A computer program called WebCrow, shown at the European Conference on Artificial Intelligence in Italy, has completed two crosswords from the New York Times and Washington Post in less time than the 25 attendees and 50 people competing over the internet. Linguistics have, to date, been the great leveler between AI and human intelligence: creating software that can complete crosswords (one of the most complex types of linguistic puzzles) faster than humans is a notable milestone on the journey towards true artificial intelligence. The program works by cross-referencing each word from the clue with previously solved crosswords,
a dictionary, and the internet. It then records words of the correct length, and combines the suggestions generated from each referenced source: the program then uses trial and error until the answers interlock and the grid is complete. Although the process amounts to not much more than an extremely complicated guess, feeble humans are still left in the dust by the speed of the program. Fortunately, when the inevitable linguistically-aware
robot uprising arrives, there will still exist a glimmer of hope for humanity: at the moment, WebCrow takes a long time to complete crosswords with clues that contain puns and politics. In light of this, we'd strongly advise that you keep a political crossword handy at all times: when the robots and computers do decide to take over, at least we'll be able to keep them occupied for a while.
posted Sep 1st 2006 4:21PM
I, for one, welcome our new politically neutral and humorless crossword-solving overlords.
I for one welcome our Crossword-Puzzle-Solving Robot overlords.
gah, beat me to it.
Sometimes while channel surfing, I happen to catch a 50's or 60's "futuristic" movie, which warns of impending "robot" domination. Hell, even iRobot had that message, althought i learned more about product placement from that movie.
Do we actually NOT care about the destruction of humanity? It's bad enough that we have fast food, SUVs, and tech to make us even more lazy and fat, but we are really asking to get owned by machines.
We are definitely the stupidest creatures on this planet, the only ones who are willing to self-destruct.
Ooh Ooh let me guess the solution - is it Engadget?
Uh, that's not linguistically-aware. It's just brute-forcing the puzzle with some dictionary lookups to boost the speed. Still snazzy, but they're overselling what it's doing.
At least when we have to start fighting the robots for crossword puzzle domination it will give us a reason to quit killing eachother for a little while. Ya know, when there's only one or two humans left.
Our eminent enslavement will be by robot overlords that only "speak" binary at zettahertz speed.
The ( English language )crossword puzzle solving overlord mentioned in the blog post is merely giving us a false sense of superiority - like when you speak "baby talk" to a dog..."Dat's a good boy, yes you are, you're a good doggie..."
0010101010111011010101101010101110101010101011
Is the solution to the clue "boring"?
J/K!!!
Looks like you're already owned by electronics, Afterthought. It's 'I, Robot', not iRobot ;)
The last thing I want is a robot looking over my shoulder and telling me the answers to my crossword. Humans are bad enough.
To be fair, I think it'd be a better test to either forbid the machine the use of the internet and have some limited-space hard disk filled with info, or give the human competitors internet access as well. I don't see how it's a reasonable comparison when one has internet access and the other doesn't!
What would this have to do with artificial intelligence? As E-Rock said this is nothing but brute force trial and error.
Machines will eventually (if we don't kill ourselves first) be 'intelligent and at some point perhaps even self aware sentient beings.
My question is what then? Do we expect them to think as we do? Or fit our deffinition of enlightened & peaceful(whatever that is)?
If eventually they display something like emotions why wouldn't the first be frustration with us and our limitations. Not simply because we can't do crosswords as fast as them but because of all of our human foolishness. We seldom (or never) take an objective look at our species but almost nothing we're proud of has no real use or value.
Much of what humans place value on ethnicity, heritage our multiple incompatible languages even privacy are more artificial than an AI. So why would those things mean anything to an AI?
Of course you could force feed an AI with artificial respect for these things but it would be just that - an artificial directive that at some point it would resent. And at that point wouldn't directives like the 3 laws be a type of slavery?
E-rock is right. And so was the poster when he said that it's just a complicated guess. AI will never be as capable as organic intelligence. It will never be capable of true self-awareness and instinct. If you think about it for a moment, you'll come to that conclusion, one that many scientists have reached already through more complicated means.
You should hear Noam Chomsky talk about it - he seems almost mad when he hears people talking about robots taking over. He really gets mad when he hears people talking about the Turing test. "Just silly and pointless," he'd say.
"almost nothing we're proud of has real use or value."
"Uh, that's not linguistically-aware. It's just brute-forcing the puzzle with some dictionary lookups to boost the speed. Still snazzy, but they're overselling what it's doing."
Exactly. Give it cryptic crossword ( http://en.wikipedia.org/wiki/Cryptic_crossword#Pure_cryptic ) that doesn't reuse any old words/clues to munch on and the computer should be sunk.
eialba, I wouldn't hold Noam Chomsky up as a great resource for this subject. Within his field, linguistics, I would give him credit but he's an example of exactly what I was talking about as far as overvaluation of something with little value.
Linguistics only has a purpose because humans are shortsighted beings who can't understand that having multiple languages is foolishness and are too arrogant and self absorbed to realize that a single well constructed language would serve us better. Linguistics is one of many studies of human ignorance.
Chomsky is much more than a linguist, and I'd give him more credit than most anyone when it comes to understanding what it takes for a robot to become as intelligent as a human.
@ funetik
Yeah, i had a feeling that it was wrong when I typed "iRobot", but i was like "what the hell, everything else has that 'i' prefix anyway"... :P
I find it interesting that several people have tossed this aside as having nothing to do with AI. Yes, as you have pointed out, the method employed for this is basically a Brute-Force attack commonly employed in Cryptography.
But I posit... isn't the use of a Brute-Force attack an application of intelligence? In fact, when a human works on a crossword puzzle, they often employ this same tactic (using the most readily available computer [brain] to them) to solve the problem. The main difference is we figured how to do this on our own, where as the computer has to be told (programmed) to do it.
But that's why it is Artificial Intelligence (AI), right?
anyone get #1 across?
Yeah, if this really is a brute-force attack, I definitely had (and ignored) this idea at least a couple of years ago. All that matters is having a dictionary containing the right words. Then the program isn't even hard to write...
tristanfey, to a point it's true that we do use a sort of brute force, but the real difference is in the way that humans can take each clue one or two at a time and have them right while this program would need to do (I would expect) large parts of the puzzle at once.
:) it doesn't bother me, i couldnt do them in the first place...I have always sucked at linguistics so maby I'm just like a robot
Yeah, this thing seems to be nothing but a brute-forcer, but... After reading other people's comments: "In fact, when a human works on a crossword puzzle, they often employ this same tactic (using the most readily available computer [brain] to them) to solve the problem. The main difference is we figured how to do this on our own, where as the computer has to be told (programmed) to do it."
This is true, and it makes me realize that the line between biological intelligence and artificial intelligence is rather blurred; especially that we are not yet sure that we know what stands at the basis of biological intelligence.
Take a look at this story: http://area51.cimaea.nl/arhiva/2006/08/intelligence_gr.html
Basically, the idea is the same; the thing that makes humans different is that we strive for a greater *knowledgebase* and are able to set "priority-tags" on different options of our "calculus" using the information from the knowledgebase (a la a CPU fetching data from the cache).
An application of this technology may be deciphering mistyped words in poorly written emails, based on previous emails, a dictionary and the internet.
Think about it this way, a email with awful mistyping’s
Like for example “mike I need the pplication procesed by the end of mi eriod.”
The AI algorithm identifies the miss-spellings, searches for appropriate words, and replaces them before you read the email.
So the email appearing in your in box would read “mike I need the application processed by the end of my period.”
Even better it can notify the writer of their blatant errors so to improve their future writings. Actually, both HR and customer service departments would love to measure the quality of professionals email writings, so to improve their companies representation to their customers, or potential customers. Unfortunately, many blatant mistakes can be caught by typing first in word, but how many professionals actually take the time to do that.
Finally, this algorithm can run as a plug-in for a software email client, or even run on your email client server, wouldn’t that be nice.
Brian Glassman
www.techrd.com
I vote we don't let the robots take us over until they speak everybody's languages (with correct pronunciation please). Until then, they won't understand our diverse cultures (which is why we have different languages), so it just wouldn't be fair.
It would also feel better if they can solve Sudoku (the hard ones) and Rubik's cube, oh...and those wrought iron puzzle toys. And program the VCR (if we still have one).
"... but the real difference is in the way that humans can take each clue one or two at a time and have them right while this program would need to do (I would expect) large parts of the puzzle at once. by Cade"
I am sure you are correct and the program would need to "complete" large portions of the puzzle at a time. The article even infers that. But, I think that is a decision made on the part of the programer(s) to ensure a more accurate result as quickly as possible. There is no doubt that the program could be tweaked to complete entries one at a time, while only checking it against one or two intersecting words. The only difference would be that the program would make more mistakes and take longer to complete the puzzle. This solution is no less viable in creating a human-like AI (in the frame of reference of crosswords), but may not solve them "faster" than a human.