PDA

View Full Version : First chess, then crosswords, now Jeopardy! may succumb to AI


tdvance
27-April-2009, 02:07 AM
http://www.nytimes.com/2009/04/27/technology/27jeopardy.html?_r=1


IBM plans to unveil AI software that can play Jeopardy against humans (and win, presumably).

hhEb09'1
27-April-2009, 02:35 AM
I've heard (from contestants) that Jeopardy! is all about the button, so as long as the machine has a relative running the board, I'd say the contest was rigged. :)

Ronald Brak
27-April-2009, 05:34 AM
I guess Jepordy would be a tough one to program for as I presume the computer would have to understand spoken questions, but it's not an insumountable challenge to overcome as reasonably effective voice recognition already exists.

Jay200MPH
27-April-2009, 11:15 AM
I watched a bit of Jeopardy the other day. It'd been a few years...

Alex Trebek has shaved his moustache.

The whole show was ruined.

- J

hhEb09'1
27-April-2009, 01:20 PM
I guess Jepordy would be a tough one to program for as I presume the computer would have to understand spoken questions, but it's not an insumountable challenge to overcome as reasonably effective voice recognition already exists.Couldn't it read the board as well?

Moose
27-April-2009, 02:14 PM
I'd make a John Henry/Skynet joke about now, but I don't want to drop an unintended spoiler in this perfectly innocent thread.

Ronald Brak
27-April-2009, 03:49 PM
Couldn't it read the board as well?

Jeopardy has a board? I used to play Jeopardy in Japan. The rules may have been slightly different.

Swift
27-April-2009, 04:02 PM
Shouldn't all the posts in this thread be in the form of a question?

Ronald Brak
27-April-2009, 04:23 PM
What are you getting at?

Moose
27-April-2009, 04:24 PM
Am I going to wind up having to merge this thread with the F&G Questions Only thread?

Fazor
27-April-2009, 04:27 PM
Am I going to wind up having to merge this thread with the F&G Questions Only thread?

What is, BAUT moderation.

I'll take "Fazor's Follys" for $500, Alex.

jfribrg
27-April-2009, 06:07 PM
Why is it that whenever there is a "Bodies of Water" category, one of the responses is always "Lake Baikal"?

novaderrik
27-April-2009, 06:18 PM
i'll be impressed when they make a computer that can beat a human at the Newlywed Game.
of course, if we get to that point, we have truly reached a really, really weird place as a society.

Swift
27-April-2009, 06:50 PM
i'll be impressed when they make a computer that can beat a human at the Newlywed Game.
of course, if we get to that point, we have truly reached a really, really weird place as a society.
I think that would only be possible with some major breakthroughs in Artificial Stupidity (AS).

tdvance
27-April-2009, 09:39 PM
I guess Jepordy would be a tough one to program for as I presume the computer would have to understand spoken questions, but it's not an insumountable challenge to overcome as reasonably effective voice recognition already exists.

Actually, IBM got a concession from the Jeopardy people to allow the computer to get an ASCII version of the questions. Then, it competes with the contestants to come up with an answer and then "click". It uses a voice synthesizer to speak the answer..I mean, question, whatever, and if correct, will choose the next topic/dollar amount.

I wonder how it would do on the "stupid answers" category....

Ronald Brak
28-April-2009, 04:19 AM
Actually, IBM got a concession from the Jeopardy people to allow the computer to get an ASCII version of the questions. Then, it competes with the contestants to come up with an answer and then "click". It uses a voice synthesizer to speak the answer..I mean, question, whatever, and if correct, will choose the next topic/dollar amount.

I wonder how it would do on the "stupid answers" category....

Well, I suppose it would only be fair to give a deaf contestant written questions, but deafness is easily cured in computers by plugging in a microphone. This is more like translating questions into a language the contestant understands. It is unusual that the computer contest will answer in English, but some people with brain damage are able to speak a language without being able to understand it.

Personally I would have preferred it if Jeopardy hadn't made that concession. Then IBM might have put a bit more effort into developing more effective voice recognition.

HenrikOlsen
28-April-2009, 01:30 PM
Well, I suppose it would only be fair to give a deaf contestant written questions,. . .
But the split second timing is king on Jeopardy, so how soon would you show it?

The whole reason why the show works as entertainment is because the viewers get the written answer before the contestants hear it so they have more time to get the question and can therefore feel smart while it's such a short time that most don't notice that they have an advantage.

If you have a group of deaf people getting the answers in writing you'd have to show them with a noticeable delay which would spoil the smartness illusion for the viewers.

tofu
28-April-2009, 04:00 PM
the viewers get the written answer before the contestants hear it

I believe the question appears on the contestant's monitor. If you watch carefully, you'll often see them moving their heads, bringing them up, when the camera cuts back to the studio view after showing you the question. The contestants have just finished looking down and reading the question themselves.

I've heard (from contestants) that Jeopardy! is all about the button

Another thing that you often see is contestants holding the clicker up and very obviously pressing the button repeatedly, as if to say, "this thing must be broken because I'm clicking it!" What they don't realize is that one of the other players read the question more quickly than they did, and pressed the button a second or more earlier. Trebek continues reading the question, even though someone has buzzed in.

Jeopardy is not about the button. It's about speed-reading.

Ronald Brak
28-April-2009, 04:01 PM
But the split second timing is king on Jeopardy, so how soon would you show it?

I guess the simplest way would be for a screen to reveal the words at the same time the host speaks the question.

EDIT: Or as Tofu has explained, it appears people can already read the questions.

tofu
28-April-2009, 04:07 PM
Actually, wikipedia says I'm wrong (http://en.wikipedia.org/wiki/Jeopardy#Ringing_in) about being able to buzz in early. Weird. I still think the question appears on their monitor though.

Donnie B.
28-April-2009, 04:11 PM
I was under the impression that the contestants' buttons were locked out until the host finished reading the answer. Also, once the button is pressed, an immediate re-press is ignored for some short interval. Does anyone know if these are true?

tdvance
28-April-2009, 08:49 PM
I believe the question appears on the contestant's monitor. If you watch carefully, you'll often see them moving their heads, bringing them up, when the camera cuts back to the studio view after showing you the question. The contestants have just finished looking down and reading the question themselves.



Another thing that you often see is contestants holding the clicker up and very obviously pressing the button repeatedly, as if to say, "this thing must be broken because I'm clicking it!" What they don't realize is that one of the other players read the question more quickly than they did, and pressed the button a second or more earlier. Trebek continues reading the question, even though someone has buzzed in.

Jeopardy is not about the button. It's about speed-reading.

Unless it changed recently, if you click before Alex finishes the question, it doesn't count--not activated till the end of the question. On the other hand, clicking causes a short delay before another click is accepted, to prevent someone from just clicking repeatedly as the question is being asked. So, timing the click is important.

tdvance
28-April-2009, 08:51 PM
I see others said the same thing--

the delay before taking a re-press, Alex himself said that was the case on one episode where a contestant was clicking like mad then griping that the clicker is broken.

Chuck
29-April-2009, 12:08 AM
Jeopardy is a good test of natural language processing but I think a computer Diplomacy (http://en.wikipedia.org/wiki/Diplomacy_(game)) player would be a more interesting test of artificial intelligence. The machine would have to negotiate with the other players, not just compete against them.

Ivan Viehoff
30-April-2009, 11:02 AM
Who says computers can do crosswords? All crosswords, or only a few special ones?

geonuc
30-April-2009, 12:15 PM
Are crossword puzzles difficult for computers?

Ronald Brak
30-April-2009, 12:53 PM
Are crossword puzzles difficult for computers?

Yes. At least at the beginning. After they have filled in a few words their ability to wiz through the dictionary at high speed gives them an advantage But understanding the clues is very difficult. For example, if I look at the first clue on an online crossword, 5 letters - Survivalist's stash. Hey! That would be a cache! Wow! That's like the first time I've gotten a crossword clue ever!

Now where was I? Anyway, understanding the clues is the hard part.

geonuc
30-April-2009, 01:19 PM
Yeah, I can see that, now. Especially with crossword puzzles that have really obscure clues.

Ivan Viehoff
30-April-2009, 02:05 PM
Here are 1 across and 1 down (same first letter) from the crossword I shall be trying to do on the train later today. As I start to write this, I have no idea of the answer to these clues.

1A Maybe the elder Bush ****ed about end of war (5)
1D Jetsam ovaries? Shows contribution of little boiler (7)

I think if computers could solve clues like these, they would have demonstrated the level of understanding necessary to make good translations from one language to another.

These are not just definitions like Ronald's example, they are like little games. There are some general rules, and common structures, but it cannot be totally relied upon. Usually part of the clue is a definition, and the rest of the clue an indirect indication, often a charade, anagram, hidden word, or pun. But it doesn't always quite work out that way. Also there are usually many possible ways to interpret them.

The elder Bush could be the politician, but it could also be a plant for example (capital B should be ignored); more likely (my guess just now) elder is in the definition part and Bush in the charade, so they aren't connected at all. In fact I can see now the answer is "shrub" - anagram (indicated by "****ed") of bush with r (end (=last letter) of "war") put in (indicated by "about"). "Maybe the elder" is the definition part that confirms shrub. The fact that a bush may also be a shrub is irrelevant, in fact weakens the clue because it put me on the right track. Now that initial S helps me see the answer to 1D is samovar, a hidden word clue, indicated by "shows contribution". Two pretty easy clues in reality.

I like to solve crosswords without recourse to a computerised word-searcher, though sometimes in frustration I will reach for it. Occasionally the answer is a word I've never seen before. Sometimes the answer isn't something the computer knows about, (someone's name, some slang or neologism) so I find nothing useful. Occasionally the setter has made a factual error or spelling mistake (very satisfying when you solve such a clue). But sometimes, even when the computer has told me only these few words can fit, and in fact one of them is right, it takes me quite a long time to work out why any of them solves that clue.

I just don't see computers as being very good at this kind of very fuzzy reasoning. You need fuzzy reasoning to play Bridge (the card game). Computers are getting better at bridge rather slowly, but are nowhere near their chess/backgammon strength, and still make occasional absurd errors no half-competent human would make. (Actually I still catch good backgammon programs making clear-cut errors a competent human wouldn't make. But this is quite rare, and in general backgammon programs are now stronger than any human.)

Ivan Viehoff
30-April-2009, 02:06 PM
I find it curious that BAUT ****'d a word that occurs in the bible (King James edition).

tdvance
30-April-2009, 11:03 PM
Who says computers can do crosswords? All crosswords, or only a few special ones?

A few years ago, someone released some program that used "google" to get 85% success rate solving crossword puzzles. It fared worst on the kind that use lots of puns and several levels of indirection of knowledge.

Ivan Viehoff
01-May-2009, 10:42 AM
A few years ago, someone released some program that used "google" to get 85% success rate solving crossword puzzles. It fared worst on the kind that use lots of puns and several levels of indirection of knowledge.

Thanks. I'll continue in my presumption that computers can't do "The Times" Crossword then. (A standard for cryptic crosswords in Britain, though certainly not the hardest one.)

tdvance
02-May-2009, 12:36 AM
oh yeah, "the Times Crossword" means something different in the UK.