On Windows, I can buy whatever mouse/keyboard/monitor/CPU/GPU/sound card/USB headphones/USB drive/etc. that I want, plug it into my PC, and it will work correctly 90% of the time. In fact, most of the time I won't even have to think about it. The other 10% of the time, I'll need to go to Windows Update (or, in very rare cases, the manufacturer's website) to get the driver... which will install itself completely automatically.
On Linux, I have to carefully research which hardware works and which doesn't; which config files I need to edit by hand; and -- if I'm feeling adventurous -- which kernel flags I need to unset to get it all to work. If I'm very lucky, my new hardware will work in some capacity; it will almost never be 100% functional, but maybe I can get it to the point where it's good enough.
This is a massive problem for everyone, but especially for gamers, who absolutely must have their GPU, monitor, audio, and network working correctly and at peak performance. This is also absolutely not a problem for ML developers, network admins, etc., who operate on clusters of 1,000 machines and couldn't care less about all of the peripherals. Guess what, though... there are way more regular users out there than AI network admins.
Now...What should be the purpose of an artificial intelligence? What should it accomplish in its lifetime?
Same as the purpose of a real intelligence, I'd imagine. While it is true that we're programmed to survive and procreate, that's not all we do; and it's not even a major portion of what we do (especially here on Slashdot, I might add). I don't see why the AI should be different.
But, if you're looking for practical applications, then I can think of some off the top of my head:
Machine Rranslation. We need something better than Babelfish to translate Mandarin Chinese to Russian in a way that is not merely comprehensible (and we don't even have that now), but actually good. Today, not many humans can do this. The task is difficult; however, competent translation is basically the Turing Test with an extra language on top.
Tech Support. Currently, it's outsourced to people in India, whose sole purpose is to pass a version of the Turing Test where, instead of pretending to be human, they are pretending to be engineers who care about you, the customer. An AI would be even cheaper.
Ubiquitous Voice Recognition. In general, any modern device could benefit from robust voice recognition. Instead of clicking around the menus on your Blackberry, you'd simply tell it, "What are my appointments for today ?", and it would read them to you. We almost have it today, but it's not nearly as robust as it needs to be.
Content Filtering. Netflix, TiVo, and Google are all investing impressive sums of money into developing an intelligence that knows what kind of content (movies, music, email) you'd want to see.
Gaming. Starcraft cheats (at least, on "Hard" mode it does). Mobs and NPCs in most MMOs are fairly static. It would be more interesting to fight against opponents (or, alternatively, with allies) who were at least as smart as an average human.
I just made up that list on the fly; I'm sure there are many other practical applications.
Even if it managed to develop "true" intelligence, which I very much doubt, how could we expect it to be anything other than dangerously insane from our perspective?
As I'd mentioned above, one way to do it would be to take shortcuts. Instead of waiting for desirable behaviors and thought patterns to arise spontaneously, we could program in the behaviors we wanted, or we could bias the learning engine to make it difficult to learn the behaviors we don't want.
Even if our goal is to create an AI that thinks and acts like a human, it doesn't mean that we need to develop it in the exact same way as biological humans develop in nature.
That said, I would agree that you shouldn't have to give a machine anything more than basic resources to begin its process of learning...
That depends on what your goal is. If your goal is to reproduce the process of human mental development, from a child to an adult, in silico, then I agree. However, if your goal is merely to produce an intelligence that can think at least as well as a human can, then you can take shortcuts -- such as supplying the intelligence with a ready-made database of knowledge, or a built-in library of common tasks ("I know Kung Fu"), etc. As long as the intelligence is as capable of learning and evolving as an average human, I see no harm in starting it off with something it can use.
Or, put it this way: adult humans take 18 years or so to mature; that's a pretty long development cycle. If you're building an AI, you might as well accelerate it as much as you can.
This computer can have no BIOS, OS, no programming at all. When it learns to use its own hardware, figures out network protocols and starts downloading web pages and porn, you have true AI.
That's like saying, "take a human baby, put him in front of an Internet kiosk. Make sure the baby has no nervous system or brain of any kind. Once he figures out how to use his eyes and fingers, and starts googling for porn, you have true natural intelligence". Your requirements are way too restrictive; no human would pass them.
Ok, let's say that we've managed to get the original optics. We've discovered that, due to slight focusing error (which is fixed) and atmospheric effects (which are chaotic), each point on the photographic plate was blurred by 0.01 mm. How do you recover the "infinite" precision from this data ?
This is exactly how transformers work (the power converters, not the robots); it's also similar to how the radio in your car works. Faraday was doing this stuff in the 1800s, I don't see what the big deal is.
Has anyone tried using drobo ? I'm contemplating upgrading my own storage, and they seem to have the least painful solution (as opposed to managing Raid and dealing with disks of different sizes, recovery, etc.), but I have no idea whether their product actually works.
Whichever calculator you get, make sure that it's programmable, and that whatever programming language it uses is convenient to use. I would not have been able to finish any of my labs without my trusty TI-85 -- at least, not without spending several additional hours each lab on things like error propagation - but I ran up against its pseudo-BASIC-esque limitations fairly quickly.
Keep in mind that there is no better way to learn some concept, than writing a program for it. Period. Once you are able to express your knowledge in code (er, in code that actually works, I mean), you have it down cold, or down kold, even.
Could they, though ? At this point, it will just be an arms race between ??AA and Torrent users, and I'm not sure who'd win. In other words, I don't know whether it is physically (or, rather, mathematically) possible to build a tamper-proof peer-to-peer network. It's an interesting question.
but subject to a DRM scheme that allowed full legitimate usage (format shifting, time shifting, playback on different devices, etc.) and only blocked illicit usage (illegal copying)
I object to DRM because I don't believe that it is physically possible to create a DRM scheme that can permit legitimate usage, but block illicit usage, unless the DRM is telepathic. Otherwise, how will it know what I intend to do with the copy I make ?
Well, torrent-poisoning would not even exist if it did not lead to the lawsuits. They can't bring down the entire network with just poisoning, and they know it.
Actually, in the total absence of regulation, this would not be a problem. The MPAA could upload whatever bogus file they wanted; their bogus uploads would be filtered out and everyone would move on. The problem is that, under the current regulatory regime, the MPAA can use their fake torrents to sue people into oblivion.
Yes, it would be nice if all software "just worked", but, until we develop Strong AI (such as HAL... hmm...), this is not possible. Since a user can't just tell his computer, "do my taxes for me, pronto !", the user will have to use his own intelligence to instruct the computer in what to do. This means that they will have to learn how to talk to the computer in its own language. The best software strikes a balance between the steepness of the learning curve, and the power it exposes to the user.
Apple tends to take the more user-friendly road: expose as little power as possible, but make the UI as simple as possible. This is a valid choice. UNIX takes the opposite approach: it's all power, all the time, if you can remember 12 different config files and env variables in your head. Given UNIX's audience, this is also a valid approach.
However, letting your users design your software from the ground up is a terrible idea, because the users are not aware of the limitations of modern technology, nor are they aware of the complexity of their own field (most of the time). In 90% of the cases, what the user truly wants is a button that says "do my work for me"... and we're right back where we started, at the beginning of this post.
Couldn't you level the same allegations at any human, though ? I.e., we can look at the entire employment history of CutCo/EdgeCom going back 10 years, see that they hired more single people than married people, then sue them for discrimination. After all, even though they don't explicitly have a policy that considers marital status, clearly their HR managers are using this as part of their judgement... right ?
But a human would never tell the computer to explicitly consider marriage. Instead, the computer would be trained (or, rather, train itself) to draw conclusions from all kinds of disparate data, which could amount to inferring whether the applicant is married or not. What happens then ?
I think the grandparent poster's main idea was not to defer more of our decisions to scientists, but for everyone to become more scientifically-minded. We live in a world where each individual wields an almost unimaginable amount of power at his fingertips -- just crash your car into a wall at 90 mph to see what I mean -- and it could really help if everyone understood a little about how this power truly works.
How does Yahoo compare to Experts-Exchange ? I used to like that site a lot, and I still use it, but their semi-recent redesign (filthy inline ads, sidebar ads, popup ads, light blue on slightly lighter blue text, ads between comments) has sent me searching for something better. But, I have to admit that the quality of the answers I get on Experts-Exchange is still superb.
IMO Azureus is the best. uTorrent has a way smaller memory footprint, and a much faster startup time, but these things simply don't matter much to me. Azureus's memory use is still reasonable (compared to say, Visual Studio, *shudder*), and its startup time is irrelevant to me because I never shut it down. It doesn't use any measurable amount of CPU, either, but it comes with a plethora of features and plugins, notably DHT, that blow uTorrent out of the water.
Be warned, though: I hear that Azureus is going to re-write their client to act primarily as an on-demand movie viewer, with some weird convoluted UI that makes no sense. So, enjoy it while it lasts...
Um, did these guys just invent a thermocouple ? Those things have been around for quite some time, and are used to generate power on satellites (using a radioactive heat source, because in space, who cares ?), and to make small, inefficient refrigerators that can fit in a car... And of course, to measure temperature pretty much everywhere.
Slashdot Mirror
On Windows, I can buy whatever mouse/keyboard/monitor/CPU/GPU/sound card/USB headphones/USB drive/etc. that I want, plug it into my PC, and it will work correctly 90% of the time. In fact, most of the time I won't even have to think about it. The other 10% of the time, I'll need to go to Windows Update (or, in very rare cases, the manufacturer's website) to get the driver... which will install itself completely automatically.
On Linux, I have to carefully research which hardware works and which doesn't; which config files I need to edit by hand; and -- if I'm feeling adventurous -- which kernel flags I need to unset to get it all to work. If I'm very lucky, my new hardware will work in some capacity; it will almost never be 100% functional, but maybe I can get it to the point where it's good enough.
This is a massive problem for everyone, but especially for gamers, who absolutely must have their GPU, monitor, audio, and network working correctly and at peak performance. This is also absolutely not a problem for ML developers, network admins, etc., who operate on clusters of 1,000 machines and couldn't care less about all of the peripherals. Guess what, though... there are way more regular users out there than AI network admins.
For the good of all of us.
(except the ones who are dead).
Same as the purpose of a real intelligence, I'd imagine. While it is true that we're programmed to survive and procreate, that's not all we do; and it's not even a major portion of what we do (especially here on Slashdot, I might add). I don't see why the AI should be different.
But, if you're looking for practical applications, then I can think of some off the top of my head:
I just made up that list on the fly; I'm sure there are many other practical applications.
As I'd mentioned above, one way to do it would be to take shortcuts. Instead of waiting for desirable behaviors and thought patterns to arise spontaneously, we could program in the behaviors we wanted, or we could bias the learning engine to make it difficult to learn the behaviors we don't want.
Even if our goal is to create an AI that thinks and acts like a human, it doesn't mean that we need to develop it in the exact same way as biological humans develop in nature.
That depends on what your goal is. If your goal is to reproduce the process of human mental development, from a child to an adult, in silico, then I agree. However, if your goal is merely to produce an intelligence that can think at least as well as a human can, then you can take shortcuts -- such as supplying the intelligence with a ready-made database of knowledge, or a built-in library of common tasks ("I know Kung Fu"), etc. As long as the intelligence is as capable of learning and evolving as an average human, I see no harm in starting it off with something it can use.
Or, put it this way: adult humans take 18 years or so to mature; that's a pretty long development cycle. If you're building an AI, you might as well accelerate it as much as you can.
What was it, out of curiosity ? I took an AP Programming class, but it was a looooong time ago, and I don't recall any BigInt questions.
How's it better than random.org or hotbits ?
Ok, let's say that we've managed to get the original optics. We've discovered that, due to slight focusing error (which is fixed) and atmospheric effects (which are chaotic), each point on the photographic plate was blurred by 0.01 mm. How do you recover the "infinite" precision from this data ?
This is exactly how transformers work (the power converters, not the robots); it's also similar to how the radio in your car works. Faraday was doing this stuff in the 1800s, I don't see what the big deal is.
Has anyone tried using drobo ? I'm contemplating upgrading my own storage, and they seem to have the least painful solution (as opposed to managing Raid and dealing with disks of different sizes, recovery, etc.), but I have no idea whether their product actually works.
> You're about six years too late for that.
No, he did specify English-speaking, so he's still early.
Whichever calculator you get, make sure that it's programmable, and that whatever programming language it uses is convenient to use. I would not have been able to finish any of my labs without my trusty TI-85 -- at least, not without spending several additional hours each lab on things like error propagation - but I ran up against its pseudo-BASIC-esque limitations fairly quickly. Keep in mind that there is no better way to learn some concept, than writing a program for it. Period. Once you are able to express your knowledge in code (er, in code that actually works, I mean), you have it down cold, or down kold, even.
Could they, though ? At this point, it will just be an arms race between ??AA and Torrent users, and I'm not sure who'd win. In other words, I don't know whether it is physically (or, rather, mathematically) possible to build a tamper-proof peer-to-peer network. It's an interesting question.
Well, torrent-poisoning would not even exist if it did not lead to the lawsuits. They can't bring down the entire network with just poisoning, and they know it.
Actually, in the total absence of regulation, this would not be a problem. The MPAA could upload whatever bogus file they wanted; their bogus uploads would be filtered out and everyone would move on. The problem is that, under the current regulatory regime, the MPAA can use their fake torrents to sue people into oblivion.
Yes, it would be nice if all software "just worked", but, until we develop Strong AI (such as HAL... hmm...), this is not possible. Since a user can't just tell his computer, "do my taxes for me, pronto !", the user will have to use his own intelligence to instruct the computer in what to do. This means that they will have to learn how to talk to the computer in its own language. The best software strikes a balance between the steepness of the learning curve, and the power it exposes to the user.
Apple tends to take the more user-friendly road: expose as little power as possible, but make the UI as simple as possible. This is a valid choice. UNIX takes the opposite approach: it's all power, all the time, if you can remember 12 different config files and env variables in your head. Given UNIX's audience, this is also a valid approach.
However, letting your users design your software from the ground up is a terrible idea, because the users are not aware of the limitations of modern technology, nor are they aware of the complexity of their own field (most of the time). In 90% of the cases, what the user truly wants is a button that says "do my work for me"... and we're right back where we started, at the beginning of this post.
Couldn't you level the same allegations at any human, though ? I.e., we can look at the entire employment history of CutCo/EdgeCom going back 10 years, see that they hired more single people than married people, then sue them for discrimination. After all, even though they don't explicitly have a policy that considers marital status, clearly their HR managers are using this as part of their judgement... right ?
But a human would never tell the computer to explicitly consider marriage. Instead, the computer would be trained (or, rather, train itself) to draw conclusions from all kinds of disparate data, which could amount to inferring whether the applicant is married or not. What happens then ?
I think the grandparent poster's main idea was not to defer more of our decisions to scientists, but for everyone to become more scientifically-minded. We live in a world where each individual wields an almost unimaginable amount of power at his fingertips -- just crash your car into a wall at 90 mph to see what I mean -- and it could really help if everyone understood a little about how this power truly works.
How does Yahoo compare to Experts-Exchange ? I used to like that site a lot, and I still use it, but their semi-recent redesign (filthy inline ads, sidebar ads, popup ads, light blue on slightly lighter blue text, ads between comments) has sent me searching for something better. But, I have to admit that the quality of the answers I get on Experts-Exchange is still superb.
IMO Azureus is the best. uTorrent has a way smaller memory footprint, and a much faster startup time, but these things simply don't matter much to me. Azureus's memory use is still reasonable (compared to say, Visual Studio, *shudder*), and its startup time is irrelevant to me because I never shut it down. It doesn't use any measurable amount of CPU, either, but it comes with a plethora of features and plugins, notably DHT, that blow uTorrent out of the water.
Be warned, though: I hear that Azureus is going to re-write their client to act primarily as an on-demand movie viewer, with some weird convoluted UI that makes no sense. So, enjoy it while it lasts...
Um, did these guys just invent a thermocouple ? Those things have been around for quite some time, and are used to generate power on satellites (using a radioactive heat source, because in space, who cares ?), and to make small, inefficient refrigerators that can fit in a car... And of course, to measure temperature pretty much everywhere.