|
Termyn8or -> RE: Watson followup. (2/17/2011 12:50:57 AM)
|
FR
The last question was US cities. This should be easy for a google like database. Yet the wager was $948 IIRC.
The actual "answer is" was a city that had it's largest airport named for a war hero and something named for a battle or battle field, something like that. I think the clue said "major" US city. The answer was Chicago. Well "What is Chicago".
Major US city - pop over one million
War hero - easy to correlate
Second part of clue, after vetting the possible responses which conform to part one.
First of all, the category itself shouldn't have shaken the poor thing. This is pretty much what databases are for. Therefore it should wagered the maximum possible amount while assuring the win.
The fact that Watson actually missed the question is, therefore, a bit befuddling to me. It's a perfect question for a computer. Hard fact, very little misinterpretation likely.
It's possible that the engineers, while building the thing, had the language database in place. Without the knowledge database though, there would be not much to say of course. Perhaps the were trying to make inroads into true artificial intelligence, by teaching Watson verbally, the way humans learn quite a bit of their knowledge, except for some people here who need a video and three affidavits to prove how many pieces of toilet paper I used to wipe my ass the last time I took a dump.
That would explain such a deficiency I think, because if it were a matter of a USB port, the data would be there.
Now the final wager is interesting. When you're in a poker game for example, you never bet your pocket. In fact you don't even bet your hand. You got three choices, no bet, the "normal" bet and the maximum bet. You don't bet $26.29 based on what is in your hand because it is telling to the other players if they know what they're doing. That dynamic does not exist here. If the Jeopardy players could see the others' wagers, that would be different.
So the only dynamics are the amount of money and the odds of success. The second place contender had about five grand, and Watson had about twenty. It could've easily wagered ten grand and be assured of the win. Not as puzzling was it's wager on a daily double. I think it was $1,248. I can see how a computer could take all the answers given by the opponents, all the money left on the board and mathematically figure out that amount. This would be beyond many humans' capabilities of course. Yes it could be done by many, scan the board, see which categories have how much money left and come up with a best/worst case wager. I can see that and someone possibly coming up with $1,200 or something like that. But $1,248 ?
However the wager of $948 in final jeopardy throws me off here. Why not $1,000 ? Does $12 matter that much ? Or, is it even fathomable that the engineers had a pool going to see how much their new "buddy" would win ? Then one of them gets in there and writes a bit of code. Guaranteed of a victory almost, a huge one, Watson could've been programmed to actually win more than the target amount, and then purposely lose the correct amount in the final round to make the exact figure become the result. That would be one hell of an office pool, especially with such "cheating" allowed. If I were managing that department, I would've allowed it, and in the end made sure that everyone knew that we operate as a team.
Nobody is in business to lose money. Maybe it was part of the sales pitch. I could be wrong, but I think that even if they gave the open source code for all the programming in Watson it would take a team of software engineers years to figure it out. By then they'll have sold these things to schools everywhere, or something. They didn't build it to keep it.
Just some thoughts. Trade ya.
T^T
|
|
|
|
