John Henry Loses Again
Tim Moran Watson, the Jeopardy-playing IBM supercomputer we first covered back in June, has now made its debut against real people and -- spoiler alert! -- won. The machine competed against Ken Jennings (74 Jeopardy wins in a row) and Brad Rutter ($3.25 million in winnings) and beat them by $1,000. A couple of interesting details: the machine has to hit the same signaling buzzer that human Jeopardy! contestants must use before it answers; and key to Watson's winning ways is that it does not answer if it isn't sure it's right, which could give it an advantage over humans, who can lose money with a wrong guess. (More on of how Watson works and the story behind its creation can be found at our previous blog post on this subject, linked above, and on IBM's own Watson site. In brief, the 80-teraflops supercomputer's "brain" scans 200 million-plus pages of data--in less than three seconds--to arrive at its answer; it is not connected to the Internet.) As video of the contest shows, Watson isn't infallible; while it gave no wrong answers, it struggled with fill-in-the-blanks questions that required the parsing of English text. What I found most interesting about the scene was the sight of a black rectangle sporting spinning lights sitting in between two humans on the Jeopardy! set--and it really didn't look all that unusual. What does that say about it. . . and us? |
Comments (13)
Okay, so it beat Ken and Brad. A: How many games would it have to win to make back the cost of building, programming and loading data into it?
B: How many games would Ken and Brad keep coming back to play?
C: How many of those games would anybody watch? D: How many Joe Averages could be recruited to
play against Watson after Ken and Brad quit?
E: How many 'Watson vs Joe' games can Jeopardy's producers sell advertising for?
I'll bet that A >> (B + E).
Posted by Burly | January 28, 2011 10:05 AM
Actually, Watson gave a couple of "wrong" answers the first night. One in particular where Ken answered first with 1920 as a wrong response followed by Watson answering 1920, still wrong.
Posted by Miickie | February 15, 2011 8:56 AM
Eniac cost $500,000 ($6 million adjusted for inflation). Although Watson won't be sold as a component in the home version of Jeopardy, the point of it is that there will exist popular applications that will drive down the cost.
Today you can pick up a tablet PC for a few hundred dollars that is significantly faster and more powerful than the mighty Eniac of yore. What will computers be capable of 65 years from now? Your children will carry with them readily accessible copies of mankind's accumulated knowledge, and use that information in ways yet undreamed.
Posted by Paul | February 15, 2011 1:15 PM
"Burly" misses the point in several respects. The idea is not that Watson is going to be playing jeopardy. The idea was to pit a machine against humans in the context of an interesting cognitive and logical game. The amazing thing is the machine is "understanding" questions and retrieving correct answers.
Burly is correct about one thing: Don't allow yourself to be suckered into the superiority of machines. The cost of making this machine perform this one task is staggering, yet humans can do it naturally. Dozens of people worked for years to get the machine to where it can compete with some "cheap carbon units", and that should inspire people...that humans have amazing capabilities baked right in!
Too often today, machines displace people, instead of using them to augment people. Increase production, don't throw away all the jobs...otherwise, who's going to buy the product "genius"? That's what we are see in America now. A-holes shipped living wage jobs overseas, now Americans can't pay their mortgage/bills and can't buy "stuff", so companies shrink.
We are all connected. Greed is NOT good. Do what is in your best interest AND the best interest of everyone else...it is possible.
Posted by vlad | February 15, 2011 1:44 PM
What amazes me is that IBM just through some silica and germanium in a pot, heated it up, and over a period of time Watson evolved with no interference or plans
Posted by Charles E. Buchanan Jr. | February 15, 2011 2:28 PM
Charles E.,
I just love the double entendre posit of your comment.
Very cool.
I don't thinks this heralds the advent of SkyNet but who knows. It is a fairly significant achievement, but I would like to see the end of the net war we currently have with purveyors of malware and those of us who want to do good computing.
Posted by Stephen S. Wolfe | February 16, 2011 11:45 AM
Watson is no P1 (who's adolesence was written about in the 1970s). What I most object to is that Jeopardy has become nothing but a long commercial for IBM. I wonder how much IBM paid Ken and Brad to play along...
Posted by joe | February 16, 2011 1:55 PM
Charles E. Buchanan,
That's lost on quite a few people, I can assure you.
After all, in Armonk the silica and germanium grew into Watson. In the Valley, it grew into the "i".
No one could have possibly planned the biodiversity of the Galapagos. No one planned the computo-diversity of integrated circuits.
I rest my case.
Posted by Jim Cole | February 16, 2011 8:11 PM
That was such a cool answer, and it didn't take a computer genius to see what implications you were making. :)
Posted by Dean Lyles | February 17, 2011 11:26 AM
I didn't watch. But, I'm curious how much of the programming was tailored for the show; if it were set on the street could it answer questions? Does it really understand the context of language and the multitude of directions which could be inferred, or was that restricted for the show? I have been in the electronics business for years watching the growth. We have come far and yet, we are still a long way away from machines which are cost effective and useful to interact in society. It took years for the computer to beat a chess master and now this, but don't look for your home robot to show up at Best Buy anytime soon!
Posted by Mitch Hilger | February 17, 2011 11:54 AM
When will there be a www.watson.com website where we can ask questions and get the answers.
Posted by Steven Stein | February 18, 2011 4:50 PM
Dear Watson:
In my country, India, we have several commissions that have to go through thousands of papers to investigate all kinds of scams - fodder scam;Satyam scam;spectrum scam;rice export scam,kerosene scam,landgrab scams etc., etc.
Can you help reading all this data and using your analytic powers to identify the reasons behind the scams ? If yes, you shall do a great social service. If no, you are not smarter than the scamsters.
j.
Posted by roameri | February 19, 2011 4:50 AM
Well done, IBM, but I think Mitch Hilger's point above is really sound.
Watson is clearly a really neat step forward: we're on the verge of computers that can figger out whose luggage is whose, just by looking in phone books, or which Galicia is which by pairing up an atlas with a guess at the user's surname. Still the fact of Watson getting "1920" wrong after it had already been tried is a sign that there's still some work to be done.
The knowledge handling of Watson needs to be married to the strategizing of Deep Blue.
Then of course the price has to come down by a factor of 10^6~7, but there's no doubt that'll happen.
-dlj.
Posted by David Lloyd-Jones | February 21, 2011 1:35 PM