One simple, illustrative observation would be that increases in the performance of artifical intelligence have not been described by Moore's Law. Why not? Speculation on the answer could only be informative.
Reminds me of something Robert Heinlein once wrote:
Suppose we had an AI system as smart as a dog and increased it's computational resources a thousand times. What would happen?
We would get a system that in practically no time at all decides to sniff your butt:-)
...I did find The Shining better than King's book though
Me too, but apparently King himself didn't. He even participated in doing a remake some years ago. Never heard of it? Guess that's because it's pretty lame compared to the Kubrick classic.
You're missing like, oh I don't know, about 20 million steps in there.
That's where the principle of natural selection steps in. Variants of the blobs that are better than the average at copying and sustaining themselves gain an advantage. During the copying there are occasional errors that result in mutations. Most of the mutations are likely to be bad, but some are improvements...
Religions are just plain silly, and large swathes of the human population have outgrown them, thankfully.
Unfortunately one belief is often just replaced by another, i.e. religion gets replaced by exaggerated confidence in science or in authorities. Today we also see an increase in "new age" movements, as well as bogus alternative medicine (not all alternative medicine is bogus thoug) that boils down to nothing but placebo. It's a great step forward to rid ourselves from religion, but we still have a long way to go.
Introns are just junk DNA thrown in there....They aren't comments...
In a sense they are, it depends on what you mean by a comment. According to your own statement they work just like chunks of code that is commented out.
Brown, purple and yellow all (can) have complete saturation (e.g. there is no grey in there). Therefore, they can be "pure" colours, consisiting of only one wavelength. The whole world is not an RGB monitor.
Sorry, you're partly right. Yellow is a spectral colour. But, show me a spectrum that contains brown or purple:-) Purple is the text-book example of a colour that does not correspond to one single wavelength.
... are they saying that the brain identifies the dress as being green because it adapts it's color table to the ambient lighting or is it that the brain knows and understands that the dress is green but it only looks different because of the ambient lighting.
From what I've read, the colour constancy phenomenon depends both on low-level adaption, and higher cognitive skills, such as remembering colours of objects, inferring that a certain patch of colour is shadowed etc.
Sorry, but I'm afraid I'm far from being an expert on color constancy. It's an interesting phenomenon though:-)
But enough rambling, red is red is red, a specific wavelength on the spectrum. And red is not the experience of red; experiences are finite and distinct, red is not, it is red, as is defined.
There is one slight problem with this definition of red. You cannot tell if something has a certain wavelength unless you use a spectrometer:-)
Since I don't want to buy one I will keep refering to my subjective experience of red as red.
To a philosopher maybe. But you could always have your eyes examined and find out that it is for real:-)
The colour sensitive cells in your eyes (red, green and blue cones) can be distinguished nowadays. If you lack one kind of cone you have to be colourblind. There are also objective tests that check if you can distinguish colours, there even was one in the article. If you cannot distinguish between, say red and green, you are per definition colour blind as well.
A distinct type of blue is a distinct wavelength of light.
Nope. A distinct type of blue can correspond to many different reflectance spectra. Reflectance functions that appear the same to us but are different are called metamers.
If you have a solid color object, only one wavelength of light will be reflected from that.
Not necessarily. Brown, Purple, Yellow etc are also colours, but they correspond to several wavelengths.
I'd prefer to skip these arbitrary definitions, & just say, for example, "Hey, I really like your lamda = 750 nm dress!"
Hmm... colour is NOT the same as wavelength.
For instance, that dress probably won't reflect just one wavelength, but is more accurately described by a reflectance function of the incoming light spectrum, still we say that it has ONE colour. The reflected wavelengths would also change, for instance when you go indoors/outdoors. So your lambda would change although the dress stays the same. A green dress is always green, and most of the time (due to colour constancy) it also looks green. Colour tells more about an object than the reflected wavelengths do.
It also would then expose the fact that the crude level of security in Unix-like operating systems is inadequate, in that the only information left vulnerable is the only information with value. If the User's data is destroyed, anything else on the hard drive is just the stuff that came off the CD-ROM.
How about the password file? And root access would allow you to do much more damage. Besides, User data is usually backed up anyway....
It's a sad day indeed when the best SciFi force fields reference they can come up with is Star Trek Voyager.
I'm sure there are better references, but none as well known. Little point in making a reference to something most people have never heard of...
But they really should have omitted the "Voyager" part of the name. "The Next Generation" is much better IMHO. Voyager is too much political correctness and Picard is missing:-)
Actually, the problem you are describing is due to the size of the image, not the resolution.
You cannot say that it's due to the size, and not the resolution, if you are interested in actually viewing the image details. You can easily realize this, if you imagine shrinking an image to a tenth of its original size, while at the same time increasing the resolution tenfold:-)
Yes, if we had better resolution displays, a 14" or maybe even a 12" screen could be useful, but you would have to sit closer to it.
If the pixels were drawn on the screen at a higher density than usual, then you would be able to see the whole image at once, without having to move your eyes around.
This would depend very much on what kind of image you are watching. It would certainly have to be a very simple one for this to be true. The human visual system only has sharp vision in 1 deg. of the visual field. This is usually about the size of a single letter in level 1 headings. This, and the limited amount of cones in the fovea limits the amount of detail you can squeeze into your image.
And that extra resolution would not be wasted, since your eyes are plenty capable of resolving details at greater than the standard 72dpi or 100 dpi of a monitor (at a distance of about 18" to 20").
OK, I admit that screens a little bit sharper would be beneficial, but we are not far from the limit at present, especially considering that most people don't have perfect vision.
As much as I would like to have better screens, I don't think merely higher resolution will solve the problem of viewing a hi-res image all at once without loosing resolution. In fact, you eye cannot see full detail on all of a 17" screen at once anyway, you have to change gaze (lots of times) to see all details.
GIMP 1.1 has a very nice feature that makes panning of large images a very pleasant experience:-). It gives you a small pop-up showing the entire image at lower resolution, and lets you move around the region of interest in the miniature with the mouse. If you have enough RAM is actually quite quick, even on large images.
BTW I also work with large images: aerial photograph databases which I probably never will be able to see on screen at full detail at once:-)
It's interesting that a blurry (compared to a monitor) tv can be just as effective at smoothing blocky low resolution stuff as the best, most intensive anti-aliasing effects.
Actually anti-aliasing on images often happens after the image has been rendered (for best results it should be applied before), and at that point there is nothing else to do but apply a blur filter... so your observation is probably correct:-)
I think that 'evolving' methodologies, and neural nets etc provide our best chance with this technology. It would be a good idea for the honda team to try a similar approach as Elvis on the control side. Perhaps add padding so it doesn't damage itself when learning to walk.
I agree that evolving, and learning are more promising. I would even say that this is the way we have to go. If we want to develop systems that interact with a complex environment (such as reality) there is simply no way to program for all the situations that can occur.
My objection was merely a semantic one. I would rather divide "improvement through experience" into selection type (genetic programming etc.) and adaption type (neural nets etc.).
I wonder what a person would do faced with a perfectly horizontally symmetrical scene...move their head and eyes around in an attempt to establish binocular disparity?
Probably. Unless the person is a couch-potato ofcourse:-)
Surprisingly this is something fairly new in Machine Vision. We are actually experiencing one of Thomas Kuhn's famous paradigm shifts today... away from the old Sense->Analyze->Act scheme, and onto Sense->Act->Analyze. If you're interested, do a search on "Active Vision".
and what reason, pray tell, is there for aligning the eyes horozontaly? as opposed to say, 3 'eyes' in a triangle?
My guess is that three cameras will cost you more in computations than you will gain in better estimations.
Besides, there is even a solution to the problem that doesn't involve an extra camera. You could simply look at the frames from a few seconds ago, and compute the disparity using them. The old frames will be displaced along the other dimension if the robot has moved forward.
Genetic programming is much more promising than programmed control, but I wouldn't call it learning. It's a means of finding an algorithm that works without writing it yourself.
If genetic programming was learning, you should also be able to say things like "Mankind has learned to have two arms and two legs.", and to me that sounds wrong... more correct to say that Elvis has evolved to walk.
Slashdot Mirror
One simple, illustrative observation would be that increases in the performance of artifical intelligence have not been described by Moore's Law. Why not? Speculation on the answer could only be informative.
Reminds me of something Robert Heinlein once wrote:
Suppose we had an AI system as smart as a dog and increased it's computational resources a thousand times. What would happen?
We would get a system that in practically no time at all decides to sniff your butt :-)
Me too, but apparently King himself didn't. He even participated in doing a remake some years ago. Never heard of it? Guess that's because it's pretty lame compared to the Kubrick classic.
Although the EU laws cover all of the EU, don't local laws override that?
Unfortunately it's the other way round. i.e. EU laws have precedence.
You're missing like, oh I don't know, about 20 million steps in there.
That's where the principle of natural selection steps in. Variants of the blobs that are better than the average at copying and sustaining themselves gain an advantage. During the copying there are occasional errors that result in mutations. Most of the mutations are likely to be bad, but some are improvements...
Religions are just plain silly, and large swathes of the human population have outgrown them, thankfully.
Unfortunately one belief is often just replaced by another, i.e. religion gets replaced by exaggerated confidence in science or in authorities. Today we also see an increase in "new age" movements, as well as bogus alternative medicine (not all alternative medicine is bogus thoug) that boils down to nothing but placebo. It's a great step forward to rid ourselves from religion, but we still have a long way to go.
Introns are just junk DNA thrown in there. ...They aren't comments ...
In a sense they are, it depends on what you mean by a comment. According to your own statement they work just like chunks of code that is commented out.
640 Neurons is enough for anyone!!
He he.... :-)
but if 640 should become a limitation, couldn't they simply divide?!
Brown, purple and yellow all (can) have complete saturation (e.g. there is no grey in there). Therefore, they can be "pure" colours, consisiting of only one wavelength. The whole world is not an RGB monitor.
Sorry, you're partly right. Yellow is a spectral colour. But, show me a spectrum that contains brown or purple :-) Purple is the text-book example of a colour that does not correspond to one single wavelength.
From what I've read, the colour constancy phenomenon depends both on low-level adaption, and higher cognitive skills, such as remembering colours of objects, inferring that a certain patch of colour is shadowed etc.
Sorry, but I'm afraid I'm far from being an expert on color constancy. It's an interesting phenomenon though :-)
But enough rambling, red is red is red, a specific wavelength on the spectrum. And red is not the experience of red; experiences are finite and distinct, red is not, it is red, as is defined.
There is one slight problem with this definition of red. You cannot tell if something has a certain wavelength unless you use a spectrometer :-)
Since I don't want to buy one I will keep refering to my subjective experience of red as red.
ANYway, colorblindness is just a subjective idea!
To a philosopher maybe. But you could always have your eyes examined and find out that it is for real :-)
The colour sensitive cells in your eyes (red, green and blue cones) can be distinguished nowadays. If you lack one kind of cone you have to be colourblind. There are also objective tests that check if you can distinguish colours, there even was one in the article. If you cannot distinguish between, say red and green, you are per definition colour blind as well.
A distinct type of blue is a distinct wavelength of light.
Nope. A distinct type of blue can correspond to many different reflectance spectra. Reflectance functions that appear the same to us but are different are called metamers.If you have a solid color object, only one wavelength of light will be reflected from that.
Not necessarily. Brown, Purple, Yellow etc are also colours, but they correspond to several wavelengths.
I'd prefer to skip these arbitrary definitions, & just say, for example, "Hey, I really like your lamda = 750 nm dress!"
Hmm... colour is NOT the same as wavelength.
For instance, that dress probably won't reflect just one wavelength, but is more accurately described by a reflectance function of the incoming light spectrum, still we say that it has ONE colour. The reflected wavelengths would also change, for instance when you go indoors/outdoors. So your lambda would change although the dress stays the same. A green dress is always green, and most of the time (due to colour constancy) it also looks green. Colour tells more about an object than the reflected wavelengths do.
It also would then expose the fact that the crude level of security in Unix-like operating systems is inadequate, in that the only information left vulnerable is the only information with value. If the User's data is destroyed, anything else on the hard drive is just the stuff that came off the CD-ROM.
How about the password file? And root access would allow you to do much more damage. Besides, User data is usually backed up anyway....
It's a sad day indeed when the best SciFi force fields reference they can come up with is Star Trek Voyager.
I'm sure there are better references, but none as well known. Little point in making a reference to something most people have never heard of...
But they really should have omitted the "Voyager" part of the name. "The Next Generation" is much better IMHO. Voyager is too much political correctness and Picard is missing :-)
This of course begs the question "Where can I buy a Lamp with LEDs which I can put on my desk?".
They might be here soon. Here in Sweden most traffic lights have been changed to use LED lamps.
Mr. McGuire: "Plasma."
Wasn't that "Plastic" in the original movie? :-)
Actually, the problem you are describing is due to the size of the image, not the resolution.
You cannot say that it's due to the size, and not the resolution, if you are interested in actually viewing the image details. You can easily realize this, if you imagine shrinking an image to a tenth of its original size, while at the same time increasing the resolution tenfoldYes, if we had better resolution displays, a 14" or maybe even a 12" screen could be useful, but you would have to sit closer to it.
If the pixels were drawn on the screen at a higher density than usual, then you would be able to see the whole image at once, without having to move your eyes around.
This would depend very much on what kind of image you are watching. It would certainly have to be a very simple one for this to be true. The human visual system only has sharp vision in 1 deg. of the visual field. This is usually about the size of a single letter in level 1 headings. This, and the limited amount of cones in the fovea limits the amount of detail you can squeeze into your image.And that extra resolution would not be wasted, since your eyes are plenty capable of resolving details at greater than the standard 72dpi or 100 dpi of a monitor (at a distance of about 18" to 20").
OK, I admit that screens a little bit sharper would be beneficial, but we are not far from the limit at present, especially considering that most people don't have perfect vision.
As much as I would like to have better screens, I don't think merely higher resolution will solve the problem of viewing a hi-res image all at once without loosing resolution. In fact, you eye cannot see full detail on all of a 17" screen at once anyway, you have to change gaze (lots of times) to see all details.
GIMP 1.1 has a very nice feature that makes panning of large images a very pleasant experience :-). It gives you a small pop-up showing the entire image at lower resolution, and lets you move around the region of interest in the miniature with the mouse. If you have enough RAM is actually quite quick, even on large images.
BTW I also work with large images: aerial photograph databases which I probably never will be able to see on screen at full detail at once :-)
It's interesting that a blurry (compared to a monitor) tv can be just as effective at smoothing blocky low resolution stuff as the best, most intensive anti-aliasing effects.
Actually anti-aliasing on images often happens after the image has been rendered (for best results it should be applied before), and at that point there is nothing else to do but apply a blur filter... so your observation is probably correct :-)
I think that 'evolving' methodologies, and neural nets etc provide our best chance with this technology. It would be a good idea for the honda team to try a similar approach as Elvis on the control side. Perhaps add padding so it doesn't damage itself when learning to walk.
I agree that evolving, and learning are more promising. I would even say that this is the way we have to go. If we want to develop systems that interact with a complex environment (such as reality) there is simply no way to program for all the situations that can occur.
My objection was merely a semantic one. I would rather divide "improvement through experience" into selection type (genetic programming etc.) and adaption type (neural nets etc.).
Erm... you've probably skipped some of the words I wrote. Or do you plan to use genetic programming on your child?! ;-)
I wonder what a person would do faced with a perfectly horizontally symmetrical scene...move their head and eyes around in an attempt to establish binocular disparity?
Probably. Unless the person is a couch-potato ofcourse :-)
Surprisingly this is something fairly new in Machine Vision. We are actually experiencing one of Thomas Kuhn's famous paradigm shifts today... away from the old Sense->Analyze->Act scheme, and onto Sense->Act->Analyze. If you're interested, do a search on "Active Vision".
and what reason, pray tell, is there for aligning the eyes horozontaly? as opposed to say, 3 'eyes' in a triangle?
My guess is that three cameras will cost you more in computations than you will gain in better estimations.
Besides, there is even a solution to the problem that doesn't involve an extra camera. You could simply look at the frames from a few seconds ago, and compute the disparity using them. The old frames will be displaced along the other dimension if the robot has moved forward.
Genetic programming is much more promising than programmed control, but I wouldn't call it learning. It's a means of finding an algorithm that works without writing it yourself.
If genetic programming was learning, you should also be able to say things like "Mankind has learned to have two arms and two legs.", and to me that sounds wrong... more correct to say that Elvis has evolved to walk.