In practice, you don't want to spend TOO much time optimizing most code, because it tends not to get run a zillion times.
True, but what if you store the results of your optimizations back to disk (say, as attributes of the executable file) so that they can be re-used the next time the executable is run? Then instead of having to re-optimize for the same code every time the program is launched, you only have to optimize the first time... after that, you'll get better-than-compiled performance every time, with no slowdowns due to optimization.
Ambition, desire, love, hate, or anything else. I didn't use the term "independently" in a very strict sense
And what are those emotions to be based on? In humans, they occur because they evolved -- because they provided survival value to the ancestors of the people who have them. In a robot, they would occur either because the programmer put them there, or because the programmer put in other goals (such as survival) and the robot's learning algorithm discovered that these emotions (or whatever you want to call them) were a useful way to meet an existing goal. But in either case, the goals were still pre-established by the designer.
It's true that any kind of goal-setting can be traced back to some kind of external stimulus, but in the case of humans, the process is not entirely rational, or based on logic.
Au contraire, the process is very rational and logical -- it's just not the Descartes-style rationality and logic that most people think about. It's the unconscious algorithms collected in our genes and culture over a few million years of breeding, fighting, gathering food, living, or dying. It's implemented in the deeper, less accessible regions of our brains (or in some cases, directly in the body itself), and not readily available for conscious introspection, but it's all still algorithms nonetheless. If you want similar things in an AI, you're going to have to program them all in yourself, or devise a method (probably evolutionary) to generate them.
You're telling me that you honestly believe that there's been noone that has ever stuck a stick of dynamite up their ass or pussy?
He's not saying that nobody tried.... just that nobody succeeded. Seriously, anything more explosive than pop rocks wouldn't be described as "gratifying", no matter how twisted your fetish.
But what does that mean, "arms"? Does it mean just guns? Surely it doesn't. It means whatever weapons deemed necessary to overthrow a corrupt government.
Clearly, it means nukes. Only with the force of Mutually Assured Destruction on our side can we be sure that we could, if push came to shove, defeat our nuclear-armed government. Which is why I advocate providing one free nuclear device to each American citizen on his/her 18th birthday. Only then can we have the violence-free utopian society we've all dreamed of.
A true AI would be able to independently set goals for itself
Based on what? Ambition? Ambition to do what? I don't think even human beings are able to "independently set goals for themselves"... they can set sub-goals (e.g. "I think I will go to college so I can get a better job") but even those can always be traced back to a goal that was hard-coded into them by nature (e.g. obtaining warmth/shelter/comfort/food/sex). Presumably an AI would be the same... it might do surprising things, but only things that were (intentionally or unintentionally) results of the nature that you originally gave it.
I'm sure the article was meant to be insightful, but are we really that stupid a species that we need to make all of our tools entirely fool proof?
Depends on who is going to be using the tools. If the tool is a $500,000 industrial welding robot and will be used only by trained professionals, no. If the tool is a $5,000 cutesy humanoid thingy whose job is to tuck Grandma into bed and keep her from falling down the stairs... yes, absolutely.
There's nothing more "robotic" about a mechanical arm in a factory then any mechanical device someone could be a twit and die around. This story is sensationalist drivel.
Except that the story isn't about industrial robots in 1981, it's about consumer robots that Japan, Korea, etc, expect to have working in people's homes, caring for the elderly, etc, in a few years. You can call it "sensationalist drivel" and enjoy your smug feeling of superiority over those dumb journalists, but the fact remains that if robots are going to be working in people's homes (where safety procedures are routinely ignored), there will be significant safety issues to resolve.
Why? Once you get to the point of requiring willful stupidity to get killed by a machine, any deaths are a good thing.
When your company has to pay $50 million in a wrongful death lawsuit (whether you think it was justified or not), your accountants will feel differently.
Even if we had the capabilities to create AI capable of following the three laws, putting it into a robot would dramatically increase the costs involved
If you're going to go blue-sky on us and imagine a scenario where we have the ability to make strong AI robots, you might as well assume that we can make them cheaply as well. Hell, if nothing else, you could make some robots that know how to build more robots... that ought to cut costs pretty quick.
Just be sure to make them so that they enjoy welding cars....
Any organisation beyond a certain size inevitably becomes pathological in its behaviour.
Agreed... when the company is below a certain size, everybody can exist within the same monkeysphere, and several hundred thousand years of social evolution help things along. In much larger organizations, multiple monkeyspheres form, leading to indifference and inefficiency at best, or low-level tribal warfare at worst.
ECHELON was a Clinton era problem, don't blame W for this intrusion because you don't like his politics.
This program is not ECHELON. This program was developed by NSA under orders from the Bush Administration. So why shouldn't I blame Bush for this program?
The spy program you say was "above the law" was briefed to congressmen of both parties and nobody raised the issue. Dems came out against spying on terrorists only after the news broke, although many new of the program earlier.
The part you're missing is that all congressmen who were briefed on the program were also sworn to secrecy -- it was illegal for them to talk about it until after it was already public knowledge. And I'm sure if they had talked about it you would be damning them as a bunch of traitorous leakers for doing so.
Not to mention that the law is defined as "what's on the law books", not "whatever the party in power can get away with".
Thus, either is Microsoft producing PDF-s that open and print in Reader, or their PDF support will just be useless
... at which point the conventional wisdom would soon become "don't use.PDF format, that format doesn't work reliably anymore. Use Microsoft's format foo instead, it always works correctly.". You can see why Adobe would not want that to happen.
In you first paragraph you state that 'graphics designers use Macintoshes extensively...' but then in your second paragraph you maintain 'Platform non-neutrality is shooting yourself in the foot in the printing business.'
Which of the two assertions do you support? You're contradicting yourself.
No, he isn't -- a platform-neutral format would support any platform (Macs, Windows, Linux, mainframes, whatever) equally well. That was his point. He was just using the popularity of Macs as an example of why a Windows-only solution is insufficient.
I don't think ants would be very interested in diet coke... it's sugar they like, and diet soda doesn't have any. Perhaps they would go for the Mentos though... nobody likes an ant with bad breath.
ACK! Hydrochloric acid? That sounds like a Darwin Award waiting to happen. Liquid Nitrogen makes a pretty "cool" bang without having to handle any acids
Dry ice will work as well, and is probably easier to get/use than either liquid nitrogen or hydrochloric acid.
Back when I was single-digit aged, I thought it would be pretty cool to "program" a CYOA book into our Vic20.
Heh, I tried the exact same thing on my TI 99/4A. I discovered that not only does the TI only have enough room for about 17 pages of text (rendered as PRINT statements), but it also gets very very slow as you near the memory-full point...
If I'm to apply the same logic to regular mail, well, regular mail is doomed too; it's full of phishing, spam, and spoofing.
Phishing, really? I don't think I've ever gotten any letters pretending to be from my bank, that were really from some Nigerian criminal.
And why is that? Because in order for it to be profitable to the criminal, they'd need me to reply (and eventually send them money, or information that would allow them to get my money). And since faking a plausible physical address is difficult and dangerous, they don't bother.
Email, on the other hand, is much easier to forge, and so I get fake emails from "banks" every day.
So why not make email harder to forge, instead of forcing every grandmother and 6-year-old child in the world to become an email forgery expert?
How exactly does this new email system stop phishing? Oh, right, it can't. Have a link, go to a malicious website, etc. How exactly does this new email system stop users from clicking executables thinking that they are going to see nudie pictures of Katie Holmes? They don't. How does this new email stop virii? It won't
Nothing is perfect, but having reliable source authentication (so that everyone can easily tell which emails are really from PayPal and which are from criminals pretending to be PayPal) would go a long way towards minimizing the problems caused by phishing.
The clicking-on-executables problem could be addressed by tagging executable that arrived via unauthenticated email as "untrusted", and either refusing to run them, or allowing them to be run only in a secure/sandboxed environment.
Sorry, but to be taken seriously, you'd at least have to have a basic framework already thought out.
So nobody is allowed to point out the email has problems until the solutions are already known? But if nobody is allowed to discuss the problems, how will the solutions ever be found?
Go back, think about it and then write a real article
This article is useful in that it gets people thinking about the problem. Now some clever person can come up with a proposed solution and post an article about it. That's how it works.
Slashdot Mirror
True, but what if you store the results of your optimizations back to disk (say, as attributes of the executable file) so that they can be re-used the next time the executable is run? Then instead of having to re-optimize for the same code every time the program is launched, you only have to optimize the first time... after that, you'll get better-than-compiled performance every time, with no slowdowns due to optimization.
There is one reason: it's the blindingly obvious thing to call people who... live in America.
Be that as it may, you'd still be out $50 million. You have to design products for the world you live in, not the world you wished you lived in.
And what are those emotions to be based on? In humans, they occur because they evolved -- because they provided survival value to the ancestors of the people who have them. In a robot, they would occur either because the programmer put them there, or because the programmer put in other goals (such as survival) and the robot's learning algorithm discovered that these emotions (or whatever you want to call them) were a useful way to meet an existing goal. But in either case, the goals were still pre-established by the designer.
It's true that any kind of goal-setting can be traced back to some kind of external stimulus, but in the case of humans, the process is not entirely rational, or based on logic.
Au contraire, the process is very rational and logical -- it's just not the Descartes-style rationality and logic that most people think about. It's the unconscious algorithms collected in our genes and culture over a few million years of breeding, fighting, gathering food, living, or dying. It's implemented in the deeper, less accessible regions of our brains (or in some cases, directly in the body itself), and not readily available for conscious introspection, but it's all still algorithms nonetheless. If you want similar things in an AI, you're going to have to program them all in yourself, or devise a method (probably evolutionary) to generate them.
He's not saying that nobody tried.... just that nobody succeeded. Seriously, anything more explosive than pop rocks wouldn't be described as "gratifying", no matter how twisted your fetish.
Clearly, it means nukes. Only with the force of Mutually Assured Destruction on our side can we be sure that we could, if push came to shove, defeat our nuclear-armed government. Which is why I advocate providing one free nuclear device to each American citizen on his/her 18th birthday. Only then can we have the violence-free utopian society we've all dreamed of.
Based on what? Ambition? Ambition to do what? I don't think even human beings are able to "independently set goals for themselves"... they can set sub-goals (e.g. "I think I will go to college so I can get a better job") but even those can always be traced back to a goal that was hard-coded into them by nature (e.g. obtaining warmth/shelter/comfort/food/sex). Presumably an AI would be the same... it might do surprising things, but only things that were (intentionally or unintentionally) results of the nature that you originally gave it.
Dude, you have the code. All three lines of it.
Of course, finding a compiler that can correctly compile that code may be tricky...
Depends on who is going to be using the tools. If the tool is a $500,000 industrial welding robot and will be used only by trained professionals, no. If the tool is a $5,000 cutesy humanoid thingy whose job is to tuck Grandma into bed and keep her from falling down the stairs... yes, absolutely.
Except that the story isn't about industrial robots in 1981, it's about consumer robots that Japan, Korea, etc, expect to have working in people's homes, caring for the elderly, etc, in a few years. You can call it "sensationalist drivel" and enjoy your smug feeling of superiority over those dumb journalists, but the fact remains that if robots are going to be working in people's homes (where safety procedures are routinely ignored), there will be significant safety issues to resolve.
When your company has to pay $50 million in a wrongful death lawsuit (whether you think it was justified or not), your accountants will feel differently.
If you're going to go blue-sky on us and imagine a scenario where we have the ability to make strong AI robots, you might as well assume that we can make them cheaply as well. Hell, if nothing else, you could make some robots that know how to build more robots... that ought to cut costs pretty quick.
Just be sure to make them so that they enjoy welding cars....
Agreed... when the company is below a certain size, everybody can exist within the same monkeysphere, and several hundred thousand years of social evolution help things along. In much larger organizations, multiple monkeyspheres form, leading to indifference and inefficiency at best, or low-level tribal warfare at worst.
Better buy both a male and a female, or you may find that the offspring isn't so hypoallergenic after all. Or figure out how to clone her...
How about this one? Older technology, but it's only half the price...
This program is not ECHELON. This program was developed by NSA under orders from the Bush Administration. So why shouldn't I blame Bush for this program?
The spy program you say was "above the law" was briefed to congressmen of both parties and nobody raised the issue. Dems came out against spying on terrorists only after the news broke, although many new of the program earlier.
The part you're missing is that all congressmen who were briefed on the program were also sworn to secrecy -- it was illegal for them to talk about it until after it was already public knowledge. And I'm sure if they had talked about it you would be damning them as a bunch of traitorous leakers for doing so.
Not to mention that the law is defined as "what's on the law books", not "whatever the party in power can get away with".
Which of the two assertions do you support? You're contradicting yourself.
No, he isn't -- a platform-neutral format would support any platform (Macs, Windows, Linux, mainframes, whatever) equally well. That was his point. He was just using the popularity of Macs as an example of why a Windows-only solution is insufficient.
I don't think ants would be very interested in diet coke... it's sugar they like, and diet soda doesn't have any. Perhaps they would go for the Mentos though... nobody likes an ant with bad breath.
Dry ice will work as well, and is probably easier to get/use than either liquid nitrogen or hydrochloric acid.
Heh, I tried the exact same thing on my TI 99/4A. I discovered that not only does the TI only have enough room for about 17 pages of text (rendered as PRINT statements), but it also gets very very slow as you near the memory-full point...
I vaguely remember the days when culture had something to do with people, not just competing marketing departments...
Phishing, really? I don't think I've ever gotten any letters pretending to be from my bank, that were really from some Nigerian criminal.
And why is that? Because in order for it to be profitable to the criminal, they'd need me to reply (and eventually send them money, or information that would allow them to get my money). And since faking a plausible physical address is difficult and dangerous, they don't bother.
Email, on the other hand, is much easier to forge, and so I get fake emails from "banks" every day.
So why not make email harder to forge, instead of forcing every grandmother and 6-year-old child in the world to become an email forgery expert?
They don't. How does this new email stop virii? It won't
Nothing is perfect, but having reliable source authentication (so that everyone can easily tell which emails are really from PayPal and which are from criminals pretending to be PayPal) would go a long way towards minimizing the problems caused by phishing.
The clicking-on-executables problem could be addressed by tagging executable that arrived via unauthenticated email as "untrusted", and either refusing to run them, or allowing them to be run only in a secure/sandboxed environment.
So nobody is allowed to point out the email has problems until the solutions are already known? But if nobody is allowed to discuss the problems, how will the solutions ever be found?
Go back, think about it and then write a real article
This article is useful in that it gets people thinking about the problem. Now some clever person can come up with a proposed solution and post an article about it. That's how it works.