Elementary, my dear Watson

[Tunnelcat]

This should be interesting. A computer pitted against 2 human champions on the game show Jeopardy. It's on 2/14/2011 (tonight) through 2/16/2011 (Wednesday).

http://washingtontechnology.com/article ... -uses.aspx

[Isaac]

Ok! that's really cool!

So it can analyze a large database of information, figure out what it means, and you can ask questions about it? Yeah, this could bring costs way down in the medical industry.

[Richard Cranium]

Ya, and computers can land an airplane... in the Hudson River.

[Isaac]

Ya, and computers can land an airplane... in the Hudson River.

Yeah, computers and the Romans... What did they ever do for us? /mp

[Tunnelcat]

I don't ever watch Jeopardy, but I'll be doing it this time. This outta be good!

[Richard Cranium]

Romans... What did they ever do for us?

Public Education, Roads, The Aqueduct

[Tunnelcat]

Yeah, computers and the Romans... What did they ever do for us? /mp

Gladiators and bloodsport games.

[Isaac]

Romans... What did they ever do for us?

Public Education, Roads, The Aqueduct

Well, when you edit out my sarcasm it's hard to be sarcastic.

[Flatlander]

"All right, but apart from the sanitation, medicine, education, wine, public order, irrigation, roads, the fresh water system and public health, what have the Romans ever done for us?"

[Isaac]

"All right, but apart from the sanitation, medicine, education, wine, public order, irrigation, roads, the fresh water system and public health, what have the Romans ever done for us?"

You got the reference! A+

[Gekko71]

"All right, but apart from the sanitation, medicine, education, wine, public order, irrigation, roads, the fresh water system and public health, what have the Romans ever done for us?"

...brought Peace?

[snoopy]

I watched last night.

It was interesting. For a bit, at the beginning, I thought that Watson was going to completely clean up.

Later, though, he did some funky/funny things. There was one case where he got the right answer, but didn't phase it correctly: The answer had to do with what particular physical feature a certain Olympic athlete had- the question was that he was missing a leg- Watson just responded "what is leg" and was wrong.

In a second case, Watson responded with the same incorrect question that had just been provided by another one of the contestants.

Also, it seemed that Watson was quite quick on the trigger- there were few times that the contestants buzzed in before him.

I'll watch the next two days to see how the two matches end. I'd bet that IBM and ABC will consider it a successful venture, because it seems to be garnering attention.

[Tunnelcat]

It seems there might be a built in problem for Watson that gives it a slight disadvantage. If one of the human contestants gets a wrong answer and Watson is called second, it has no way of knowing what wrong answer was said beforehand by the human. The example I give is that the first answer given by the human was wrong in one instance, so when Watson was given the second opportunity to give it's answer, it gave the exact same answer because it was at the top of it's confidence list. It had no opportunity to change it's answer to the second best based on new input because it never heard that the first one was wrong. That's the opportunity that second or third responders get, because each successive time a wrong answer is said, that's one more chance someone gets to come up with the correct one owing to the process of elimination.

[Spidey]

According to the NOVA story, they were going fix that little problem.

[Flatlander]

"All right, but apart from the sanitation, medicine, education, wine, public order, irrigation, roads, the fresh water system and public health, what have the Romans ever done for us?"

...brought Peace?

"Oh, peace - shut up!"



Back on topic, I did watch Jeopardy last night (first time in a very long time) - it was interesting. Wonder how it would do on a Turing test?

[Burlyman]

Okay, this is freaky. I thought to myself "this should be interesting" when I clicked to see what this thread is about. -.-

This should be interesting. A computer pitted against 2 human champions on the game show Jeopardy. It's on 2/14/2011 (tonight) through 2/16/2011 (Wednesday).

http://washingtontechnology.com/article ... -uses.aspx

I don't see how this is going to be a competition. How could a computer fail at looking up the answer in a database?

[Isaac]

I don't see how this is going to be a competition. How could a computer fail at looking up the answer in a database?

By not being able to understand the question.

[snoopy]

I don't see how this is going to be a competition. How could a computer fail at looking up the answer in a database?

By not being able to understand the question.

And, it's database is in the form of about a million books- so it has to intelligently interpret it's own database, too.

[Isaac]

Trying to find what language the AI is in...
edit: I bet watson could tell me..

[snoopy]

It seems there might be a built in problem for Watson that gives it a slight disadvantage. If one of the human contestants gets a wrong answer and Watson is called second, it has no way of knowing what wrong answer was said beforehand by the human. The example I give is that the first answer given by the human was wrong in one instance, so when Watson was given the second opportunity to give it's answer, it gave the exact same answer because it was at the top of it's confidence list. It had no opportunity to change it's answer to the second best based on new input because it never heard that the first one was wrong. That's the opportunity that second or third responders get, because each successive time a wrong answer is said, that's one more chance someone gets to come up with the correct one owing to the process of elimination.

It also seems to be a very easy disadvantage to solve- give the avatar a microphone and speech recognition software, so it can recognize the responses given by the other contestants, and give them a zero probability.

[Richard Cranium]

It's Big Blue all over again.

[Nightshade]

Meh. Nothing to see here really.

Basically typing in search terms for a database query. Watson is merely a huge database in a fast computer that can parse English phrases.

You might as well go up against Google.

[Isaac]

Meh. Nothing to see here really.

Basically typing in search terms for a database query. Watson is merely a huge database in a fast computer that can parse English phrases.

You might as well go up against Google.

Clearly you're the expert.

[Sirius]

Parsing English phrases is actually the difficult part. Data mining isn't trivial itself, but it's a lot easier to do to some degree of reliability - though from the responses I've read about, it seems that some questions were still able to stump it. These were usually the sort of questions that you couldn't just Google and get the answer either.

[Nightshade]

from the responses I've read about, it seems that some questions were still able to stump it. These were usually the sort of questions that you couldn't just Google and get the answer either.

Exactly. These are the questions that would "stump" a computer database search.

[Tunnelcat]

Watson stumbled a bit on the third day. It seems that a computer has trouble when it has to combine several language or idea concepts together to formulate an answer, so on that count, humans win. Fuzzy logic is not quite ready for prime time, yet. But when it comes to encyclopedic knowledge and data mining, the computer wins.

The two human contestants seemed to have trouble beating the computer to the button press, another computer advantage against human reaction times. I noticed Jennings getting a little flustered because he couldn't push his button fast enough to beat Watson to the punch.

As entertaining as all this was, it was a one shot deal. Who's going to watch a computer that can win every time? Once it's been done, boring.

[snoopy]

Watson stumbled a bit on the third day. It seems that a computer has trouble when it has to combine several language or idea concepts together to formulate an answer, so on that count, humans win. Fuzzy logic is not quite ready for prime time, yet. But when it comes to encyclopedic knowledge and data mining, the computer wins.

The two human contestants seemed to have trouble beating the computer to the button press, another computer advantage against human reaction times. I noticed Jennings getting a little flustered because he couldn't push his button fast enough to beat Watson to the punch.

As entertaining as all this was, it was a one shot deal. Who's going to watch a computer that can win every time? Once it's been done, boring.

I was noticing that, too. It would be interesting if you forced Watson to "read" the clue by viewing the same image that the contestants do. I don't think it'd make a difference, though. You could also feed the clue to Watson a word at a time, as Trebek reads them, but that would give the contestants an unfair advantage of being able to read ahead.

Maybe the fairest way to do it would be to force both the contestants and Watson to listen to Trebek as he reads the clue, and only make it visible after he's finished reading it.

I really do think that how well or poorly Watson would do is closely tied to his ability to click in before or after the other contestants.

[Tunnelcat]

Trying to keep his thumb poised for the strike, Jennings looked like he was literally getting a cramp in his thumb from holding the button so hard! He kept grimacing too, like he already knew the answer before Watson.

[Thenior]

I was noticing that, too. It would be interesting if you forced Watson to "read" the clue by viewing the same image that the contestants do. I don't think it'd make a difference, though. You could also feed the clue to Watson a word at a time, as Trebek reads them, but that would give the contestants an unfair advantage of being able to read ahead.

Maybe the fairest way to do it would be to force both the contestants and Watson to listen to Trebek as he reads the clue, and only make it visible after he's finished reading it.

I really do think that how well or poorly Watson would do is closely tied to his ability to click in before or after the other contestants.

You presented two possible options:

1) Read the clue from the image
2) Listen to the clue as it is read

Did it already have to do one of these?

[SirWinner]

I missed the 3 Jeopardy shows with Watson on it.

Did get to see a special where they overviewed what a LOT of people and trainees at IBM did getting Watson ready for the show.

Over 4 years of taking a bank of computers and training them for ONLY doing Jeopardy... Just think what they could have done had they used it to solve more real world issues!

Talk about a big expense of computer time, people time, etc used in getting Watson ready for Jeopardy!

Wheeeee!

[Burlyman]

Hey, man, people waste more time and money in other things that don't matter, like space exploration, for example. ^_~

[Lothar]

Over 4 years of taking a bank of computers and training them for ONLY doing Jeopardy... Just think what they could have done had they used it to solve more real world issues!

Just think of all the people who waste time doing things like "math" for its own sake, instead of using it to solve real-world problems!

It turns out, it's very important to do research even when applications aren't obviously on the horizon. Training a computer to do Jeopardy might not have any immediate real-world use, but the process -- learning how to teach computers to learn, for example -- is quite likely to be useful in the future.

[snoopy]

I think it's a guise for the newest super spy computer. Now it will even be able to understand our puns.

The next step is to implant a robot who is linked the the supercomputer behind the curtain as president of the U.S. Beware of any candidates with a monotone voice.

[Tunnelcat]

If they could just get past the point of only number and data crunching and evolve to the point where human language nuances and meanings could be understood in their proper contexts within a computer mind, then we might have something. Right now, we are at the "Spock" stage of logic with computers, fast, cold and efficient. When we finally reach the "Bones" stage of logic, or in other words emotional logic, we will have either created something useful, functional and impressive, or a monster.

[snoopy]

I was noticing that, too. It would be interesting if you forced Watson to "read" the clue by viewing the same image that the contestants do. I don't think it'd make a difference, though. You could also feed the clue to Watson a word at a time, as Trebek reads them, but that would give the contestants an unfair advantage of being able to read ahead.

Maybe the fairest way to do it would be to force both the contestants and Watson to listen to Trebek as he reads the clue, and only make it visible after he's finished reading it.

I really do think that how well or poorly Watson would do is closely tied to his ability to click in before or after the other contestants.

You presented two possible options:

1) Read the clue from the image
2) Listen to the clue as it is read

Did it already have to do one of these?

No. The clues were "fed to him in a plain text format"

[SirWinner]

I wasn't knocking the long term uses of the learning that people did getting the low tech Artificial Intelligence added to allow the Computer to Play Jeopardy.

They had a technical competition many years ago where the goal was to write a software program for a "mouse" to navigate a maze. Someone had a major flaw in their code... the mouse amazingly ran the maze faster... then when people realized the loophole, they too made a dumber mouse program... moral of the story: the dumbest mouse won the competition!

Wheeeeeee!