Cartoon Aficionado / Contributor
Posted:
Feb 14, 2011, 6:06 AM
|
|
Elementary, My Dear Watson.
|
|
|
Starting this week, Jeopardy contestants will be competing against a computer named Watson. If they can beat the machine, they'll be very lucky.
-------------------------
"Never walk alone in a danger zone."
|
|
Key Animator / Contributor
Posted:
Feb 14, 2011, 11:20 PM
|
Not necessarily, Tammi. Although the machine might have an impressive database of trivia with a heuristic index, speech parsing is still processor-intensive. Remember, Alex Trebek will only be speaking the "answers" aloud to the contestants; there will be no text-based means of transmitting the answers to the machine.
I have IBM's ViaVoice on one of my computers. Although it does an admirable job of converting spoken speech into ASCII text, there's a discernible lag while the program tries to match the analog-to-digital microphone input against its database of vocabulary homophones. The human brain does speech-to-context conversion at astonishing speeds, so human contestants will already be retrieving the correct question from their memories while Watson is still deciphering the answer.
In relation to cartoons, however, I can point you to a cartoon that illustrates gorgeously how difficult programming simple tasks to a mindless machine can be: the Jetsons, their seventh episode: Rosey's Boyfriend. "Open door ..."
|
|
Supervising Animator / Moderator
Posted:
Feb 15, 2011, 5:58 AM
|
Round One on Jeopardy! ended last night in a virtual draw between Watson and humanity. Round Two happens tonight.
-------------------------
"If it's hot, don't sit on it." —NYC Mayor Michael Bloomberg
|
|
Supervising Animator / Contributor
Posted:
Feb 15, 2011, 11:26 AM
|
Remember, Alex Trebek will only be speaking the "answers" aloud to the contestants; there will be no text-based means of transmitting the answers to the machine. I gather that in fact Watson is fed the "answer" in text form. But what it does not have is the ability to profit from the incorrect replies of the live contestant.
-------------------------
Back by popular demand -
"La-la-La-la.. I can't hear you!"
|
|
Zinc Saucier
Posted:
Feb 15, 2011, 4:44 PM
|
see, here's the thing... in the various specials leading up to the actual games (airing this week), it covered both the way Watson gets the "answers"- transmitted via text at the same time they're read aloud by the host- and how it was a major problem for the development crew to get the program to a point where it DOES benefit from incorrect answers given by the contestants. However, last night, Watson did repeat a previously given incorrect answer, so I'm wondering if that part was finalized or if it's still more hit and miss.
(This post was edited by krisAHQ on Mar 1, 2011, 3:46 PM)
|
|
Cartoon Aficionado / Contributor
Posted:
Mar 1, 2011, 5:24 AM
|
Well, it did beat the all-time Jeopardy champs, so I guess I'm vindicated.
-------------------------
"Never walk alone in a danger zone."
|
|
Zinc Saucier
Posted:
Mar 1, 2011, 3:54 PM
|
See, that's kind of misleading though. Yes, the program won, but how exactly? Same question applies to Deep Blue beating Kasparov some years back. The program itself didn't do too terribly much except accomplish what they wanted it to on a somewhat acceptable basis, parse (what they called) "natural language" and come to a more often than not correct solution. But the fact remains that the only reason it HAD those answers was because it was quite possibly one of the biggest (for lack of a better analogy) encyclopedias on the planet. It didn't know or think of anything on its own, it just parsed the question and pulled up a factual answer that it already had. I bet, given a T3 line and Google access at their podiums, both Ken and Whatshisname would have also had every answer correct within 10 seconds or so. Ditto for Deep Blue; the program didn't beat Kasparov, the knowledge of the 5 former Chess Grandmasters that Deep Blue had access to was the key. Essentially, all the program did was visually identify the move Kasparov made and then search/compare it with the literally thousands of moves and strategies it was programmed with. From there it's a simple case of matching the right move with the right strategy.
The real test is going to be when a program can actually "think" for itself- without the benefit of the equivalent of the entire Library of Congress already loaded and ready to search for whatever terms happen to match.
|
|
Apprentice
Posted:
Mar 12, 2011, 5:24 AM
|
Well, it did beat the all-time Jeopardy champs, so I guess I'm vindicated.
Yes, this is great saying of yours.. And the best is that I am all with you.. :)
(This post was edited by krisAHQ on Mar 13, 2011, 6:04 PM)
|
|
|
|
