Watson wins!


Agonus

 

Posted

I LOLd at Ken jennings written comment on the last final Jeopardy.

The whole thing was intriguing to me. I really hoped to hear about Watsons betting logic, and why it chose what categories it chose. The betting was in odd numbers, and if watson had a betting strategy at all. Thats what makes the daily doubles so exiting because sometimes the player will only vote a little, and sometimes theyll roll the dice and make it a true daily double.


So you mean you'll put down your rock, and I'll put down my sword; and we'll try and kill each other like civilized people?

Dubbed first knight of pep-istan by her majesty Queen Pepcat. first catmonaut to walk onna moon.

PENGUIN!!!...(^)>
...............C(...)D
.................m.m

 

Posted

Quote:
Originally Posted by Lucky666 View Post
Smartest computer ever or new overlord?
And the Technological Singularity takes another baby step towards reality...


Loth 50 Fire/Rad Controller [1392 Badges] [300 non-AE Souvenirs]
Ryver 50 Ele� Blaster [1392 Badges]
Silandra 50 Peacebringer [1138 Badges] [No Redside Badges]
--{=====> Virtue ♀

 

Posted

Is anyone really surprised that a talking search engine beat a human at a quiz show?


- CaptainFoamerang

Silverspar on Kelly Hu: A face that could melt paint off the wall *shivers*
Someone play my AE arc! "The Heart of Statesman" ID: 343405

 

Posted

I caught a part of the 2nd round of Day 1 of that tourny.

Watson was taking the gents to town big time. It just was wrong seeing it run the board.

Thank you for the time...


@Travlr (Main) / @Tymers Realm (Test)

Arc 5299: Magic, Mystery, and Mayhem Updated!! 09/15/09

 

Posted

Quote:
Originally Posted by CaptainFoamerang View Post
Is anyone really surprised that a talking search engine beat a human at a quiz show?
Quote:
Originally Posted by Tymers_Realm View Post
Watson was taking the gents to town big time. It just was wrong seeing it run the board...
Computers have been beating chess masters for years now.
I just see Watson as another step in the process of machines "evolving" into whatever they are destined to become.


Loth 50 Fire/Rad Controller [1392 Badges] [300 non-AE Souvenirs]
Ryver 50 Ele� Blaster [1392 Badges]
Silandra 50 Peacebringer [1138 Badges] [No Redside Badges]
--{=====> Virtue ♀

 

Posted

Quote:
Originally Posted by CaptainFoamerang View Post
Is anyone really surprised that a talking search engine beat a human at a quiz show?
Actually, Jeopardy is more than just a quiz show esp. with it's word play and references to pop culture. That's why Watson's success is so surprising. It took a dedicated team 4 years to teach it 'common sense' ie. water is wet, which was key in making it a success at the game.

Of course, now that they've done it we're all obsolete.


Craft your inventions in AE!!

Play "Crafter's Cafe" - Arc #487283. A 1 mission, NON-COMBAT AE arc with workable invention tables!

 

Posted

Quote:
Originally Posted by BackFire View Post
Actually, Jeopardy is more than just a quiz show esp. with it's word play and references to pop culture. That's why Watson's success is so surprising. It took a dedicated team 4 years to teach it 'common sense' ie. water is wet, which was key in making it a success at the game.

Of course, now that they've done it we're all obsolete.
I don't find it to be all that surprising, actually. Just seems like a matter of teaching the software to recognize key words.


- CaptainFoamerang

Silverspar on Kelly Hu: A face that could melt paint off the wall *shivers*
Someone play my AE arc! "The Heart of Statesman" ID: 343405

 

Posted

Exactly. I saw this and instantly thought, "Why is Google allowed to do this exactly?" I mean, it's not Google, but functionally wouldn't it be doing something very similar? I don't see the technical marvel here.



 

Posted

It was fun to watch, and I liked the bar at the bottom that showed Watson's top three answers and his confidence level in them. It was interesting to see what questions threw him off, especially in the first game's Final Jeopardy. Toronto is not a U.S. city, Watson.


 

Posted

Quote:
Originally Posted by Genesis Man View Post
Exactly. I saw this and instantly thought, "Why is Google allowed to do this exactly?" I mean, it's not Google, but functionally wouldn't it be doing something very similar? I don't see the technical marvel here.
While yes I agree somewhat about it just being a talking search engine it's also so much more. I was actually paying some attention to this project on a show called daily planet and alot of programming went into this thing just to teach it common sense.

Even so to google something pick the right answer from possibly thousands of results and do it before the top two jeopardy players ever to play the game is I gotta say impressive.


 

Posted

Just to clarify for those who didnt catch it. Watson was not connected to the internet at all for the matches. So his answers were compeltely self generated.


So you mean you'll put down your rock, and I'll put down my sword; and we'll try and kill each other like civilized people?

Dubbed first knight of pep-istan by her majesty Queen Pepcat. first catmonaut to walk onna moon.

PENGUIN!!!...(^)>
...............C(...)D
.................m.m

 

Posted

What I was curious about was how Watson handled the more strategic aspects of the game -- choosing where to go next on the board, or how much to wager in the Daily Doubles. What formula or criteria was it following?


99458: The Unbearable Being of Lightness
191775: How the Other Half Lives
My Webcomics

 

Posted

Quote:
Originally Posted by JKCarrier View Post
What I was curious about was how Watson handled the more strategic aspects of the game -- choosing where to go next on the board, or how much to wager in the Daily Doubles. What formula or criteria was it following?
I think Watson liked to choose categories that have already been started. The categories sometimes confused him at first, but he would figure it out after a couple questions. That's why Ken Jennings kept going for the harder questions first, to get them before Watson could catch on.

I have no idea how Watson chose his wagers, though. He made some weird ones.


 

Posted

I, for one, welcome our new AI overlords...


 

Posted

Quote:
Originally Posted by Vox Populi View Post
I have no idea how Watson chose his wagers, though. He made some weird ones.
Quote:
Originally Posted by Panzerwaffen View Post
I, for one, welcome our new AI overlords...
Ghosts in the machine...
By the time we fully figure out what "motivates" a machine like Watson to do what it does it'll be far too late for us.


Loth 50 Fire/Rad Controller [1392 Badges] [300 non-AE Souvenirs]
Ryver 50 Ele� Blaster [1392 Badges]
Silandra 50 Peacebringer [1138 Badges] [No Redside Badges]
--{=====> Virtue ♀

 

Posted

Quote:
Originally Posted by Vox Populi View Post
I think Watson liked to choose categories that have already been started. The categories sometimes confused him at first, but he would figure it out after a couple questions. That's why Ken Jennings kept going for the harder questions first, to get them before Watson could catch on.
...
Jennings seemed to catch on that Watson had trouble with pop culture, but there were still more than enough categories based around pure facts for Watson to mostly (Toronto?!) steamroll through.


Tales of Judgment. Also here, instead of that other place.

good luck D.B.B.

 

Posted

I'd expect a computer as sophisticated as this to win on Jeopardy. It's the sort of thing machines would excel at; that's why it's such a challenge for humans.

I'd like to see Watson win on the $100,000 Pyramid.

EDIT: Or Family Feud.


 

Posted

Can just imagine it now. AI is next judge of American Idol in Season 20 and 0-PRA is first AI to host a talk show. Then we have to worry about AI is our elections. Although, they would do a better job than most politicians assuming they have the right program. Have to get rid of that pesky DESTROY ALL HUMANS bug that seems to be present in most AIs.


The first step in being sane is to admit that you are insane.

 

Posted

Quote:
Originally Posted by starphoenix View Post
Can just imagine it now. AI is next judge of American Idol in Season 20 and 0-PRA is first AI to host a talk show. Then we have to worry about AI is our elections. Although, they would do a better job than most politicians assuming they have the right program. Have to get rid of that pesky DESTROY ALL HUMANS bug that seems to be present in most AIs.
. . . You mean Jon Kerry wasn't a robot? >.>


- CaptainFoamerang

Silverspar on Kelly Hu: A face that could melt paint off the wall *shivers*
Someone play my AE arc! "The Heart of Statesman" ID: 343405

 

Posted

The key thing about Watson isn't its database or search algorithms. It's the ability to process casual language and derive "meaning" from it. Especially when you consider the convoluted answer-and-question format that Jeopardy! uses.

Watson's betting strategy was probably just derived around an algorithm incorporating its own score, its competitors' scores and its history of correct responses.

Apply a few iterations of Moore's Law to Watson and before too long you'll have everyday-use computer interfaces that respond to natural language instructions in conversational form.


 

Posted

Quote:
Originally Posted by Panzerwaffen View Post
I, for one, welcome our new AI overlords...
Go ahead and suck up meat bag, just like the rest of us, they will decide your fate in an instant too!



------->"Sic Semper Tyrannis"<-------

 

Posted

Quote:
Originally Posted by Starjammer View Post
Apply a few iterations of Moore's Law to Watson and before too long you'll have everyday-use computer interfaces that respond to natural language instructions in conversational form.
Sweet, I'll be ready to buy a new Mustang in a few years and hopefully Ford's SYNC will be filled with this goodness.

Not that I care for distracted driving, but the Mini brand is set to introduce an option where you can tweet from your car.

Combine these technologies though and I will finally be able to realize a dream (provided the car ahead of me can do the same) of mine - the ability to tell the person riding along in the left lane 15 miles under the speed limit how I really feel in real time.

"Car please contact liscence plate ST XXX-XXXX, I have something I really want to tell them"



------->"Sic Semper Tyrannis"<-------

 

Posted

Quote:
Originally Posted by ebon3 View Post
Sweet, I'll be ready to buy a new Mustang in a few years and hopefully Ford's SYNC will be filled with this goodness.

Not that I care for distracted driving, but the Mini brand is set to introduce an option where you can tweet from your car.

Combine these technologies though and I will finally be able to realize a dream (provided the car ahead of me can do the same) of mine - the ability to tell the person riding along in the left lane 15 miles under the speed limit how I really feel in real time.

"Car please contact liscence plate ST XXX-XXXX, I have something I really want to tell them"
Really, that's not the cool possibility. Here's the cool possibility:

A Watson-like computer can solve word problems.

Think about it. Pure math is no problem for computers, provided you tell it what equation you want to solve, what the values/ranges are and so forth. Knowing how to set up the problem becomes the hardest part and the one that people most commonly screw up. Garbage in, garbage out.

With a Watson-like system, you can simply describe the problem you need solved and it will interpret your request, look up the appropriate equations and data and then do the math. Then it can provide the answer in colloquial language and elaborate the parts you may not understand.

With enough "experience" it can even anticipate the misconceptions of the person asking the question and avoid any errors inherent in the query. Or it might ask for or even offer clarification.

Next likely goal: See if Watson 2.0 (or whatever the next generation is) can "earn" an advanced degree through standard education and testing.


 

Posted

Quote:
Originally Posted by Starjammer View Post
Really, that's not the cool possibility. Here's the cool possibility:

A Watson-like computer can solve word problems.

Think about it. Pure math is no problem for computers, provided you tell it what equation you want to solve, what the values/ranges are and so forth. Knowing how to set up the problem becomes the hardest part and the one that people most commonly screw up. Garbage in, garbage out.

With a Watson-like system, you can simply describe the problem you need solved and it will interpret your request, look up the appropriate equations and data and then do the math. Then it can provide the answer in colloquial language and elaborate the parts you may not understand.

With enough "experience" it can even anticipate the misconceptions of the person asking the question and avoid any errors inherent in the query. Or it might ask for or even offer clarification.

Next likely goal: See if Watson 2.0 (or whatever the next generation is) can "earn" an advanced degree through standard education and testing.
That is pretty lofty and would be neat indeed, I think I still would settle for Watson 1.01 if it allowed me to communicate with any sort of device easily.

As a problem solving tool, Watson 2 would be awesome though.



------->"Sic Semper Tyrannis"<-------