I'm not really sure I approve of the police doing this any more than a random citizen. At the very least it should require approval by three separate courts and a public notice (which the target, of course, anonimized). And public notice doesn't mean a posting in some inaccessible place, but listing on a web page, something like: 2018/02/18 15:27 warrant approved until 2018/02/25 15:30 to (stalk?..need better description) (some explicit description of what is to be surveilled). The explicit description of the target should probably by a value generated by a hash function of the IP address.
Intel not only made dangerously broken CPUs which had been predicted to be dangerously broken (without a definite exploit) before they were designed, but if they didn't already know about how to exploit it, they were informed at least 6 months before the public notice, and appear to have taken no steps to mitigate the problem prior to public notice. We can't really know, but the patches that they rushed out after notice was made public were so poor that they probably hadn't done anything.
Etc.
I'm willing for Intel to prove that they were acting in a reasonable and ethical manner, but the preponderance of the evidence seems against that assumption.
Actually, asteroids *do* occasionally hit each other. It doesn't happen every day, but we don't really have a real good idea of how often. Often enough that we can observe the after effects, but we can only see that when two large chunks collide. For the smaller pieces...we can be sure it happens because there are a lot more of them, and even the larger pieces collide, but we can't really observe.
OTOH, frequency... it's probably rather rare, and most of the collisions are at low relative speeds. Fender-benders are a lot more common then head on collisions. But occasionally things will hit each other at a large enough angle to shatter. I don't know what occasionally means, it might be once every thousand years for something the mass of a car. Or it might be less frequently.
Yes, space is big, and there's a lot of space on the average between things. But most things are about in the plane of rotation, and resonance effects tend to shift things into similar orbits and... well, when things are nearly co-planar and the eccentricities are different, eventually a collision is likely. Of course, eventually can take a very long time, but with millions of pieces of stuff orbiting around they do happen.
I don't think "liberal" administrations are any less given to spying or centralization of power than are "conservative" administrations. They just spend a bit more on the below median income citizen. You can say this is good or bad, and there are reasonable arguments either way, but neither is "small government" or "get the government off the backs of the people" or any of the similar slogans that have been used in past campaigns. They only tend to lie about how they'll do that when they get power again when they're out of power.
So there's no reason to be surprised that Obama's administration was caught spying...except, of course, that they shouldn't have been caught.
Another question: What phones are "made in USA"? I sure can't think of any. So to me this sounds like "They won't add *our* backdoors, only their own." where apparently with most phones it's both sides have backdoors.
Well, since what they said was only that 7 out of 13 "have been shown" it's not that surprising. It could easily be 13 our of 13, and that would still be honest. It's like the ISP promise of a connection speed of "up to 10 mbps". It's perfectly honest even if you can't raise the carrier at all.
Generally these days "Liberal" just seems to mean people I don't like with different opinions. I haven't heard anyone self-identify as a liberal for nearly a decade. Part of the reason is it never had a solid definition (in the US)...or at least I don't remember it having one.
And these days most of the people who self-identify as conservatives tend to be xenophobes. The ideological conservatives of several stripes seem to have abandoned the term.
Now that's odd. I often scan the NYT headlines...and I tend to think that they have a right wing slant, though not a strong one. I'll admit I never read their stories, and perhaps if I did I'ld have a different opinion.
FWIW, in my opinion newspapers highly process all their stories to make them more exciting. Using them as "sources of reputable information" isn't viable...but then there *aren't* any accurate sources of politically important information. And it seems plausible to me that there never were. It was said a long time ago that the power of the press belongs to the man who owns one (paraphrase of A. J. Liebling https://en.wikipedia.org/wiki/...). And this may be the true reason that control of ISPs has been centralized.
So would you consider an autonomous fork-lift a robot? If I understand you correctly, the answer is yes. And also a self-driving car, even if it's legally required to carry an official "driver"?
I think of those as actual robots, albeit primitive ones. And the robot car, at least, is required to have "actual situational evaluation". The fork-lift might not be except for things like bumper guards, which can be "pre-scripted", but only if it is restricted to operating in an extremely simplified environment. Say, a warehouse with lots of coded storage bins, and paths laid out along the floor for it to travel. So I don't consider a Roomba, at least as I've seen it described, as a robot. It's merely an automated vacuum cleaner. There are, however, reasonable arguments that it's at the very low end of actual robot, largely because the word doesn't seem well defined. It's definitely not just a telefactor, but I'd think of it as an automated cleaner rather than as a robot.
I gave a usable definition of intelligence. What's your's? Saying it's "not software" doesn't help much. And that some things with intelligence wouldn't have very much of it is pretty much inherent in anything that comes in variable quantities.
I'll agree that the definition I have isn't intuitively obvious, and that it will actually draw boundaries that many people disagree with. But it does allow bounds to be drawn, and presents an objectively quantifiable measure. Perhaps there's a better definition, if so I'd be happy to adopt it, as the one I offered while objective doesn't address, e.g., how effectively the entity solves the problem it's dealing with. But it needs to be something that is, in principle measurable by and objective test over a very large range of degrees. So, e.g., it can't do things like requiring that you manage language, because mice have a certain level of intelligence. I personally believe that there exist minimal forms of intelligence, i.e. instances which are as simple as it is possible to be, but this wouldn't be a feature of all possible definitions.
Do you distinguish between a robot and an automated tool? What's you're definition? I often see the term robot used in a context where telefactor would be more appropriate. What distinction do you make between a robot and a numerically controlled tool?
To me a robot must have significant intelligence. Not just internal logic, but actual situational evaluation. Behavior that has been called "Sphexish" (like that of a particular wasp) doesn't qualify as intelligence, as it appears to be all preprogrammed responses to situational cues. (In another place I argue that this kind of thing is minimally intelligent, and I'll stand by that, but the emphasis is on minimal.)
A definite point, and since this is about manufacturing work, a significant one that's unanswerable.
For some jobs a robot can be worth 100,000 workers. For many others worth less than one. E.g., how many workers do you think it would take to replace an automated crane that can lift 5 tons 5 stories in half an hour? But generally an automated crane wouldn't be able to do anything else...or at least not much else.
The USPTO shouldn't be punish, it should be abolished, and all it's prior decisions invalidated.
Then, perhaps, it might be advisable to start a new patent office based on the law that set up the first US one. With STRICT limitations on the length of patent validity related to the legitimate up-front costs of development, a 5 year renewal, arbitrarily renewable for additional 5 year terms, but the cost of each renewal the square of the cost of prior one measured in cents, and a first renewal fee of less than a dollar forbidden.
Sorry, that battle was lost decades ago, and I doubt there's anyway to reverse it. The only way I can imagine is Hollywood don't a blockbuster film with an idealized hacker as the hero. And that probably wouldn't work.
I know they didn't invent it. It goes back to at least the "company towns" with their own money, and the stores that would sell things for company currency. Probably further. It was evil then, too.
That doesn't work because: 1) Wire/cable access to customer property requires a government approved easement over lots of neighbors' property. 2) Wire/cable access is a natural monopoly. You don't want every new would-be ISP stinging wires and cables throughout the area.
The reasonable approach is for every locality to manage its own hardware layer, but most don't have the technical capability, so it would need to be contracted out. Which it is. The problem is the company stringing the wires/cables feels that it has a right to control access to those wires and cables at above the hardware layers...and to an *extent* it needs to. But the upper layers of the 7-layer stack should be open. Which doesn't mean free. You can't allow everyone to pick their IP address as they choose. But it means the people who run the hardware layer shouldn't run the IP layer.
Back when the perceptrons were in vogue chips were expensive, so as much as possible things were put together out of discrete components. Then Minsky and Papert wrote this paper called "Perceptrons" which showed some limitations that they had, and this was misunderstood by almost everyone to be much more extreme than they had actually claimed.
They were right, but they were describing a network with only one layer...and they showed that no single layer network of perceptrons could learn to implement XOR. But this was misunderstood and reported as "perceptron networks can't learn to implement XOR". So just about everyone lost interest in them for around 20 years.
When you hear the term "deep learning" what they're talking about is essentially multi-layer perceptron networks. So they're back in vogue, and now chip designs are easier to fabricate, and discrete components are rare...so now they're starting to build chips for multi-layered perceptron networks. But this is a bit tricky, since a part of the program of a perceptron network is in the layout of connections between the perceptrons. So different groups are experimenting with different designs. It's believed that "recurrent networks" are important, and what this means is that some cells on the upper layers feed signals to the lower levels, not just up. And there's also the question of how much horizontal signal passing is best..and which. (And it seems to depend on what you're doing. For visual field analysis horizontal passing is extremely important. For other things much less so.)
While I agree with your words, I think I give them a rather different meaning.
There's intelligence, and there's the things it's embedded in. Some intelligence is embedded in artificial devices, but it's not an artificial intelligence.
Currently the purposely designed intelligences are rather weak, but they are still real intelligences.
And actually there are artificial intelligences. Robby in "Forbidden Planet" was an artificial intelligence. The actual intelligence moved the body that appeared in the movie, but the body, i.e. the robot, was either a disguise or a telefactor, not a real intelligence of any level. The real intelligence was always hidden. So Robby was an artificial intelligence. Similarly Bugs Bunny is an artificial intelligence. the character has no real intelligence, the appearance of intelligence is created by people making drawings that are projected in sequence, to create the impression of an intelligence, so it is an artificial intelligence.
OTOH, the thermostat that adjusts the temperature to fall within a particular range is an actual intelligence. It's perhaps the simplest intelligence possible, but it's a real intelligence. I.e. it manages a feed-forwards/feed-back loop to achieve particular ends. (Not always successfully, but intelligence isn't always successful, so that's not a counter argument.) That it didn't choose its ends is almost irrelevant. You didn't choose most of yours, either. You only chose ways to accomplish them.
If it can respond to "the wake word" then you know it's listening all the time. Just like if a computer can "wake on net" you know the off switch doesn't turn off the power.
That's not clear. The description is quite vague. It could be full of various forms of dynamically alterable persistent memory, which AI find very useful, but which normal chips haven't been designed for. It could have TensorFlow built into a ROM component. There's lots of ways it could be different. If you mean it won't be a separately intelligent chip, sure, that's right. That that doesn't mean it will be a normal chip. One guy is talking about essentially building an analog computer with persistent memory in a chip as a component of an AI. (I think it operated on gradations of voltage rather than on and off, but the article wasn't detailed enough to be sure.)
Still, it could just be a custom chip of a rather conventional design. But the source doesn't appear to be explicit enough to make that a valid deduction.
There are ways in which people are difficult to predict, but there are lots of ways in which they are quite predictable. We'd waste a huge amount of time trying to, say, figure out which way we were going to tie our shoes otherwise. The interesting thing is we often guess wrong about which of our actions are hard to predict and which are easy. That's part of how con-men and stage magicians make their living.
It's also true that the very concept of "free will" isn't one that people can really define without subscribing to magic or equating it with randomness. It doesn't really seem to be either. What it seems to be is the result of a tremendous amount of compactly coded information compressed in a lossy fashion which is the basis on which decisions are made. If you knew enough it would probably be totally predictable, if you knew enough, except for some small amount of error which would be random. But the compression algorithm seems to be unique for each person, and the result of developmental changes.
I probably should have thrown in a few more "seems" and "probably"s in the prior paragraph, because there's lots of places where there is more evidence needed. But that's the way it looks to me, and that view is consistent with all the evidence I know of.
Slashdot Mirror
I'm not really sure I approve of the police doing this any more than a random citizen. At the very least it should require approval by three separate courts and a public notice (which the target, of course, anonimized). And public notice doesn't mean a posting in some inaccessible place, but listing on a web page, something like:
2018/02/18 15:27 warrant approved until 2018/02/25 15:30 to (stalk?..need better description) (some explicit description of what is to be surveilled).
The explicit description of the target should probably by a value generated by a hash function of the IP address.
Intel not only made dangerously broken CPUs which had been predicted to be dangerously broken (without a definite exploit) before they were designed, but if they didn't already know about how to exploit it, they were informed at least 6 months before the public notice, and appear to have taken no steps to mitigate the problem prior to public notice. We can't really know, but the patches that they rushed out after notice was made public were so poor that they probably hadn't done anything.
Etc.
I'm willing for Intel to prove that they were acting in a reasonable and ethical manner, but the preponderance of the evidence seems against that assumption.
Actually, asteroids *do* occasionally hit each other. It doesn't happen every day, but we don't really have a real good idea of how often. Often enough that we can observe the after effects, but we can only see that when two large chunks collide. For the smaller pieces...we can be sure it happens because there are a lot more of them, and even the larger pieces collide, but we can't really observe.
OTOH, frequency... it's probably rather rare, and most of the collisions are at low relative speeds. Fender-benders are a lot more common then head on collisions. But occasionally things will hit each other at a large enough angle to shatter. I don't know what occasionally means, it might be once every thousand years for something the mass of a car. Or it might be less frequently.
Yes, space is big, and there's a lot of space on the average between things. But most things are about in the plane of rotation, and resonance effects tend to shift things into similar orbits and ... well, when things are nearly co-planar and the eccentricities are different, eventually a collision is likely. Of course, eventually can take a very long time, but with millions of pieces of stuff orbiting around they do happen.
Think of all the free advertising mileage he got/ is getting out of it.
It's probably a net savings, and the advertising is a lot less annoying.
I don't think "liberal" administrations are any less given to spying or centralization of power than are "conservative" administrations. They just spend a bit more on the below median income citizen. You can say this is good or bad, and there are reasonable arguments either way, but neither is "small government" or "get the government off the backs of the people" or any of the similar slogans that have been used in past campaigns. They only tend to lie about how they'll do that when they get power again when they're out of power.
So there's no reason to be surprised that Obama's administration was caught spying...except, of course, that they shouldn't have been caught.
Another question: What phones are "made in USA"? I sure can't think of any. So to me this sounds like "They won't add *our* backdoors, only their own." where apparently with most phones it's both sides have backdoors.
Well, since what they said was only that 7 out of 13 "have been shown" it's not that surprising. It could easily be 13 our of 13, and that would still be honest. It's like the ISP promise of a connection speed of "up to 10 mbps". It's perfectly honest even if you can't raise the carrier at all.
Generally these days "Liberal" just seems to mean people I don't like with different opinions. I haven't heard anyone self-identify as a liberal for nearly a decade. Part of the reason is it never had a solid definition (in the US)...or at least I don't remember it having one.
And these days most of the people who self-identify as conservatives tend to be xenophobes. The ideological conservatives of several stripes seem to have abandoned the term.
Now that's odd. I often scan the NYT headlines...and I tend to think that they have a right wing slant, though not a strong one. I'll admit I never read their stories, and perhaps if I did I'ld have a different opinion.
FWIW, in my opinion newspapers highly process all their stories to make them more exciting. Using them as "sources of reputable information" isn't viable...but then there *aren't* any accurate sources of politically important information. And it seems plausible to me that there never were. It was said a long time ago that the power of the press belongs to the man who owns one (paraphrase of A. J. Liebling https://en.wikipedia.org/wiki/...). And this may be the true reason that control of ISPs has been centralized.
So would you consider an autonomous fork-lift a robot?
If I understand you correctly, the answer is yes. And also a self-driving car, even if it's legally required to carry an official "driver"?
I think of those as actual robots, albeit primitive ones. And the robot car, at least, is required to have "actual situational evaluation". The fork-lift might not be except for things like bumper guards, which can be "pre-scripted", but only if it is restricted to operating in an extremely simplified environment. Say, a warehouse with lots of coded storage bins, and paths laid out along the floor for it to travel. So I don't consider a Roomba, at least as I've seen it described, as a robot. It's merely an automated vacuum cleaner. There are, however, reasonable arguments that it's at the very low end of actual robot, largely because the word doesn't seem well defined. It's definitely not just a telefactor, but I'd think of it as an automated cleaner rather than as a robot.
I gave a usable definition of intelligence. What's your's? Saying it's "not software" doesn't help much. And that some things with intelligence wouldn't have very much of it is pretty much inherent in anything that comes in variable quantities.
I'll agree that the definition I have isn't intuitively obvious, and that it will actually draw boundaries that many people disagree with. But it does allow bounds to be drawn, and presents an objectively quantifiable measure. Perhaps there's a better definition, if so I'd be happy to adopt it, as the one I offered while objective doesn't address, e.g., how effectively the entity solves the problem it's dealing with. But it needs to be something that is, in principle measurable by and objective test over a very large range of degrees. So, e.g., it can't do things like requiring that you manage language, because mice have a certain level of intelligence. I personally believe that there exist minimal forms of intelligence, i.e. instances which are as simple as it is possible to be, but this wouldn't be a feature of all possible definitions.
IQ doesn't work at all, not even between humans.
Do you distinguish between a robot and an automated tool?
What's you're definition? I often see the term robot used in a context where telefactor would be more appropriate.
What distinction do you make between a robot and a numerically controlled tool?
To me a robot must have significant intelligence. Not just internal logic, but actual situational evaluation. Behavior that has been called "Sphexish" (like that of a particular wasp) doesn't qualify as intelligence, as it appears to be all preprogrammed responses to situational cues. (In another place I argue that this kind of thing is minimally intelligent, and I'll stand by that, but the emphasis is on minimal.)
A definite point, and since this is about manufacturing work, a significant one that's unanswerable.
For some jobs a robot can be worth 100,000 workers. For many others worth less than one. E.g., how many workers do you think it would take to replace an automated crane that can lift 5 tons 5 stories in half an hour? But generally an automated crane wouldn't be able to do anything else...or at least not much else.
The USPTO shouldn't be punish, it should be abolished, and all it's prior decisions invalidated.
Then, perhaps, it might be advisable to start a new patent office based on the law that set up the first US one. With STRICT limitations on the length of patent validity related to the legitimate up-front costs of development, a 5 year renewal, arbitrarily renewable for additional 5 year terms, but the cost of each renewal the square of the cost of prior one measured in cents, and a first renewal fee of less than a dollar forbidden.
And, as per your analysis, a patent case coming to a decision that seems reasonable is newsworthy. At least one involving a technical matter.
What the constitution defined, and what the patent office is, have very little relation to each other.
Sorry, that battle was lost decades ago, and I doubt there's anyway to reverse it. The only way I can imagine is Hollywood don't a blockbuster film with an idealized hacker as the hero. And that probably wouldn't work.
Not if Comcast could select the arbitrator.
I know they didn't invent it. It goes back to at least the "company towns" with their own money, and the stores that would sell things for company currency. Probably further. It was evil then, too.
That doesn't work because:
1) Wire/cable access to customer property requires a government approved easement over lots of neighbors' property.
2) Wire/cable access is a natural monopoly. You don't want every new would-be ISP stinging wires and cables throughout the area.
The reasonable approach is for every locality to manage its own hardware layer, but most don't have the technical capability, so it would need to be contracted out. Which it is. The problem is the company stringing the wires/cables feels that it has a right to control access to those wires and cables at above the hardware layers...and to an *extent* it needs to. But the upper layers of the 7-layer stack should be open. Which doesn't mean free. You can't allow everyone to pick their IP address as they choose. But it means the people who run the hardware layer shouldn't run the IP layer.
Well....
Back when the perceptrons were in vogue chips were expensive, so as much as possible things were put together out of discrete components. Then Minsky and Papert wrote this paper called "Perceptrons" which showed some limitations that they had, and this was misunderstood by almost everyone to be much more extreme than they had actually claimed.
They were right, but they were describing a network with only one layer...and they showed that no single layer network of perceptrons could learn to implement XOR. But this was misunderstood and reported as "perceptron networks can't learn to implement XOR". So just about everyone lost interest in them for around 20 years.
When you hear the term "deep learning" what they're talking about is essentially multi-layer perceptron networks. So they're back in vogue, and now chip designs are easier to fabricate, and discrete components are rare...so now they're starting to build chips for multi-layered perceptron networks. But this is a bit tricky, since a part of the program of a perceptron network is in the layout of connections between the perceptrons. So different groups are experimenting with different designs. It's believed that "recurrent networks" are important, and what this means is that some cells on the upper layers feed signals to the lower levels, not just up. And there's also the question of how much horizontal signal passing is best..and which. (And it seems to depend on what you're doing. For visual field analysis horizontal passing is extremely important. For other things much less so.)
While I agree with your words, I think I give them a rather different meaning.
There's intelligence, and there's the things it's embedded in. Some intelligence is embedded in artificial devices, but it's not an artificial intelligence.
Currently the purposely designed intelligences are rather weak, but they are still real intelligences.
And actually there are artificial intelligences. Robby in "Forbidden Planet" was an artificial intelligence. The actual intelligence moved the body that appeared in the movie, but the body, i.e. the robot, was either a disguise or a telefactor, not a real intelligence of any level. The real intelligence was always hidden. So Robby was an artificial intelligence. Similarly Bugs Bunny is an artificial intelligence. the character has no real intelligence, the appearance of intelligence is created by people making drawings that are projected in sequence, to create the impression of an intelligence, so it is an artificial intelligence.
OTOH, the thermostat that adjusts the temperature to fall within a particular range is an actual intelligence. It's perhaps the simplest intelligence possible, but it's a real intelligence. I.e. it manages a feed-forwards/feed-back loop to achieve particular ends. (Not always successfully, but intelligence isn't always successful, so that's not a counter argument.) That it didn't choose its ends is almost irrelevant. You didn't choose most of yours, either. You only chose ways to accomplish them.
If it can respond to "the wake word" then you know it's listening all the time. Just like if a computer can "wake on net" you know the off switch doesn't turn off the power.
That's not clear. The description is quite vague. It could be full of various forms of dynamically alterable persistent memory, which AI find very useful, but which normal chips haven't been designed for. It could have TensorFlow built into a ROM component. There's lots of ways it could be different. If you mean it won't be a separately intelligent chip, sure, that's right. That that doesn't mean it will be a normal chip. One guy is talking about essentially building an analog computer with persistent memory in a chip as a component of an AI. (I think it operated on gradations of voltage rather than on and off, but the article wasn't detailed enough to be sure.)
Still, it could just be a custom chip of a rather conventional design. But the source doesn't appear to be explicit enough to make that a valid deduction.
There are ways in which people are difficult to predict, but there are lots of ways in which they are quite predictable. We'd waste a huge amount of time trying to, say, figure out which way we were going to tie our shoes otherwise. The interesting thing is we often guess wrong about which of our actions are hard to predict and which are easy. That's part of how con-men and stage magicians make their living.
It's also true that the very concept of "free will" isn't one that people can really define without subscribing to magic or equating it with randomness. It doesn't really seem to be either. What it seems to be is the result of a tremendous amount of compactly coded information compressed in a lossy fashion which is the basis on which decisions are made. If you knew enough it would probably be totally predictable, if you knew enough, except for some small amount of error which would be random. But the compression algorithm seems to be unique for each person, and the result of developmental changes.
I probably should have thrown in a few more "seems" and "probably"s in the prior paragraph, because there's lots of places where there is more evidence needed. But that's the way it looks to me, and that view is consistent with all the evidence I know of.