You might look into using something like the Pareto distribution or Zipf's Law for estimating the probability of low probability, high-magnitude loads on the system. A little math like this in conjunction with a limited sample size of observed input load levels will help you guesstimate the 99.9999...% level of loading.
I find that randomness helps me enjoy songs for a greater number of plays -- I don't get sick of songs as quickly when they are decontextualized. In album format, each track prompts too much memory of the succeeding tracks. And if the album has "bad" songs, then I find the memory of the bad song taints my enjoyment of the preceding song.
I'm sure music people don't want tactics that increase the number of enjoyable plays. Its in the music industry's interests for customers to become tired of the music so people go buy more.
Conformal change vibration and resonance analysis
on
The Sound of Cells
·
· Score: 5, Insightful
Whenever a protein or enzyme in a cell changes shape, it should induce characteristic vibrations in the surround media. Each enzyme would emit its own characteristic vibrations when it undergoes a change in shape as it catalyzes a reaction or does its business.
For example, I'd bet nerve cells give off sounds as the propagating impulse causes cell-surfane ion channels to pop open and closed. The ion pumps that restore ion concentrations would also emit a hum with characteristic frequencies. For membrane-embedded enzymes (e.g., the channels on nerve cells), interferometry off the membrane surface might help to detect these minute vibrations. I wonder if one could even detect the sound of prions forming when a protein is warped into the misshaped conformation that characterizes conditions like BSE -- sound of a brain going mad.
I'd bet that one could also analyze protein/enzyme states with a fine-grained analysis of the sound transfer function for a cell. Depending on the physical state of each protein species and its concentration, a cell would attenuate or resonate with particular acoustic frequencies. Large cell structures (e.g. mitochondria) might also have their own characteristic acoustical modulation functions that depend on the size and membrane structure. If analyzing the transfer function for a live, wet cell is too hard, I suspect that flash-freezing the cell might create a better acoustical specimen.
IANAL, but I suspect that if you intentionally demonstrate the insecurity of the system, you will be sent to jail. Ask a lawyer, but I suspect that their advice wil be to not do anything that involves you breaking into the system.
On the otherhand, until somebody at the school gets their identity stolen AND they can prove the school was at fault, nothing will change.
At most, I would document the problem WITHOUT breaking any laws (again IANAL). Even documenting the problem that might get you in hot water for the terrorist crime of "hacking."
This sounds exactly like what M$ (motto: "All your devices are belong to us") is trying to do - PCs, office software, servers, enterprise software, XBox, PocketPC, media formats, online music sales, tablet computers, MSN, etc. I wonder who will win the interface definition standardization game? A bunch of really smart people at MIT or an even larger bunch of better funded smart people at Microsoft? (Note: at $6 billion dollars, Microsoft's R&D department has more than 4 times the money of ALL of MIT.)
If you have multiple samples of the same species of beasty in a uniform chunk of rock, then you could reconstruct them by mapping homologous points between the beasties and finding the best-fit deformation that brings all the specimens into congruence. IANAG, but I suspect that the basic fossil deformation equation has only 4 relevant unknowns in it (magnitude of flattening, direction of flattening, magnitude of shear, and direction of shear). A fifth unknown, overall shrinkage is irrelevant to reconstructing shape and would be difficult to estimate without a yardstick -- some fossil with a known or incompressible size. (Note: more complex rock deformations might have 8 or 9 unknowns). The point is that cross mapping a handful of points across a handful of fossils lets you find the best fit values for the unknowns and then undeform the fossils.
For intra-fossil deformations, I see two possible solutions. First, one could create a more complex model of the deformation that breaks up the fossil into polyhedral voxels and assumes that soft and hard tissue deformations follow certain invariants across sample fossils. A second solution is to study the microstructure of the rock matrix. If parts of the fossil deform differentially, I would expect that to leave some evidence in the spacing and orientation of any grains in the matrix - the grains would be looser and show evidence of flowing into the area around zones that shrank more (e.g., decayed soft tissues).
Why not make the prices fully variable and a function of the rate of downloading. All music would start at 0.99 per song. If the rate of downloading is high, the price would creep upwards until the rate of downloading slows. If the rate of downloading is low, the price would subside. Maybe the good songs are worth 2.99, maybe the sucky one are worth only 0.25 -- let the rate of downloading set the price.
And if you really want to use a market mechanism, then let people put in bids. When the price of the song drops to the bid price, the bidder gets the song. If the bidder wants the song sooner, then they will have to up their bid.
I'm sure that blue LEDs will fade in time. They were cool because they were new and rare. But novelty, by defintion, cannot last. Just wait a few years and everyone will think that blue LED are just so so early 2000's.
Of course, by then we'll have some other over-used new display technology. Perhaps consumer electronics makers will use OLEDs to form a glowing full-color brand name logos. Then the space around our desktops and dens will look like a miniture cityscape with tiny glowing neonesque billboards for all the brands that we buy.
Oh, and wait 20-40 years and blue LEDs will be back as a retro fad. The aging youth of today will look back to this time and will revel in the glory days when devices only had a single simple little blue light.
Although I sympathize with the protester's motives, I'm not sure how wise a website blackout is in the long-run. Were I an enterprise IT manager, I would not want my suppliers to be taking denial of service actions based on political issues that I don't necessarily care that much about. Blackouts will damage OSS' reputation with businesses.
If KDE, GNU, and Gimp want enterprise adoption, they would do well to maintain 100% uptime -- appearing to be reliable business-like providers of quality software.
I certainly didn't want to make it sound easy. That is why I said _if_ we could teach computers the basics.
Actually, you and I agree on this, sorry if my responding post seemed like an attack. The point of my orginal post was to mark the vast difference between chess and the real world. Thus, a deeper understanding of chess may not actually be as useful as we think because chess lacks so many of the features that makes real-world decision making so hard.
Obviously discovering just what the "basics" are and getting a computer to work by those rules would be very difficult, but ultimatly I'd think it would work.
Good point. But perhaps the real problem is that the rules are not immutable on two levels. First the rules or contraints of the world can change. Second, the rules for how to think might need changing depending on the ratios of various external and internal variables (e.g., the ratio of the rate at which the world changes vs. the rate of instruction execution).
In contrast, a chess program never has to worry about the rules changing and that leads to highly specialized rigid "thought
processes" in the top chess programs. Decision making where the rules or structure of the constraints can change is very different than decision making under fixed-rules. An open, changing world implies expending resources to map the boundaries of the rules/constraints and occssionally re-exploring those limits. Yet this exploration process must be bounded by fear of the unknown unknowns -- how would you play chess if it were possible (with unknown probability) that moving the Queen backward results in an automatic forfeit or where the opponent might suddenly introduce a new-improved Knight that can go up to four-steps forward and up to 2 steps over or where a third opponent could appear on the board or where you own pieces could mutiny?
As for the problem of decompositional limits, making assumptions, etc... This is what humans do all the time. We break things down until we can understand.
This, too, is interesting because it is not clear just how well we humans do this. The fact that so few people are adept at science and engineering suggests that most people can't do this. Just as an airplane does not emulate a bird, perhaps computers should not emulate humans (not the we understand human thought well enough to even emulate).
I do think, like you, that we can create computers that think. I only argue that we need to consider how machines should think in an open, mutable (not-chess-like) world if we are to avoid brittle machine intelligences.
Basically everything in life is just a series of much smaller problems, requiring a finite number of operations.
Well, no. Real world situations tend to have nasty nonlinear coupling -- you hit a decompostional limit that forces to either deal with the whole big system, make assumptions that discard parts of the problem, or use iterative approaches that may not converge. For example the N-body gravity problem cannot be accurately reduced to a set of 2-body problems. The fact that so many human decisions have unintended consequences also illustrates this fact well.
If we could teach computers the "basics", they should be able to handle any situation on their own.
Easier said than done. When I did work on sensor management, the decision making problems were often deeply intractable because of second-order uncertainties. Not only were we not sure whether we had detected or classified an object correctly, but we had no easy way for computing the probabilities of detection/classification because the sensor and environment were not well characterized and not immediately accessible to us. Although we could easily create decision trees for making decisions, we could not easily populate the model with accurate values for the probabilities of the branches.
On the one hand, chess is a very interesting realm for understanding the realms of human and machine intelligence. It is an interesting microworld with enough complexity that it lacks brute force or close-form solutions. Thus it provides a useful test case for understanding rational action. And blitz chess is useful for looking at reasoning under time constraints.
On the other hand, chess is closed - a King will always be limited to moving one square in any direction. With chess, no new moves, new pieces, new board locations can ever appear. Chess is also certain -- there are no ambiguites in the locations of the pieces. With chess the rules and positions are fully known before hand by the exactly two players who adhere to the constraints of the game.
By contrast, the field of human affairs evolves continuously to create new scenarios, new possible movements, new roles, and new players. Everyday slashdot has articles about the novel activities of people (from scammers using TTY relays to new chipsets to new laws). I would argue that decision making under conditions that are uncertain, open-ended, massively multiplayer, and subject to changes in the rules are a bit different.
They say one must learn to crawl before learning to walk. In some ways, learning about the intelligence required to play chess is like learning to crawl. That even the decision making underpinnings of playing chess is so hard to understand says something about how hard it will be to understand true intelligence in open-ended situations the poeple deal with every day.
finite TTL
An exceelent point, if you can't crack the code with the TTL, there is no point in trying.
Given that, there are problems that, although possible to solve in theory, are impossible in practice - take, for example, the problem of playing perfect chess by fully computing the game tree. With enough bits encryption becomes unbruteforcable too.
The problem with that logic is that it assume brute force attack -- that a chess program must compute all branches of the game tree or that a cracker must compute and try all possible key values. Under that assumption, adding a 1 bit to the key doubles the work required.
In the true brute force case, the time required for cracking is C*2^N, where N is the number of bits and C is the time required to try one key value. Faster computers reduce the value of C. But this only means we have to add 1 more bit every 18 months to keep up with Moore's Law. Distributed cracking networks also reduce C by an amount roughly related to the number of computers (M) in the network. Combating this means we need to add log2(M) bits to lengthen the cracking time back to greater than the TTL. For brute force, it does not take many more bits to make the problem uncrackable.
But the theory used for cracking code implies that the problem has more structure than a naive brute-force cracker might think. Thus, the decrypter only need to try a small subset of possible keys.
So, instead of the cracking time being t = C*2^N, we have t = C*B^N, where 1 is less than B is less than or equal to 2. If B is small, then we need to add many more bits to regain practical uncrackability (t > TTL). For example if B=1.1, we need to add 7 bits to the key every 18 months to redouble the cracking effort. And if B=1.01, we need to add 69 bits to the key every 18 months to redouble the cracking effort.
What makes all of these cryptosystems insecure in a deep sense is that we have no gaurantee that B has some lower bound with which to define a secure level of N. It does not take much of a change in B to make cracking much much easier. For the case of the 109-bit code discussed here, get a B =1.8 makes the cracking problem 97177 times easier. And a B=1.7 makes cracking 49 million times faster. The biggest issue is that we have no idea of the value of C or B that the NSA has achieved. Until there is a crytosystem with a proven lower bound for B, at least, we are vulnerable.
The goal of the government in selling rights to pollute or log is more than just a matter of granting the right to do whatever the auction-winner wishes to do. The goal of the government, in part, is to encourage economic activity that creates jobs, exportable goods, and additional tax revenues. If someone buys the right to pollute, mine, or log, but does not use it, they are , at some level, not compensating the government and public for the full impact of their withholding of that resource from economic use.
The implicit social contract is that the buyer will exercise these rights for an economic gain that benefits others too. Its analogous to the platform ecosystem business model -- you have a platform that others can create products around. You sell access to the platform but let entrants extract value too. The goal of the creating platform or in auction public resources is to enlarge the economic pie for all.
One solution might be to limit the term of the right. Rather than granting in-perpetuity ownership to a pollution right or old-growth forest logging right, the term would be limited to some reasonable length of time. For instance, five years might be sufficient time to encourage peope to buy the right and make the needed invetsment to use the right. Every 5 years, that right would be reauctioned. This ensures that one group or company can't lock-in and inefficiently use these rights. If the former owner is not making money off the right they won't have money to buy the next 5 years worth. If another group has a better use, then they can take over for a better price.
I wonder if writers will start using obscure words and literary allusions in order to confuse these ad-words systems (And Google's GMail). Deft use of langugage should help both elevate reader's vocabulary and muddle the automated systems.
Time to go see if the Amazon ranking for Thesauri are up.
One day in class, the wacky department head at the engineering school I went to told us a little story from his youth. He said that while in chemistry class he discovered that if you dip a nickel (a 5 cent piece for international/.ers) into a pool of mercury it gets very shiny. And if you put it in your mouth, it tastes funny! That story definitely explained why the fellow was almost as mad as a hatter.
Its possible that chimp langauges might include phonetic variations that will be impossible for adult humans to hear. For example, some human languages (Navaho is one IIRC) involve phonemes that must be learned in infancy - if one doesn't hear these sounds while the brain is plastic, one never can learn these sounds. Once a baby is older than 18 months, they lose the ability to hear the differences in phonemes. The same could be true with chimps.
We adults may not even be hearing the differences in all the sounds that chimps can make (and mean). And I doubt anyone is going to let a human infant be raised by chimps to properly learn their language.
If the WTO wants Americans to gamble onlne, I wonder if they will insist that Americans have the right to buy from international image vendors of their choice too?
This might help solve the problem of getting patches on to a recently reformatted machine that is vulnerable. Rather than connect the unpatched machine to the internet, you go to CompUSA to get a $9.95 patch disk that fixes known exploits of the OS.
It would solve the chicken and egg problem -- can't get the patches without going online, shouldn't attach an unpatched machine to the internet.
In the open scenes of Ringworld, Louis Wu travels around the Earth for his 200th birthday -- using transporter booths to jump to the next timezone and have a 48-hour long birthday party. In the very rare first edition of the book, he travels from West to East, which is the wrong direction. Later versions corrected this.
To get a full charge in 30 seconds, you would need a charging current of 600 amps (!!)
But those numbers came out of my ass. We need real values...
Yes, the recharge current for a 5 Ahr battery would be at least 600 Amps. If the laptop battery runs at 14 volts, that means that one would need at least a 8400 watt recharger - a solid 70 Amps on a 120 AC circuit.
As for heat, its more likely that the battery will dissipate a percentage of the input as heat. My understanding is that batteries are only about 80% efficient during a recharge -- suggesting that the fast-charge batteries will dissipate at least 1680 watts. In reality, cell resistance will make this even worse.
Ultra fast charging is a nice idea, but, except with very small batteries, this is not at all practical.
Is the law still working for he amount of power used? I understand that the newest chips do use more power, but shouldn't that be the approach.
Good point, but there are other factors at work. What the article ignores is that the current law is still working fine. The amount of power used by an x-MHz chips has been steadily declining. I remember when a machine with a 16 MHz processor needed a 200W power supply -- now we have machines that are orders of magnitude faster and still use a 200W power supply.
What has changed is that people insist on having desktop power in a laptop form factor. Another change has been a move to architectures that do more per clock cycle -- old processors often took 4 or more cycles per instruction. New processors, using pipelined and superscalar architectures, dispatch and retire multiple instructions per clock cycle. Technologies like hyperthreading boosts utilization of the CPU's logical units and thus increase power. Modern CPUs draw more power because they do more per cycle.
Moore's Law is fine. That mobile devices lag desktop devices in performance (or suffer from poor battery life) has nothing to do with Moore. The computing power of a 1 watt processor continues its steayd rise.
Why do I get the feeling that this turns rich web pages into bit-size powerpoint bullets? (Confession, I have no windows machines so have no way of testing this thing). Maybe they will create a version that converts webpages into Flash animations -- showing... you... one... word... at... a.... time.
On the other hand, this type of content decompostion technology highlights the superiority of markup langages (e.g., HTML) over page layout languages (e.g. PDF). HTML retains more of the meaning of the content while PDF is basically a fancy way of converting content into a screenshot. Try extracting sentences from a PDF, what a PITA.
That's too narrow definition of chaotic system, because Lyaponov coeefficients and strange attractors realted only to dynamical systems which have a topology - that is some underlying continuity.
Good point. Most of the chaos-control research that I have seen focuses on physical/dynamic systems.
discrete objects like cellular automata, which have no notion of divergence
Yes and no. With CA's the divergence can be expressed in terms of the state difference between initially similar configurations. (XOR and count if the CA is binary). Also, CAs are actually broader than chaotic systems considering that only 1 of the 4 Wolfram classes of CA is characterized by chaotic behavior.
The real killer app will be when Home & Garden's magazine zooms in on your home and analyzes your landscaping and house. Different people might get different covers and articles on rejuvenating dead lawns, trimming overgrown trees, or xeriscaping. You might even discover you've won the contest for most beautiful garden with an aerial view.
And they could even analyze your house & land for marketing opportunities. If the satellite veiw is oblique and the paint is peeling, they could forward your name to the local aluminum siding company or house painters.
Slashdot Mirror
You might look into using something like the Pareto distribution or Zipf's Law for estimating the probability of low probability, high-magnitude loads on the system. A little math like this in conjunction with a limited sample size of observed input load levels will help you guesstimate the 99.9999...% level of loading.
I find that randomness helps me enjoy songs for a greater number of plays -- I don't get sick of songs as quickly when they are decontextualized. In album format, each track prompts too much memory of the succeeding tracks. And if the album has "bad" songs, then I find the memory of the bad song taints my enjoyment of the preceding song.
I'm sure music people don't want tactics that increase the number of enjoyable plays. Its in the music industry's interests for customers to become tired of the music so people go buy more.
Whenever a protein or enzyme in a cell changes shape, it should induce characteristic vibrations in the surround media. Each enzyme would emit its own characteristic vibrations when it undergoes a change in shape as it catalyzes a reaction or does its business.
For example, I'd bet nerve cells give off sounds as the propagating impulse causes cell-surfane ion channels to pop open and closed. The ion pumps that restore ion concentrations would also emit a hum with characteristic frequencies. For membrane-embedded enzymes (e.g., the channels on nerve cells), interferometry off the membrane surface might help to detect these minute vibrations. I wonder if one could even detect the sound of prions forming when a protein is warped into the misshaped conformation that characterizes conditions like BSE -- sound of a brain going mad.
I'd bet that one could also analyze protein/enzyme states with a fine-grained analysis of the sound transfer function for a cell. Depending on the physical state of each protein species and its concentration, a cell would attenuate or resonate with particular acoustic frequencies. Large cell structures (e.g. mitochondria) might also have their own characteristic acoustical modulation functions that depend on the size and membrane structure. If analyzing the transfer function for a live, wet cell is too hard, I suspect that flash-freezing the cell might create a better acoustical specimen.
IANAL, but I suspect that if you intentionally demonstrate the insecurity of the system, you will be sent to jail. Ask a lawyer, but I suspect that their advice wil be to not do anything that involves you breaking into the system.
On the otherhand, until somebody at the school gets their identity stolen AND they can prove the school was at fault, nothing will change.
At most, I would document the problem WITHOUT breaking any laws (again IANAL). Even documenting the problem that might get you in hot water for the terrorist crime of "hacking."
I feel for you. Be careful.
This sounds exactly like what M$ (motto: "All your devices are belong to us") is trying to do - PCs, office software, servers, enterprise software, XBox, PocketPC, media formats, online music sales, tablet computers, MSN, etc. I wonder who will win the interface definition standardization game? A bunch of really smart people at MIT or an even larger bunch of better funded smart people at Microsoft? (Note: at $6 billion dollars, Microsoft's R&D department has more than 4 times the money of ALL of MIT.)
Can me bitter, but I fear that with billion in R&D and hundreds of millions of dollars for marketing, M$ will win this game unless they commit suicide.
If you have multiple samples of the same species of beasty in a uniform chunk of rock, then you could reconstruct them by mapping homologous points between the beasties and finding the best-fit deformation that brings all the specimens into congruence. IANAG, but I suspect that the basic fossil deformation equation has only 4 relevant unknowns in it (magnitude of flattening, direction of flattening, magnitude of shear, and direction of shear). A fifth unknown, overall shrinkage is irrelevant to reconstructing shape and would be difficult to estimate without a yardstick -- some fossil with a known or incompressible size. (Note: more complex rock deformations might have 8 or 9 unknowns). The point is that cross mapping a handful of points across a handful of fossils lets you find the best fit values for the unknowns and then undeform the fossils.
For intra-fossil deformations, I see two possible solutions. First, one could create a more complex model of the deformation that breaks up the fossil into polyhedral voxels and assumes that soft and hard tissue deformations follow certain invariants across sample fossils. A second solution is to study the microstructure of the rock matrix. If parts of the fossil deform differentially, I would expect that to leave some evidence in the spacing and orientation of any grains in the matrix - the grains would be looser and show evidence of flowing into the area around zones that shrank more (e.g., decayed soft tissues).
Why not make the prices fully variable and a function of the rate of downloading. All music would start at 0.99 per song. If the rate of downloading is high, the price would creep upwards until the rate of downloading slows. If the rate of downloading is low, the price would subside. Maybe the good songs are worth 2.99, maybe the sucky one are worth only 0.25 -- let the rate of downloading set the price.
And if you really want to use a market mechanism, then let people put in bids. When the price of the song drops to the bid price, the bidder gets the song. If the bidder wants the song sooner, then they will have to up their bid.
I'm sure that blue LEDs will fade in time. They were cool because they were new and rare. But novelty, by defintion, cannot last. Just wait a few years and everyone will think that blue LED are just so so early 2000's.
Of course, by then we'll have some other over-used new display technology. Perhaps consumer electronics makers will use OLEDs to form a glowing full-color brand name logos. Then the space around our desktops and dens will look like a miniture cityscape with tiny glowing neonesque billboards for all the brands that we buy.
Oh, and wait 20-40 years and blue LEDs will be back as a retro fad. The aging youth of today will look back to this time and will revel in the glory days when devices only had a single simple little blue light.
Although I sympathize with the protester's motives, I'm not sure how wise a website blackout is in the long-run. Were I an enterprise IT manager, I would not want my suppliers to be taking denial of service actions based on political issues that I don't necessarily care that much about. Blackouts will damage OSS' reputation with businesses.
If KDE, GNU, and Gimp want enterprise adoption, they would do well to maintain 100% uptime -- appearing to be reliable business-like providers of quality software.
I certainly didn't want to make it sound easy. That is why I said _if_ we could teach computers the basics.
Actually, you and I agree on this, sorry if my responding post seemed like an attack. The point of my orginal post was to mark the vast difference between chess and the real world. Thus, a deeper understanding of chess may not actually be as useful as we think because chess lacks so many of the features that makes real-world decision making so hard.
Obviously discovering just what the "basics" are and getting a computer to work by those rules would be very difficult, but ultimatly I'd think it would work.
Good point. But perhaps the real problem is that the rules are not immutable on two levels. First the rules or contraints of the world can change. Second, the rules for how to think might need changing depending on the ratios of various external and internal variables (e.g., the ratio of the rate at which the world changes vs. the rate of instruction execution).
In contrast, a chess program never has to worry about the rules changing and that leads to highly specialized rigid "thought processes" in the top chess programs. Decision making where the rules or structure of the constraints can change is very different than decision making under fixed-rules. An open, changing world implies expending resources to map the boundaries of the rules/constraints and occssionally re-exploring those limits. Yet this exploration process must be bounded by fear of the unknown unknowns -- how would you play chess if it were possible (with unknown probability) that moving the Queen backward results in an automatic forfeit or where the opponent might suddenly introduce a new-improved Knight that can go up to four-steps forward and up to 2 steps over or where a third opponent could appear on the board or where you own pieces could mutiny?
As for the problem of decompositional limits, making assumptions, etc... This is what humans do all the time. We break things down until we can understand.
This, too, is interesting because it is not clear just how well we humans do this. The fact that so few people are adept at science and engineering suggests that most people can't do this. Just as an airplane does not emulate a bird, perhaps computers should not emulate humans (not the we understand human thought well enough to even emulate).
I do think, like you, that we can create computers that think. I only argue that we need to consider how machines should think in an open, mutable (not-chess-like) world if we are to avoid brittle machine intelligences.
Basically everything in life is just a series of much smaller problems, requiring a finite number of operations.
Well, no. Real world situations tend to have nasty nonlinear coupling -- you hit a decompostional limit that forces to either deal with the whole big system, make assumptions that discard parts of the problem, or use iterative approaches that may not converge. For example the N-body gravity problem cannot be accurately reduced to a set of 2-body problems. The fact that so many human decisions have unintended consequences also illustrates this fact well.
If we could teach computers the "basics", they should be able to handle any situation on their own.
Easier said than done. When I did work on sensor management, the decision making problems were often deeply intractable because of second-order uncertainties. Not only were we not sure whether we had detected or classified an object correctly, but we had no easy way for computing the probabilities of detection/classification because the sensor and environment were not well characterized and not immediately accessible to us. Although we could easily create decision trees for making decisions, we could not easily populate the model with accurate values for the probabilities of the branches.
On the one hand, chess is a very interesting realm for understanding the realms of human and machine intelligence. It is an interesting microworld with enough complexity that it lacks brute force or close-form solutions. Thus it provides a useful test case for understanding rational action. And blitz chess is useful for looking at reasoning under time constraints.
On the other hand, chess is closed - a King will always be limited to moving one square in any direction. With chess, no new moves, new pieces, new board locations can ever appear. Chess is also certain -- there are no ambiguites in the locations of the pieces. With chess the rules and positions are fully known before hand by the exactly two players who adhere to the constraints of the game.
By contrast, the field of human affairs evolves continuously to create new scenarios, new possible movements, new roles, and new players. Everyday slashdot has articles about the novel activities of people (from scammers using TTY relays to new chipsets to new laws). I would argue that decision making under conditions that are uncertain, open-ended, massively multiplayer, and subject to changes in the rules are a bit different.
They say one must learn to crawl before learning to walk. In some ways, learning about the intelligence required to play chess is like learning to crawl. That even the decision making underpinnings of playing chess is so hard to understand says something about how hard it will be to understand true intelligence in open-ended situations the poeple deal with every day.
finite TTL An exceelent point, if you can't crack the code with the TTL, there is no point in trying. Given that, there are problems that, although possible to solve in theory, are impossible in practice - take, for example, the problem of playing perfect chess by fully computing the game tree. With enough bits encryption becomes unbruteforcable too.
The problem with that logic is that it assume brute force attack -- that a chess program must compute all branches of the game tree or that a cracker must compute and try all possible key values. Under that assumption, adding a 1 bit to the key doubles the work required.
In the true brute force case, the time required for cracking is C*2^N, where N is the number of bits and C is the time required to try one key value. Faster computers reduce the value of C. But this only means we have to add 1 more bit every 18 months to keep up with Moore's Law. Distributed cracking networks also reduce C by an amount roughly related to the number of computers (M) in the network. Combating this means we need to add log2(M) bits to lengthen the cracking time back to greater than the TTL. For brute force, it does not take many more bits to make the problem uncrackable.
But the theory used for cracking code implies that the problem has more structure than a naive brute-force cracker might think. Thus, the decrypter only need to try a small subset of possible keys.
So, instead of the cracking time being t = C*2^N, we have t = C*B^N, where 1 is less than B is less than or equal to 2. If B is small, then we need to add many more bits to regain practical uncrackability (t > TTL). For example if B=1.1, we need to add 7 bits to the key every 18 months to redouble the cracking effort. And if B=1.01, we need to add 69 bits to the key every 18 months to redouble the cracking effort.
What makes all of these cryptosystems insecure in a deep sense is that we have no gaurantee that B has some lower bound with which to define a secure level of N. It does not take much of a change in B to make cracking much much easier. For the case of the 109-bit code discussed here, get a B =1.8 makes the cracking problem 97177 times easier. And a B=1.7 makes cracking 49 million times faster. The biggest issue is that we have no idea of the value of C or B that the NSA has achieved. Until there is a crytosystem with a proven lower bound for B, at least, we are vulnerable.
The goal of the government in selling rights to pollute or log is more than just a matter of granting the right to do whatever the auction-winner wishes to do. The goal of the government, in part, is to encourage economic activity that creates jobs, exportable goods, and additional tax revenues. If someone buys the right to pollute, mine, or log, but does not use it, they are , at some level, not compensating the government and public for the full impact of their withholding of that resource from economic use.
The implicit social contract is that the buyer will exercise these rights for an economic gain that benefits others too. Its analogous to the platform ecosystem business model -- you have a platform that others can create products around. You sell access to the platform but let entrants extract value too. The goal of the creating platform or in auction public resources is to enlarge the economic pie for all.
One solution might be to limit the term of the right. Rather than granting in-perpetuity ownership to a pollution right or old-growth forest logging right, the term would be limited to some reasonable length of time. For instance, five years might be sufficient time to encourage peope to buy the right and make the needed invetsment to use the right. Every 5 years, that right would be reauctioned. This ensures that one group or company can't lock-in and inefficiently use these rights. If the former owner is not making money off the right they won't have money to buy the next 5 years worth. If another group has a better use, then they can take over for a better price.
I wonder if writers will start using obscure words and literary allusions in order to confuse these ad-words systems (And Google's GMail). Deft use of langugage should help both elevate reader's vocabulary and muddle the automated systems.
Time to go see if the Amazon ranking for Thesauri are up.
One day in class, the wacky department head at the engineering school I went to told us a little story from his youth. He said that while in chemistry class he discovered that if you dip a nickel (a 5 cent piece for international /.ers) into a pool of mercury it gets very shiny. And if you put it in your mouth, it tastes funny! That story definitely explained why the fellow was almost as mad as a hatter.
Its possible that chimp langauges might include phonetic variations that will be impossible for adult humans to hear. For example, some human languages (Navaho is one IIRC) involve phonemes that must be learned in infancy - if one doesn't hear these sounds while the brain is plastic, one never can learn these sounds. Once a baby is older than 18 months, they lose the ability to hear the differences in phonemes. The same could be true with chimps.
We adults may not even be hearing the differences in all the sounds that chimps can make (and mean). And I doubt anyone is going to let a human infant be raised by chimps to properly learn their language.
If the WTO wants Americans to gamble onlne, I wonder if they will insist that Americans have the right to buy from international image vendors of their choice too?
This might help solve the problem of getting patches on to a recently reformatted machine that is vulnerable. Rather than connect the unpatched machine to the internet, you go to CompUSA to get a $9.95 patch disk that fixes known exploits of the OS.
It would solve the chicken and egg problem -- can't get the patches without going online, shouldn't attach an unpatched machine to the internet.
In the open scenes of Ringworld, Louis Wu travels around the Earth for his 200th birthday -- using transporter booths to jump to the next timezone and have a 48-hour long birthday party. In the very rare first edition of the book, he travels from West to East, which is the wrong direction. Later versions corrected this.
To get a full charge in 30 seconds, you would need a charging current of 600 amps (!!)
But those numbers came out of my ass. We need real values...
Yes, the recharge current for a 5 Ahr battery would be at least 600 Amps. If the laptop battery runs at 14 volts, that means that one would need at least a 8400 watt recharger - a solid 70 Amps on a 120 AC circuit.
As for heat, its more likely that the battery will dissipate a percentage of the input as heat. My understanding is that batteries are only about 80% efficient during a recharge -- suggesting that the fast-charge batteries will dissipate at least 1680 watts. In reality, cell resistance will make this even worse.
Ultra fast charging is a nice idea, but, except with very small batteries, this is not at all practical.
Is the law still working for he amount of power used? I understand that the newest chips do use more power, but shouldn't that be the approach.
Good point, but there are other factors at work. What the article ignores is that the current law is still working fine. The amount of power used by an x-MHz chips has been steadily declining. I remember when a machine with a 16 MHz processor needed a 200W power supply -- now we have machines that are orders of magnitude faster and still use a 200W power supply.
What has changed is that people insist on having desktop power in a laptop form factor. Another change has been a move to architectures that do more per clock cycle -- old processors often took 4 or more cycles per instruction. New processors, using pipelined and superscalar architectures, dispatch and retire multiple instructions per clock cycle. Technologies like hyperthreading boosts utilization of the CPU's logical units and thus increase power. Modern CPUs draw more power because they do more per cycle.
Moore's Law is fine. That mobile devices lag desktop devices in performance (or suffer from poor battery life) has nothing to do with Moore. The computing power of a 1 watt processor continues its steayd rise.
Why do I get the feeling that this turns rich web pages into bit-size powerpoint bullets? (Confession, I have no windows machines so have no way of testing this thing). Maybe they will create a version that converts webpages into Flash animations -- showing ... you ... one ... word ... at ... a .... time.
On the other hand, this type of content decompostion technology highlights the superiority of markup langages (e.g., HTML) over page layout languages (e.g. PDF). HTML retains more of the meaning of the content while PDF is basically a fancy way of converting content into a screenshot. Try extracting sentences from a PDF, what a PITA.
That's too narrow definition of chaotic system, because Lyaponov coeefficients and strange attractors realted only to dynamical systems which have a topology - that is some underlying continuity.
Good point. Most of the chaos-control research that I have seen focuses on physical/dynamic systems.
discrete objects like cellular automata, which have no notion of divergence
Yes and no. With CA's the divergence can be expressed in terms of the state difference between initially similar configurations. (XOR and count if the CA is binary). Also, CAs are actually broader than chaotic systems considering that only 1 of the 4 Wolfram classes of CA is characterized by chaotic behavior.
The real killer app will be when Home & Garden's magazine zooms in on your home and analyzes your landscaping and house. Different people might get different covers and articles on rejuvenating dead lawns, trimming overgrown trees, or xeriscaping. You might even discover you've won the contest for most beautiful garden with an aerial view.
And they could even analyze your house & land for marketing opportunities. If the satellite veiw is oblique and the paint is peeling, they could forward your name to the local aluminum siding company or house painters.
Time to get a PO box!