A true Bayesian filter, wow. Let's face it, statistical classifiers based von Bayes' formula are not really state of the art. They make false assumptions about the data (independence of features).
More intelligent classification algorithms can solve non-linear problems far better. Check out Kernel Machines and, somewhat older, Maximum Entropy models.
wow, finally an affordable application. in europe, 3G almost ruined a major player in the telco market, and severely damaged the financial situation of the German Telekom, after the govt made some 60 billion EUR.
(Oh, as a small remark.. German Telekom's shares are in public hands.. )
Re:Just Call it a LIBRARY, Please!
on
Libraries Are 31337
·
· Score: 2, Insightful
you're completely right. somehow i got the impression, some people missed the irony in my post, pardon: in the first part of it...
call them information broker, and the jobs sounds fancy again.
librarians are old-fashioned only as long as they stick to storing information on paper instead of creating networked, digital libraries. the first will protect there jobs, probably, the latter is going to save us researchers/users/customers more and more time.
Wouldn't be as good. Apple has a (short) history of damaging Apple-based software businesses by including functionality for free. This happened with CD burning software (Toast used to be much more important...) and some smaller multimedia utils such as Winamp (there was an alpha version for OS X - it won't be cont'd probably due to iTunes).
It's essentially the same strategy MS persued when "integrating" IE into Windows.
Of course, those Apple applications are neat and a huge added-value. But at some point, Apple will hurt its industry partners - that's bad because the OS X platform needs new software and new ports.
... and this browser sucks big-time, seen from the usability-perspective. GUI Fonts look more than crappy since Anti-Aliasing seems to be shut off. Buttons react like windows active desktop ones (underlined or boxed when mouse over).
Popup menus (hit "print" or "back" to see) look like Windows garbage.
I'm not using Gecko. I'm using a browser, and unless this application is meant for web users (and not 'rendering engine users'), it will have to go with what I'm used to and with what I want because I spend like 1500 bucks more on a Powerbook (in comparison to a Wintel machine).
Oh, it doesn't feel like Windows due to a "poor understanding of XUL"??? Don't believe it.
It's the machine that should adapt to the user, not the user who should adapt to a new technical concept that puts technological advance and ease of implementing/porting in favor of GUI standardization, even though it'S good-looking (and Mozilla is).
Each OS has its interface standards, so does Windows. If you don't comply with them, you gotta have reasons that result in improvements for the user. This skinning / widget drawing technology obviously does not, since I can't use my standard skinning applications (e.g. WindowBlinds which can replace the WIndows windows with a neat Aqua interface) any more.
I can accept a non-standard user interface in highly specialized applications that use e.g. Java/Swing for economic reasons (portability), but not in a standard-everyday-use application like a web browser...
these seem to be the alternative. they offer read/write protection. USB is widely accepted by the industry, memory sticks are cheap to produce. Interchange between different OSs seems to be fairly easy and is, to my knowledge (I tried it) partially implemented (Win/OSX interchange is possible).
Hope standard-BIOSes will support booting from memory sticks soon:)
What matters is the level of abstraction. Of course, you can "learn by example". Storing all situations and finding the right decision by looking at the stored stuff is called "memory-based learning" in AI. However, the wonderful thing humans do is to generalize from learned situations. This generalization could be actually called "learning". You see a dog leave lift a leg next to a tree on the street. You see that a second time, and a third one. And you learn, that dogs just happen to do that. You will expect a dog pee whenever it lifts a leg, even if it's not next to a tree but next to your leg with your favorite pants on. You learned: the tree is not important.
You simply sorted out the not-so-relevant information. Besides, the concept of the dog can't be represented as a list. You can't store it as a conjunction of elements (4 legs, brown hair, barks every now and then). If a poor dog in Jerusalem gets hit by a bomb and looses two legs, it's still a dog and you would recognize the animal as a dog. If a dog does not bark, it's still a dog. You don't store the dog as a picture either: Children do recognize the dog in a cartoon as such, even though Goofy doesn't really ressemble the dog on the street.
Knowledge is much more than lists. Maybe you can imagine it as a network of concepts, linked with each other by relation such as "is-a" (as in "a dog" -is-a- "animal") or "has" ("a dog" -has- "legs").
The original question aimed at linguistic knowledge, and that's also knowledge that can't be represented as lists. That's why Alicebot is indeed a little primitive and not really state-of-the-art AI, as I see it. (See&rate my question regarding Deep / Shallow analysis.)
The Cyc project aims to collect world knowledge ("common sense"). However, many AI tasks show that this job is probably too huge to do it manually.
Do you think we will eventually get to a point were an AI system is able to gather common sense knowledge from a giant corpus, such as the web? What are the problems we will have to solve?
"Understanding" an utterance usually means to perform various analysis steps. This involves a tremendous amount of (linguistic and) world knowledge.
A big issue among language technology researchers is whether this is necessary at all when bringing speech to computers. Is a dialog (or just a single natural language utterance) supposed to be deeply analyzed in terms of syntactic structure and its semantic and rhetorical contribution? The alternative is to apply statistical models and rather simple knowledge. Up to now, the latter systems are known to give quicker results.
RW, how much does a computer really need to know to make it a good replacement for a, say, sales clerk in a web shop?
yes, Darwin is available for x86 platforms. However, you'll miss the great Aqua user interface that comes with OS X. You could run KDE3, which is not as much fun...
AI systems for inference or classification (as this is the case here) work either rule-based or statistical-based. In both cases, the knowledge contained in the database must be acquired somehow. Automatically training such a database is a hard thing to do, because you need valid data in machine-readable form: that is a database full of found symptoms and the diagnosed disease. Such data probably exists for the simple-to-diagnose diseases that are well-known, but not for rare diseases as mentioned in the article. So, we end up squeezing our knowledge into the computer. But - who does that?
Can the doctors who enter their knowledge into the machine be held responsible for mistakes? Even for mistakes that occured simply because the knowledge is overcome?
In AI research, more and more people acknowledge that we can't fit our full knowledge into a database. One reason for that is that we can't even spell out 'tacit knowledge' (or: implicit knowledge) the way we need to when giving the computer rule-based knowledge to solve a problem.
A doctor learns to have a feeling about a patient: this intuition can't be replaced, at least not within the next 20 years...
One day Bill complained to his friend, "My elbow really hurts, I guess I should see a doctor."
His friend said, "Don't do that. There's a computer at the drug store that can diagnose anything quicker and cheaper than a doctor. Just put in a sample of your urine and the computer will diagnose your problem and tell you what you can do about it. It only costs ten dollars."
Bill figured he had nothing to lose, so he filled a jar with a urine sample and went to the drug store. Finding the computer, he poured in the sample and deposited the $10. The computer started making some noise and various lights started flashing. After a brief pause out popped a small slip of paper on which was printed:
You have tennis elbow. Soak your arm in warm water. Avoid heavy lifting. It will be better in two weeks.
Late that evening while thinking how amazing this new technology was and how it would change medical science forever, he began to wonder if this machine could be fooled. He decided to give it a try. He mixed together some tap water, a stool sample from his dog, and urine samples from his wife and daughter. To top it off, he masturbated into the concoction. He went back to the drug store, located the machine, poured in the sample, and deposited the $10. The computer made the usual noise and printed out the following message:
Your tap water is too hard. Get a water softener.
Your dog has worms. Get him vitamins.
Your daughter is using cocaine. Put her in a rehabilitation clinic.
Your wife is pregnant with twin girls. They aren't yours. Get a lawyer.
And if you don't stop jerking off, your tennis elbow will never get better."
It's quite normal for a company actually to follow the old policy of the carrot and the stick when it comes to their relationship with the media.
As a journalist focussing on 'digital lifestyle', of course I get free media passes, rebates and free goodies from some companies. From others I don't. Just two days ago I was informed that my radio show won't get to give away a brand-new PDA (as a prize), because another show will do a marketing-like interview with the product manager - that's when I'm out. I won't let them buy me. From Apple, e.g., I'd can get new hardware for a test - but unfortunately, I have to return it. That is clearly o.k.. But some other companies won't even talk to us, because our reporting was not in favor of them.
Such a negative behavior is clearly un-professional from the perspective of the Public Relations guys, because it will just worsen the pictures media will draw about their company.
It is known that Steve Jobs usually goes far in planning every detail of a press briefing. It's all in his story book: even the brand of mineral water everyone gets to drink there is part of the company C.I..
"Technology without the price" was their slogan in the late 80's, and despite Atari's great success in gaming, the company sold A LOT of their 68k-based machines in Europe.
I happily remember my first Atari, a 1040STF that I bought as a teen. The better machines (like the Mega ST, later: TT030 with astonishing 16, then 32 MHz CPU!) were beyond my reach, but I could afford an extension set to boost the black&white 640x400 display to ~690x450. What a gain!
The GUI based OS ("TOS") was quite good for the time - it really kicked ass in comparison to Win 3.11. Even without the price, we got a machine that we could use to actually get some work done. That was different with Commodore's Amiga machines, which weren't too expensive - but they sucked because of their crappy color display.
Oh, if you want to know what Jack Tramiel (our hero!) is doing today... Electric Escape has a feature article.
Why? Non-North-American countries which have a lot of developers (Poland, Germany, India) find it a lot easier to buy into hardware that runs X11.
Good idea to downgrade the Germans to the group of 2nd world countries. We can now enjoy int'l aid, subsidiaries from the European Union (instead of only paying them) and, finally, we can get rid of those stupid Daimler Benz company that's not so much fun any more since they bought out Chrysler.
Oh, and I'm going to return my fast DSL line (which I'm getting for 10 Euros a month here in Berlin) to demonstrate my commitment to fulfill your Germany-is-about-like-India prophecy.
Here's an answer to the question of the day: Apple supports Cocoa, because they want to create a good user experience that's exactly the same for all applications. And that's why the Mac OS (9/X) GUI is ahead of common Linux GUIs...
ps.: Berlin is the capital of Germany. That's some country in Europe.
Suse has a bootable 'LiveEval' CD ROM (available as ISO). It has everything one needs; config is saved on some HD partition (this does not destroy anything). Of course, it won't boot as fast a floppy-based solution does.
Haven't seen such a poor interview technique for a long time. Your questions were lacking coherence; the things touched range from some very special stuff taken from special kernel versions up to what Alan does in his garden. Oh goodie.
Slashdot Mirror
More intelligent classification algorithms can solve non-linear problems far better. Check out Kernel Machines and, somewhat older, Maximum Entropy models.
Enough nerd talk for today :-)
wow, finally an affordable application. in europe, 3G almost ruined a major player in the telco market, and severely damaged the financial situation of the German Telekom, after the govt made some 60 billion EUR.
(Oh, as a small remark.. German Telekom's shares are in public hands.. )
you're completely right. somehow i got the impression, some people missed the irony in my post, pardon: in the first part of it...
call them information broker, and the jobs sounds fancy again.
librarians are old-fashioned only as long as they stick to storing information on paper instead of creating networked, digital libraries. the first will protect there jobs, probably, the latter is going to save us researchers/users/customers more and more time.
Wouldn't be as good. Apple has a (short) history of damaging Apple-based software businesses by including functionality for free. This happened with CD burning software (Toast used to be much more important...) and some smaller multimedia utils such as Winamp (there was an alpha version for OS X - it won't be cont'd probably due to iTunes).
It's essentially the same strategy MS persued when "integrating" IE into Windows.
Of course, those Apple applications are neat and a huge added-value. But at some point, Apple will hurt its industry partners - that's bad because the OS X platform needs new software and new ports.
... and this browser sucks big-time, seen from the usability-perspective. GUI Fonts look more than crappy since Anti-Aliasing seems to be shut off. Buttons react like windows active desktop ones (underlined or boxed when mouse over).
Popup menus (hit "print" or "back" to see) look like Windows garbage.
I'm not using Gecko. I'm using a browser, and unless this application is meant for web users (and not 'rendering engine users'), it will have to go with what I'm used to and with what I want because I spend like 1500 bucks more on a Powerbook (in comparison to a Wintel machine).
"This theme simulates the appearance of previous Netscape versions for the Windows operating system".
Sad, but almost true.
Oh, it doesn't feel like Windows due to a "poor understanding of XUL"??? Don't believe it.
It's the machine that should adapt to the user, not the user who should adapt to a new technical concept that puts technological advance and ease of implementing/porting in favor of GUI standardization, even though it'S good-looking (and Mozilla is).
Each OS has its interface standards, so does Windows. If you don't comply with them, you gotta have reasons that result in improvements for the user. This skinning / widget drawing technology obviously does not, since I can't use my standard skinning applications (e.g. WindowBlinds which can replace the WIndows windows with a neat Aqua interface) any more.
I can accept a non-standard user interface in highly specialized applications that use e.g. Java/Swing for economic reasons (portability), but not in a standard-everyday-use application like a web browser...
these seem to be the alternative. they offer read/write protection. USB is widely accepted by the industry, memory sticks are cheap to produce. Interchange between different OSs seems to be fairly easy and is, to my knowledge (I tried it) partially implemented (Win/OSX interchange is possible).
:)
Hope standard-BIOSes will support booting from memory sticks soon
does not really work well in language. As Chomsky formulated:
:)
"Colorless green ideas sleep furiously."
You've never encountered one word after the other before, but you can perfectly understand it.
Humans are better than Markov models
You simply sorted out the not-so-relevant information. Besides, the concept of the dog can't be represented as a list. You can't store it as a conjunction of elements (4 legs, brown hair, barks every now and then). If a poor dog in Jerusalem gets hit by a bomb and looses two legs, it's still a dog and you would recognize the animal as a dog. If a dog does not bark, it's still a dog. You don't store the dog as a picture either: Children do recognize the dog in a cartoon as such, even though Goofy doesn't really ressemble the dog on the street.
Knowledge is much more than lists. Maybe you can imagine it as a network of concepts, linked with each other by relation such as "is-a" (as in "a dog" -is-a- "animal") or "has" ("a dog" -has- "legs").
The original question aimed at linguistic knowledge, and that's also knowledge that can't be represented as lists. That's why Alicebot is indeed a little primitive and not really state-of-the-art AI, as I see it. (See&rate my question regarding Deep / Shallow analysis .)
Do you think we will eventually get to a point were an AI system is able to gather common sense knowledge from a giant corpus, such as the web? What are the problems we will have to solve?
"Understanding" an utterance usually means to perform various analysis steps. This involves a tremendous amount of (linguistic and) world knowledge.
A big issue among language technology researchers is whether this is necessary at all when bringing speech to computers. Is a dialog (or just a single natural language utterance) supposed to be deeply analyzed in terms of syntactic structure and its semantic and rhetorical contribution? The alternative is to apply statistical models and rather simple knowledge. Up to now, the latter systems are known to give quicker results.
RW, how much does a computer really need to know to make it a good replacement for a, say, sales clerk in a web shop?
yes, Darwin is available for x86 platforms. However, you'll miss the great Aqua user interface that comes with OS X. You could run KDE3, which is not as much fun...
AI systems for inference or classification (as this is the case here) work either rule-based or statistical-based. In both cases, the knowledge contained in the database must be acquired somehow. Automatically training such a database is a hard thing to do, because you need valid data in machine-readable form: that is a database full of found symptoms and the diagnosed disease. Such data probably exists for the simple-to-diagnose diseases that are well-known, but not for rare diseases as mentioned in the article.
So, we end up squeezing our knowledge into the computer. But - who does that?
Can the doctors who enter their knowledge into the machine be held responsible for mistakes? Even for mistakes that occured simply because the knowledge is overcome?
In AI research, more and more people acknowledge that we can't fit our full knowledge into a database. One reason for that is that we can't even spell out 'tacit knowledge' (or: implicit knowledge) the way we need to when giving the computer rule-based knowledge to solve a problem.
A doctor learns to have a feeling about a patient: this intuition can't be replaced, at least not within the next 20 years...
One day Bill complained to his friend, "My elbow really hurts, I guess I should see a doctor."
His friend said, "Don't do that. There's a computer at the drug store that can diagnose anything quicker and cheaper than a doctor. Just put in a sample of your urine and the computer will diagnose your problem and tell you what you can do about it. It only costs ten dollars."
Bill figured he had nothing to lose, so he filled a jar with a urine sample and went to the drug store. Finding the computer, he poured in the sample and deposited the $10. The computer started making some noise and various lights started flashing. After a brief pause out popped a small slip of paper on which was
printed:
You have tennis elbow.
Soak your arm in warm water.
Avoid heavy lifting.
It will be better in two weeks.
Late that evening while thinking how amazing this new technology was and how it would change medical science forever, he began to wonder if this machine could be fooled. He decided to give it a try. He mixed together some tap water, a stool sample from his dog, and urine samples from his wife and daughter. To top it off, he masturbated into the concoction. He went back to the drug store, located the machine,
poured in the sample, and deposited the $10. The computer made the usual noise and printed out the following message:
Your tap water is too hard.
Get a water softener.
Your dog has worms.
Get him vitamins.
Your daughter is using cocaine.
Put her in a rehabilitation clinic.
Your wife is pregnant with twin girls.
They aren't yours.
Get a lawyer.
And if you don't stop jerking off, your
tennis elbow will never get better."
As a journalist focussing on 'digital lifestyle', of course I get free media passes, rebates and free goodies from some companies. From others I don't. Just two days ago I was informed that my radio show won't get to give away a brand-new PDA (as a prize), because another show will do a marketing-like interview with the product manager - that's when I'm out. I won't let them buy me. From Apple, e.g., I'd can get new hardware for a test - but unfortunately, I have to return it. That is clearly o.k.. But some other companies won't even talk to us, because our reporting was not in favor of them.
Such a negative behavior is clearly un-professional from the perspective of the Public Relations guys, because it will just worsen the pictures media will draw about their company.
It is known that Steve Jobs usually goes far in planning every detail of a press briefing. It's all in his story book: even the brand of mineral water everyone gets to drink there is part of the company C.I..
"Technology without the price" was their slogan in the late 80's, and despite Atari's great success in gaming, the company sold A LOT of their 68k-based machines in Europe. I happily remember my first Atari, a 1040STF that I bought as a teen. The better machines (like the Mega ST, later: TT030 with astonishing 16, then 32 MHz CPU!) were beyond my reach, but I could afford an extension set to boost the black&white 640x400 display to ~690x450. What a gain! The GUI based OS ("TOS") was quite good for the time - it really kicked ass in comparison to Win 3.11. Even without the price, we got a machine that we could use to actually get some work done. That was different with Commodore's Amiga machines, which weren't too expensive - but they sucked because of their crappy color display.
Oh, if you want to know what Jack Tramiel (our hero!) is doing today... Electric Escape has a feature article.
Why? Non-North-American countries which have a lot of developers (Poland, Germany, India) find it a lot easier to buy into hardware that runs X11.
Good idea to downgrade the Germans to the group of 2nd world countries. We can now enjoy int'l aid, subsidiaries from the European Union (instead of only paying them) and, finally, we can get rid of those stupid Daimler Benz company that's not so much fun any more since they bought out Chrysler.
Oh, and I'm going to return my fast DSL line (which I'm getting for 10 Euros a month here in Berlin) to demonstrate my commitment to fulfill your Germany-is-about-like-India prophecy.
Here's an answer to the question of the day: Apple supports Cocoa, because they want to create a good user experience that's exactly the same for all applications. And that's why the Mac OS (9/X) GUI is ahead of common Linux GUIs...
ps.: Berlin is the capital of Germany. That's some country in Europe.
Suse has a bootable 'LiveEval' CD ROM (available as ISO). It has everything one needs; config is saved on some HD partition (this does not destroy anything). Of course, it won't boot as fast a floppy-based solution does.
Haven't seen such a poor interview technique for a long time. Your questions were lacking coherence; the things touched range from some very special stuff taken from special kernel versions up to what Alan does in his garden. Oh goodie.