Artificial eyes do exist, but I don't know where exactly they connect to. Actually primary visual cortex is quite well understood (pray for souls of 100s of macaque monkeys died for the deed, if you are so inclined) and is not too complicated. The eyes send signals quite like an unfocused camera with moderate amount of noise, the first stages of processing also quite nicely coincides with the artifical vision models for robots (contour detection, line detection, movement detection etc. are somewhat modularized and follow basically same route in both cases.) Models for visual functions do exist too, and an interesting property of them is gven the task of seeng, artifical neural networks tend to organize themselves into modules resembling real ones. So sending visual sensetion of a chair would not be too hard tomorrow, to any primary level.
But object recognition (which is a prerequisite for experineces related to vision) is another thing entirely. It is not like that eyes see a chair and somewhere along the line of the visual path, a module become aware of seeing a chair and sending information to later processes "I'm looking at a chair. My experience is seeing a chair." The information is represented in a distributed manner (that is a population of neurons contribute to a representation by assuming different levels of activation for each item, but no single neuron define an object, or an attribute) which makes very hard to link concepts to activities of population of neurons, and impossible to link activities of single neurons. So an experience is not represented anywhere by itself. The transition from sensation to experience is gradual too.
There are also encouraging findings. It seems it is possible to differentiate between a small list of predetermined items a person is seeing/imagining by monitoring activites of relatively small number of neurons. Distinguishing members of a closed set is far from manipulating activites to feed whatever experinece you like. But that is a step in the right direction.
Please don't confuse feeding experiences with feeding information. In the first case, it is sufficient that experience occurs, in the second case it must also make sense to the host, and preserved over time. First is very difficult, second is a few orders of magnitude more difficult. Sort of like the difference between copying a previously unknown ancient glyph to paper and understanding what that glyph means.
Not all cognitive scientists are stuck in 60ties to 80ties paradigm of "brain==computer", "language==computation." Actually, save for cognitive linguistics field where majority is following that idiot "I said so. QED" Chomsky, computer methaphor is usually not taken seriously. Cognitive psychology has quite reliable findings that human mind does not operate like a computer, executing algorithms on data collection, at all.
OTOH neural approaches (artifical models and neurocognition) are plagued by lack of explanatory(sp?) power. I believe the process based explanations, based on experimental data from cogs.psych field, rather than pure theory from phil or cogs.ling fields, will prove to be the most useful findings in future; both by explaning how natural intelligence works, and helping how to build artificial ones.
Even though I like cognitive psychology more, my line of research is artificial life because there are few options here to go for experimental route:(
Nothing is stopping you from augmenting your brain with extra senses or extra limbs. Go ahead, you can do it now, with today's technology. You might not be able to learn to control/make sense of them (since you are old enough to write on slashdot) but then again, you might.
I have been talking about impossibility (or let me sugar coat it, since the idea doesn't seem to be well received: extreme improbability) of symbolic knowledge feeding to/extraction from brain.
I think as humans we greatly underestimate the capibility of our own mind...
Oh, I got it wrong. I thought you were talking about technology. Saying something like "we can invent things that can overcome these problems, although we now have no idea how to attack them. we are clever."
Your greatest flaw is comparing our brain to large chunk of transistors... the human brain has much more capibility than that...
That is my job, in a sense. I'm a cognitive scientist (and a chemical engineer, so what?)
What are the initials for ROTFL in german? Please wait 3 hours (counted from posting of this message) before posting initials for MAO part; that might offend German government, as all of us know very well from the last century's experiences, is not a good idea at all.
Early in life, the brain is highly adaptible. Perhaps we could just decide on a simple protocol, then allow the brain to learn to use it? IIRC, you aren't born knowing exactly what parts of your brain hook up to your eyes, ears, nose, etc.
That is partly true, the brain is not a homogenous lump of neurons even at birth. Cell differentiation is already done, and which parts get which neurons is determined by genome almost exclusivly. Sensory input to the brain is also not to random locations, visual information goes to back side, auditory information goes to middle, inner side etc. But the barin is higly plastic, if in early life (a few weeks at best) the feeding locations are altered, or if they were wrong to begin with, the brain can adopt to changes, usually without any visible consequences. Children born with a whole hemisphere of brain missing can become normal individuals, that is the extend of amazing plasticity of brain.
If you consider it carefully, though, it works against external information feeding. It would be much easier if people had fixed locations for differents functions. Discover once, use for all. Since that is not the case, something must be done either to dechipher individual brains, or as you suggest make different brains look alike. But a single interface would not solve the problem, the interfaces to the external interface would still be higly variable. To solve that you have to standardize whole brain, which means controlling precisely what information goes into brain in early life, from which locations it goes, and no genetic differences in the brain encoding genes. I'm not sure that is a good idea. I would rather have a fool child than an identical child.
I think as humans we greatly underestimate the capibility of our own mind...
With huge amount of data collected from different people, it will be known what parts of cognitive functions share a common basis among individuals, and what parts are highly variable. For example we know that the concept of "red" is perceptually grounded and somewhat constant concept among individuals, we can look for redness encoded in the brain, starting from visual fields and going to deeper parts. But what about "grandmother"? Does its encoding share anything at all among individuals? The answer is probably not.
If it turns out that variables are so much higher in number than constants, symbolic uploading of information into the brain is impossible without a machine that can already do much more than an augmented brain. The first part of this if statement is an open question, second part is a logical consequence. It may turn out that even though people can make super AI machines, they still prefer to do thinking themselves. Then your dream may come true. I wouldn't bother with thinking if something else is doing it provably better than me though.
For example, if someone had a lesion which interfered with their ability to see, presumably, we could eventually "rewire" that part of the visual cortex back up.
An organic replica of the damaged region is probably easier and more compatible. Rewiring brain does not have to be made with perfect accuracy, neurons are capable of learning new connections and destroying unused ones. Just provide appropriate mix of neuron types, and connections to outside, brain will heal itself. Stem cells extracted from adults is the way to go. If that proves unfeasible, other ways of introducing correct type of fresh neurons should be considered before going non-organic.
But a lot of people seem to think this sort of thing would be useful, so what the hey...
If they are fast, who cares what they are made of. I can't think of a reason to prefer an organic computer to a non-organic one, except for coolness rating, if they perform more or less the same. I'd rather halt than kill my computer.
Downloadin and uploading information directly to brain has much bigger hurdles than easily connecting to neurons. Our understanding of brain strongly suggests (but does not prove) that information is encoded in a unique way for each individual. Therefore any attept at uploading data to brain should start with complete decryption of his/her brain structure and information encoding first. This must be done as the brain changes (albeit slowly), and without access to much of the information already encoded (how can you think about everything you know while someone is scanning you?) I would say that it is impossible. Motor and sensory areas may possibly be augmented, but not cognitive functions without hardware so powerful that, existance of it would render the goal of augmenting people's brains useless.
Nobody who can't extract his own would have a use for them until someone writes an Xbox emulator. Even then I guess the emulation can be done with a few device drivers running on nt, instead of writing a whole machine emulation. Bios image may be required in that case too, but probably not.
One of my friends recommended glory season as "Left hand of darkness done right." Nothing could be further from the truth.
The similarity between the two are rather superficial as "glory season" fails to explore prejudices (spelling?) of society and sexual identity matters in any depth. Sure, the conflict is there, but how does brin explore it? Well main characters have to struggle, that is basically it. Poor little swan, but my mom already told me that story when I was 4, sorry Brin.
I'm not much of a fan of Brin, I have to admit. I really love Earth, and most of his short fiction is good. IMHO both uplift series and glory season are overly long, medicore pieces of sci-fi. (yeah, I hate robert jordan too. If your ideas can fill 100 pages, I think 101 pages is much too much.We are lucky that PKD, Heinlein, LeGuin etc. were born long before word processors were invented.)
She is one of my favorite authors, and Left Hand of Darkness is my favorite piece, but I didn't need to say that, did I?
Anyway, my friends usually don't like LeGuin, especially the Earthsea Trilogy (which is another "a trilogy in five parts" now) is bashed all too often. But Dispossesed and Left Hand of Darkness is liked by almost everyone. I strongly suggest these two who have not read any LeGuin yet.
It is also interesting that she can write the most brilliant pieces of science fiction with minimal, almost nonexistant understanding of hard sciences, and without any futuristic stuff.
Not at all, kinematic viscosity of air and consequently the drag force is not enough to carry "sand" aloft, though it can carry very small particles with ease. Mars happens to have a fair amount dust made of small particles (no rain to wash them away), this is what gives foggy and dark looks to Martian storms. It is not like an desert storm on earth.
Ofcourse "no problems except visibility" is a bit off the mark. There would be many problems with loss communication, static, dust seeping in the suits and possibly breathing apparatus, mechanical failures of instruments due to dust accumulation, unavaliability of air transorts etc. Let me put it this way, there is practically no difference between a 200kph martian storm and a 500kph one for someone on Martian surface.
Haven't they had enough problems controlling fairly large machines on that planet?
No they haven't. AFAIK, all crafts that succeded in landing on the surface performed well beyond their design. Mars was russians nightmare, they send countless probes, and none (or a few) got there. States has only two major failures with crafts bound to Mars, it is just that those two were the last two missions to Mars.
Martian storms are frequent, fast and long-lasting (years are not unheard of), but they are not "powerful" in the sense that storms push things around. The reason is air density, it is too low. You can walk in a 500kph martian storm, and have no problems except loss of visibility.
Perhaps this doesn't apply at all to this robots, I mean, how can it fly when its surface is too low compared to its weight such that it is invulnerable to storms?
If you like Mr. Sterling but don't like him so much that you want to spare a good 15 minutes to a pointless monologue, don't read it. Go read something like "hacker crackdown" or "the difference engine." I'm very disappointed, must be one of his worst talks.
Really, I don't think there is much to see there compared to mars, moons of jupiter and saturn. The only reason to go there when there is so much to learn in the near bodies seems to be the fact that pluto is comparitevly near to sun right now, and it won't be there for the next hundered years (I thought a pluto year is much longer, why should a planet that far go that fast?)
Kuiper belt is a lot more interesting though. NASA is downplaying it possibly because they will fit the craft mainly for pluto-charon system and won't be able to do much about the belt.
I don't think a god exists. How arrogant is that? Why is it arrogant when he suggests (maybe not, maybe he just thinks god left evolution alone knowing in advance it would eventually produce humans)/I claim that there is no god but it is perfectly acceptable you claim that there is a God? If you are offended by people expressing different views about religious matters, maybe you should try to convert Osama to christianity. He would be a perfect ally.
Err... why? GP has produced lots of impressive solutions to problems that consist of better code than produced by any human before.
What sort of problems were those? Have they ever been succesful on a task like "give us a good vm program"? The problem of "write a good kernel piece" is not like usual problems thrown at GPs. (At least not to my knowledge, which as you observe, is not much on GP side of the things.) Therefore I assume careful construction of representation and mapping is required to get a useful result (this is why NFL applies.)
Linus's point was that there wasn't much planning involved! I was working from that assumption, though I'm not sure to what degree I agree with him.
I wasn't referring to planning at a meta level (eg.where is linux going?) Rather, I was thinking about planning structures and interfaces. Evolving an algorithm is easy, but interoperation of different algorithms are hard, because it is hard to find an objective function that can differentiate between good and bad interfaces. Maybe this doesn't apply to GPs.
Linux does not evolve like species at all
on
The Evolution of Linux
·
· Score: 2, Interesting
OK, linux is evolving (=changing incrementally) and not controlling tightly how it evolves is a nice idea, but this is how far the analogy goes. Linus is taking the analogy too far and use biological evolution out of its context. People do design pieces of code they submit, linus do control which ones are released in the main tree. Both of these facts, especially the latter one, make evolution of linux fundementally different than natural evolution. If you agree with linus please carefully state what do you agree with. Do you agree that any complicated engineering project can not be designed in advance? Or the fact that linux is not particularly directed to a defined goal is a good thing? Or natural evolution is a proof that not designing linux is a good idea? I agree with first two, but third one is plain wrong.
It would not be possible to evolve chunks bigger than a small function. Even then, it won't
better than human designed code. GAs are not good at "planning" kind of tasks. Actually how succesful such a system would be almost completely determined by genotype->phenotype mapping. Start with bad mapping (such as genotype consists of ascii characters and phenotype is a copy of genotype until first EOF) and you wouldn't even be able to get any compilable code in a lifetime. I can't think of a promising mapping though. You should check "no free lunch theorem."
1- Get latest kernel source
2- Open a random file, go to a random location in that file
3- Roll a d100, use this table:
0-10 insert a random C keyword
11-20 delete nearest statement
21-50 define a new variable
51-80 delete a variable declaration
81-85 change your keyboard layout to some language, switch off monitor and start typing headlines of slashdot. Stop when you feel like it
86-90 delete rand(20) characters
91 delete file
92 merge file with other random
93 copy file with a new name
94 move file to another location
95+ merge file with a random C source from net
4- try building source. If all goes well,submit a patch. Otherwise roll d6, on 1-5 return to step 2, on a 6 return to step 1.
Slashdot Mirror
But object recognition (which is a prerequisite for experineces related to vision) is another thing entirely. It is not like that eyes see a chair and somewhere along the line of the visual path, a module become aware of seeing a chair and sending information to later processes "I'm looking at a chair. My experience is seeing a chair." The information is represented in a distributed manner (that is a population of neurons contribute to a representation by assuming different levels of activation for each item, but no single neuron define an object, or an attribute) which makes very hard to link concepts to activities of population of neurons, and impossible to link activities of single neurons. So an experience is not represented anywhere by itself. The transition from sensation to experience is gradual too.
There are also encouraging findings. It seems it is possible to differentiate between a small list of predetermined items a person is seeing/imagining by monitoring activites of relatively small number of neurons. Distinguishing members of a closed set is far from manipulating activites to feed whatever experinece you like. But that is a step in the right direction.
Please don't confuse feeding experiences with feeding information. In the first case, it is sufficient that experience occurs, in the second case it must also make sense to the host, and preserved over time. First is very difficult, second is a few orders of magnitude more difficult. Sort of like the difference between copying a previously unknown ancient glyph to paper and understanding what that glyph means.
OTOH neural approaches (artifical models and neurocognition) are plagued by lack of explanatory(sp?) power. I believe the process based explanations, based on experimental data from cogs.psych field, rather than pure theory from phil or cogs.ling fields, will prove to be the most useful findings in future; both by explaning how natural intelligence works, and helping how to build artificial ones.
Even though I like cognitive psychology more, my line of research is artificial life because there are few options here to go for experimental route :(
Huh? The data always enters the brain the same way, but once inside, it's up to the individual brain to assimilate the data.
Yes, but I don't remember objecting to that?
I have been talking about impossibility (or let me sugar coat it, since the idea doesn't seem to be well received: extreme improbability) of symbolic knowledge feeding to/extraction from brain.
Oh, I got it wrong. I thought you were talking about technology. Saying something like "we can invent things that can overcome these problems, although we now have no idea how to attack them. we are clever." Your greatest flaw is comparing our brain to large chunk of transistors... the human brain has much more capibility than that...
That is my job, in a sense. I'm a cognitive scientist (and a chemical engineer, so what?)
What are the initials for ROTFL in german? Please wait 3 hours (counted from posting of this message) before posting initials for MAO part; that might offend German government, as all of us know very well from the last century's experiences, is not a good idea at all.
Nobody needs 20 seconds to write that, why doesn't /. complain?
That is partly true, the brain is not a homogenous lump of neurons even at birth. Cell differentiation is already done, and which parts get which neurons is determined by genome almost exclusivly. Sensory input to the brain is also not to random locations, visual information goes to back side, auditory information goes to middle, inner side etc. But the barin is higly plastic, if in early life (a few weeks at best) the feeding locations are altered, or if they were wrong to begin with, the brain can adopt to changes, usually without any visible consequences. Children born with a whole hemisphere of brain missing can become normal individuals, that is the extend of amazing plasticity of brain.
If you consider it carefully, though, it works against external information feeding. It would be much easier if people had fixed locations for differents functions. Discover once, use for all. Since that is not the case, something must be done either to dechipher individual brains, or as you suggest make different brains look alike. But a single interface would not solve the problem, the interfaces to the external interface would still be higly variable. To solve that you have to standardize whole brain, which means controlling precisely what information goes into brain in early life, from which locations it goes, and no genetic differences in the brain encoding genes. I'm not sure that is a good idea. I would rather have a fool child than an identical child.
With huge amount of data collected from different people, it will be known what parts of cognitive functions share a common basis among individuals, and what parts are highly variable. For example we know that the concept of "red" is perceptually grounded and somewhat constant concept among individuals, we can look for redness encoded in the brain, starting from visual fields and going to deeper parts. But what about "grandmother"? Does its encoding share anything at all among individuals? The answer is probably not.
If it turns out that variables are so much higher in number than constants, symbolic uploading of information into the brain is impossible without a machine that can already do much more than an augmented brain. The first part of this if statement is an open question, second part is a logical consequence. It may turn out that even though people can make super AI machines, they still prefer to do thinking themselves. Then your dream may come true. I wouldn't bother with thinking if something else is doing it provably better than me though.
An organic replica of the damaged region is probably easier and more compatible. Rewiring brain does not have to be made with perfect accuracy, neurons are capable of learning new connections and destroying unused ones. Just provide appropriate mix of neuron types, and connections to outside, brain will heal itself. Stem cells extracted from adults is the way to go. If that proves unfeasible, other ways of introducing correct type of fresh neurons should be considered before going non-organic.
But a lot of people seem to think this sort of thing would be useful, so what the hey...
If they are fast, who cares what they are made of. I can't think of a reason to prefer an organic computer to a non-organic one, except for coolness rating, if they perform more or less the same. I'd rather halt than kill my computer.
Downloadin and uploading information directly to brain has much bigger hurdles than easily connecting to neurons. Our understanding of brain strongly suggests (but does not prove) that information is encoded in a unique way for each individual. Therefore any attept at uploading data to brain should start with complete decryption of his/her brain structure and information encoding first. This must be done as the brain changes (albeit slowly), and without access to much of the information already encoded (how can you think about everything you know while someone is scanning you?) I would say that it is impossible. Motor and sensory areas may possibly be augmented, but not cognitive functions without hardware so powerful that, existance of it would render the goal of augmenting people's brains useless.
Nobody who can't extract his own would have a use for them until someone writes an Xbox emulator. Even then I guess the emulation can be done with a few device drivers running on nt, instead of writing a whole machine emulation. Bios image may be required in that case too, but probably not.
The similarity between the two are rather superficial as "glory season" fails to explore prejudices (spelling?) of society and sexual identity matters in any depth. Sure, the conflict is there, but how does brin explore it? Well main characters have to struggle, that is basically it. Poor little swan, but my mom already told me that story when I was 4, sorry Brin.
I'm not much of a fan of Brin, I have to admit. I really love Earth, and most of his short fiction is good. IMHO both uplift series and glory season are overly long, medicore pieces of sci-fi. (yeah, I hate robert jordan too. If your ideas can fill 100 pages, I think 101 pages is much too much.We are lucky that PKD, Heinlein, LeGuin etc. were born long before word processors were invented.)
Anyway, my friends usually don't like LeGuin, especially the Earthsea Trilogy (which is another "a trilogy in five parts" now) is bashed all too often. But Dispossesed and Left Hand of Darkness is liked by almost everyone. I strongly suggest these two who have not read any LeGuin yet.
It is also interesting that she can write the most brilliant pieces of science fiction with minimal, almost nonexistant understanding of hard sciences, and without any futuristic stuff.
Ofcourse "no problems except visibility" is a bit off the mark. There would be many problems with loss communication, static, dust seeping in the suits and possibly breathing apparatus, mechanical failures of instruments due to dust accumulation, unavaliability of air transorts etc. Let me put it this way, there is practically no difference between a 200kph martian storm and a 500kph one for someone on Martian surface.
No they haven't. AFAIK, all crafts that succeded in landing on the surface performed well beyond their design. Mars was russians nightmare, they send countless probes, and none (or a few) got there. States has only two major failures with crafts bound to Mars, it is just that those two were the last two missions to Mars.
Perhaps this doesn't apply at all to this robots, I mean, how can it fly when its surface is too low compared to its weight such that it is invulnerable to storms?
Thanks for ref. I'm not impressed with the results though.
Hey mod, I'm just trying to help my fellow /.ers.
Kuiper belt is a lot more interesting though. NASA is downplaying it possibly because they will fit the craft mainly for pluto-charon system and won't be able to do much about the belt.
I don't think a god exists. How arrogant is that? Why is it arrogant when he suggests (maybe not, maybe he just thinks god left evolution alone knowing in advance it would eventually produce humans)/I claim that there is no god but it is perfectly acceptable you claim that there is a God? If you are offended by people expressing different views about religious matters, maybe you should try to convert Osama to christianity. He would be a perfect ally.
What sort of problems were those? Have they ever been succesful on a task like "give us a good vm program"? The problem of "write a good kernel piece" is not like usual problems thrown at GPs. (At least not to my knowledge, which as you observe, is not much on GP side of the things.) Therefore I assume careful construction of representation and mapping is required to get a useful result (this is why NFL applies.)
Linus's point was that there wasn't much planning involved! I was working from that assumption, though I'm not sure to what degree I agree with him.
I wasn't referring to planning at a meta level (eg .where is linux going?) Rather, I was thinking about planning structures and interfaces. Evolving an algorithm is easy, but interoperation of different algorithms are hard, because it is hard to find an objective function that can differentiate between good and bad interfaces. Maybe this doesn't apply to GPs.
OK, linux is evolving (=changing incrementally) and not controlling tightly how it evolves is a nice idea, but this is how far the analogy goes. Linus is taking the analogy too far and use biological evolution out of its context. People do design pieces of code they submit, linus do control which ones are released in the main tree. Both of these facts, especially the latter one, make evolution of linux fundementally different than natural evolution. If you agree with linus please carefully state what do you agree with. Do you agree that any complicated engineering project can not be designed in advance? Or the fact that linux is not particularly directed to a defined goal is a good thing? Or natural evolution is a proof that not designing linux is a good idea? I agree with first two, but third one is plain wrong.
It would not be possible to evolve chunks bigger than a small function. Even then, it won't better than human designed code. GAs are not good at "planning" kind of tasks. Actually how succesful such a system would be almost completely determined by genotype->phenotype mapping. Start with bad mapping (such as genotype consists of ascii characters and phenotype is a copy of genotype until first EOF) and you wouldn't even be able to get any compilable code in a lifetime. I can't think of a promising mapping though. You should check "no free lunch theorem."
1- Get latest kernel source 2- Open a random file, go to a random location in that file 3- Roll a d100, use this table: 0-10 insert a random C keyword 11-20 delete nearest statement 21-50 define a new variable 51-80 delete a variable declaration 81-85 change your keyboard layout to some language, switch off monitor and start typing headlines of slashdot. Stop when you feel like it 86-90 delete rand(20) characters 91 delete file 92 merge file with other random 93 copy file with a new name 94 move file to another location 95+ merge file with a random C source from net 4- try building source. If all goes well,submit a patch. Otherwise roll d6, on 1-5 return to step 2, on a 6 return to step 1.