SoylentNews is people
SoylentNews
It's always hard to take your eyes off Serena Williams. But it’ll be especially tough at this year’s U.S. Open, where the tennis champ is currently working toward a single season Grand Slam. She’s just so darn good. But what is it, exactly that makes her so good?
Sure, we can all speculate—it’s her power, her serve, her stamina, the way she controls a point. But we can’t calculate precisely what makes her game so special. IBM believes it can.
Since 1990, IBM has been working with the United States Tennis Association to support the technological infrastructure of the U.S. Open. Back in the day, that meant generating scores and keeping the website up and running. Today, it means doing those things while also analyzing millions of data points about every player, every stat, every point, in every tournament, extending back for decades to derive insight about how a given match—or career—will play out.
http://www.wired.com/2015/09/ibm-us-open-serena-williams-data/
Original Submission
(Score: 2, Interesting) by Anonymous Coward on Friday September 04 2015, @01:08PM
Of course skills cannot be reduced only to algorithm. You also need to have the proper hardware (if your taste is gone, you'll never be a good chef), and the proper data (part of becoming a good chef is learning how things taste).
And of course, being a good chef is a special case because there' no objective measure (unlike in sports, where you can clearly say more goals win, or first in finish wins), but the goal is to match the unconscious evaluation done by human brains. A real chef has a real human brain that works the same as other human brains (with variations, of course, but then, such variations will also be between different guests of the restaurant, so unless the chef is an outlier, it still can be relied on). On the other hand, a computer program would need to reverse engineer it (or use a reverse engineered algorithm provided by the programmer), and with reverse-engineering, you are never sure when you hit an incompatibility (which in this case would equate to preparing a meal that doesn't taste well).
The language translation problem is, of course, based on the fact that the translators simply lack of knowledge. Also humans can be trapped by lack of knowledge.
Indeed, I myself got trapped into it:
I once read about a nice example, where a computer meant to analyse sentences was given the sentence "time flies like an arrow", and then could not decide if it was a special type of flies that happen to like arrows, or if it was time that is flying, just like an arrow does.
Now, when they cited that sentence, they didn't say what it meant, because "of course" everyone knows it anyway. Now I'm not a native English speaker, and hadn't come across that sentence before. However I didn't notice that I had no idea what it means because the meaning seemed clear to me: I thought it refers to the arrow of time, that is, the fact that the past is different than the future.
Later, I read again about that, in a German text. There they explained what this sentence means, by translating it into the equivalent German saying (which doesn't feature an arrow). It was at that point that I recognized that the "clear" sentence wasn't that clear at all to me.
Note that training is also just a way to pass on information. And it is indeed quite common that programs are trained (think of spam filters for example; no one would be that silly to try to write down a strict rule about what spam is, but also Watson is trained).
Also, if you've even been in a language learning class, you certainly have experienced how different the translation attempts can be from the actually correct translation, even with a common human knowledge background aiding you. Indeed, I'd not be surprised if an American beginning learner of Russian or a Russian beginning learner of English would produce the very same translation you gave (especially if he happens not to be a Christian and didn't ever hear about this bible quote).
Also I've once came across an example of double-translation by humans that also changed the meaning (although quite subtly). I don't recall the details, unfortunately, but I had read a German translation of a French book (the German title is "Zufall und Notwendigkeit"; probably the English title is then something like "Randomness and Necessity"), and then I came across a quote in the German translation of an English book where the author quoted a sentence from that book; being an English book, he surely quoted it in English.
Now I was curious if the translator of the English book looked up the quoted sentence in the German translation of the French book, or translated it himself. So I looked up the sentence in my copy of the German translation of the French book, and compared. At that point I noticed the subtly different meaning. My conclusion is that the quote in the German translation of the English book is the translation of the English translation from the French book, and it was this double translation that caused the slight change in meaning compared to the direct translation. (I was, of course, acting on the hypothesis that all involved translators were up to the task; otherwise the discrepancy could have been explained by lack of competence)
So my conclusion is that all those examples of where machines fail are nothing machine-specific, but just amplified versions of the same errors human make, with the amplification being caused by the fact that the computer programs and data are not yet advanced enough.