anybody else following [s]Skynet's[/s] Watson's progress?

[Elder Iptuous]

I'm pretty impressed!
for anyone that is unfamiliar with the story, Watson is an IBM computer.  after they did the Deep Blue thing and beat Kasparov at chess, they wanted a bigger challenge.  the guys that were involved were at a restaurant mulling over what would be a good challenge for a super computer to beat a human at, and a bunch of patrons crowded in to the bar.  upon investigating what the hub bub was, they found it was Ken Jennings doing his famous Jeopardy! run.  a-ha!  there's the answer....

so this machine is pretty damned impressive at processing natural language and answering questions.  er.... at questioning answers, that is.

it's a three day game, and yesterday was the second day with Watson kicking Jennings and Rutter's collective ass. 

in between rounds they have little bumpers where the IBM guys talk about the Watson software and capabilities, and how they hope it will help humanity.
Like i said, i'm pretty damned impressed.

[Jasper]

So does it actually interpret spoken word, or was somebody offstage typing everything to Watson?  My phone does passable speech to text, but nowhere near perfect.

Also does it just do jeopardy, or can it hold a conversation?

[LMNO]

How does it do with puns and wordplay?

[Jasper]

I'll bet it's GREAT with sarcasm.

[Elder Iptuous]

it does the wordplay pretty fucking impressively.
idioms, puns, etc...

i don't have a source, but i believe that it interprets the spoken word.
it has a robot finger to press the button.
on the show they show the top three answers that Watson comes up with for each question and it's percent certainty.
it's pretty neat. i suggest catching the last round tonight.

i don't know if it does conversational. i just caught it, and now want to dig deeper into it.

ETA: i don't think it does conversation, as it is developed under the IBM 'DeepQA' project to research AI question answering...

[Jasper]

I want to talk to it and make smarmy comments that are indirectly bigoted to robots.  

"Wow!  You must be pleased with your success, given your handicap!"

[Elder Iptuous]

one of the big uses they are hoping for it to go towards is medical.
they say that diagnosticians have this huge burden because, not only is there just a shit ton of information out there that they are supposed to be basing their decisions on (which isn't humanly possible to fully utilize), but new shit is being generated at such an incredible rate that it's just not possible for a doctor to keep up.  if all this information was in machine readable code, we would already have some system that could parse it all to get better diagnoses and save lives, but it's all in natural language.  so now Watson software can parse this mountain of information to help out, they are hoping. (plus, being a machine, it wouldn't be susceptible to the pitfall of diagnostic anchoring like a human doctor is, where they are reticent to change previous diagnosis based on new information that would have led them to a different conclusion had it been there in the first place.  that kills people all the time.)

[Jasper]

Sorry, I shouldn't joke.  I'm very excited about applications for this, and not especially worried that it will become sentient (or anything like it).

[Elder Iptuous]

but the jokes; they write themselves!  who can blame, huh?
and seeing as how this is a machine purpose made just for open domain question answering rather than cognizance, i don't think anyone is speculating on sentience, despite the fact that it 'learns' algorithmically and is getting better at doing what it does do.  our intelligence is cobbled together from a bunch of different pieces, not just one magic bullet.  and this could be a job well done on an important piece of a true AI, though...

it answers all these obscure questions right, and then every so often it makes some answer that is wrong, or doesn't even make sense, and when asked why it blew it, the creators honestly have no idea. as they didn't teach it the answers, they just taught it how to learn.

it's funny that they've got it programmed to speak with some Jeopardy specific idiosyncrasies that make it sound very human for something that's got a text-to-speech voice coming from a monitor with an animated icon for a physical presence.

also it's funny because when it does the daily doubles, it always picks some weird ass amount to wager.  "i'll got for twelve seventy six, Alex"  on the final round it blew the question (i can't recall what it was) after just trouncing the opponents and getting very few incorrect answers, and then when the wager was revealed, it had only bet $957 despite having a kit of over 35 grand, while the guy in second had only 5000 something...

[The Good Reverend Roger]

So, let me get this straight.  They built the world's biggest, most badass computer, and it...plays quiz shows.

OH, GODDAMMIT!

I hate people.  I fucking hate them.  Even the smart ones.  ESPECIALLY the smart ones.

[AFK]

How does it do with puns and wordplay?

Now there's a challenge.  Someone set me up and I'll do a pun-off with ole' Watson. 

[Luna]

The DJ on my morning show is crapping himself over this, and talking "Rise of the Machines."

[AFK]

but the jokes; they write themselves!  who can blame, huh?
and seeing as how this is a machine purpose made just for open domain question answering rather than cognizance, i don't think anyone is speculating on sentience, despite the fact that it 'learns' algorithmically and is getting better at doing what it does do.  our intelligence is cobbled together from a bunch of different pieces, not just one magic bullet.  and this could be a job well done on an important piece of a true AI, though...

it answers all these obscure questions right, and then every so often it makes some answer that is wrong, or doesn't even make sense, and when asked why it blew it, the creators honestly have no idea. as they didn't teach it the answers, they just taught it how to learn.

it's funny that they've got it programmed to speak with some Jeopardy specific idiosyncrasies that make it sound very human for something that's got a text-to-speech voice coming from a monitor with an animated icon for a physical presence.

also it's funny because when it does the daily doubles, it always picks some weird ass amount to wager.  "i'll got for twelve seventy six, Alex"  on the final round it blew the question (i can't recall what it was) after just trouncing the opponents and getting very few incorrect answers, and then when the wager was revealed, it had only bet $957 despite having a kit of over 35 grand, while the guy in second had only 5000 something...

I caught the last round last night.  Over the three days it did pretty much take the two meat-contestants to town.  Though, I do have to say that when it flashed the answers on the screen, some of the answers it was mulling were completely non-sensical.  And on its last Daily Double it made a tiny wager that was basically inconsequential even though it got the answer correct.  

I will finish by adding that the saddest commentary is that of the four entities on the stage, including Alex Trebek, Watson probably had the most personality and character.  

[Luna]

So, let me get this straight.  They built the world's biggest, most badass computer, and it...plays quiz shows.

OH, GODDAMMIT!

I hate people.  I fucking hate them.  Even the smart ones.  ESPECIALLY the smart ones.

Two reasons.  One, it's an interesting way to field test it.  And, two, it's three days of free advertising, not even counting the fact that people are going to be talking about the thing for months.

At least they didn't name it Hal.

[LMNO]

Here's a Slate article about how it would be more impressive to teach Watson how to play no-limit Texas Hold 'Em against multiple opponents.