Originally posted by inimalist
36000?jesusWTFingchrist
For reference, though, the record for a single game is 77000 (set in 2010). http://www.denverpost.com/nationworld/ci_16076411
They have one more game to play. If Watson can set a record, that would be really amazing, though the way it plays it could probably keep playing forever without losing a game.
Originally posted by dadudemon
"Watson" could have safely wagered $25,199 and still would not have lost if Rutler bet it all and won. I wonder what the logic was behind him wagering just a little bit? (I did not watch it, just going off of what Sym said.)
It has algorithms gauge its own confidence and, given its comically specific Daily Double selections, must have an algorithm for optimizing how much money it bets. So Watson must have had very low confidence on its knowledge of cities, possibly adjusted for knowing it was Final Jeopardy and the question would be especially hard.
Interestingly both Rutter and Jennings got the answer right.
Originally posted by Symmetric Chaos
It has algorithms gauge its own confidence and, given its comically specific Daily Double selections, must have an algorithm for optimizing how much money it bets. So Watson must have had very low confidence on its knowledge of cities, possibly adjusted for knowing it was Final Jeopardy and the question would be especially hard.Interestingly both Rutter and Jennings got the answer right.
What was the question?
Additionally, they need to adjust the algorithm to be more "encompassing" so it can take into consideration the other's votes. But, that makes sense that they would adjust the algorithm to only bet on a magnitude of confidence, maximizing "earnings."
Originally posted by Symmetric Chaos
For reference, though, the record for a single game is 77000 (set in 2010). http://www.denverpost.com/nationworld/ci_16076411They have one more game to play. If Watson can set a record, that would be really amazing, though the way it plays it could probably keep playing forever without losing a game.
I was talking with a friend in my lab about this
so amazing, like seriously amazing. It is weird though, he was saying that questions that use sarcasm or puns tended to trick the computer, because these things are much harder to understand than "simple" language. I'm sure they could skew the questions such that it had a much harder time with them (or, by chance alone, we'd expect that on any given week).
still, so amazing 🙂
Originally posted by dadudemon
What was the question?Additionally, they need to adjust the algorithm to be more "encompassing" so it can take into consideration the other's votes. But, that makes sense that they would adjust the algorithm to only bet on a magnitude of confidence, maximizing "earnings."
if im not mistaken, the Watson program is more about being able to interact with humans in a linguistic manner
the type of probabilistic max-min stuff that it would take to do that is trivial for programming, isn't it? Like, an algorhythm that said "X has 2000 dollars, do not bet so much as to go below 4001 dollars" shouldn't be hard, in theory? (no idea about this stuff personally, lol)
Originally posted by Symmetric ChaosSpoiler:
Both humans said "Chicago". Watson said "What is Toronto?????"
Lester B Pearson just decided to haunt that machine
Originally posted by Robtard
LoL, fail. A Canadian war hero? WTF was it thinking.
The answer is:
The nationality of the man who shot down the Red Barron
or
The nationality of the man who single handedly rescued Holland from Nazis
or
The nationality of the only Western military leader who wanted to end the genocide in Rwanda
😉
also: soooooo mad!
Originally posted by inimalistI believe that's exactly what this is about. It's [how fast can humans respond] vs [how fast can Watson understand]. Otherwise, this really would be a slaughter.
if im not mistaken, the Watson program is more about being able to interact with humans in a linguistic manner
Originally posted by inimalist
if im not mistaken, the Watson program is more about being able to interact with humans in a linguistic mannerthe type of probabilistic max-min stuff that it would take to do that is trivial for programming, isn't it? Like, an algorhythm that said "X has 2000 dollars, do not bet so much as to go below 4001 dollars" shouldn't be hard, in theory? (no idea about this stuff personally, lol)
Yes. It would be what some call a "subroutine."
It's a miniature program within a program.
To do the "money" bet thing.
It would be similar to this:
Retrieve "current earnings."
Wager "current earnings"*"probability value" such that x*"probability value is < or = "current earnings."
Then, you have a separate routine (massive) for the probability value calculator (that's the one that shows 3 answers with a weight for each). And a separate program (actually, multiple) for the current earnings.
That math portion is very easy...the hard part is the building of the probability program (which would be massive amounts of code because it is the "meat" of the entire program" and the "current earnings" portion because that would be comprised of multiple smaller programs.)
Here's a cheat Watson is capable of that a lot of people missed. It has figured out the Daily Double patterns that Jeopardy! uses based on all the games it has memorized. Human players have tried the same tactic but Watson managed to hit a pretty incredibly number during its first game.
http://live.washingtonpost.com/jeopardy-ken-jennings.html#question-17
Originally posted by Symmetric ChaosI bet he is connected to a network during the game.
Here's a cheat Watson is capable of that a lot of people missed. It has figured out the Daily Double patterns that Jeopardy! uses based on all the games it has memorized. Human players have tried the same tactic but Watson managed to hit a pretty incredibly number during its first game.http://live.washingtonpost.com/jeopardy-ken-jennings.html#question-17
Originally posted by chomperx9
I bet he is connected to a network during the game.
It was mentioned, in very explicit terms, that he is not connected to the internet.
However, he is definitely "networked." He's connected to his own network. How else is information going to get to and from each "smaller" server?
Originally posted by dadudemonwell we dont know how fast he is capable of scanning the question and searching for the answer online, if he was connected to the net. He can probably google something 100 times faster than we can.
It was mentioned, in very explicit terms, that he is not connected to the internet.However, he is definitely "networked." He's connected to his own network. How else is information going to get to and from each "smaller" server?
Originally posted by chomperx9
well we dont know how fast he is capable of scanning the question and searching for the answer online, if he was connected to the net. He can probably google something 100 times faster than we can.
He's not connected to the internet at all. He would be much slower if he had to do that.
Originally posted by chomperx9
well we dont know how fast he is capable of scanning the question and searching for the answer online, if he was connected to the net. He can probably google something 100 times faster than we can.
thats the thing, he is built using search algorithms far superior to anything google uses, and much more specific to the understanding of language.
google might be able to match words or sounds, but it can't interpret the meaning of a sentence, which is what Watson needs to do. If you typed a jeopardy question verbatim into a google search bar, I highly doubt you would get answers superior to human contestants most times.
so, I watched a clip of the final jeopardy question...
Watson guessed Toronto to a question that was specifically about US cities. This means that Watson either cannot form related connections between concepts like "this city is in this nation", or, is unable to filter items by particular qualities...
I know it is designed more as a language understanding machine, but that seems like a huge flaw... I'd be interested in the process it went through to guess "Toronto"... and why it couldn't restrict its search to only cities located in the US. All I mean, is given the category of "US cities", Toronto has a 100% chance of being the wrong answer.