IBM To Ken Jennings: You Will Beg For Mercy

[Manic!]

http://blogs.forbes.com/bruceupbin/2...mepagelighttop

Watson, IBM’s Jeopardy-playing supercomputer, is going to open up a can of massively parallel whup-ass on an unsuspecting world in February, when the powerful Q&A machine goes up against the two greatest Jeopardy! champions of all time, Ken Jennings and Brad Rutter, on national television.

Jennings, you’ll recall, won 74 games in a row in the 2004-2005 season, a DiMaggian record that will likely never be broken. Rutter, who has never lost a J! match, won $3.25 million in prize money, also a record, by taking first place in three different Tournaments of Champions. These guys are the best there are at what they do–unlike yours truly, who was doing really well until crashing hard in my one and only J! appearance.

Something tells me that Jennings and Rutter are going to go down, too, if not this February, then eventually. As anyone who has seen 2001: A Space Odyssey or the Terminator series knows, computers don’t give up. Nor does IBM when its reputation is on the line. Fifteen years ago IBM finally beat chess master Garry Kasparov with its 1.4-ton Deep Blue supercomputer.

This time around IBM built an entirely different kind of machine, a marvel of natural language processing that can apply up to 1,000 algorithms to the info-equivalent of millions of books. It understands anagrams, puns, word play and has memorized every Shakespeare soliloquy, major river and world capital on Earth. A team of a 20 to 25 IBMers have spent the past four years tuning Watson’s hundreds of algorithms to speed up its analysis, inferencing and answer retrieval. In 2006 it was rarely confident enough to buzz in and was correct only 15% of the time. Now it is smart enough to have spent much of this autumn in Yorktown Heights, N.Y., sparring successfully with former J! champions in preparation for the Jennings-Rutter throwdown.

David Ferrucci, the IBM scientist running the project, won’t say what Watson’s record was this fall, but insists it is ready for Ken and Brad (and Alex Trebek). You can track the results of those matches here. Watson, named after IBM’s founder Thomas J. Watson, harnesses a massively parallel network of two to three thousand Power 7 computing cores in six refrigerator-sized containers. IBM has fed Watson any and all licensed and publicly available content to create a huge semantic index. When clues are read by the host they are fed electronically to Watson, which parses the text, formulates hunches and checks all the evidence it can retrieve to test its hunches and then generates its five best answers, assigning each a confidence level before deciding whether to buzz in. It can assess the dollar amounts left on the J! board, how far behind or ahead it is, how well it is doing in a given category and adjust its confidence level to suit. Ferrucci says if Watson is way behind it will accept a lower confidence level of, say, 40%, in an attempt to “go for it,” but it can also think to itself “I’m so far ahead why take the risk?” and refuse to buzz in despite a confidence level of, say, 75%.

Watson, however, is functionally blind and deaf, so it cannot answer audio or video Daily Double. Those are foregone in Watson games. Ferrucci says Watson also has trouble with clues that deal strictly in the realm of human experience, such as “Look in this direction and you will see the wainscoting.” Watson answered “Wall” when the answer was, “What is down?” Watson may also get thrown at first by a category such as Whirled Capitals, which is looking for anagrams of nations’ capitals, but after the first couple of clues it will think to itself, “Aha, we’re doing anagrams now” and proceed to nail it.

The Jeopardy challenge is a fun promotion for Watson’s more serious applications in natural language processing, one of computer science’s thorniest problems. A Watson-like machine will be a powerful tool for diagnosing disease, automating customer call centers, handling technical support online and parsing vast tracts of legal documents. “It overcomes the recall blindness of finding documents you forgot you had read,” said Ferrucci. That said, his entire experience with Watson is underscored by a deep appreciation for the power of the human brain, which can do in seconds using material that can fit into a shoebox what it takes six refrigerators worth of machinery to do.

Richard Doherty, research director at the Envisioneering Group in Seaford, N.Y., got an early look at Watson and was floored. “This is way beyond the database and spreadsheet era we are now leaving behind. People want answers,” he says. He sees the potential for Watson cousins customized with analytics for cancer diagnosis, traffic pattern optimization and emergency planning devoid of the emotions that bedevil mere people. Charles King of PUND-IT in Concord, Mass., expects that “we’ll eventually see Watson-like technologies cropping up in commercial computing solutions before too long.” The Watson Jeopardy! shows will air February 14, 15 and 16. The grand prize is $1 million with second place earning $300,000 and third place earning $200,000. Rutter and Jennings will donate half of their winnings to charity. IBM will donate all of its winnings to charity. May the best circuitry win.

[Slifer]

The Rainman will own all of them..

[TheKingdom2000]

Originally Posted by Slifer

The Rainman will own all of them..

rainman would lose. he wouldn't be able to push the button fast enough.

and how will the computer push a button?
it's not really fair since a computer can fire an impulse 100000 times faster than the human nervous system can.

[Presto]

So it's a computer that can Google? I think this is stupid. Unlike Deep Blue, which had to calculate its moves, this is pretty much a search through a database. It doesn't even need to know the question before it hits the buzzer. By the time it's time to respond, it'll have its response ready. I can see that it's more than this, but putting it in a competition is just cheese.

[JordanLee]

Watson the super computer

+



=

[dangonay]

Originally Posted by mx703

rainman would lose. he wouldn't be able to push the button fast enough.

and how will the computer push a button?
it's not really fair since a computer can fire an impulse 100000 times faster than the human nervous system can.

The computer doesn't just "push a button" super fast for every question. It only "pushes" when it's "confident" it knows the answer. Simply being the first to buzz doesn't mean you win - if you get the answer wrong you lose money. And since this computer isn't right %100 of the time, it can't simply buzz in every question expecting to have the answer before time runs out.
Posted via RS Mobile

[CanadaGoose]

Cyberdyne in the making lol

Somebody better call the governator

[Nightwalker]

Originally Posted by Presto

So it's a computer that can Google? I think this is stupid. Unlike Deep Blue, which had to calculate its moves, this is pretty much a search through a database. It doesn't even need to know the question before it hits the buzzer. By the time it's time to respond, it'll have its response ready. I can see that it's more than this, but putting it in a competition is just cheese.

This is infinitely harder than chess.

[MindBomber]

So, this is basically a supped up version of Googles "I'm Feeling Lucky" button.

I don't even see how this is supposed to be entertaining, its just a database search. Deep Blue had to make a series of calculations simultaneously trying to predict it's opponents moves, while planning it's own attacks, countering its opponents attacks and adjusting its strategy. Not to mention, Deep Blue was built in a different era of computer sciences, when duke nukem was the latest thing..

[slammer111]

This is actually pretty freaking hard. It's not just a database. It's understanding sentence structure, idioms, plays on words etc. I'd agree with buddy up top. Chess is infinitely easier, since you KNOW what possible moves are there, and then you're really just determining the optimal path out of the gazillion combinations or so. You CANNOT do that with a language.

There's a reason we still can't just "talk" to computers like we do with people.