IBM BlueGene to Compete on Jeopardy

jadebox

Illuminator
Joined
Sep 26, 2004
Messages
3,450
Location
USA
NY Times: "Computer Program to Take On Jeopardy!"

http://www.nytimes.com/2009/04/27/technology/27jeopardy.html?_r=2&hp

Quote:

"Under the rules of the match that the company has negotiated with the 'Jeopardy!' producers, the computer will not have to emulate all human qualities. It will receive questions as electronic text."


Apparently, the following is the program they are going to use:

10 INPUT A$
20 PRINT A$
30 GOTO 10

-- Roger
 
When I fist read this I was unimpressed - we all know that jeopardy would be no challenge at all to even an idiot when armed with an internet connection and a search engine.

The more I thought about it, though, the more intriguing it sounded. The breakthroughs in data storage, categorization and retrieval could be very interesting.
The nice thing about "Jeopardy" is that you really have no idea what the questions might be.

There are far more possibilities to take account of than there are in chess. Watson isn't merely going to have to think ahead. He's going to have to interpret and understand. Which, as we know in our own dealings with humans, is a terribly tricky affair.

And without an internet connection, too.

I think this challenge is even more daunting than chess (remember Deep Blue?). To win at Jeopardy it will have to master a broad range of subject matter, provide accurate answers quickly, and have the ability to analyze clues containing irony, riddles and other subtleties.

I'm intrigued.. and it may be the first time in a decade I watch Jeopardy.



What do you think - great marketing move for IBM to showcase brilliant software design, or a fiasco where IBM gets its butt handed to it by a mere mortal like Ken Jennings?
 
Are they going to have Blue Gene "ring in" first, and then search for the correct response (like many champions seem to do), or is Blue Gene going to search for and find the correct response before ringing in?
 
Are they going to have Blue Gene "ring in" first, and then search for the correct response (like many champions seem to do), or is Blue Gene going to search for and find the correct response before ringing in?

Good question. I bet they eliminate or at least modify the buzz-in protocol. Otherwise, If the programmers are confident, they can set up the machine to buzz in for every question. I doubt a human could compete with the response time. Then, it would simply be down to the IBM computer being able to answer enough of the questions in the time allotted. It'd be boring to watch.

But, if the programmers aren't confident of being able to steamroll the game, they'll have to employ some interesting game theory. The computer will have to be able to choose categories it likes, avoid those it doesn't, steer opponents away from their strengths, etc...
 
Do you think this particular computer will be more intelligent actually than humans?

BTW: Do you really think making A.I.'s smarter than humans is a good idea? I personally don't.
 
According to the NYT article, the questions will be supplied to the software electronically. Given that advantage, even I could beat Ken Jennings!

Seriously, though. I suspect the producers of the show are going to be asked to avoid some kinds of categories to give the computer any chance of competing.

-- Roger
 
According to the NYT article, the questions will be supplied to the software electronically. Given that advantage, even I could beat Ken Jennings!

You mean, like if you had a tap right into your brain a la The matrix? Yeah, perhaps. But, you'd still actually need to have the info handy to retrieve and spit out in a cogent manner. That's the hard part, I think.
 
Are they going to have Blue Gene "ring in" first, and then search for the correct response (like many champions seem to do), or is Blue Gene going to search for and find the correct response before ringing in?

Looks like it will look for the answer first:
IBM Guy from video said:
It's going to have to very rapidly compute a confidence that it's correct so that it buzzes in in time and can get the questions it believes it knows.
Remember that wrong answers or no answer loses points so getting in first on every question doesn't seem like a viable strategy unless they are incredibly confident.



Do you think this particular computer will be more intelligent actually than humans?

BTW: Do you really think making A.I.'s smarter than humans is a good idea? I personally don't.
I don't think this system is really very intelligent at all. It's just a special purpose data retrieval system. It won't really understand English for example. It's just going to be looking for key words and dates. It's probably been fed a database of old questions and answers from old Jeopardy shows.
 
I wonder if that Watson program they mentioned is backended by Cyc or OpenCyc.
 
Do you think this particular computer will be more intelligent actually than humans?
Depending on what you mean by intelligence... of course not.
But it may be better at this particular task.

BTW: Do you really think making A.I.'s smarter than humans is a good idea? I personally don't.

I do. I think having computers that are better than us at performing particular tasks is very very useful.
And think that's true whether the task the computer is performing is designing better transportation infrastructure or writing novels.
And so long as we use our intelligence when programming them, I don't see any danger.
 
The computer will definitely not forget to put the answer in the form of a question.
 
You mean, like if you had a tap right into your brain a la The matrix? Yeah, perhaps. But, you'd still actually need to have the info handy to retrieve and spit out in a cogent manner. That's the hard part, I think.

On Jeopardy, the questions are the correct responses. The host gives the answer to which the contestant must reply with the question. If they give the questions to the computer, as the article said they would, that would be cheating.

-- Roger
 
Last edited:
By getting the question in electronic form, the researchers can ignore problems of speech recognition, and focus on problems of natural language interpretation, which is clearly their current goal.

This will be very interesting to see. Natural language interpretation is a huge milestone in AI development.
 
On Jeopardy, the questions are the correct responses. The host gives the answer to which the contestant must reply with the question. If they give the questions to the computer, as the article said they would, that would be cheating.

-- Roger

:) I completely missed that. As I said, its been a decade since I watched.
 
Bump for update nearly two years later: news outlets are reporting that a Jeopardy show featuring I.B.M.'s project (now named "Watson") versus Jeopardy record-holders Ken Jennings and Brad Rutter will air February 14-16.

eweek article

IBM's Watson PR page

Respectfully,
Myriad
 
Bump for update nearly two years later: news outlets are reporting that a Jeopardy show featuring I.B.M.'s project (now named "Watson") versus Jeopardy record-holders Ken Jennings and Brad Rutter will air February 14-16.

eweek article

IBM's Watson PR page

Respectfully,
Myriad

Does this mean the actual game has already taken place? If so, the fact that it extends over 3 days means the computer did well.
 

Back
Top Bottom