Is this IBM's successor to Deep Blue?

[sje]

Is this IBM's successor to Deep Blue?

http://news.cnet.com/8301-17852_3-10227842-71.html

[Nick C]

A comparison of Watson and Deep Blue.

http://www.research.ibm.com/deepqa/deepblue.shtml

AFAIK there are no ASICs in the Watson hardware, it's a standard Blue Gene/P, and I can imagine that the eval will be really more important than ever at guiding the search for Jeopardy

[Dann Corbit]

Playing Jeopardy is just regurgitation of facts.
The IBM computer can just load the encyclopedia, index it, and barf up the answers.

I literally can't imagine any task less impressive for computers to accomplish.

Anyone who thinks winning at Jeopardy is a marvelous feat for computers does not understand what computers are good at.


I suppose that there may be some AI if the machine needs to recognize human speach, but that vocabulary will be limited.

All in all, very unimpressive to me.

[Terry McCracken]

Playing Jeopardy is just regurgitation of facts.
The IBM computer can just load the encyclopedia, index it, and barf up the answers.

I literally can't imagine any task less impressive for computers to accomplish.

Anyone who thinks winning at Jeopardy is a marvelous feat for computers does not understand what computers are good at.


I suppose that there may be some AI if the machine needs to recognize human speach, but that vocabulary will be limited.

All in all, very unimpressive to me.

Dan I think IBM is trying to do more than what you claim.

It has to understand the questions as you would before it could begin to look it up and in cases it doesn't have the answer it will have to try and solve it by inference etc. The semantics in such a case alone with human language has remained out of the domain of computers thus far.

I agree parts of it will be very easy if the question is properly understood and it's already in it's internal database but there's much more to it then that.

The idea of course is one day a computer could work with people directly with normal human language and rely on it's ability to comprehend what it hears and sees and the compute the best responses similar to you thinking and giving the best response.

The computer will have to do a lot more than just checking a few indices.

[Dann Corbit]

Playing Jeopardy is just regurgitation of facts.
The IBM computer can just load the encyclopedia, index it, and barf up the answers.

I literally can't imagine any task less impressive for computers to accomplish.

Anyone who thinks winning at Jeopardy is a marvelous feat for computers does not understand what computers are good at.


I suppose that there may be some AI if the machine needs to recognize human speach, but that vocabulary will be limited.

All in all, very unimpressive to me.

Dan I think IBM is trying to do more than what you claim.

It has to understand the questions as you would before it could begin to look it up and in cases it doesn't have the answer it will have to try and solve it by inference etc. The semantics in such a case alone with human language has remained out of the domain of computers thus far.

I agree parts of it will be very easy if the question is properly understood and it's already in it's internal database but there's much more to it then that.

The idea of course is one day a computer could work with people directly with normal human language and rely on it's ability to comprehend what it hears and sees and the compute the best responses similar to you thinking and giving the best response.

The computer will have to do a lot more than just checking a few indices.

It will have to understand the grammar of a short fact statement in the English language.

Since IBM already has written speach recognition software, and the grammar needed for this tiny subset of the English language is small, I do not see it as any great feat.

Oviously, with zero reaction time it can also always win the buzzer war.

[Edmund]

Reminds me very much on the Chatterbox challenge: http://www.chatterboxchallenge.com/

Some of the programs are already very advanced in talking nonsense.

I think its not far off these projects to built a program to form a question to an answer.

[Terry McCracken]

Playing Jeopardy is just regurgitation of facts.
The IBM computer can just load the encyclopedia, index it, and barf up the answers.

I literally can't imagine any task less impressive for computers to accomplish.

Anyone who thinks winning at Jeopardy is a marvelous feat for computers does not understand what computers are good at.


I suppose that there may be some AI if the machine needs to recognize human speach, but that vocabulary will be limited.

All in all, very unimpressive to me.

Dan I think IBM is trying to do more than what you claim.

It has to understand the questions as you would before it could begin to look it up and in cases it doesn't have the answer it will have to try and solve it by inference etc. The semantics in such a case alone with human language has remained out of the domain of computers thus far.

I agree parts of it will be very easy if the question is properly understood and it's already in it's internal database but there's much more to it then that.

The idea of course is one day a computer could work with people directly with normal human language and rely on it's ability to comprehend what it hears and sees and the compute the best responses similar to you thinking and giving the best response.

The computer will have to do a lot more than just checking a few indices.

It will have to understand the grammar of a short fact statement in the English language.

Since IBM already has written speach recognition software, and the grammar needed for this tiny subset of the English language is small, I do not see it as any great feat.

Oviously, with zero reaction time it can also always win the buzzer war.

Dann without knowing all the details I think this is an unfair reaction.

This is more complex then speech recognition software that you have to teach to understand your voice.

The whole idea behind this project is to eventually bring it up to a level unheard of in the field of computers and programming, that a machine may be able to verbally interact with a human with understanding.

That is a might lofty goal! This step may seem trivial to you but as I said there is more to it then you give credit. The goal is enormous but you have to start with simpler steps before you can produce a HAL 9000 which in the end is the goal which will take decades.

[Will Singleton]

DC,

Watch any typical game of Jeopardy and you'll see it would be impossible for a computer to compete. What they will do is dumb-down the game, eliminating many classes of questions, including those that depend on video, audio, rhyming, word play, questions that require mixing responses, etc etc.

With a simplified game, it should be very interesting to gauge the computer's performance. Doubtless they'll run tests to ensure the game will be competitive, so we can judge the performance level just by assessing the difficulty of the questions.

In any case, such an effort by IBM is extremely interesting whether it yields valid results or not. Judging by your reaction, I doubt you're a Jeopardy aficionado. Or perhaps you're not very good at it.

Will

[Dann Corbit]

DC,

Watch any typical game of Jeopardy and you'll see it would be impossible for a computer to compete. What they will do is dumb-down the game, eliminating many classes of questions, including those that depend on video, audio, rhyming, word play, questions that require mixing responses, etc etc.

With a simplified game, it should be very interesting to gauge the computer's performance. Doubtless they'll run tests to ensure the game will be competitive, so we can judge the performance level just by assessing the difficulty of the questions.

In any case, such an effort by IBM is extremely interesting whether it yields valid results or not. Judging by your reaction, I doubt you're a Jeopardy aficionado. Or perhaps you're not very good at it.

Will

I watch it all the time, and usually know the right answers.

There is a board game version. When my wife was in the hospital once, I was in a waiting room for 8 hours or so, playing the board game version of Jeopardy with 6-7 others. I won by a colossal score.

Are you unaware that the answers posted on the board are DIRECT QUOTES from the Encyclopedia Brittanica? How hard do you think it is to recognize a direct quote?

I guess that they could leave in all the video questions etc, and the computer will still win.

I work in data retrieval. I could write a program that will recognize the given sentence and return the answer (from the Encyclopedia database) in less than one millisecond.

[Dann Corbit]

DC,

Watch any typical game of Jeopardy and you'll see it would be impossible for a computer to compete. What they will do is dumb-down the game, eliminating many classes of questions, including those that depend on video, audio, rhyming, word play, questions that require mixing responses, etc etc.

With a simplified game, it should be very interesting to gauge the computer's performance. Doubtless they'll run tests to ensure the game will be competitive, so we can judge the performance level just by assessing the difficulty of the questions.

In any case, such an effort by IBM is extremely interesting whether it yields valid results or not. Judging by your reaction, I doubt you're a Jeopardy aficionado. Or perhaps you're not very good at it.

Will

Imagine a game where 50% of the questions are not the simple written kind.

If the computer cannot get any correct, then the computer will gather at worst half of the points.
If either competitior fails to gather the full 50% of all the non-written type points then they cannot win in Final Jeopardy because the computer will have 50% of the full points from the board. Being able to bet the entire sum in the last round, it cannot be beaten and at best only tied.

And (as you know) most of the questions are the written-out, simple kind.

The computers will also be able to rhyme a trillion times faster than you can.
http://www.rhymer.com/

For auditory clues referring to music, the data can be passed though an fft and from there converted to MIDI (there are even free programs you can get which will do this). Then, the note pattern can be compared against a database (again, faster than any human can possibly hope to compete).

The video questions would be a challenge for the computer. I guess at most there are three of these per game (6 total -- I never saw one in Final Jeopardy and I am sure it would not be allowed in this particular contest since it is the only way possible for the machine to lose).