Almost - depends on the OS, but with Windows/Solaris/Linux you just put in live/install CD which has it's own root password and gives you access to the filesystems on disk.
Unless they are totally incompetent they'll have contacted the system vendor and got in by now - no need to hold the passowrd holder out the window by his ankles 'til he squeals.
If you want to distinguish between "none" and "zero" (as a more specific mathematical concept) then I'm not going to argue.
However, I don't believe the OP's claim that the Romans "had no expression for the number zero" was making that fine dinstinction. The fact is that the Romans did have a concept of none, could talk and write about it, and while their numeric system didn't have or need a zero digit, they did indeed use the word "nulla" where they needed to specify a quantity of "none".
I would also question your assertion that the lack of a zero digit is what caused the Romans to adopt a numeric system that used context-independent digits. Use of specific symbols for 10, 100, etc is simply a more obvious and simpler system which naturally arose first (the Greeks using it before the Romans, and I assume other cultures before that).
The invention of the Arabic numeric system with it's greater ease of calculation (at least with pen & paper - the abacus was well matched to math with roman numbers) doesn't really require a zero digit - it only needs a positional placeholder to indicate there were no (vs 1, 2, 3...) multiples of that power of 10 in the number, and it's interesting to note that the early indo-arabic glyph for zero was "." which suggests exactly that - a placeholder. There's no reason to suppose that the meaning of the Arabic "." was anything more sophisticated than a quantity of "none", and in fact I'd assert that is still the meaning of the modern digit "0", notwithstanding any more refined mathematical definition of "zero" which is something that came later and doesn't retroactively usurp the original simple meaning of a digit (i.e. symbol used in context of a number) for "zero/none".
SMT (aka hyperthreading - Intel P4 HT) is as real as multi-threading per core gets. If you instead duplicated the entire execution pipeline you'd just have another entire core!
The execution pipeline of a modern CPU consists of many specialized decode, load/store, integer/logic (ALU), floating point (FPU), etc components (often more than one of each type) that all run independently, maybe idle at any given time if not needed or maybe waiting for operand results of an earlier step in the pipeline, or waiting for memory access (latency).
What SMT essentially does is add mutiple decode streams feeding into the pipeline so that the pipeline and it's constituent components are genuinely exectuting the instruction streams from multiple threads simultaneously. Despite the fact that they are still competing for the shared components there is still genuine parallelism since in a single threaded scenarion many of the units may be sitting idle waiting for operands or memory access. Using SMT means that the execution units are more fully utilized and it prevents memory latency (cache misses) from stailling the core by allowing the pipeline to utilize the execution units (ALU, FPU, etc) for instructions for which the operands are already available.
In summary, SMT is true multi-threaded core multi-threading. The only better (but more wasteful) type of multithreading is to use multiple cores, but they are still not fully indpendent since (at least in a desktop architecture) they are going to be competing for scarce memory and bus bandwidth.
Null and 0 are not the same things - null is a concept and zero is a number.... and "nulla" is a concept/word that **was** used in a numeric context to mean zero. This is recorded fact, not speculation.
The latin language does have a word "nulla" for zero/nothing was used in numeric context.
I think you mean that the roman numeral system doesn't use a zero digit, but this wasn't becuase they had no concept of zero, it was because their numeric system didn't need it. Zero's are only needed in a system such as our where digit value is context specific (i.e. the "1" in "100" means something different than the "1" in "10") - the roman numeric system doesn't work this way.
Seeing as the researchers arn't that sure about it, maybe they want to learn the language fluently before they go off publishing papers about it.
The insight that word categories are not intrinsic to language is something I'd put in the "well, duh!" category. Discrimination and labelling are not the same thing, even if they are bidirectionally related. Consider how babies can learn any language and discriminate the sounds (phonemes) in them, but once you've settled on a language (or few) you all but lose the ability to discriminate among the sound categories your language doesn't use - they get binned to useful categories.
It may be (notwithstanding the need for a less speculative understanding of the language!) that these laid back chappies really have no words in their language for number concepts more accurate than "a few" or "whatever, dude", but I'm sure they are able to discriminate between the categories one, two, three, etc, up to some small number - a capability which does appear to be innate - and I'll even bet they have some way to express it also, if only in some laborious forms such as "one pig, and another pig, and another pig". It's probably no coincidence that there are only a handful of native speakers left though, since a language like this isn't going to be very useful ("how many enemy did you see heading our way?" "a few" (reality: 300)).
The existence of a god that has any power to affect things in the realm of science (i.e. the entire natural world, including you and your brain) is disproven a zillion times a day by the laws of science being upheld. The experiment "the outcome will be according to god, not science" consistently fails.. He might exist, and might have an ability to control something, but the evidence certainly says it's highly unlikely. This also covers god as proposed intelligent designer of species.
God as creator of the universe is only possible if the universe has a beginning, which I don't think is current scientific understanding (in this context taking universe to mean out universe or the infinite regress of what begat that).
I doubt it would make much difference there either. The biggest latencies for VoIP are likely also either your local connection (esp. if cable or DSL), local computer (Windows isn't exactly a real-time OS!) or the far end. Note incidently that you need 50-100ms (depends whether echo cancellation is being applied) to have an audible delay.
For that matter, note that TFA is only about switching speed - but a router still needs to decide what switch to make. Your latency is only as good as the slowest link, so a photonic switch preceded by a silicon switching decision would be of little benefit.
I don't see any problem (or benefit!) in teaching ID, but since it's not science it should not be taught as part of science.
If Louisiana really have more time in the school day than they need for more useful subjects (who'd have guessed their school system was so efficient?), then by all means offer ID, Astrology & Palm Reading as a filler. I went to school in the UK and we had religion offered as a filler class - it never harmed me, since I chose to take cooking with the girls instead!
In practical terms this means nothing for internet users since download speeds are invariably limited by the server or your local internet connection - not the backbone.
Even in a hypothetical situation where you have a fiber (e.g. FiOS) internet connection and were connected to a server that required you to have a photonic home router to keep up, you'd still be ultimately throttled by your NIC, mobo bus, or CPU.
Faster/cheaper switches may be exciting to backbone providers, but really that's about it.
It's interesting to consider that even Joe Average could get crazy wealthy just due to the power of compound interest. If Joe's parents put $1000 in a bank account for him at birth, earning 5% compounded annually, then at age 1000 that would have grown to a staggering $1.5 * 10^24 . Of course with this much money swooshing around there'd be killer inflation to boot. It's hard to imagine what the financial world would look like!
I wonder if someone with a few hundres years of life experience would even have any interest in things like grad school level studies - you might develop a "seen it done it" total boredom with the world and/or such a high level degree of abstraction born of so much life experience that the trifles of the specifics of any field would not interest you. Maybe the young "fresh brains" would turn out to be massively in demand as the only ones with enough LACK of life experience to still have an interest in actually acomplishing anything.
The societal changes brought about by a 1000 year lifetime would no doubt be extrememly pround and extrememly difficult to predict.
Is is actually more economical to pay the Russians or us Eurotrash to send them to space rather than the cost of maintaining and flying the shuttle?
I guess that depends on how many people you're sending up. At this point the incremental cost for the Shuttle is about $60M/flight (notwithstanding that the total cost, including per-flight development amortization, comes to about $1B/flight), so you can compare that to the Russian's $20M per-person ISS round-trip tourist tickets.
Given the massive cost of developing new launch launch capabilities and spacecraft, it would seem to make sense to outsource as much as possible on a permanent ongoing basis. Even future planetary exploration should be planned around using commercial heavy lift (human rated if necessary, but robotic is better) services. It doesn't make sense for NASA to spend it's extremely limited budget on developing it's own lauch capabilities when it can be had cheaper commercially. Let the military budget (which is massively greater than NASA's) pay for anything that is a matter of national security where outsourcing doesn't cut it.
To be honest I am a bit surprised - I assumed they were at least making **some** effort to balance privacy vs desire to retain data. It seems they are making no effort at all - they just keep a raw history of all your activity. Maybe if the Google search logs get breeched and we read about some Senator's embarassing search history then it'll change.
They may have assumed that there was no downside to keeping the raw logs "just in case", but hopefully they'll suffer sufficient loss of trust from this being exposed that it'll make them reconsider.
I think this piss-poor attitude to user privacy will bite them in the ass in the end though... Google started out with such a positive image, but they are rapidly losing that and their "do no evil" mantra has become a joke.
They don't need that level of detail though. The data is used for serving advertisements, so classes of interests is enough. They need to know that use Joe is interested in, say, the category "girls", not specifically where, when and for how long he watched a specific video called boobie jiggle.
It makes a big difference - would you prefer, should the data ever be compromized, someone else to know these general (ad targetting) categories you have demonstarated interest in ("porn" would/should be seperate from "girls" or "sexy", and no-one really needs to know about your furry fetish), or are you equally OK with your minute-by-minute activities being logged?
Not much Google can do now if ordered by a court to turn it over, but I'm a bit (lot) creeped out by why they think they need to keep this type of detailed personalized history in the first place. Having the history means that a court/government/disgruntled employee/future employer/hacker may get access to it.
If Google really need to target advertising to what we're watching, searching for, etc, etc (big brother - yuck), then couldn't they at least do so in a more anonymous form - rather than storing "user Joe just watched Viacom's boobie-jiggle on YouTube on July 3rd", just do user[joe].interests[interest-class[boobie-jiggle]]++.
I guess Google don't consider being big brother incarnate to be "evil", but I think most of their user base do!
I've got news for you - you didn't learn to learn speech! That speech learning capability is part of the brain architecture that you were born with, and it's not very speech specific either. Now, if you knew what that built-in learning-capable architecture was, you could simulate it, and do so at many different levels from the individual neuron on up to the highest level at which your understand the architecture. Guess which would be most efficient? You'd not be losing anything by simulating it at the highest level possible, since the architecture itself isn't something that we learn - it is fixed - we only learn within the framework of the genetically encoded fixed brain architecture we are born with.
In other news, the brain does not learn by trail and error or goal seeking! Those are subjective psychological descriptions of one type of learning process, but at the level of the brain itself it's essentially a matter of extracting the regularities and interrelationships of sensory patterns and sequences. One of the things we are all inevitably bound to learn is a sense of self, and it is only that "self" to which we ascribe goals.
I'm not going to continue this any further - it's going nowhere fast.
You are talking about a HUGE leap in technology. AI will not be to the point where it could come even close to that sort of reasoning. When you think about the amount of processing power an organic brain has in its size compared to computer chips, you'd know that we'd need the biggest super computer in existence to emulate the brain of a hamster... maybe.
If your talking a neuron-level simulation, or even a cortical mini-column simulation, then that is certainly true (and anyways we don't know the connection map, so it is anyways impossible!).
OTOH it's not at all obvious that a **behavioral** simulation of human brain is necessarily out of reach of modern computers. I strongly believe that we're very close to the point where that will be possible. Consider for example that we can do realtime speech recognition on a desktop PC even though doing it via a neuron level simulation of the auditory cortex would be computationally impossible. I think the same thing likely applies of the neocortex that's responsible for cognition - neuron level is impossible, but I think we understand the struture well enough to simlate it at a FAR higher and more efficient level.
Of course there will always be that trade-off, but for many applications the robot/AI would be an easy win. I'm sure that if it were mass produced something like Asimo could easily retail for $25K or less, and adding AI hardly affects the production cost. Now consider that for this $25K one-time cost a company could replace, say, a programmer that's costing them $100K+ PER YEAR, and it starts to look pretty attractive. Ditto for almost any office job that doesn't require any special physical capabilities.
Piloting a space ship is exactly the sort of application where using an AI makes sense! It doesn't eat, breath or crap, doesn't need to sleep, doesn't care if it's sent on a 10yr one-way mission, etc, etc.
Bear in mind that once AI software and AI robots are available, they can be bought off the shelf for retail prices. The developemnt cost it not an issue for the consumer. NASA will probably go to Circuit City, or maybe Walmart, to pick up the first robotic space ship pilot headed for Mars.
Sure - quit your job, buy yourself a few acres in the boonies someplace, and become self-sufficient. Or go build a cabin up in Alaska and live off the land.
Not as such, no, but one of the reasons why we employ computers is an effort to remove human error. If we design computers that are just as fallible humans, then what is the point?
As I stated in my initial reply, I don't think the goal of AI is merely to match human capability - it is to exceed it, and one specific, and very useful, way to exceed it (as I again stated in my original reply) is to (in addition to it's human-like behaviors) also endow it with the traditional computer benefits of meory, computation, high speed communiction, etc. This of this computational layer as an additional shell around the existing human brain.
However, even if an AI was "merely" to match human capability, with the potential fallibility that entails, I think you are being a bit shortsighted not to see the tremendous utility. Think of the most capable humans - computer scientists, nuclear engineers, surgeons, but now available to work for you 24x7 for no pay, always gaining experience, never getting older, never getting sick... You seem to think that "human fallibility" makes us useless as a species (and therefore as a target for emulation), but I think you are wrong.
You seem to be under the impression that we don't understand how the brain works. This is a false assumption. We know quite a bit about how the brain learns.
I'm well aware of that. I am building an AI;-)
The issue is the "fuzzy" aspect of learning machines is that "fuzzy" is not always appropriate. You don't want a "fuzzy" calculator balancing your check book. You want a rock solid deterministic calculator.
I'd be quite happy having someone with the skill of an accountant or Math PhD balancing my checkbook, or planning my retirement investments for that matter. I'd also be very glad if they weren't just doing it as a rote task, but realized that an entry for a $2000 oil change probably had a missing decimal point and went and checked that, etc. Whether that math PhD is Carbon or Silicon based I could care less.
Your whole argument against a human-emulation AI is that it would be fallible therefore useless. You seem to be judging the utility of it based on assuming that the implementation is flawed. What if the implementation wasn't flawed - what if it was indeed the exact equivalent of the smartest most highly trained professional you know, and maybe also augmented in the ways I suggested with super-human capabilities of computer like massive perfect recall, execution of stored programs (self programming, self debugging of course)...
In this whole thread you have basically set up your own straw man of what AI would be (sub-human - fallible to the point of being largely unreliable and useless), and then shoot that down (essentially "it'd be fallible, so it's be fallible"). What about assuming that AI is what everybody actually has as their goal and arguing against that instead?!
Your objections apply equally to humans as to an AI implemented in a human-like manner. Do you also think humans are useless?!
Rather than panning AI based on a simplistic and inadequate model of how it may be achieved, it would be more logical to try to understand how we ourselves transcend our fuzzy underpinnings, then realize that those same mechanisms can be implemented in an AI. The trivial proof-of-concept for an AI would just be to reimplement our own brain architecture (notwithstanding that it'll be a while before we've mapped it well enough to do that). Of course in practice behavioral models will be more practical and efficient than neuron level simulations, and indeed alternative architectures with equivalent capabilities may also be conceived.
The key to understanding how to build an AI isn't to tackle it as, or conceive of it as, a computing task, but rather to think in terms of architecture and how we ourselves, from birth on, adapat to the world and achieve what we do. Unless you are a dualist, you must realize that our mind is just out brain in action, and unless you are a pessimist you must realize that one day we will indeed understand how the brain is wired and by extension how it does what it does (we already do understand quite a large piece of it). Human level AI will therefore certainly be achieved one day - one can only argue about when. To argue that AI will be useless is either to argue that humans are useless or to argue that AI will never be achieved.
The most interesting prospect of AI isn't in matching human capability - it's in exceeding it. Not just in speed of thought or memory capacity, but in every way possible.
Why would you go to all the trouble of building an AI then deny it capabilites that computers already posess (memory, computation, communication) that give it a post-human edge?
Achieving human-level capabilty is the hard part and therefore the initial goal, but the real dream is something more like a human that has evolved another layer to our brain - the outer computational cortex that has computer-like memory and computational capabilities. Think Spock, or HAL, as the goal, not a human.
TFA is the usual drivel that gets trotted out every few years. Yeah, we havn't achieved AI (= general purpose human-level intelligence) yet, but we have achieved engineered (not learnt) solutions to many useful problems that people have in the past listed among the goals of an AI. So what?
I don't think there's even that many people trying to achieve AI, especially since most pure-research labs have gone the way of the Dodo or changed to a short term more results-based mandate. Academia isn't much better off since it's always been based on incremental building on others work and focused research grants. Step-by-step will eventually get us to AI, but not as fast as a pure researh lab with that as its proximate goal.
Slashdot Mirror
Almost - depends on the OS, but with Windows/Solaris/Linux you just put in live/install CD which has it's own root password and gives you access to the filesystems on disk.
Unless they are totally incompetent they'll have contacted the system vendor and got in by now - no need to hold the passowrd holder out the window by his ankles 'til he squeals.
If you want to distinguish between "none" and "zero" (as a more specific mathematical concept) then I'm not going to argue.
However, I don't believe the OP's claim that the Romans "had no expression for the number zero" was making that fine dinstinction. The fact is that the Romans did have a concept of none, could talk and write about it, and while their numeric system didn't have or need a zero digit, they did indeed use the word "nulla" where they needed to specify a quantity of "none".
I would also question your assertion that the lack of a zero digit is what caused the Romans to adopt a numeric system that used context-independent digits. Use of specific symbols for 10, 100, etc is simply a more obvious and simpler system which naturally arose first (the Greeks using it before the Romans, and I assume other cultures before that).
The invention of the Arabic numeric system with it's greater ease of calculation (at least with pen & paper - the abacus was well matched to math with roman numbers) doesn't really require a zero digit - it only needs a positional placeholder to indicate there were no (vs 1, 2, 3...) multiples of that power of 10 in the number, and it's interesting to note that the early indo-arabic glyph for zero was "." which suggests exactly that - a placeholder. There's no reason to suppose that the meaning of the Arabic "." was anything more sophisticated than a quantity of "none", and in fact I'd assert that is still the meaning of the modern digit "0", notwithstanding any more refined mathematical definition of "zero" which is something that came later and doesn't retroactively usurp the original simple meaning of a digit (i.e. symbol used in context of a number) for "zero/none".
SMT (aka hyperthreading - Intel P4 HT) is as real as multi-threading per core gets. If you instead duplicated the entire execution pipeline you'd just have another entire core!
The execution pipeline of a modern CPU consists of many specialized decode, load/store, integer/logic (ALU), floating point (FPU), etc components (often more than one of each type) that all run independently, maybe idle at any given time if not needed or maybe waiting for operand results of an earlier step in the pipeline, or waiting for memory access (latency).
What SMT essentially does is add mutiple decode streams feeding into the pipeline so that the pipeline and it's constituent components are genuinely exectuting the instruction streams from multiple threads simultaneously. Despite the fact that they are still competing for the shared components there is still genuine parallelism since in a single threaded scenarion many of the units may be sitting idle waiting for operands or memory access. Using SMT means that the execution units are more fully utilized and it prevents memory latency (cache misses) from stailling the core by allowing the pipeline to utilize the execution units (ALU, FPU, etc) for instructions for which the operands are already available.
In summary, SMT is true multi-threaded core multi-threading. The only better (but more wasteful) type of multithreading is to use multiple cores, but they are still not fully indpendent since (at least in a desktop architecture) they are going to be competing for scarce memory and bus bandwidth.
Null and 0 are not the same things - null is a concept and zero is a number. ... and "nulla" is a concept/word that **was** used in a numeric context to mean zero. This is recorded fact, not speculation.
The latin language does have a word "nulla" for zero/nothing was used in numeric context.
I think you mean that the roman numeral system doesn't use a zero digit, but this wasn't becuase they had no concept of zero, it was because their numeric system didn't need it. Zero's are only needed in a system such as our where digit value is context specific (i.e. the "1" in "100" means something different than the "1" in "10") - the roman numeric system doesn't work this way.
Seeing as the researchers arn't that sure about it, maybe they want to learn the language fluently before they go off publishing papers about it.
The insight that word categories are not intrinsic to language is something I'd put in the "well, duh!" category. Discrimination and labelling are not the same thing, even if they are bidirectionally related. Consider how babies can learn any language and discriminate the sounds (phonemes) in them, but once you've settled on a language (or few) you all but lose the ability to discriminate among the sound categories your language doesn't use - they get binned to useful categories.
It may be (notwithstanding the need for a less speculative understanding of the language!) that these laid back chappies really have no words in their language for number concepts more accurate than "a few" or "whatever, dude", but I'm sure they are able to discriminate between the categories one, two, three, etc, up to some small number - a capability which does appear to be innate - and I'll even bet they have some way to express it also, if only in some laborious forms such as "one pig, and another pig, and another pig". It's probably no coincidence that there are only a handful of native speakers left though, since a language like this isn't going to be very useful ("how many enemy did you see heading our way?" "a few" (reality: 300)).
The existence of a god that has any power to affect things in the realm of science (i.e. the entire natural world, including you and your brain) is disproven a zillion times a day by the laws of science being upheld. The experiment "the outcome will be according to god, not science" consistently fails.. He might exist, and might have an ability to control something, but the evidence certainly says it's highly unlikely. This also covers god as proposed intelligent designer of species.
God as creator of the universe is only possible if the universe has a beginning, which I don't think is current scientific understanding (in this context taking universe to mean out universe or the infinite regress of what begat that).
I doubt it would make much difference there either. The biggest latencies for VoIP are likely also either your local connection (esp. if cable or DSL), local computer (Windows isn't exactly a real-time OS!) or the far end. Note incidently that you need 50-100ms (depends whether echo cancellation is being applied) to have an audible delay.
For that matter, note that TFA is only about switching speed - but a router still needs to decide what switch to make. Your latency is only as good as the slowest link, so a photonic switch preceded by a silicon switching decision would be of little benefit.
I don't see any problem (or benefit!) in teaching ID, but since it's not science it should not be taught as part of science.
If Louisiana really have more time in the school day than they need for more useful subjects (who'd have guessed their school system was so efficient?), then by all means offer ID, Astrology & Palm Reading as a filler. I went to school in the UK and we had religion offered as a filler class - it never harmed me, since I chose to take cooking with the girls instead!
In practical terms this means nothing for internet users since download speeds are invariably limited by the server or your local internet connection - not the backbone.
Even in a hypothetical situation where you have a fiber (e.g. FiOS) internet connection and were connected to a server that required you to have a photonic home router to keep up, you'd still be ultimately throttled by your NIC, mobo bus, or CPU.
Faster/cheaper switches may be exciting to backbone providers, but really that's about it.
It's interesting to consider that even Joe Average could get crazy wealthy just due to the power of compound interest. If Joe's parents put $1000 in a bank account for him at birth, earning 5% compounded annually, then at age 1000 that would have grown to a staggering $1.5 * 10^24 . Of course with this much money swooshing around there'd be killer inflation to boot. It's hard to imagine what the financial world would look like!
I wonder if someone with a few hundres years of life experience would even have any interest in things like grad school level studies - you might develop a "seen it done it" total boredom with the world and/or such a high level degree of abstraction born of so much life experience that the trifles of the specifics of any field would not interest you. Maybe the young "fresh brains" would turn out to be massively in demand as the only ones with enough LACK of life experience to still have an interest in actually acomplishing anything.
The societal changes brought about by a 1000 year lifetime would no doubt be extrememly pround and extrememly difficult to predict.
If I gave you a lab rat today, how long could you extend his life?
What about me - is there anything I can do (other than a healthy lifestyle), or could have done, today, to start extending my life?
How long before the answers to either of these questions change significantly? 5 years? 10? 20?
Is is actually more economical to pay the Russians or us Eurotrash to send them to space rather than the cost of maintaining and flying the shuttle?
I guess that depends on how many people you're sending up. At this point the incremental cost for the Shuttle is about $60M/flight (notwithstanding that the total cost, including per-flight development amortization, comes to about $1B/flight), so you can compare that to the Russian's $20M per-person ISS round-trip tourist tickets.
Given the massive cost of developing new launch launch capabilities and spacecraft, it would seem to make sense to outsource as much as possible on a permanent ongoing basis. Even future planetary exploration should be planned around using commercial heavy lift (human rated if necessary, but robotic is better) services. It doesn't make sense for NASA to spend it's extremely limited budget on developing it's own lauch capabilities when it can be had cheaper commercially. Let the military budget (which is massively greater than NASA's) pay for anything that is a matter of national security where outsourcing doesn't cut it.
If you want to build computing into a utility, you need large real-time systems running as sufficiently as possible.
But if you want to build sprockets into a weasel you need small batch-mode systems running as necessarily as possible.
If the poster had anything interesting to say (I'd guess not, but who knows!), it was totally obscured by his lack of grasp of the English language.
To be honest I am a bit surprised - I assumed they were at least making **some** effort to balance privacy vs desire to retain data. It seems they are making no effort at all - they just keep a raw history of all your activity. Maybe if the Google search logs get breeched and we read about some Senator's embarassing search history then it'll change.
They may have assumed that there was no downside to keeping the raw logs "just in case", but hopefully they'll suffer sufficient loss of trust from this being exposed that it'll make them reconsider.
Yeah, sadly it seems you're right.
I think this piss-poor attitude to user privacy will bite them in the ass in the end though... Google started out with such a positive image, but they are rapidly losing that and their "do no evil" mantra has become a joke.
They don't need that level of detail though. The data is used for serving advertisements, so classes of interests is enough. They need to know that use Joe is interested in, say, the category "girls", not specifically where, when and for how long he watched a specific video called boobie jiggle.
It makes a big difference - would you prefer, should the data ever be compromized, someone else to know these general (ad targetting) categories you have demonstarated interest in ("porn" would/should be seperate from "girls" or "sexy", and no-one really needs to know about your furry fetish), or are you equally OK with your minute-by-minute activities being logged?
Not much Google can do now if ordered by a court to turn it over, but I'm a bit (lot) creeped out by why they think they need to keep this type of detailed personalized history in the first place. Having the history means that a court/government/disgruntled employee/future employer/hacker may get access to it.
If Google really need to target advertising to what we're watching, searching for, etc, etc (big brother - yuck), then couldn't they at least do so in a more anonymous form - rather than storing "user Joe just watched Viacom's boobie-jiggle on YouTube on July 3rd", just do user[joe].interests[interest-class[boobie-jiggle]]++.
I guess Google don't consider being big brother incarnate to be "evil", but I think most of their user base do!
I've got news for you - you didn't learn to learn speech! That speech learning capability is part of the brain architecture that you were born with, and it's not very speech specific either. Now, if you knew what that built-in learning-capable architecture was, you could simulate it, and do so at many different levels from the individual neuron on up to the highest level at which your understand the architecture. Guess which would be most efficient? You'd not be losing anything by simulating it at the highest level possible, since the architecture itself isn't something that we learn - it is fixed - we only learn within the framework of the genetically encoded fixed brain architecture we are born with.
In other news, the brain does not learn by trail and error or goal seeking! Those are subjective psychological descriptions of one type of learning process, but at the level of the brain itself it's essentially a matter of extracting the regularities and interrelationships of sensory patterns and sequences. One of the things we are all inevitably bound to learn is a sense of self, and it is only that "self" to which we ascribe goals.
I'm not going to continue this any further - it's going nowhere fast.
You are talking about a HUGE leap in technology. AI will not be to the point where it could come even close to that sort of reasoning. When you think about the amount of processing power an organic brain has in its size compared to computer chips, you'd know that we'd need the biggest super computer in existence to emulate the brain of a hamster... maybe.
If your talking a neuron-level simulation, or even a cortical mini-column simulation, then that is certainly true (and anyways we don't know the connection map, so it is anyways impossible!).
OTOH it's not at all obvious that a **behavioral** simulation of human brain is necessarily out of reach of modern computers. I strongly believe that we're very close to the point where that will be possible. Consider for example that we can do realtime speech recognition on a desktop PC even though doing it via a neuron level simulation of the auditory cortex would be computationally impossible. I think the same thing likely applies of the neocortex that's responsible for cognition - neuron level is impossible, but I think we understand the struture well enough to simlate it at a FAR higher and more efficient level.
Of course there will always be that trade-off, but for many applications the robot/AI would be an easy win. I'm sure that if it were mass produced something like Asimo could easily retail for $25K or less, and adding AI hardly affects the production cost. Now consider that for this $25K one-time cost a company could replace, say, a programmer that's costing them $100K+ PER YEAR, and it starts to look pretty attractive. Ditto for almost any office job that doesn't require any special physical capabilities.
Piloting a space ship is exactly the sort of application where using an AI makes sense! It doesn't eat, breath or crap, doesn't need to sleep, doesn't care if it's sent on a 10yr one-way mission, etc, etc.
Bear in mind that once AI software and AI robots are available, they can be bought off the shelf for retail prices. The developemnt cost it not an issue for the consumer. NASA will probably go to Circuit City, or maybe Walmart, to pick up the first robotic space ship pilot headed for Mars.
Sure - quit your job, buy yourself a few acres in the boonies someplace, and become self-sufficient. Or go build a cabin up in Alaska and live off the land.
Not as such, no, but one of the reasons why we employ computers is an effort to remove human error. If we design computers that are just as fallible humans, then what is the point?
As I stated in my initial reply, I don't think the goal of AI is merely to match human capability - it is to exceed it, and one specific, and very useful, way to exceed it (as I again stated in my original reply) is to (in addition to it's human-like behaviors) also endow it with the traditional computer benefits of meory, computation, high speed communiction, etc. This of this computational layer as an additional shell around the existing human brain.
However, even if an AI was "merely" to match human capability, with the potential fallibility that entails, I think you are being a bit shortsighted not to see the tremendous utility. Think of the most capable humans - computer scientists, nuclear engineers, surgeons, but now available to work for you 24x7 for no pay, always gaining experience, never getting older, never getting sick... You seem to think that "human fallibility" makes us useless as a species (and therefore as a target for emulation), but I think you are wrong.
You seem to be under the impression that we don't understand how the brain works. This is a false assumption. We know quite a bit about how the brain learns.
I'm well aware of that. I am building an AI ;-)
The issue is the "fuzzy" aspect of learning machines is that "fuzzy" is not always appropriate. You don't want a "fuzzy" calculator balancing your check book. You want a rock solid deterministic calculator.
I'd be quite happy having someone with the skill of an accountant or Math PhD balancing my checkbook, or planning my retirement investments for that matter. I'd also be very glad if they weren't just doing it as a rote task, but realized that an entry for a $2000 oil change probably had a missing decimal point and went and checked that, etc. Whether that math PhD is Carbon or Silicon based I could care less.
Your whole argument against a human-emulation AI is that it would be fallible therefore useless. You seem to be judging the utility of it based on assuming that the implementation is flawed. What if the implementation wasn't flawed - what if it was indeed the exact equivalent of the smartest most highly trained professional you know, and maybe also augmented in the ways I suggested with super-human capabilities of computer like massive perfect recall, execution of stored programs (self programming, self debugging of course)...
In this whole thread you have basically set up your own straw man of what AI would be (sub-human - fallible to the point of being largely unreliable and useless), and then shoot that down (essentially "it'd be fallible, so it's be fallible"). What about assuming that AI is what everybody actually has as their goal and arguing against that instead?!
Your objections apply equally to humans as to an AI implemented in a human-like manner. Do you also think humans are useless?!
Rather than panning AI based on a simplistic and inadequate model of how it may be achieved, it would be more logical to try to understand how we ourselves transcend our fuzzy underpinnings, then realize that those same mechanisms can be implemented in an AI. The trivial proof-of-concept for an AI would just be to reimplement our own brain architecture (notwithstanding that it'll be a while before we've mapped it well enough to do that). Of course in practice behavioral models will be more practical and efficient than neuron level simulations, and indeed alternative architectures with equivalent capabilities may also be conceived.
The key to understanding how to build an AI isn't to tackle it as, or conceive of it as, a computing task, but rather to think in terms of architecture and how we ourselves, from birth on, adapat to the world and achieve what we do. Unless you are a dualist, you must realize that our mind is just out brain in action, and unless you are a pessimist you must realize that one day we will indeed understand how the brain is wired and by extension how it does what it does (we already do understand quite a large piece of it). Human level AI will therefore certainly be achieved one day - one can only argue about when. To argue that AI will be useless is either to argue that humans are useless or to argue that AI will never be achieved.
Not sure how this got modded as interesting...
The most interesting prospect of AI isn't in matching human capability - it's in exceeding it. Not just in speed of thought or memory capacity, but in every way possible.
Why would you go to all the trouble of building an AI then deny it capabilites that computers already posess (memory, computation, communication) that give it a post-human edge?
Achieving human-level capabilty is the hard part and therefore the initial goal, but the real dream is something more like a human that has evolved another layer to our brain - the outer computational cortex that has computer-like memory and computational capabilities. Think Spock, or HAL, as the goal, not a human.
TFA is the usual drivel that gets trotted out every few years. Yeah, we havn't achieved AI (= general purpose human-level intelligence) yet, but we have achieved engineered (not learnt) solutions to many useful problems that people have in the past listed among the goals of an AI. So what?
I don't think there's even that many people trying to achieve AI, especially since most pure-research labs have gone the way of the Dodo or changed to a short term more results-based mandate. Academia isn't much better off since it's always been based on incremental building on others work and focused research grants. Step-by-step will eventually get us to AI, but not as fast as a pure researh lab with that as its proximate goal.