Can Artificial Intelligence Beat Jeopardy?

[The_root_of_all_evil]

Can Artificial Intelligence Beat Jeopardy?

Artificial Intelligence has been making leaps and bounds since its first thrashing of humanity in a chess game, but can it cope with being put on Jeopardy?

This week saw IBM unveil its Question Answering system, codenamed Watson, that it hopes can complete the greatest challenge of all, defeating humanity.

While Deep Blue resoundingly finished off the best humanity had to offer [http://www.research.ibm.com/deepblue/], IBM was looking for a much greater test for the second wave and so came up with the idea of "Jeopardy".

For those not acquainted with Alex Trebek's show, or the various SNL skits [http://en.wikipedia.org/wiki/Celebrity_Jeopardy!_(Saturday_Night_Live)], to play you are given the answer to a question and you have to guess the question involved.

This makes a real challenge for a computer, which needs to not only needs to know history, politics, film, and pop culture, but also how to decipher grammar and construct meaningful sentences.

As if that's not tough enough, you then have to try and teach them things like irony, riddles, double entendres, sarcasm and other features of language that often puzzle us mere mortals.

"Progress on the (project) will be important in the quest to understand and build 'intelligent computing systems' capable of cooperating with humans in language-related tasks previously out of reach for computers," said Dr. David Ferrucci, leader of the IBM Watson project team.

There has been no mention of when the challenge will be set or whether it will be televised, but Alan Turing's birthday (June 23rd) would seem to be the best idea, given he is widely seen as the father of Artificial Intelligence.

Source: Monsters and Critics [http://www.monstersandcritics.com/tech/news/article_1473374.php/IBMs_next_frontier_for_computing_TV_game_shows_]



Permalink

 

[TaborMallory]

I imagine any program can thwart any challenge if given enough programming. (Correct me if I'm wrong)

As far as I know, we still have a hell of a time programming recognition skills for things such as sarcasm and humor, though.
Take a look at any hollywood film involving robots. No matter how far into the future they are, the robots still don't recognize sarcasm and emotion.

 

[oliveira8]

Actually Deep Blue only won one match versus Kasparov, lost 3 and there was two ties.

And Deep Blue vs Kasparov Game 2 is believed to have been rigged.

So..AI hasnt really trashed humanity in chess.

 

[BobisOnlyBob]

I imagine any program can thwart any challenge if given enough programming. (Correct me if I'm wrong)

A program can thwart any specific, pre-defined challenge given appropriate programming. Something like speech recognition is plausible given time, specific calibration, tuning and a vast phonetic->syllabic dictionary. Parsing grammar is a whole separate field, as is then turning that grammar into a database query, finding the appropriate answer and then displaying or vocalising it in the form of a query. That's some extremely heavy challenges to overcome, and to do them fast and efficiently, it's going to take more than substantial effort.

On the other hand, this is just a specially-designed chatbot hooked up to a database. They've been attempting natural language processing for years, but not with a motivation as clear and specific as this before.

 

[oliveira8]

Actually Deep Blue only won one match versus Kasparov, lost 3 and there was two ties.

And Deep Blue vs Kasparov Game 2 is believed to have been rigged.

So..AI hasnt really trashed humanity in chess.

And they never will they lack the instinct and the heart to become masters of chess.

The ironic part was that Kasparov beated Deep Blue twice using the same tactic and trap. And this is were the "Game 2 was rigged" talk starts. Some folk say that IBM changed DB programing so he wouldnt fall for the same tactic over and over.

That isnt AI.

 

[Lord_Ascendant]

I dunno if I should be scared or curious.

 

[super_smash_jesus]

Its actually just Ken Jennings in a robot suit, answering questions.

 

[The_root_of_all_evil]

The ironic part was that Kasparov beated Deep Blue twice using the same tactic and trap. And this is were the "Game 2 was rigged" talk starts. Some folk say that IBM changed DB programing so he wouldnt fall for the same tactic over and over.

That isnt AI.

Interesting points, I remember Kasparov throwing an absoloute fit over game 2, but it'd be churlish of me to suggest rigging. In the link there's an interview with him that it might be interesting reading.

As for the change in programming, wouldn't that be similar to your mother telling you off for something you keep doing. Surely true AI would need a similar "parent" figure to help it learn?

Take a look at any hollywood film involving robots. No matter how far into the future they are, the robots still don't recognize sarcasm and emotion.

Keanu Reeves isn't actually a robot.

 

[Altorin]

Actually Deep Blue only won one match versus Kasparov, lost 3 and there was two ties.

And Deep Blue vs Kasparov Game 2 is believed to have been rigged.

So..AI hasnt really trashed humanity in chess.

And they never will they lack the instinct and the heart to become masters of chess.

I wouldn't say never

but I did have some objection to the idea that AI has "Resoundingly finished off humanity at chess"

that's not the case at all.

I say, let Deep Blue compete in a full tournament as a person.. let the engineers tweak him between each tournament, sort out any obvious issues.. Don't just say "Lol, can we set the machine against one of the worlds finest players?"

Machine or Not - Deep Blue has to earn the right to sit at that table like everyone else.

As for the topic at hand.. I don't think a robot could grasp jeopardy right now.. I can see why it would be a goal.. it's generally considered to be a smart man's game show.. The "Chess" of game shows.. But I think there is too much language involved.. If the game were switched, where questions were given and answers were required, the computer would have man beat there, no problem - program enough information into a computer and it can be a search engine of itself.

But to form the correct question to a randomly formed answer, especially one involving sarcasm or irony, would be a tough call for a computer.

But one day, maybe.

 

[oliveira8]

The ironic part was that Kasparov beated Deep Blue twice using the same tactic and trap. And this is were the "Game 2 was rigged" talk starts. Some folk say that IBM changed DB programing so he wouldnt fall for the same tactic over and over.

That isnt AI.

Interesting points, I remember Kasparov throwing an absoloute fit over game 2, but it'd be churlish of me to suggest rigging. In the link there's an interview with him that it might be interesting reading.

As for the change in programming, wouldn't that be similar to your mother telling you off for something you keep doing. Surely true AI would need a similar "parent" figure to help it learn?

Considering that its a game, and Kasparov chess tactics are professional clean moves, if they block the machine from not falling into traps, they not making true AI, they just making a regular program that recognizes chess moves and avoids, without any intelligence behind.

Their goal was to make a machine that can "think" and play chess without human hand(apart from programing), when humans insert a program that makes the machine never fall into traps, they defeating the purpose of a "thinking machine".

Deeper Blue(Deep Blue upgraded version for game 2) was programmed to "think" 200 million chess moves per second, yet IBM had to introduce(if the cheating stories are correct.) programming so the machine wouldnt fall for Kasparov trap which had falled twice.

 

[The_root_of_all_evil]

Deeper Blue(Deep Blue upgraded version for game 2) was programmed to "think" 200 million chess moves per second, yet IBM had to introduce(if the cheating stories are correct.) programming so the machine wouldnt fall for Kasparov trap which had falled twice.

Having a think on this, shouldn't the basis of AI be "Once bitten, twice shy"? There really ought to have been a AI routine to look for traps.

 

[sonidraw]

Calculating and thinking are two completely different things. There are plenty of computers that can calculate odds and calculate the effects of various moves in some games. But I don't know if any of them are actually thinking.

Complicated games where there are thousands of possible moves every turn, each with their own outcome and influence on all previous and future moves, are still too difficult for computers when facing off against good human opponents.

Plus, if they aren't learning from their mistakes without humans reprogramming them, then they aren't really intelligent. Heck, there are millions of species of different animals that are smarter than a computer that needs to be reprogrammed to avoid a single trap it keeps on falling into.

 

[oliveira8]

Deeper Blue(Deep Blue upgraded version for game 2) was programmed to "think" 200 million chess moves per second, yet IBM had to introduce(if the cheating stories are correct.) programming so the machine wouldnt fall for Kasparov trap which had falled twice.

Having a think on this, shouldn't the basis of AI be "Once bitten, twice shy"? There really ought to have been a AI routine to look for traps.

Well...the machine can process 200 million chess moves per second, I think it should be able to see a trap coming.

The machine should be able to see the trap without having the "Avoid trap" program in it. If IBM encoded every way to avoid all traps in a chess game the machine would be unbeatable.

This is chess not a beat 'em up were you press buttons randomly and win.

 

[The_root_of_all_evil]

The machine should be able to see the trap without having the "Avoid trap" program in it. If IBM encoded every way to avoid all traps in a chess game the machine would be unbeatable.

Possibly, but with Chess there are a number of moves that can seem beneficial (Queen takes Rook, which would take her) but lead to Queen checkmates King, as the far away Queen can't block the move.

Working on a situation like that would require an analysis that needs to postulate what the next move would be, rather than could be, which is where the real test of "humanity" is.

As for the "avoid trap" program, don't you have a similar thought process when you come across a stack of ammo and health in an FPS? You KNOW that something is amiss without having any real clue as to what.

 

[oliveira8]

The machine should be able to see the trap without having the "Avoid trap" program in it. If IBM encoded every way to avoid all traps in a chess game the machine would be unbeatable.

As for the "avoid trap" program, don't you have a similar thought process when you come across a stack of ammo and health in an FPS? You KNOW that something is amiss without having any real clue as to what.

But a chess trap unless you a pro chess player, you only notice when you lost the game.

OR!

You're this guy

Spoiler