Welcoming our Robot Overlords (?)
Feb. 18th, 2011 02:45 am![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
capri0mniSo. Over the last four-something years, IBM has been working on "The next grand challenge" in A.I.: getting a computer to understand natural human language, and parse out what a question is actually asking for, rather than just pinging onto specific keywords.
Back in the 1990s, they worked on computer-planning-ahead programming by coming up with Deep Blue, and challenging Kasparov to a chess match. This time, they decided to put "Watson" up against two humans on a TV quiz show, Jeopardy! (Because the questions are written with humans and human humor in mind, instead of being coded for the machine).
Now, comes the not-really-a-spoiler spoiler. Watson won. It helps to win a trivia game if you've spent the last three years doing nothing but study every word of Wikipedia, IMdB.com and Google's library of digitized books, and have a literal one-track mind (Watson was disconnected from the Internet at the time of gameplay, though).
One interesting thing they did, though, is flash a graphic showing Watson's top three choices for a correct answer, along with how confident he was about said choice, so you could get a glimpse of how it was thinking. And although "Watson" got nearly every question correct (he did not get all of them, though), seeing his second and third choices left me with the notion that he may be on the way toward intelligence, but he's not thinking like a human -- yet.
Take this example: Under the category "Hedgehog-podge" (for $1600):
(quote) Hedgehogs are covered in quills or spines, which are hollow hairs made stiff by this protein (unquote).
Watson buzzed in and replied: "What is keratin?" And he got his $1600. The graphic at the bottom of the screen showed this:
keratin -- 99% certain
Porcupine -- 36% certain
fur -- 8% certain.
I'd be more convinced that he actually can parse human languge if his second and third choices were also proteins -- or at least, biological compounds.
------------------------------------
Androids may not count electric sheep, but they get excited about their seventh anniversaries.
At certain points in the game, the players are asked to risk a portion of what they've won so far. Humans generally pick multiples of 10 ... or 100. Watson had no such emotional attatchment to round numbers, and to us, the amounts he wagered seemed silly. But it almost seemed to me as if he did have a fondness for some numbers more than others -- notably, the number seven (7). Among the wagers Watson made in the second games were:
$2,227 (he got it wrong -- giving the answer "Dorothy Parker" instead of The Elements of Style [picked up on "1959," "brevity" and "wit" but missed the clue "New York Times Book Review"])
$367 (he got it right)
And when the money he'd won over the course of the two games was added together? It came to: $77,147
...I may be anthropomorphising prematurely, but that kind of calculation, coming up with that final result -- I think he likes seven.
D'awww....
Back in the 1990s, they worked on computer-planning-ahead programming by coming up with Deep Blue, and challenging Kasparov to a chess match. This time, they decided to put "Watson" up against two humans on a TV quiz show, Jeopardy! (Because the questions are written with humans and human humor in mind, instead of being coded for the machine).
Now, comes the not-really-a-spoiler spoiler. Watson won. It helps to win a trivia game if you've spent the last three years doing nothing but study every word of Wikipedia, IMdB.com and Google's library of digitized books, and have a literal one-track mind (Watson was disconnected from the Internet at the time of gameplay, though).
One interesting thing they did, though, is flash a graphic showing Watson's top three choices for a correct answer, along with how confident he was about said choice, so you could get a glimpse of how it was thinking. And although "Watson" got nearly every question correct (he did not get all of them, though), seeing his second and third choices left me with the notion that he may be on the way toward intelligence, but he's not thinking like a human -- yet.
Take this example: Under the category "Hedgehog-podge" (for $1600):
(quote) Hedgehogs are covered in quills or spines, which are hollow hairs made stiff by this protein (unquote).
Watson buzzed in and replied: "What is keratin?" And he got his $1600. The graphic at the bottom of the screen showed this:
keratin -- 99% certain
Porcupine -- 36% certain
fur -- 8% certain.
I'd be more convinced that he actually can parse human languge if his second and third choices were also proteins -- or at least, biological compounds.
------------------------------------
Androids may not count electric sheep, but they get excited about their seventh anniversaries.
At certain points in the game, the players are asked to risk a portion of what they've won so far. Humans generally pick multiples of 10 ... or 100. Watson had no such emotional attatchment to round numbers, and to us, the amounts he wagered seemed silly. But it almost seemed to me as if he did have a fondness for some numbers more than others -- notably, the number seven (7). Among the wagers Watson made in the second games were:
$2,227 (he got it wrong -- giving the answer "Dorothy Parker" instead of The Elements of Style [picked up on "1959," "brevity" and "wit" but missed the clue "New York Times Book Review"])
$367 (he got it right)
And when the money he'd won over the course of the two games was added together? It came to: $77,147
...I may be anthropomorphising prematurely, but that kind of calculation, coming up with that final result -- I think he likes seven.
D'awww....
no subject
Date: 2011-02-19 02:55 am (UTC)no subject
Date: 2011-02-19 03:10 am (UTC)And then, the thought occured to me that maybe, in binary code -- how computers count -- seven is a round number. Ten is only round to us because we (typically) have ten fingers. But fingers mean nothing to Watson.