Watson "wins" at Jeopardy

[MatthewV]

I just watched it and found it rather interesting. There are a few amusing moments.
I was under the impression that the entire text was fed to Watson instead of word-by-word. Do the contestants get to see a text version or do they just get to listen?

[Samadhi]

[LordKinbote]

[extro...*]

[Zag]

He was sent the entire text as soon as it was shown. Note that no decent player listens for Alex Trebek to read it, he reads it on his own. Many people (including Ken Jennings, I know, and probably the other guy, as well) read a whole sentence at a time, and have read the entire clue within an instant of it being shown.

The painful prevalence of the younger generation to write "your" for "you're" and "would of" for "would have" (and similar mistakes) is evidence that they don't have this talent, and they are silently "reading aloud" in their heads.

[wordcross]

So, extro, in light of that information, still think the "commonplace OCR software" would give him an advantage? The players all have to wait until the buzzer is available anyway.
_________________
Has anyone really been far even as decided to use even go want to do look more like?

[jesternl]

I'm guessing the episodes will also be on Hulu.com,. legality not as much in question as Torrents

[extro...*]

[MatthewV]

I know I read aloud in my head. Very slowly. And because I respond much better to listening, I usually listen to the question. I also am very poor at Jeopardy!

Because of the way the show is produced, it is hard for me to know if the text version is just for TV viewers and contestants. I guess as long as contestants can see the entire question instantly it is fair enough to feed the computer text.

It would have been more elegant if they had taken the video-to-text approach. It wouldn't have been difficult and would have quelled the notion of an unfair advantage.

[Scurra]

Hmmm. I was impressed, but in the same way that Deep Blue impressed me - fun but not really relevant because it wasn't doing what it appeared to be doing.

Sure there were some clear advances in context and pattern recognition, but I would hope there had been in the years and years that people have been working at this. And even then it was clearly not able to deal with some very simple things (that Final question being a great example.)

Still it was an entertaining piece of television, and I guess that's all you could really ask for.
_________________
_{still Quiz Olympiad champion. Must get a life.
New definitions: COFFEE - someone who is coughed upon}

[extro...*]

[Chaz]

[Scurra]

[*casinopete*]

I realize I'm several days late, but this:

[Zag]

Meh. OK. The theoretical number of arrangements is 3^(19^2), which is, in fact, 1.74e+172. So you're right that you would have to do a ton of pruning, which means that you'll never have it "completely solved." However, the game lends itself well to such an approach, and the likelihood of missing relevant combinations is minuscule.

[Nsof]

i wonder how much time until my desktop can do the same as Watson?
unrelated:

_________________
Will sell this place for beer

[ctrlaltdel]

i play chess a lot, use a comp to analyze a lot, know how the chess progs have grown in their ability. they are better than humans. no contest.

but i am more impressed with this watson comp than with rybka or firebird beating a grand master. chess is not complicated, it is very simple - but its too complex. ehrm, that didnt make sense, did it? rules of chess are much simpler than rules of natural language. theres no contexts, hidden meanings, multiple meanings, irony (whereby it is understood from the tone of the sentence that we actually mean the complete opposite!), puns, etc...

i understand watson didnt have to deal with ALL of those, but a chess move is a chess move. Nf3. it is very simple for a computer to understand what just happened. the rest is just tweeking the algorithms to perform best, since bruteforcing is not yet within the computational horizon of machines. (it might be one day, who knows.)

btw, would you say that winning thru algorithms is more "valuable" than winning thru nalimov tablebases? either way, the comp doesnt "know" it just computes 1s and 0s.

so i am very impressed with how well watson understood the questions. its akin to making a chess program that would know a move when i input "Nf3" and also when i say "you know, i made the second move of the italian opening" or "i put my knight two squares ahead of the funny looking piece with a pointed hat".

[Thok]

[ctrlaltdel]

i think in Go, as oppose to chess theres a much bigger margin of moves that are "ok" especially in the beginning of the game. moves that dont necesserily win or lose or significantly add to either extreme - they are "playable". in chess (and especially in the opening) the moves are almost forced if you dont want to give the opponent a significant advantage.

[Nsof]

Just Months After Jeopardy!, Watson Wows Doctors With Medical Knowledge
Not there yet but i can certainly see doctors routinely using these kind of support systems within 20 years or so.
_________________
Will sell this place for beer

[Death Mage]

"Don't wanna sound like a dick or nothing, but it says on your chart that you're fucked up. You talk like a fag, and your shit's all retarded."
_________________
* These senseless ramblings brought to you by Insanity™. If you just can't figure the dang thing out, it must be Insanity™.
[YOUR AD HERE!]

[LordKinbote]