I'm afraid you're right. But hey, there's still the Supreme Court to stop it...
Anyway, why would my post have been labeled "flamebait"? Perhaps someone doesn't like publicity for the idea that his (her is not likely) country is going to jail a few innocent bystanders in the War Against ?
A colleague of mine (we work at the FC Donders Centre for Cognitive Neuroimaging, http://www.ru.nl/fcdonders), told me that the anterior cingulate (which is found in these trials), is involved in sensosomatory processes, such as respiration, heart beats, skin sensitiviy... That means that seeing activation there in an fMRI scan is exactly the same as finding a difference in heart rate, respiratory rate or skin resistance, except it is much more expensive.
A 90% success rate implies that 1 in every 10 innocent people gets labeled guilty. Suppose you run an investigation and you've got a group of 50 people, and you're sure one of them did it. You scan them all. Then you end up with a group of some 4 to 8 people with a 90% probability that it includes the perpetrator.
So, yes, this system is a great solution, but not for the problem "who did it?". No, it is a great solution for the problem how we can fill our prisons as quickly and subjectively as possible.
The point is: large numbers of rare events. In statistical linguistics (which is where I do/did some of my work), 1000 words can give you some clue, but anything less than a million is simply not sufficient, and studies show that every sample still misses out on rare events, since the total number of rare events is simply huge.
Another point is: representation. If you pick a thousand web sites, how can you be sure that they are representative of the entire population? With a billion, you can get accurate distributions over collections of 1000 pages instead of just a single number.
Suppose you wanted to see which character is most frequently used per top level domain. If you would take 1000 pages, chances are that your sample would only contain a few of even no pages for certain domains. That would make your statistics highly unreliable.
So, yes, 1.000.000.000 is a good place to start. And that's why it's cool...
If you know a little bit more about neuro-imaging (I guess they were using fMRI, or PET or something similar), you would know that this kind of result proves *absolutely nothing*. The fact that areas "normally associated with reasoning" do not show more activity says more about the baseline they used for their study than about political reasoning. Good fMRI experiments are extremely hard to set up, and cannot be used for drawing conclusions at this level. As a matter of fact, it is not even known which areas are responsible for reasoning, let alone to what degree they should "light up" in the said experiment.
PS I do cognitive neuro-science for a living, including fMRI studies...
Hey, don't we need gargage men, factory workers, and clerks?
This article is a lot of manure. "Unproductive pleasures pall eventually. After a while you get tired of lying on the beach. If you want to stay happy, you have to do something." He speaks for himself, obviously, as there are whole legions of people who prefer this over their work.
"So one thing that falls just short of the standard, I think, is reading books." Is this guy serious? I guess he never ever read a book, or, if he did, he didn't get it.
Why does this get mentioned on Slashdot? Just because the guy is a programmer?
What I'm missing from the discussion is a reference to the Sapir-Whorf hypothesis (see http://venus.va.com.au/suggestion/sapir.html): does our language shape/limit our thought! And this study clearly seems to contradict it...
I beg to differ. While I agree that the conclusions cannot be about innateness of geometry, it does show that you *somehow* pick up these concepts without verbal instruction. That means that at birth our brains contain a mechanism for either recognizing basic geometrical features, or a mechanism for learning it. So there you are.
Your claim about pattern recognition basically is what the study claims, only the study was done with patterns over specific features. So please be more careful in your criticism: either drop it, or give better arguments. Like this it seems "Insightful +5" to the average Slashdotter, whereas your critique is seriously flawed.
Boy, this Eric Kandel is really a clever guy. How does he think "free will" comes about? Does he think the rest of the world think it's somewhere outside the brain? Of course not. The signal that is processed is part of the same system that implements the free will. He is confusing "free will" with our concious experience of stimuli. I'll have his Nobel Prize now, thank you.
If you want an answer to these questions, I urge you to start studying neuro-biology and cognitive psychology immediately, do a PhD, get a post-doc, start your own research group and come back to inform us what you've found in 25 years.
The point I'm trying to make, is that there has been done an awful lot of research in this area, and there is still no way to answer your question.
If you want to start reading: most introductory books to neuro-cognition ("how the brain works") will give you a clue, but beware that many of the findings that are presented in such texts have been simplified and/or are controversial...
I didn't bother to read the study (so I don't know whether they did this), but it is possible to supprees the after image: you do it by replacing the original image with something called a "mask" after the desired interval (50ms in this case). Such a mask could be a gray picture, or a picture of noise or something similar. That effectively blocks the after-image effect.
I'm reasonably sure that 50ms is too short, at least it is for visual processes involving character recognition; 70ms is an absolute minimum for that. All that can be recognized in 50ms are big shapes...
One of the commenters wrote that "this guy is buried - forever". That is precisely what is wrong. If you kill someone, ok, then you can be locked up for the rest of your life. But sending a few e-mails (per capita) is not really tantamount to killing, is it? This sentence is indeed draconian.
Read before you reply. It says "If you assume...". It's a problem often used to dismiss results of voluntary studies, but that doesn't invalidate this study completely (as it can do in other studies).
Your remark that "Every single possible conclusion that one might seek to draw from this study could very well be illusory." is something that applies to every questionnaire.
Of course it is, but I dutifully reported my experiences with my laptops even though I have nothing to complain about. The point of this study is not to supply Apple with 95% accurate figures of the number of repairs (I'm sure they've got their own set of numbers), but to give users *some* insight.
WRT to the person that complained that Ric doesn't answer: I wrote him a somewhat more extensive critique of his methods in normal wordings, and he did reply. And although it is difficult to take all points into account (such as deducing the base rate for failure from a time dependent series in something that is not a simple Poisson process), he did neatly split out all results into 1 year, 2 year, 3 year and 4-or-more year failures.
If you assume respondent bias, you can expect people with more problems to respond more often, which give the results a slightly non-linear character, but the order won't change. So it will give us some insight into which machines are more prone to faults.
Whoever modded the parent as "insightful" didn't think very long...
By the way, the site still responds. Whatever happened to the slashdot effect???
One of the posters pointed out larger magical squares in Dutch, and I've written another program just for the heck of it, and now it does find 6x6 and 7x7 solutions, even using relatively common words. From the 23 7x7 solutions, the one with most frequent words it found (in just 3 minutes, computers have become a lot faster) was:
b e s l i s t e n t e n t e s t a r t e r l e r a r e n i n t r e d e s t e e d s e t e r n e e r
Without compounding, it does not produce an 8x8 square. Lowering the frequency threshold to the hapax level takes very, very long. I'll post the result when and if the program finds one...
Can't reconstruct it, as the code is long buried, but the complexity would be O(k^N), where k is the number of words of length N. This is however not an good upper bound for an efficient implementation, as the first word limits the choice of the second word with a factor 26 (well, the alphabet size of first positions, to be precise) on average. Choosing the second word will limit the choice of the third word with a factor 26^2, since the first two letters are now fixed, etc. For choosing the last word, a single lookup would suffice (but that won't list all posibilities). So a better estimation would be O(k * k / 26 * k / 26^2...) = O(k^N / 26^((N-1)(N-2)/2)).
That actually demonstrates the sparseness problem: the second term grows much faster than the first.
It's a sparseness problem. The space of two letter words is pretty full, but as the length of the words increases, the number of words does not increase as fast as the number of possible combinations.
I've actually written a program to generate the Dutch solutions to the 5x5 puzzle somewhere around 1990, and it found several good solutions with a 210,000 word dictionary. However, it didn't find solutions for the 6x6 square. So I would expect that the 10x10 square is near impossible, unless wacky compounds would be allowed, since they are the only thing that can keep the letter combination filled...
Slashdot Mirror
Well, relying on reference counting is just as insane: one cycle and you're memory can fill up pretty quickly, potentially crashing your application.
I'm afraid you're right. But hey, there's still the Supreme Court to stop it...
Anyway, why would my post have been labeled "flamebait"? Perhaps someone doesn't like publicity for the idea that his (her is not likely) country is going to jail a few innocent bystanders in the War Against ?
A colleague of mine (we work at the FC Donders Centre for Cognitive Neuroimaging, http://www.ru.nl/fcdonders), told me that the anterior cingulate (which is found in these trials), is involved in sensosomatory processes, such as respiration, heart beats, skin sensitiviy... That means that seeing activation there in an fMRI scan is exactly the same as finding a difference in heart rate, respiratory rate or skin resistance, except it is much more expensive.
A 90% success rate implies that 1 in every 10 innocent people gets labeled guilty. Suppose you run an investigation and you've got a group of 50 people, and you're sure one of them did it. You scan them all. Then you end up with a group of some 4 to 8 people with a 90% probability that it includes the perpetrator.
So, yes, this system is a great solution, but not for the problem "who did it?". No, it is a great solution for the problem how we can fill our prisons as quickly and subjectively as possible.
Nope. If I hadn't written this, someone would have replied: what the hell do you know? Is "I hold a post-doc in the field" better?
Anyway, I fail to see your logic...
The point is: large numbers of rare events. In statistical linguistics (which is where I do/did some of my work), 1000 words can give you some clue, but anything less than a million is simply not sufficient, and studies show that every sample still misses out on rare events, since the total number of rare events is simply huge.
Another point is: representation. If you pick a thousand web sites, how can you be sure that they are representative of the entire population? With a billion, you can get accurate distributions over collections of 1000 pages instead of just a single number.
Suppose you wanted to see which character is most frequently used per top level domain. If you would take 1000 pages, chances are that your sample would only contain a few of even no pages for certain domains. That would make your statistics highly unreliable.
So, yes, 1.000.000.000 is a good place to start. And that's why it's cool...
If you know a little bit more about neuro-imaging (I guess they were using fMRI, or PET or something similar), you would know that this kind of result proves *absolutely nothing*. The fact that areas "normally associated with reasoning" do not show more activity says more about the baseline they used for their study than about political reasoning. Good fMRI experiments are extremely hard to set up, and cannot be used for drawing conclusions at this level. As a matter of fact, it is not even known which areas are responsible for reasoning, let alone to what degree they should "light up" in the said experiment.
PS I do cognitive neuro-science for a living, including fMRI studies...
Hey, don't we need gargage men, factory workers, and clerks?
This article is a lot of manure. "Unproductive pleasures pall eventually. After a while you get tired of lying on the beach. If you want to stay happy, you have to do something." He speaks for himself, obviously, as there are whole legions of people who prefer this over their work.
"So one thing that falls just short of the standard, I think, is reading books." Is this guy serious? I guess he never ever read a book, or, if he did, he didn't get it.
Why does this get mentioned on Slashdot? Just because the guy is a programmer?
What I'm missing from the discussion is a reference to the Sapir-Whorf hypothesis (see http://venus.va.com.au/suggestion/sapir.html): does our language shape/limit our thought! And this study clearly seems to contradict it...
I beg to differ. While I agree that the conclusions cannot be about innateness of geometry, it does show that you *somehow* pick up these concepts without verbal instruction. That means that at birth our brains contain a mechanism for either recognizing basic geometrical features, or a mechanism for learning it. So there you are.
Your claim about pattern recognition basically is what the study claims, only the study was done with patterns over specific features. So please be more careful in your criticism: either drop it, or give better arguments. Like this it seems "Insightful +5" to the average Slashdotter, whereas your critique is seriously flawed.
Boy, this Eric Kandel is really a clever guy. How does he think "free will" comes about? Does he think the rest of the world think it's somewhere outside the brain? Of course not. The signal that is processed is part of the same system that implements the free will. He is confusing "free will" with our concious experience of stimuli. I'll have his Nobel Prize now, thank you.
If you want an answer to these questions, I urge you to start studying neuro-biology and cognitive psychology immediately, do a PhD, get a post-doc, start your own research group and come back to inform us what you've found in 25 years.
The point I'm trying to make, is that there has been done an awful lot of research in this area, and there is still no way to answer your question.
If you want to start reading: most introductory books to neuro-cognition ("how the brain works") will give you a clue, but beware that many of the findings that are presented in such texts have been simplified and/or are controversial...
I didn't bother to read the study (so I don't know whether they did this), but it is possible to supprees the after image: you do it by replacing the original image with something called a "mask" after the desired interval (50ms in this case). Such a mask could be a gray picture, or a picture of noise or something similar. That effectively blocks the after-image effect.
I'm reasonably sure that 50ms is too short, at least it is for visual processes involving character recognition; 70ms is an absolute minimum for that. All that can be recognized in 50ms are big shapes...
One of the commenters wrote that "this guy is buried - forever". That is precisely what is wrong. If you kill someone, ok, then you can be locked up for the rest of your life. But sending a few e-mails (per capita) is not really tantamount to killing, is it? This sentence is indeed draconian.
No, this might be justice, it isn't right.
For those of you who don't understand Spanish, this is the translation: For end I may type by without wear one dictionary.
Read before you reply. It says "If you assume...". It's a problem often used to dismiss results of voluntary studies, but that doesn't invalidate this study completely (as it can do in other studies).
Your remark that "Every single possible conclusion that one might seek to draw from this study could very well be illusory." is something that applies to every questionnaire.
Of course it is, but I dutifully reported my experiences with my laptops even though I have nothing to complain about. The point of this study is not to supply Apple with 95% accurate figures of the number of repairs (I'm sure they've got their own set of numbers), but to give users *some* insight.
WRT to the person that complained that Ric doesn't answer: I wrote him a somewhat more extensive critique of his methods in normal wordings, and he did reply. And although it is difficult to take all points into account (such as deducing the base rate for failure from a time dependent series in something that is not a simple Poisson process), he did neatly split out all results into 1 year, 2 year, 3 year and 4-or-more year failures.
If you assume respondent bias, you can expect people with more problems to respond more often, which give the results a slightly non-linear character, but the order won't change. So it will give us some insight into which machines are more prone to faults.
Whoever modded the parent as "insightful" didn't think very long...
By the way, the site still responds. Whatever happened to the slashdot effect???
One of the posters pointed out larger magical squares in Dutch, and I've written another program just for the heck of it, and now it does find 6x6 and 7x7 solutions, even using relatively common words. From the 23 7x7 solutions, the one with most frequent words it found (in just 3 minutes, computers have become a lot faster) was:
b e s l i s t
e n t e n t e
s t a r t e r
l e r a r e n
i n t r e d e
s t e e d s e
t e r n e e r
Without compounding, it does not produce an 8x8 square. Lowering the frequency threshold to the hapax level takes very, very long. I'll post the result when and if the program finds one...
Let me check, just for educational purposes.
I can only check Dutch for the moment. It has the following distribution:
1 letter 26 words 100% coverage
2 letters 150 words 22% coverage
3 letters 1774 words 10% coverage
4 letters 6028 words 1% coverage
5 letters 10698 words 0.09% coverage
6 letters 18188 words 0.006% coverage
The one letter coverage is an artefact of the dictionary (CELEX), but there you are.
Just for your information: where I live, a CD easily costs 20 euro, which is about $23.50...
Can't reconstruct it, as the code is long buried, but the complexity would be O(k^N), where k is the number of words of length N. This is however not an good upper bound for an efficient implementation, as the first word limits the choice of the second word with a factor 26 (well, the alphabet size of first positions, to be precise) on average. Choosing the second word will limit the choice of the third word with a factor 26^2, since the first two letters are now fixed, etc. For choosing the last word, a single lookup would suffice (but that won't list all posibilities). So a better estimation would be O(k * k / 26 * k / 26^2 ...) = O(k^N / 26^((N-1)(N-2)/2)).
That actually demonstrates the sparseness problem: the second term grows much faster than the first.
It's a sparseness problem. The space of two letter words is pretty full, but as the length of the words increases, the number of words does not increase as fast as the number of possible combinations.
I've actually written a program to generate the Dutch solutions to the 5x5 puzzle somewhere around 1990, and it found several good solutions with a 210,000 word dictionary. However, it didn't find solutions for the 6x6 square. So I would expect that the 10x10 square is near impossible, unless wacky compounds would be allowed, since they are the only thing that can keep the letter combination filled...
Can you imagine what would happend if they print Wikipedia with this error in it and hand it out in third world countries?
78k? Wow, if those developers would squeeze just a little bit more, it might run on a Commodore 64...