Man vs. Machine: Will Watson Win Jeopardy?

Quote: Nareed

I haven't seen the show in months :) It's on when I'm at work.

DVR is the answer :) I can't remember last time I watched live TV :)

Quote:

When I do watch I don't play along with a clicker. I just say the questions aloud if I know them or can think of them. Usuaully I read faster than Trebek, so I know the right question before I could click.

I could, probably, read it faster too, but I did not have my glasses on, so I had to listen :)
But even if I read it with my eyes, the point is that I would still know that I will know the question before I finished reading the clue, but I would not know what it was until I finished it (and even a bit later, because it takes some time to understand the meaning of what you've just read too).

I am just saying that there are two different events happening here - first I realize, that I know the question, and then I find out what it is, and I could click after the first even occurred, without necessarily waiting for the second.

Quote: boymimbo

I would just let Watson ring in and let it try.

That would be a good strategy, if you were the only one human contestant.

Quote: weaselman

I am just saying that there are two different events happening here - first I realize, that I know the question, and then I find out what it is, and I could click after the first even occurred, without necessarily waiting for the second.

I get what you are saying now, and, yes, ringing in because you have some certainty you will be able to come up with the answer is something the humans do and Watson is not doing [I'll take your word for that]

I assume Watson is also fed what value answers have been taken, so he knows what remains on the board, unless they are using optical recognition for that aspect. Sure it is SMOP (simple matter of programming) but it was neat when Watson said (I paraphrase) "Let's finish the category for 1600", indicating he knew one answer was left.
Again, just guessing, but I bet they analyzed how fast a human can react to a trigger and determined that Watson responding in 4ms would give the humans a fair chance, and at 5ms the humans would ring in first every time. There were many times where you could see the other two trying to ring in, and you can be certain they also know most all the answers.

I think I would like to see the contest changed slightly. I like the Jeopardy answer/question format, and the word associations that are required to produce the correct question. I don't like the ring in aspect of the game when it is man vs machine, as man is at a distinct disadvantage.
I say hook them all up to alpha scanners, do away with the buzzers, and let the humans think the answer. Watson gets the text file when the answer is revealed and whoever arrives at the correct answer fastest gets the points. Think of a wrong answer, and lose points.

Quote: weaselman

I could, probably, read it faster too, but I did not have my glasses on, so I had to listen :)

I'm a fast reader. I read automatically, too. I can't look at writing and not read it (I had to learn to ignore subtitles at the movies).

Quote:

But even if I read it with my eyes, the point is that I would still know that I will know the question before I finished reading the clue, but I would not know what it was until I finished it (and even a bit later, because it takes some time to understand the meaning of what you've just read too).

Sometimes you can get the wrong question by thinking you're done with the clue too soon. for example, "This 'I, Robot' author despite-" You may think "Isaac Asimov," but the clue goes on to say "having never flown was a regular contributor to this in-flight magazine." And that's an entirely different question. Ok, my phrasing may be off, but sometimes you need to identify two things, not just one.

I don't have DVR. Many of the shows I watch are repeated on weekends, when I'm free, so I'd rather save me the expense of hiring DVR services.

Quote: Nareed

Ok, my phrasing may be off, but sometimes you need to identify two things, not just one.

The recognition I was talking about was at a somewhat higher level - I did not know the answer, and just knew that whatever the question is, I will know it once I understand the question. When I said that "Harry Potter" popped in my head when I heard the beginning of the question, I did not mean I thought it was the right answer - I did not know what the answer was, but I knew it was about the HP book, and therefore, I will almost definitely know it once I hear the whole clue.

Quote:

I don't have DVR. Many of the shows I watch are repeated on weekends, when I'm free, so I'd rather save me the expense of hiring DVR services.

The main reason I use DVR is not even being able to watch things when I want to watch them, not when they are on, but being able to skip the commercials. I just can't stand them. If I have to wait till the commercial is over, I usually just tune out, and start doing something else, and then, when it is over, I miss a part of the show, and by the time I tune back in, it's another commercial :)

Quote: weaselman

The main reason I use DVR is not even being able to watch things when I want to watch them, not when they are on, but being able to skip the commercials. I just can't stand them. If I have to wait till the commercial is over, I usually just tune out, and start doing something else, and then, when it is over, I miss a part of the show, and by the time I tune back in, it's another commercial :)

I'm often on the computer while watching TV. That way I can play, browse or do something else when the commercials come up. Somtimes I'll switch channels and follow, or half-follow, two shows at once. I do this most on Football games; anythign you miss there will be replayed eventually. I'm used to it and it's cheaper.

The final tally was

Watson -- $77,147
Jennings -- $24,000
Rutter -- $21,600

Quote: DorothyGale

The final tally was

Watson -- $77,147
Jennings -- $24,000
Rutter -- $21,600

Which just goes to show that there may be something to this radical notion of using computers to retrieve information!


Oh, wait....

Seriously, as with Deep Blue before it, Watson's achievement is the achievement of a bunch of people, not a machine. As to what it says about the future of humanity, well, just that we can build machines that exceed our capabilities. We wouldn't expect a man to beat a Ferrari in a mile long race, would we?

Both Jennings and Rudder have to deal with being underdogs now too -- which is not normal for them. Maybe confidence levels are affecting their game?

The machine doesn't care how far it's up or down. It keeps at it as hard as ever. Unless it gets a blue screen of death. (does it do that)

The buzzer factor really ruined my enjoyment of the show. I'm not even sure what they could do to fix it. I am convinced that this did not prove that computers are better than humans at Jeopardy.

Quote: rxwine

Both Jennings and Rudder have to deal with being underdogs now too -- which is not normal for them. Maybe confidence levels are affecting their game?

The machine doesn't care how far it's up or down. It keeps at it as hard as ever. Unless it gets a blue screen of death. (does it do that)

It is running Linux, not micro-buggy, so I doubt it suffers from the BSOD. Probably doesn't have to load weekly security updates to close holes either.
Not saying, just saying.....................

Quote: FinsRule

The buzzer factor really ruined my enjoyment of the show. I'm not even sure what they could do to fix it. I am convinced that this did not prove that computers are better than humans at Jeopardy.

I absolutely agree. I actually thought it was using voice recognition technology at first. So now I am not so impressed, what with someone typing in the question [did that guy screw over Watson by being too slow or whatever? must have happened a time or two] And the advantage of a machine buzzing in, that was unacceptable IMO. I'd like to see what would have happened had Watson been actually required to give the humans enough time to buzz in. Maybe the rule should have been that if they all buzzed in within the first half second, the winner of the buzz-in could have been selected at random.

PS:

Quote: Nareed

Sometimes you can get the wrong question by thinking you're done with the clue too soon. for example, "This 'I, Robot' author despite-" You may think "Isaac Asimov," but the clue goes on to say "having never flown was a regular contributor to this in-flight magazine." And that's an entirely different question.

the guy writing the text may have been good enough to reform the question in a manner that helped Watson. In the example given, this unseen person might have typed " I, robot author disliked flying" but then realized he needed to quickly do "what in-flight magazine" to nudge Watson to the central question. Using google ourselves, certainly a similar thing, we quickly find certain words should be left out and key words matter more. Did this go on? We don't know.

Quote: odiousgambit

actually thought it was using voice recognition technology at first.

Yeah, I was just thinking how much cooler that would be. THAT would be really impressive.

Quote:

So now I am not so impressed, what with someone typing in the question [did that guy screw over Watson by being too slow or whatever? must have happened a time or two]

I don't think they were typing real time. That would be real low-tech :) I would think they just fed the clues to Watson electronically as Trebek finished reading.

Quote:

And the advantage of a machine buzzing in, that was unacceptable IMO. I'd like to see what would have happened had Watson been actually required to give the humans enough time to buzz in. Maybe the rule should have been that if they all buzzed in within the first half second, the winner of the buzz-in could have been selected at random.

I for one was more interested in just comparing what questions Watson knew compared to what the humans knew, the speed of the response doesn't really matter (as long as it is less then a minute :)). I think, it was a nice touch to show Watson's answers when he did not buzz in. I wish something similar could have been done for the humans too.

Watson received the entire clue as written and read, via a text file, the moment it is revealed. No one is typing it in at that moment, they are sent electronically, pre-typed. This gives another advantage to Watson: He has read the entire clue in milliseconds, and is already working on an answer, likely before Alex can even begin saying it aloud.

Don't cry too hard for Brad and Ken; They each earned $100,000 or more just for showing up, plus helped out two charities they believed in, for the same amounts.

Quote: weaselman

I for one was more interested in just comparing what questions Watson knew compared to what the humans knew, the speed of the response doesn't really matter (as long as it is less then a minute :)). I think, it was a nice touch to show Watson's answers when he did not buzz in. I wish something similar could have been done for the humans too.

I think something like this could easily be accomplished. Start by figuring out the average time required for a human to read the answer. Then have the humans and Watson supply the questions to every clue (like a standardized test). Watson will get the text answer after the delay that equals the average time required to read the clue, so human and Watson are both starting the search for the correct question about the same time. As soon as you supply the question, you move to the next answer of your choice. So everyone is working independently of each other.
The contestant picks the answer order, and can choose to not answer a question with no penalty. Wrong questions lose the value of that answer. Daily Doubles exist as in the regular game, except in this case, all three contestants would have a daily double. Each contestant would go at their own speed. First contestant done is awarded 3000 bonus points, second one done gets 1000 bonus points. Final Jeopardy is played same as today, so wagering and confidence in the last category still comes into play.

Using a split screen where you could see the answer each contestant was working on, their score total, and the question they supplied would help out, but I think it would make for crappy TV. However, since you would see each contestant's score climbing in real time, you would get an idea how close they were to each other. And it should feel like a race as well. What you wouldn't really know is what point values each of them was selecting.

Quote: weaselman

I don't think they were typing real time. That would be real low-tech :) I would think they just fed the clues to Watson electronically as Trebek finished reading.

Quote: Dween

Watson received the entire clue as written and read, via a text file, the moment it is revealed. No one is typing it in at that moment, they are sent electronically, pre-typed.

Makes sense. But if so why was Trebek said to have read it slower than usual? That made me think it was to give an operator time to start typing it in. If you two are right, Watson needed more time to think! [or it is not true that Trebek spoke slower]

Quote: Dween

Don't cry too hard for Brad and Ken; They each earned $100,000 or more just for showing up, plus helped out two charities they believed in, for the same amounts.

Sure, and that's great. But the issue to me is still whether machine beat man fair and square. Maybe I am too affected by an upbringing in which the John Henry song and folklore were very much emphasized. Certainly I am also not quite satisfied with the way IBM took down Kasparov in chess, IMO not being transparent enough about what really went on, and also failing to accept the challenge of truly proving the computer could compete with Grandmasters in a more normal process. Further exploration of Deep Blue vs this guy or that guy without tweaking between rounds by humans would have helped answer some questions. And just generally being forced to compete in tournaments, allowing the contestants to study the computer's games, was definitely in order. But IBM took their prize and then dismantled Deep Blue!

I think what Deep Blue and Watson have established is that computers can beat elite humans at Chess and Jeopardy! under controlled circumstances with human assistance. IBM, IMO, is guilty of a bit of dishonesty no doubt due to quite human hubris in the way they have gone about these things.

Quote: odiousgambit

the guy writing the text may have been good enough to reform the question in a manner that helped Watson. In the example given, this unseen person might have typed " I, robot author disliked flying" but then realized he needed to quickly do "what in-flight magazine" to nudge Watson to the central question. Using google ourselves, certainly a similar thing, we quickly find certain words should be left out and key words matter more. Did this go on? We don't know.

Oh, a quick-thinking typist could even write "in-flight magazine Asimov regularly wrote for." He's have to be a fast reader, too.

For that matter the guy typing the clues could also provide the question. That would be a waste of the machine's capability, but for some clues it might be worth it. Finally, I'm sure IBM could set up practice sessions for Watson to play against people. The human contestants could not.

Quote: odiousgambit

Makes sense. But if so why was Trebek said to have read it slower than usual? That made me think it was to give an operator time to start typing it in. If you two are right, Watson needed more time to think! [or it is not true that Trebek spoke slower]

I don't watch Jeopardy much, so I don't have a baseline. But I think, it's probably just the 'placebo effect' - people felt he was talking slower because of this image of somebody typing behind the scenes in their head. It is also possible that he really did spoke slower than usual, because subconsciously he was trying to compensate for Watson's "handicap"

Quote:

I think what Deep Blue and Watson have established is that computers can beat elite humans at Chess and Jeopardy! under controlled circumstances with human assistance. IBM, IMO, is guilty of a bit of dishonesty no doubt due to quite human hubris in the way they have gone about these things.

I think, it's a waste to build a supercomputer to beat human at a game of Jeopardy (chess is a bit different because of its game-theory implications). It wasn't a real competition, just a fun way to test Watson's abilities in natrual language comprehension (which, I would say is impressive).

It was a pretty stupid contest. The machine had the huge advantage of cognition speeds that were thousands of times faster than its opponents. It was like two humans running a race against a sports car.

The only surprising thing was that the game was at least relatively close. To me, that shows the deficiencies of computer-based psedo-cognition. It should have totally blown its opponents away, answering every single question before its human opponents even had time to think about it.

The speed of talking could have been easily solved without speech recognition... have the text of the questions written down in a file, but buffer the feed to Watson word by word at the same speed as Trebek.

Guys, guys, it was announced at the beginning of the first episode that Watson is deaf and was fed pre-written text files worded exactly same as the clues the humans were given, at the same time Trebek read them (whether at start of clue or end was not stated). Then they showed us Watson's button-pushing device. I detected no show-down in Trebek's reading.

Quote: mkl654321

It was a pretty stupid contest. The machine had the huge advantage of cognition speeds that were thousands of times faster than its opponents. It was like two humans running a race against a sports car.

The only surprising thing was that the game was at least relatively close. To me, that shows the deficiencies of computer-based psedo-cognition. It should have totally blown its opponents away, answering every single question before its human opponents even had time to think about it.

The point of the contest was to test the quality of that "pseudo-cognition" of the machine, compared to humans. The speed of clicking the buzzer was never in question.

Found this

Quote:

Yes, we could have handled a lot more information. We could have put more memory in each server, but once we got the answers to three seconds, we didn't need to go further."

Pearson explained that reaching the three-second answer threshold was just a matter of simple mathematics.

The original algorithm ran as a single threaded process on a single core processor took two hours to scan memory and produce an answer. So the IBM technologists just divided two hours by 2,880 CPUs, which produced the ability to answer questions in three-seconds.

here

Quote: weaselman

The point of the contest was to test the quality of that "pseudo-cognition" of the machine, compared to humans. The speed of clicking the buzzer was never in question.

That's not what I meant. The machine should have been able to digest the answer, search its database, and generate the right answer before the humans even managed to process the information. There actually wouldn't have been much difference (a few microseconds) between the speed of the computer and the speed of the humans in pressing the buzzer (or whatever equivalent action the computer did to indicate that it had figured out the answer). But the computer should have crushed the humans based on cognition speed alone. I'm very surprised that the humans were even able to answer SOME questions faster than the computer.

My local broadcast had problems with the audio signal during the broadcast. I suspect they were "time shifting" the length of the broadcast to squeeze in an extra :30 spot. If your broadcast was not compressing the program time, it could account for the perception that Trebeck was reading clues at a slower pace.

From my observation of the three top answers devised by Watson (color coded for our convenience), the computer would delay (yellow), or not buzz in (red) if its confidence in a response was below a certain threshold. This allowed Jennings and Rutter to slip in. It was interesting that the top answer displayed by Watson was usually incorrect when this happened. Perhaps Jennings and Rutter should have made a pact to not answer until Watson took a shot at every question. There was no jumping ahead when it was confident, and the humans were "saving" it from the penalty of incorrect answers by jumping in before it would answer incorrectly.

The thing that really irks me is that the computer was fed a signal when the buzzer was "live". The humans had to anticipate when that event occured; a distinct disadvantage.

Did Watson ever answer correctly after one of the other players missed? I know that it got some wrong, including repeating a previous incorrect answer.

Quote: mkl654321

But the computer should have crushed the humans based on cognition speed alone. I'm very surprised that the humans were even able to answer SOME questions faster than the computer.

That's not true, or, at least, not obvious. Computers are obviously very good at sequentially evaluating every possibility, but that is not the way human cognition works. It took ~3000 processors to allow Watson to arrive at answers in time comparable to what humans can do naturally. This is hardly trivial. When presented with a clue, we don't immediately start mining our internal "database", evaluating every grain of knowledge against it. That's (roughly) what Watson does. If we tried to do something like that, of course, we'd stand no chance to a computer, but our minds work in a mysterious and different way, being somehow able to get directly to the required information, bypassing all the irrelevant noise.

Quote: Ayecarumba

From my observation of the three top answers devised by Watson (color coded for our convenience), the computer would delay (yellow), or not buzz in (red) if its confidence in a response was below a certain threshold. This allowed Jennings and Rutter to slip in.
It was interesting that the top answer displayed by Watson was usually incorrect when this happened.

Hm. My impression is the opposite - it seems to me there were more cases when Watson had the right answer, but one of the other contestants buzzed first, then those where Watson's answer was incorrect. It also seemed to happen quite often (among those cases where a human got to answer), that Watson's answer was not only correct, but "green".

Quote:

Did Watson ever answer correctly after one of the other players missed?

Not, as far as I can remember.

Quote: weaselman

That's not true, or, at least, not obvious. Computers are obviously very good at sequentially evaluating every possibility, but that is not the way human cognition works. It took ~3000 processors to allow Watson to arrive at answers in time comparable to what humans can do naturally. This is hardly trivial. When presented with a clue, we don't immediately start mining our internal "database", evaluating every grain of knowledge against it. That's (roughly) what Watson does. If we tried to do something like that, of course, we'd stand no chance to a computer, but our minds work in a mysterious and different way, being somehow able to get directly to the required information, bypassing all the irrelevant noise.

Right, but transistors and neurons are nothing alike. The POWER7 core has 1.2B transistors, so Watson has 3.456T transistors. The human brain has somewhere around 20B neurons (in the neocortex). The difference is that each transistor has 3 connections, while each neuron has at least 1000. Moreover, the connections between transistors are fixed and etched into silicon, but neurons can rewire themselves. To wit: if you shoot your computer in the CPU, it's toast. On the other hand, Gabrielle Giffords is talking again.

That singularity that Kurzweil is predicting? You'll know it's happening when you shoot your computer in the CPU and it rewires itself. And then constructs a robotic arm, grabs the gun, and shoots you back. (insert ominous music here)

Quote: Ayecarumba

My local broadcast had problems with the audio signal during the broadcast. I suspect they were "time shifting" the length of the broadcast to squeeze in an extra :30 spot. If your broadcast was not compressing the program time, it could account for the perception that Trebeck was reading clues at a slower pace.

The pace seemed back to normal for the second and third days. But the cadence of the first day seemed slightly slower to me, and I'm a regular viewer.

I'd be more impressed with Watson if he were the size of a human brain instead of as big as he was, though. Or the size of one human, even.

Quote: mkl654321

It was a pretty stupid contest. The machine had the huge advantage of cognition speeds that were thousands of times faster than its opponents. It was like two humans running a race against a sports car.

Have you seen the Tug of War challenge between 8 athletes and a supercharged V12 Mercedes with over 600 horsepower?







The athletes won.

If you want to see how fast a man can run against car, just put him in front of the car.

Quote: P90

Have you seen the Tug of War challenge between 8 athletes and a supercharged V12 Mercedes with over 600 horsepower?

The athletes won.

Sure, but that would be a test of torque, not speed. The pulling power of a sports car at zero MPH wouldn't be all that impressive--it isn't geared to produce high torque at low speeds (the way, say, a diesel truck/SUV is).

Sports cars still take quite a bit of low-speed torque to push the car from zero to thirty in a second, particularly large-displacement engines. Such a car could probably out-tow many light trucks (SUV). The problem was insufficient grip: car's engine started spinning the wheels, and just couldn't pull against the team.

What this goes to showing is that man v machine competitions often have counterintuitive results.

Quote: P90

Have you seen the Tug of War challenge between 8 athletes and a supercharged V12 Mercedes with over 600 horsepower?

No. But I saw the Mythbusters try to separate two phone books with interleaved pages, held only by friction. They required a two tanks to pull them apart.

MathExtremist's comments were bang on.

Watson kicked ass, in Jeopardy. It had hundreds of practice rounds with real humans in order to hone its programming and to fix the relevance of words. For example, there must be programming in Watson to devalue the Category somewhat though it's very meaningful. For example, there would be no way that our brains would come up with Toronto as the answer for a U.S. Cities problem. That's because the IBM programmers had to program watson to think outside of the category somewhat when doing a search. The same thing was true with the Decades category: the computer's best answer was sometime a year, not a decade.

And finally, Watson was not connected to the internet meaning that it had millions and millions of pages of encyclopedic knowledge that the contestants did not. In the end, if Watson could understand the question, and had the page of data in its system, it could come up with the answer. And given the minute delay of the buzzer, if it knew the answer, it would ring in first, with only a couple of exceptions.

The human brain is a fantastic device though. Watson and its 2,800 or so processors could only be programmed to be a viable contestant on Jeopardy. It can't love, do dishes, or even play blackjack. Incredible.

Quote: MathExtremist

Right, but transistors and neurons are nothing alike. The POWER7 core has 1.2B transistors, so Watson has 3.456T transistors. The human brain has somewhere around 20B neurons (in the neocortex). The difference is that each transistor has 3 connections, while each neuron has at least 1000. Moreover, the connections between transistors are fixed and etched into silicon, but neurons can rewire themselves.

This is a little oversimplified. The most obvious pitfall is that human neurons serve not only as parts of the CPU, but also memory. Watson had 15 terabytes (I think) of memory in addition to those 3.5T transistors you mentioned. The humans have to share their neurons between "CPU" and "storage". They also have to control several thousand of "peripheral devices" - libms, eyes, bowels, facial muscles - etc (yes, most of those functions are in neocortex too), while Watson has it's entire "brain" dedicated to the one question.

Quote:

To wit: if you shoot your computer in the CPU, it's toast. On the other hand, Gabrielle Giffords is talking again.

Well, not Watson. If you only shoot one of its CPUs, it'll be barely noticeable (except for some "memory loss").

Quote: boymimbo

For example, there must be programming in Watson to devalue the Category somewhat though it's very meaningful. For example, there would be no way that our brains would come up with Toronto as the answer for a U.S. Cities problem. That's because the IBM programmers had to program watson to think outside of the category somewhat when doing a search.

Not quite. The reason is that the real life "taxonomy" is infinite (well, technically, it isn't, but for all practical purposes anyway) - you can't categorize everything ahead of time in every possible way. Toronto is a Canadian City, and North American City, an English Speaking City, a city, younger than 400 years, a city with more than a million people (is it?), a city, starting with "T", etc, etc.
The Jeopardy's categories cannot therefor be "attached" to every answer for Watson's quick selection. It has to consider a category just like another part of clue, with an added complication that it can't use the semantics to deduce the relative significance/weight of it. If the clue was something like "this US city has two airports ...", it would be definitively established that the answer is a US city, but the category is sorta "affixed" to the clue on the side, making it a lot harder to relate it to the answer.

That's a part of what I was saying earlier. Human brain works very differently from the computer. Even though they both arrive to the same solution to the same problem in this case, the paths they use are very very different, and that's why you can't predict the winner by simply comparing the "computing power". Watson has a lot more of it, but humans have ways to solve these problems more rationally, without requiring as much computation, and that makes it possible for them to compete with Watson (almost) head-to-head.

Quote:

And finally, Watson was not connected to the internet meaning that it had millions and millions of pages of encyclopedic knowledge that the contestants did not.

Well ... I think, the consensus here is that they did know the answer to (almost) every question, just were not as fast to ring it in.

Quote:

In the end, if Watson could understand the question, and had the page of data in its system, it could come up with the answer.

Well, this is actually the biggest challenge. Once it "understands" the question (which is roughly as trivial as just breaking it into words), how to pick the one correct answer out of the trillions of possibilities? Humans have (not entirely understood) ways of quickly navigating their memory chains directly to the right place, the computer's pretty much only choice is look at every answer (well, not exactly, but close), and evaluate it, and that takes a lot of time, even if you have 3000 CPUs, that's why on many occasions, Watson had the right answer, according to the display, but one of the humans still got to answer, because he buzzed in first.

Some years ago, when computers were not as powerful as they are now, a popular branch of engineering was analogue computing. That, I think, is a lot more close to the way humans think, compared to the "normal", digital computers. To give you an example, suppose you have a huge network of pipelines, with known pressures in some of the nodes, input and output, and you want to calculate the amount of gas (or liquid) flowing through each pipe per second, and the pressures in every node. If the network is large enough, even a modern PC (not Watson) could take, probably, hours, to perform such calculation, and a couple of decades ago, it was more like months or years, even on a fairly big machine. So, what people did instead was to assemble an electrical network of the same configuration, with the resistance of edges representing the radius of corresponding pipes, and then connect it to the voltage according to the specified conditions at edges. Now, in seconds you could now the current through any given edge by simply connecting the ammeter to its ends, or to find out the pressure at any node, you just measure the voltage between that node, and any known one.

Watson was programmed extensively to weed out the non-relevant terms of the question. In hundreds of practice rounds with real humans, the programming was altered to find the relevant terms based on the categories and to weed the search further. It wasn't the knowledge base that was adjusted, it was the programming. Watson was only presented to the Jeopardy producers when its results were competitive with the mainstream of winners in Jeopardy based on it answering real Jeopardy questions.

Even still, how Watson came up with Toronto as an answer is confounding as it couldn't possibly be the right answer given the clue.

In the second round, Rutter and Jennings only buzzed on on 14 of the 57 clues (3 daily doubles). In 13 of those cases, Watson's probability was below 90 percent. Watson won the buzz-in on all of the other clues and answered correctly on every one except for 1. There was no way therefore that Rutter and Jennings was going to beat Watson to the buzzer if Watson had the answer before the en of the reading of the question.

Watson's extensive 15 TB memory contained documents, not answers. What's amazing about Watson is that it was programmed with rules to not only interpret the Jeopardy questions, but to sift and filter through the millions of documents to come up with the right question. The fact that it did so on more than three quarters of the clues is amazing.

I still say it was profoundly unfair when all 3 contestants wanted to be the first to buzz in. With the humans, this often was because they realized they had a high percentage chance of getting the answer correct even before they actually had the answer. If they could just win that little part of the contest that, largely due to Jennings, we now know is critical. With Watson, he had to have the green light first, so to speak. So this was one area where humans had an edge, but the asinine decision to allow a machine which got a leg up [only one notified] to then dominate a contest involving mechanical quickness took this away.

Quote: boymimbo

Watson was programmed extensively to weed out the non-relevant terms of the question. In hundreds of practice rounds with real humans, the programming was altered to find the relevant terms based on the categories and to weed the search further. It wasn't the knowledge base that was adjusted, it was the programming.

Exactly, what I am saying is there is only so much you can do with programming. Interpreting Jeopardy's categories is more of an art than science. Granted, "US cities" was pretty straightforward, but a program needs to be generic enough to cover all possibilities. Such as "Legal "E"s" for example. It is not as straightforward at all as "weed out non-relevant terms of the question" sounds.

Quote:

In the second round, Rutter and Jennings only buzzed on on 14 of the 57 clues (3 daily doubles). In 13 of those cases, Watson's probability was below 90 percent.

Really? Did you read it somewhere? My impression was that there were more cases than one, when Watson's answer was green, and it still did not buzz. But I am combining both games together. It could be that just the second one, taken in isolation, was different.

Quote: weaselman

This is a little oversimplified. The most obvious pitfall is that human neurons serve not only as parts of the CPU, but also memory. Watson had 15 terabytes (I think) of memory in addition to those 3.5T transistors you mentioned. The humans have to share their neurons between "CPU" and "storage".

That's all true, but I think the real difference is that the brain is not binary. My focus is software and not hardware, but I'm pretty sure that transistors in CPUs are either on or off (digital) based on the current at the base. On the other hand, neocortical synapses have many distinguishable firing levels, and it's known that changing the intensity or timing of the input signals can change the output signals -- or even the connections between neurons themselves. This is called neuroplasticity.

Normal transistors do not exhibit plasticity. However, last year, researchers developed a transistor that *does* act like a neuron:

http://www.sciencedaily.com/releases/2010/01/100125122101.htm

Skynet hasn't become self-aware yet, but it will. 1997 was just a few decades premature.

I don't even know why there's a discussion over this. You've got the two geniuses vs. Watson. Watson feeds them their lunch. Who developed Watson, man or machine?