An accountant and an economist are walking in the park together. As they pass a pond they see a frog. The economist says to the accountant, "I'll pay you $40 to lick that frog." The accountant thinks about it for a moment, agrees to lick the frog, and the economist gives him his $40. As they continue, they see another frog, and the account says to the economist, "I'll pay you $40 to lick that frog." The economist agrees, licks the frog, and gets his $40. The accountant then says, "Well, what was the point of that? Now we've both licked a frog and have nothing to show for it!" The economist replies, "True, but the economy has seen an increase of $80!"
The economy is not a zero-sum game. Even if both the economist and accountant are completely indifferent between paying or receiving $40 as the cost of licking a frog they valued the overall experience, e.g. the utility of having walked and licked frogs was greater than the utility of doing nothing regardless of the face value of the money that was equally exchanged. Otherwise they would have just sat in the park doing absolutely nothing.
Non-algorithmic computing won't keep authentication, authorization, or privilege escalation bugs from happening. It won't prevent side-channel attacks or social engineering. It won't provide automatic segregation and compartmentalization of data. All the major problems we have today will still exist; they'll just happen synchronously.
What you should actually look into is formally verifiable software design where the model of computation and data representation can be formally analyzed and every program can be proven to adhere to a security policy that enforces capability-based access to data and resources, limited access to timing information to prevent remote timing attacks, fine-grained authentication and authorization directly tied to capabilities, and a formal proof verifier that can check the entire system from the BIOS to the running applications every time you turn the computer on. Forget trusted computing; we need provable computing.
How many more "friends" do people have now that social media is big? With only 2 or 3 degrees of separation between most people on social networks it's surprising that you don't hear about nearly *every* crime in the country.
It wasn't just RAM; it was anything on an address bus. ROM, number of ports on an expander, number of values a register could store, etc. Then there's alignment issues (hint: don't try aligning your partitions on SI prefix boundaries).
How the heck is it that all such "genius", "unappreciated" world-altering inventions go through hype, secrecy, bilked investors, and nothing ever comes out of them. Nothing. Na da.
Babbage never did get that Analytical Engine running, did he?
Re-encoding one meaningless collection of symbols in to another doesn't magically make meaning appear. Imagine translating a message in something like Morse code in to Russian. We can still compute against the higher-level symbols, but at no point will the symbols be meaningful.
Google Translate doesn't do too bad taking "meaningless" collections of symbols in one language and outputting "meaningless" symbols in another language. Obviously it's a lossy process, but so is human translation. Humans are just better at translation, for now.
Sorry, Chalmers, but sub-symbolic computation IS symbolic computation.
I meant "Yes, I have an answer to the symbol grounding problem" not "Yes, I am just reasserting computationalism in the face of evidence to the contrary"
The symbols the brain receives are grounded in specific physical interactions with the world. The brain almost certainly translates those into higher-level symbols before further processing. Therefore it is sufficient to compute with corresponding higher-level symbols in an AI to achieve the same processing ability as the brain. It's precisely the same reason mathematics does not need to be done solely in the language of ZFC set theory and can instead be done using much broader definitions representing very complex constructions in the formal language of ZFC. Higher-level definitions still have meaning and computational value even if their meaning is only grounded at a much lower level.
Yes: symbols the brain receives are lower level than "car", "boat", "plane", or "red". They might more accurately be labeled "the thing that happens when touch neuron 5002 fires" and "the thing that happens when the center of the retina of the left eyeball receives energy from a photon between 500 and 600 nm". The brain builds models out of those sensory inputs corresponding to objects and qualia but the model containing those objects and qualia is partially detached from direct representation in the lowest level symbols because we have the ability to imagine things that we haven't experienced and communicate those imaginations meaningfully to other humans who understand them without receiving those low-level symbols. That implies that if you want an AI to be an expert at human sensory data and share our imagination ability it will need to be trained on inputs that mimic all the important human sensory inputs. But if all you want is for an AI to understand human language it is sufficient to have a large enough training corpus; people born without sight can appreciate the concept of color and talk about it rationally despite never experiencing it. All AIs that want to meaningfully interact with humans will need some concept of the environment that it is supposed to be a domain expert in, but that environment need not be a complete simulation of what a human would experience to produce enough familiarity with human concepts that the things it doesn't experience can be (partially) explained to it.
"the meaning isn't in the message" and "syntax is insufficient for semantics"
You might have a point if the brain actually reached out and touched the world, but it doesn't. It's hidden behind layers that process input from the real world and only feed messages to the brain, which does just fine constructing meaning from it.
Whoever writes a functional language that understands arrays and pointers will rule the world.
Intel has a fairly good one, I'd wager. A computer is really just a big iterated function of the current state of the registers, cache, RAM, and I/O devices. If you *really* want low-level hardware access in a functional language that's what it's going to look like in a pure functional language. The CPU is what evaluates the "computer" function and writes its output back to the rest of the hardware in manageable chunks of pure side-effect. Expressing arrays and pointers in a functional language requires modeling the memory-addressing hardware of the CPU.
This is basically the heart of the issue. Allowing users to add whatever keys they want (their uncle? Their best "friend" in Russia?) to their own user accounts circumvents the traditional authentication/authorization methods. When employees leave, occasionally their accounts stick around while things get cleaned up (or if they were especially bad and running production jobs out of their home account, they just stick around forever) their/etc/secrets entry gets a * for a password hash immediately but it's not obvious to check their personal.ssh directory and clean up authorized_keys. Especially if that annoying production job happens to use ssh in a non-trivial way.
Single sign-on systems are a possible solution. OpenSSH supports Kerberos integration and that may be the way to go for big *nix enterprises.
In the paper world you have to invest significant resources to forge each paper document. In the digital world if you can forge one document with a free tool you can forge as many as you want. To raise the cost of being able to forge a digital document beyond what an attacker is willing to pay the cost of legitimate use becomes greater than the benefit.
One possible solution is a hierarchy of security where the higher layers increase both the cost of forgery and the cost of legitimate use and let the market decide how much risk to bear. The SSL world tried to do that with extended validation certificates (the green address bar) but I'm not convinced it actually improved overall security since the problem is almost always at the user level. Maybe if they started selling extended validation hardware clients whose components were fixed in epoxy and ran highly secure firmware and software it would actually work. Trusted Computing is the obvious parallel in the PC world but it fails because the cost of developing software aimed at general purpose computers to rigorous security standards is too high. It's possible that as Moore's Law shoves hardware prices through the floor banks will just send relatively cheap secure hardware home with their new customers.
Cryptography has a poor track record of trying to use NP-hard or NP-complete problems to form ciphers. Even if P = NP it may still be possible to find algorithms with constant factors that make encryption and decryption practical but brute-force decryption impractical.
We're talking a particle accelerator the diameter of the asteroid belt.
Shouldn't be too hard; put the beam deflection electromagnets in orbit and just shoot the beam through the vacuum of space. Not quite a ringworld, but close enough for me.
The concept of "benefit of society as a whole" is an illusion. There is exactly one sure argument that an action can be expected to result in a net increase in value, and that is that the action is voluntary: everyone with an interest (property right) at stake gives their free and informed consent. Because value is subjective, you can't balance the cost to one group against the benefit to another. Only when everyone benefits can the result be considered a definite improvement.
If you quantify benefits you get utilitarianism and it's fairly obvious that the "benefit of society as a whole" is some form of overall utility maximization, not an illusion. Pricing is the mechanism for objectifying values between entities and the law of diminishing marginal utility suggests that a rich person giving $1 to a poor person increases overall utility.
I've never seen a good explanation for why a corporate entity should pay a lower tax rate than a personal entity when they both have essentially the same property rights. Sure, corporations aren't eligible for Medicare or Welfare but they receive subsidies and bailouts instead. Either get rid of all taxes or tax all taxable entities in accordance with their consumption of public goods. Corporations probably consume most of the federal budget spent on legislative actions and civil courts and judges, for instance. Most of the budget of the patent and trademark office and the copyright registrar. A significant portion of law enforcement investigates civil and criminal Copyright infringement for corporate entities. The roads are used significantly by transportation for corporate entities. Why shouldn't corporations have to pay for those obvious benefits of government?
Typically, the maximum heat generated from 24 fuel assemblies stored in a cask is less than that given off by a typical home heating system in an hour.
Holy crap, can I just get one for my house? It's cold in the winter.
A deflationary currency works fine so long as the rate of deflation has a lower magnitude than the average market return on investments. If you can earn an expected (risk-adjusted) 1% in an index fund it would be inadvisable to hoard any currency with less than 1% deflation.
Answering phones will never get you enough money to buy a house without a loan, or afford an education before you're too old to make back the money invested on it.
How many "decent" jobs are there in the U.S., and how many people seeking jobs are there in the U.S.? See any potential problems with that? I've always loved the term "entry-level job". If there aren't at least as many "second-level jobs" then by definition some of those "entry-level jobs" are lifetime positions.
Seriously? Anyone who can argue "Everything has a cause, therefore if the universe has a cause then that cause is named God" hasn't apparently seen the proof that there is no greatest integer. What you get out of "Every integer has a successor (every thing has a cause), 0 exists (the universe exists)" is "there is no greatest integer (there is no first cause)" not "there is a Greatest Integer because it would seem silly otherwise!!". Attempts to separate the word "God" from the rules of logic just make it apparent that it's a farce. If anything humans can imagine must have a cause, then so must "God". If any thing can be its own cause, then the universe is just as privileged in that respect as "God". This has been a known weakness of the argument for centuries. Anyone continuing to use such an argument is either lying or unable to apply the rules of formal logic. What is actually happening is that Craig and others are arbitrarily defining "God" as an additional axiom to their system of logic to which the normal rules do not apply.
The other arguments have similar problems. Debating Craig would involve formally specifying the rules of logic and then batting down each misapplication of the formal rules. It would take weeks in debate format, all while Craig complains that atheism makes no sense because he can't get an immediate, straight answer for why he's wrong. Apparently variable name invariance isn't a good enough reason.
Slashdot Mirror
An accountant and an economist are walking in the park together. As they pass a pond they see a frog. The economist says to the accountant, "I'll pay you $40 to lick that frog." The accountant thinks about it for a moment, agrees to lick the frog, and the economist gives him his $40. As they continue, they see another frog, and the account says to the economist, "I'll pay you $40 to lick that frog." The economist agrees, licks the frog, and gets his $40. The accountant then says, "Well, what was the point of that? Now we've both licked a frog and have nothing to show for it!" The economist replies, "True, but the economy has seen an increase of $80!"
The economy is not a zero-sum game. Even if both the economist and accountant are completely indifferent between paying or receiving $40 as the cost of licking a frog they valued the overall experience, e.g. the utility of having walked and licked frogs was greater than the utility of doing nothing regardless of the face value of the money that was equally exchanged. Otherwise they would have just sat in the park doing absolutely nothing.
Non-algorithmic computing won't keep authentication, authorization, or privilege escalation bugs from happening. It won't prevent side-channel attacks or social engineering. It won't provide automatic segregation and compartmentalization of data. All the major problems we have today will still exist; they'll just happen synchronously.
What you should actually look into is formally verifiable software design where the model of computation and data representation can be formally analyzed and every program can be proven to adhere to a security policy that enforces capability-based access to data and resources, limited access to timing information to prevent remote timing attacks, fine-grained authentication and authorization directly tied to capabilities, and a formal proof verifier that can check the entire system from the BIOS to the running applications every time you turn the computer on. Forget trusted computing; we need provable computing.
How many more "friends" do people have now that social media is big? With only 2 or 3 degrees of separation between most people on social networks it's surprising that you don't hear about nearly *every* crime in the country.
Put out a sign saying "Already burgled" and don't bother buying any replacement items.
It wasn't just RAM; it was anything on an address bus. ROM, number of ports on an expander, number of values a register could store, etc. Then there's alignment issues (hint: don't try aligning your partitions on SI prefix boundaries).
How the heck is it that all such "genius", "unappreciated" world-altering inventions go through hype, secrecy, bilked investors, and nothing ever comes out of them. Nothing. Na da.
Babbage never did get that Analytical Engine running, did he?
Re-encoding one meaningless collection of symbols in to another doesn't magically make meaning appear. Imagine translating a message in something like Morse code in to Russian. We can still compute against the higher-level symbols, but at no point will the symbols be meaningful.
Google Translate doesn't do too bad taking "meaningless" collections of symbols in one language and outputting "meaningless" symbols in another language. Obviously it's a lossy process, but so is human translation. Humans are just better at translation, for now.
Do you believe in p-zombies too?
Sorry, Chalmers, but sub-symbolic computation IS symbolic computation.
I meant "Yes, I have an answer to the symbol grounding problem" not "Yes, I am just reasserting computationalism in the face of evidence to the contrary"
The symbols the brain receives are grounded in specific physical interactions with the world. The brain almost certainly translates those into higher-level symbols before further processing. Therefore it is sufficient to compute with corresponding higher-level symbols in an AI to achieve the same processing ability as the brain. It's precisely the same reason mathematics does not need to be done solely in the language of ZFC set theory and can instead be done using much broader definitions representing very complex constructions in the formal language of ZFC. Higher-level definitions still have meaning and computational value even if their meaning is only grounded at a much lower level.
Yes: symbols the brain receives are lower level than "car", "boat", "plane", or "red". They might more accurately be labeled "the thing that happens when touch neuron 5002 fires" and "the thing that happens when the center of the retina of the left eyeball receives energy from a photon between 500 and 600 nm". The brain builds models out of those sensory inputs corresponding to objects and qualia but the model containing those objects and qualia is partially detached from direct representation in the lowest level symbols because we have the ability to imagine things that we haven't experienced and communicate those imaginations meaningfully to other humans who understand them without receiving those low-level symbols. That implies that if you want an AI to be an expert at human sensory data and share our imagination ability it will need to be trained on inputs that mimic all the important human sensory inputs. But if all you want is for an AI to understand human language it is sufficient to have a large enough training corpus; people born without sight can appreciate the concept of color and talk about it rationally despite never experiencing it. All AIs that want to meaningfully interact with humans will need some concept of the environment that it is supposed to be a domain expert in, but that environment need not be a complete simulation of what a human would experience to produce enough familiarity with human concepts that the things it doesn't experience can be (partially) explained to it.
Humans are just scaled-up apes. Apes are just scaled up mammals that happened to survive whatever killed the dinosaurs.
"the meaning isn't in the message" and "syntax is insufficient for semantics"
You might have a point if the brain actually reached out and touched the world, but it doesn't. It's hidden behind layers that process input from the real world and only feed messages to the brain, which does just fine constructing meaning from it.
Whoever writes a functional language that understands arrays and pointers will rule the world.
Intel has a fairly good one, I'd wager. A computer is really just a big iterated function of the current state of the registers, cache, RAM, and I/O devices. If you *really* want low-level hardware access in a functional language that's what it's going to look like in a pure functional language. The CPU is what evaluates the "computer" function and writes its output back to the rest of the hardware in manageable chunks of pure side-effect. Expressing arrays and pointers in a functional language requires modeling the memory-addressing hardware of the CPU.
This is basically the heart of the issue. Allowing users to add whatever keys they want (their uncle? Their best "friend" in Russia?) to their own user accounts circumvents the traditional authentication/authorization methods. When employees leave, occasionally their accounts stick around while things get cleaned up (or if they were especially bad and running production jobs out of their home account, they just stick around forever) their /etc/secrets entry gets a * for a password hash immediately but it's not obvious to check their personal .ssh directory and clean up authorized_keys. Especially if that annoying production job happens to use ssh in a non-trivial way.
Single sign-on systems are a possible solution. OpenSSH supports Kerberos integration and that may be the way to go for big *nix enterprises.
In the paper world you have to invest significant resources to forge each paper document. In the digital world if you can forge one document with a free tool you can forge as many as you want. To raise the cost of being able to forge a digital document beyond what an attacker is willing to pay the cost of legitimate use becomes greater than the benefit.
One possible solution is a hierarchy of security where the higher layers increase both the cost of forgery and the cost of legitimate use and let the market decide how much risk to bear. The SSL world tried to do that with extended validation certificates (the green address bar) but I'm not convinced it actually improved overall security since the problem is almost always at the user level. Maybe if they started selling extended validation hardware clients whose components were fixed in epoxy and ran highly secure firmware and software it would actually work. Trusted Computing is the obvious parallel in the PC world but it fails because the cost of developing software aimed at general purpose computers to rigorous security standards is too high. It's possible that as Moore's Law shoves hardware prices through the floor banks will just send relatively cheap secure hardware home with their new customers.
Cryptography has a poor track record of trying to use NP-hard or NP-complete problems to form ciphers. Even if P = NP it may still be possible to find algorithms with constant factors that make encryption and decryption practical but brute-force decryption impractical.
We're talking a particle accelerator the diameter of the asteroid belt.
Shouldn't be too hard; put the beam deflection electromagnets in orbit and just shoot the beam through the vacuum of space. Not quite a ringworld, but close enough for me.
Elliptic Curve ElGamal encryption is pretty fast.
The concept of "benefit of society as a whole" is an illusion. There is exactly one sure argument that an action can be expected to result in a net increase in value, and that is that the action is voluntary: everyone with an interest (property right) at stake gives their free and informed consent. Because value is subjective, you can't balance the cost to one group against the benefit to another. Only when everyone benefits can the result be considered a definite improvement.
If you quantify benefits you get utilitarianism and it's fairly obvious that the "benefit of society as a whole" is some form of overall utility maximization, not an illusion. Pricing is the mechanism for objectifying values between entities and the law of diminishing marginal utility suggests that a rich person giving $1 to a poor person increases overall utility.
I've never seen a good explanation for why a corporate entity should pay a lower tax rate than a personal entity when they both have essentially the same property rights. Sure, corporations aren't eligible for Medicare or Welfare but they receive subsidies and bailouts instead. Either get rid of all taxes or tax all taxable entities in accordance with their consumption of public goods. Corporations probably consume most of the federal budget spent on legislative actions and civil courts and judges, for instance. Most of the budget of the patent and trademark office and the copyright registrar. A significant portion of law enforcement investigates civil and criminal Copyright infringement for corporate entities. The roads are used significantly by transportation for corporate entities. Why shouldn't corporations have to pay for those obvious benefits of government?
Even S/MIME might meet your needs in this case. Encryption is cheap enough even for mailing lists now.
Typically, the maximum heat generated from 24 fuel assemblies stored in a cask is less than that given off by a typical home heating system in an hour.
Holy crap, can I just get one for my house? It's cold in the winter.
A deflationary currency works fine so long as the rate of deflation has a lower magnitude than the average market return on investments. If you can earn an expected (risk-adjusted) 1% in an index fund it would be inadvisable to hoard any currency with less than 1% deflation.
Answering phones will never get you enough money to buy a house without a loan, or afford an education before you're too old to make back the money invested on it.
How many "decent" jobs are there in the U.S., and how many people seeking jobs are there in the U.S.? See any potential problems with that? I've always loved the term "entry-level job". If there aren't at least as many "second-level jobs" then by definition some of those "entry-level jobs" are lifetime positions.
Seriously? Anyone who can argue "Everything has a cause, therefore if the universe has a cause then that cause is named God" hasn't apparently seen the proof that there is no greatest integer. What you get out of "Every integer has a successor (every thing has a cause), 0 exists (the universe exists)" is "there is no greatest integer (there is no first cause)" not "there is a Greatest Integer because it would seem silly otherwise!!". Attempts to separate the word "God" from the rules of logic just make it apparent that it's a farce. If anything humans can imagine must have a cause, then so must "God". If any thing can be its own cause, then the universe is just as privileged in that respect as "God". This has been a known weakness of the argument for centuries. Anyone continuing to use such an argument is either lying or unable to apply the rules of formal logic. What is actually happening is that Craig and others are arbitrarily defining "God" as an additional axiom to their system of logic to which the normal rules do not apply.
The other arguments have similar problems. Debating Craig would involve formally specifying the rules of logic and then batting down each misapplication of the formal rules. It would take weeks in debate format, all while Craig complains that atheism makes no sense because he can't get an immediate, straight answer for why he's wrong. Apparently variable name invariance isn't a good enough reason.