machine, especially as arithmetic is exactly what computers are most ideally suited for. A harder world these days for folks like her than a century ago.

As for translation, while it may not be research level, the difficulty is obvious enough with a little study of language and a taste of automatic translation - even between near lingua francas. Heaven help the machine trying to translate between two tiny obscure languages. While I don't know two of them to verify, I'm not sure I would want to see the results - except maybe to gloat over human superiority while it's still possible.

I imagine the research level translators can do better than what I have seen, but one could wonder if machine vision for most anything short of art will be achieved before translation of nice, deceptively easy for a human to parse, simple text. Hmmm... maybe that's just saying the obvious that computers will manage art when computers manage art.

As for arithmetic speed, I'm at least decent with numbers, but even the cheapest calculator I've ever worked with will beat me unless the problem is ridiculously simple so it takes longer to key it in than for me to answer it. Of course, by the time I started using calculators a washing machine of storage had been compressed into a shirt pocket sized square an eighth inch thick. That makes two of us, so the sample must be at least somewhat representative

[crfriend]

While I knew well that anything of today would be rather faster, I can't say I'm terribly surprised at the age of the machine, especially as arithmetic is exactly what computers are most ideally suited for.

This is a pitfall that many fall into, and one which nearly cost me a career with the devices as I stank at math as a younster and, as everybody knows, you HAVE TO BE a whiz at math to do anything with computers. It's BS, of course, and computers are quite adept at manipulating arbitrary symbols (else we'd not have fora such as this one) and it's more down to logic and orderly thinking than mathematical ability.

I imagine the research level translators can do better than what I have seen, but one could wonder if machine vision for most anything short of art will be achieved before translation of nice, deceptively easy for a human to parse, simple text.

Machine vision is surprisingly well advanced as evidenced by Google's (and others') self-driving cars, and algorithms that make use of that are in very common use in consumer-grade kit like jitter-stabilising cameras. It's all down to application. Art, on the other hand, may well stay out of the computing sphere for a good many more years as it's a notoriously hard judgment call as to what "art" is. I'll say that computers have entered the sphere of art when I hear a proper symphony composed by a machine that is pleasing to the ear and evokes emotion; nothing less will do. Computer "painting" can be done with cameras and Photoshop, so that doesn't count.

As for arithmetic speed, I'm at least decent with numbers, but even the cheapest calculator I've ever worked with will beat me unless the problem is ridiculously simple so it takes longer to key it in than for me to answer it. Of course, by the time I started using calculators a washing machine of storage had been compressed into a shirt pocket sized square an eighth inch thick.

I predate pocket calculators, and still recall (and have) my grand-dad's HP-35 -- the first "pocket scientific calculator" (for big pockets, mind!) -- device. In light of that, I am usually more willing to reach for my slide-rule when I need something "quick and dirty" than I am to fire up a "calculator" program on any of my computers; it's just easier that way, and besides, 3 significant digits are almost always enough to prove or disprove a hypothesis.

That makes two of us, so the sample must be at least somewhat representative

Beware sampling error -- huge pitfalls lie therein.