You missed something. The unity-gain bandwidth is based upon a capacitative effect which acts as a low-pass filter, which limits the gain as the frequency through the device increases. When you hit the unity-gain bandwidth, your device will no longer amplify an input signal. Since most digital-logic requires amplification during every logic-element, this is what happens here. The problem is really not that you need the signal to travel between a lot of transistors (there are ways around this if you're clever) but just the simple fact that you usually need a gain of at least 4 or 5 to be able to use the circuit as a logic element.
The point is that the faster a transistor is rated to go, the smaller these capacitances, and the higher the unity-gain frequency.
I just wanted to point out here that they are talking about Bipolar transistors, the kind used in amplifier design. Specifically, these types of chips are designed to target the upper end of the Microwave band, since that band is fairly difficult to transmit in (due to the high clock speeds required). By the sounds of it, they are targeting this at the same market being fed by Gallium-Arsenide circuits now, which are already in this speed bracket, but suffer a few very severe drawbacks (high cost, low production rates, low gain, etc).
These have nothing to do with the MOSFET transistors used in conventional VLSI CMOS digital circuits, like processors. Although they could make something like the old TTL or the fast ECL with these transistors, I seriously doubt you will see much in this way any time. Both TTL and ECL are FAR faster than CMOS made in the same feature-size, their high power-dissipation made heat-sinking a severe problem for VLSI. Although Bipolar ECL did find some limited use in the LN2-cooled supercomputer market (they have essentially no dynamic power-dissipation, but they have a fairly high static power-dissipation, so they get *QUITE* hot).
On the other hand, this is quite likely to end up in all sorts of fun communcation and signal-processing applications.
Unless I'm missing something, energy and mass can be converted back and forth (in some instances). They're related by, like I said, E=M*c^2, so factoring for Mass, you get M=E/c^2. High-energy photons do decompose into particles, after all... Given enough energy, you certainly can create a black hole.
Well, you get more oil from soybeans then from hemp, and anybody who knows anything about soybeans will tell you that it is almost all usable for something. If you need the oil, you're still left with de-oiled vegetable protein (which is used for things like veggie-burgers and stuff, you can get the raw stuff under the name of "texturized vegetable protien" and use it sorta like hamburger) and the plant stems (which should be brewable into the methanol you need to make biodeisel).
Mind you, there's a hemp store in the mall here sellling shirts and stuff made of hemp fibre, hemp-based foodstuffs, etc. Apparently the Canadian government doesn't care about hemp growing.
I've also seen a multi-fuel reformer or two that can be attached to a gasoline engine to make it run off anything. Looked really neat.
Some company did this a while back, it was on slashdot actually. Unlike most of these, it was not vapor, there was an actual review of the thing, pictures of it, a solid explanation of the technology, etc. It had about a 10 degree field of view or so ("sweet spot"), but if you stayed in that area, it was supposedly a beautiful screen.
Course, it was like a flat LCD with a lens fused to the front or somesuch, but it did work. And the technology worked.
This makes me very sad. When I was 12, I pulled one of the worst hacks you could do to a network without erasing data (I took over admin, locked all admins out, etc). Back then, I just got a slap on the wrist, probably mainly because I was the only person qualified to fix it. Would I do it again? Hell no, I was a stupid, immature 12-year-old.
WHY did I do it? Because I loved networks and wanted to learn how they worked. When I got control, I didn't do anything particularly malicious (no mark-changing, etc), but I poked around at all the administrative functions. This was far enough back that I hadn't yet had a home-network to play with.
The point is that nobody got hurt, I changed everything back, and I learned about networking. I'm glad I don't live in (the USA in general actually) a draconian system like that. Inquisitive kids should just get a good strong slap on the wrist so they never do it again, but making an example of them in this fashion is just bad.
Capitalism is a superrealistic model of a big-buisiness environment in a rather fun game. It is probably more directly related than any of the above. They're making a new version now, Capitalism MBA, for use in an instructional environment, but it will still probably be a fun game.
The real problem is with propogation delays between the internal transistors, rather than heat. Now, as you cool the chips, the propogation delays go down, so cooling does allow you to run substantially faster.
However, on a chip where heat is not the main issue (as the Crusoe easily may be), it is possible to overclock to a point where, heatwise, the chip is fine, but in which it will never boot because signals are simply not moving fast enough for instructions to process. Cooling can help, but overheating is not the problem in such a case.
(Try this: take a 486DX33 and overclock it to 50. With cooling it will easily run, but it will not heat dangerously with just an ordinary heatsink fan. However, it will not run.)
Seriously, I use KDE on my Dual OC 550 Celeron A, but on my P75 laptop, nothing quite beats a well-configured copy of TWM. Has anybody out there actually USED this little wm? It's fast as sin.
Well, size isn't as bad as it sounds. The blue LED's were the major factor now, but the Virtual Boy had a red LED display in it (which I heard was terrible, but that's for other reasons).
LED's are just a P/N junction, today we can fit millions of these on a chip. The junction in an LED isn't really that much bigger, so theoretically these things could be built like integrated circuits, including the current-limiting resistors and the latch-and-multiplex circuits sitting at the base of the LED itself. We're only talking 20000 micron technology IC's here, which is hardly difficult by any stretch.
The current problem is the patent on the new high-intensity blue LED if I recall correctly.
Well, if an OLED is anywhere near as fast as a silicon LED (which it should be, given the same basic premise), it should actually have a much HIGHER refresh rate than LCD's.
Liquid crystal displays actually have to realign crystal cells, which is quite a lengthy process (which causes the ghosting). LED's are just a P/N junction. Turn power on -> light starts as soon as the junction triggers. Turn power off -> light dies.
OLED's should have comparable performance to LED's, which can have refresh-rates in the megahertz.
Somebody moderate this up (if I'm not the only one this deep in the thread)! I just about died when I read it, probably the funniest thing thusfar on this story!
That's right, one time I hacked AAVGA myself to be able to support the 320x240 VGA mode used by Quake2. Looked pretty cool. I've still got a few screenshots around...
First off, yes, I _HAVE_ worked with analog neural networks. They tend to produce some rather interesting results at times. But that's not what I'm referring to in particular.
Realtime and hardware are both VERY relevant. These lead to quite a bit of numerical error in the dataset, so the same network may not perform the same way even given the exact same input conditions.
Yes, I fully agree with one thing. "CONVENTIONAL" neural networks ARE statistical models, mostly because they're being done WRONG and for the WRONG reasons. Conventional neural networks have no place in a serious discussion about AI.
Lastly, we see why most people dismiss AI. The failure to understand the effect of true chaos on intelligence. No purely computational solution can ever be truly intelligent. Even when applied to a conventional neural network, the addition of a HARDWARE random number generator (read: one that generates REAL random numbers, not numerical approximations) can significantly improve response times. With something a little less conventional, the results are even more dramatic.
I don't think I'll ever succeed in convincing you that AI is possible, because you seem dead-set against it. I will point out, however, that when I said "totally unexpected conclusion", I meant to say "totally unexpected but more CORRECT conclusion". The fact of the matter is that the partially-hardware systems demonstrate the same emotional and learning capacities of the lesser biological lifeforms in behavioral tests. Fine, you can argue that this still does not make it intelligent, then you can also go and point out that for the same reasons, a ringworm is also totally unintelligent (is it?).
Anyways, I'm done with this thread. The only way I could convince you is by breaking NDA, and that's the last thing in the world I want to do. Now I'll get off the soapbox.
No, here in Canada, we don't really believe in the government serving the people. Something that the US doesn't realize is that some countries have totally ineffectual governments that do nothing but endlessly bicker.
Such is Canada. We don't worry about our government, because the only thing it's good at is infighting and buying $20000 doorknobs. They waste lots of money, yes, but what government doesn't? If that's our biggest complaint, we have little to complain about.
Now for a list of why I say this: We have free crypto (afaik), you can use radio scanners to pick up cell-phone conversations legally (whether this is good or bad is your decision... I don't do it, but some do...), very very little is censored.
There is no 'internet decency act' equivalent up here. In fact, there was a recent ruling stating that the ban against downloading child porn was unconstitiutional. (Mind you, creating such stuff is still VERY illegal.)
We all grumble about the government here, but trust is a non-issue. How many people hear about CECIS (the canadian equivalent of the CIA, possibly spelled wrong)? Not many. Why? Because it is TINY, and has almost NO power inside of the country!
Oh, and the status-quo is protected with a passion here. Any attempt at censorship results in a terrible onslaught of outraged protests, and quickly dies a silent death.
Basically, we've proven that a non-government is better than a totalitarian democracy;)
Yes, genetic algorithms and fuzzy logic ARE certainly overused buzzwords. I personally avoid them like the plague (the words at least).
However, the other two, adaptive knowledge-based systems and neural networks, are not.
First, I should define an adaptive knowledge-based system. Basically, a KBS is a big database search program. An adaptive KBS is a KBS with the ability to self-modify, either through the use of built-in matching rules or by the use of other systems (like NN).
Onwards and upwards, about neural networks, you seem to be using the 1980's definition, straight out of a textbook.
When I said "dabbles in AI professionally", I mean that I've actually done a fair amount of stuff with it professionally, but I'm not the researcher, I just code it. I've seen NN systems that fly in the face of all current models.
In fact, it's been years since I've seen a backpropogation or Bayesian NN used inside these walls. We've been using realtime systems closer to CORE, which are not simple back-to-front-and-learn-backwards networks. They are latticed in all directions, a lot of them are at least partially hardware.
You seem to have the misconception that an NN-based system outputs a number. This is only the case for what exists in the public domain. I assure you, a lot more than this exists. Like a robot that demonstrates a fear of light for instance.
NN systems are not merely a statistical system, but an approximation of the brain. In some of our recent tests (I cannot go much into specifics, but I can say that they involve pulses over time) we've seen some interesting things happen, and quite often, the NN jumps to a totally unexpected conclusion. It's only a matter of time before the approximation becomes accurate enough.
And while I agree that AI has been having a rocky road, this is due to the overwhelming feeling that it must be 'programmed' in. In fact, it must NOT be. We've seen time and again hardware systems succeed where a software model of the same model fails. Why? Chaos. Computers are terrible at calculating chaos. Hardware, on the other hand, has lots of little transmission delays.
Anyways, I've already said more than I'm probably cleared to say (aint NDA's a pain?).... But I hope that clears up some of the common misconceptions.
Okay, maybe I'm paranoid, but does anybody else see something REALLY bad on the horizon?
First off, we have countries like Australia and the US slowly eradicating privacy and freedom. This is in itself a bad thing.
Here's the kicker though. When the Y2K bug hits (which will, in and of itself, be a mostly non-event now...), the ensuing paranoia, rioting, etc will cause considerable chaos. Don't forget that Clinton still has the ability to sign exectutive orders! Does anybody else see the US becoming a dictatorship soon?
Regardless of Y2K, the other side of the coin is large, multinational corporations and conglomerates like the RIAA and MPAA. With the legislation being passed, soon these companies will have absolutely NO liability to consumers, and, in fact, to the government itself!
Then again, I oculd just be paranoid and sleep-deprived from working quadruple-overtime. Heh, whatever. Glad I live in Canada.
What about if the robot was a neural system? For me, this isn't just idle debate; I've worked extensively with neural networks, even in the context of robotic intelligence. The one thing I can tell you right now is that you DO get robots with personality disorders. In one of my preliminary tests, I had a computer vision system go into "complete negative lockdown", which is the equivalent to suicidal depression.
Granted, at the 50-neuron level, it certainly was not self-aware. However, most of my decent tests of neural systems all point to the existance of true artificial intelligence in them. Yes, it is about as intellegent as a ringworm, and possesses all the intelectual and emotional development of one.
But the point is, from the second I switch on a neural network, I hardly have the foggiest notion what it'll be like. The same matrix that went negative-lockdown stayed "sane" for about a full minute when I tried the same test with the room lights darker.
While other experimental techniques like genetic algorithms, fuzzy logic and adaptive knowledge-based systems do still have the capability to mess up in ways the designer did not intend (your favorite search engine is proof of that), neural networks (which most experts believe is the most promising currently) have the capability for a MUCH larger range of reactions.
Somehow I doubt many of us would like to have "Marvin the Paranoid Android" driving our car. But with neural systems, you cannot be sure about the differences between one neural map and another. The equations for them become exponentially complex with netsize, so much so that for a 50-neuron system, the equations are often hundreds or even thousands of pages long.
Oh, and it doesn't necessarily take an AI system to design something patentable. A simple brute-force "try everything until it works" system works quite well for any problem domain with a well-defined method of simulation (mechanical design, architecture, etc). While the calculations would take months to perform, I have seen a few of these work wonders.
Just the opinion of somebody who dabbles in AI professionally.
Not QUITE right about VNC... While it is the case that GDI calls are not intercepted by VNC in the windows version, the equivalent is done on the X versions. The thing has its own Xserver that you run in the background, and connect to using vncviewer. In fact, I'm using it right now to type this comment. VNC is amazing.
Although, just the same, VNC in windows doesn't sound as powerful overall. Ah well.
Yes, this is a violation. An EULA must be agreed to before you can use software. Most of them specifically disallow rental. In this case, you have to agree to the GPL before you use the product, even though you are renting it.
Take them to task on this one. With a blatant disregard for the GPL like this, the bare mention that you might send lawyers their way is liable to stir up some attention. Of course, since they have no legal leg to stand on, I seriously doubt that they will call the bluff.
Lucid dreaming does not necessarily imply remembering the dream... But it does tend to increase retention rate anyways. No idea on pages about this, I trained myself. As a kid, I used to have terrible nightmares every night. So I learned how to recognize a dream, then how to wake up from them.
It was only a matter of time before I learned how to do cool stuff like flying and teleporting and creating objects and stuff (some of these are actually quite difficult!). The hardest one of course is changing universes...
Now I'm trying to figure out how to extend dreams... Oh well...
Sorry, but the way that this was done in the "Thirteenth Floor" is HORRIBLY inaccurate. You do NOT "transfer" a human conciousness into a machine for VR. You simply supply new sensory data. This is closer to the way it was done in "The Matrix", although the Canadian film "Existenz" did an even better job, if you are looking for reasonably accurate portrayals.
It always bugged me that movies depicted an inability to simply "unplug" (hence "Existenz" is more accurate than "The Matrix".) Sudden disconnection would be a very dizzying, disorienting experience but NOT a fatal one.
Actually, you are VERY right here. This was my inspiration for going into the VR field. I never dream un-lucidly (okay, that's probably not the correct word...). So I know what I'm looking for.
The trick that most people miss is that the brain already has a full VR facility built in. Most proposed direct neural interfaces override real muscle controls and send responses back to real sensor inputs. This has advantages and disadvantages. The main advantage is that it is usually easier.
However, feeding signals back into the sub-conscious brain would produce much more vivid simulations. This capability is still outside of reach, but it seems to be getting closer every day. The most major benefit is that the dream interface would be able to read "commands" as well as motions.
In truth, you don't even need to make it descent slowly. Tilt the person inside gimbals so that the gravity perception is different, then keep changing the lock-heights (make them "soft"). This is actually a rather realistic simulation.
Everybody seems to be assuming that speed and timing of the controllers are particularly challenging. First of all, the controllers are already WAY faster than they have to be. Imagine a dedicated G4 or PIII or whatever chip to control nothing but your movements?
Second, there are already plenty of fast-acting powerful force-feedback actuators. 10-gauge nitinol wire (muscle wire) has an actuation time well under a tenth of a second for a standard controller, and a high-current pulse-modulated controller can actuate it even faster (the pulse-width must be kept small to avoid burning the wire). And wire of that gauge has a VERY significant pull. The only problem is the heat generated, but insulation is definately not an unsurmountable challenge.
While I'm not doing much in VR hardware, I do work with the software for VR systems. Most of this is still early development or prototype, but little of it remains in the pipe-dream stage.
A totally different idea. Concentric positional gyroscopes with the person held in the center. They never actually have their feet on the actual ground, but they wear a tactile/force feedback suit while suspended. The gyros can rotate to simulate the effects of moving-mass and gravity. The force-feedback can give the impression of actually walking. The positional readback sensors can determine the actual change in position vectors.
I work with the software for VR systems. So far, this idea seems to have the most promise, although I have not yet personally had a chance to work with it. But short of full neural interface, this is as close to real as we will get.
Slashdot Mirror
The point is that the faster a transistor is rated to go, the smaller these capacitances, and the higher the unity-gain frequency.
These have nothing to do with the MOSFET transistors used in conventional VLSI CMOS digital circuits, like processors. Although they could make something like the old TTL or the fast ECL with these transistors, I seriously doubt you will see much in this way any time. Both TTL and ECL are FAR faster than CMOS made in the same feature-size, their high power-dissipation made heat-sinking a severe problem for VLSI. Although Bipolar ECL did find some limited use in the LN2-cooled supercomputer market (they have essentially no dynamic power-dissipation, but they have a fairly high static power-dissipation, so they get *QUITE* hot).
On the other hand, this is quite likely to end up in all sorts of fun communcation and signal-processing applications.
Unless I'm missing something, energy and mass can be converted back and forth (in some instances). They're related by, like I said, E=M*c^2, so factoring for Mass, you get M=E/c^2. High-energy photons do decompose into particles, after all... Given enough energy, you certainly can create a black hole.
Mind you, there's a hemp store in the mall here sellling shirts and stuff made of hemp fibre, hemp-based foodstuffs, etc. Apparently the Canadian government doesn't care about hemp growing.
I've also seen a multi-fuel reformer or two that can be attached to a gasoline engine to make it run off anything. Looked really neat.
Course, it was like a flat LCD with a lens fused to the front or somesuch, but it did work. And the technology worked.
WHY did I do it? Because I loved networks and wanted to learn how they worked. When I got control, I didn't do anything particularly malicious (no mark-changing, etc), but I poked around at all the administrative functions. This was far enough back that I hadn't yet had a home-network to play with.
The point is that nobody got hurt, I changed everything back, and I learned about networking. I'm glad I don't live in (the USA in general actually) a draconian system like that. Inquisitive kids should just get a good strong slap on the wrist so they never do it again, but making an example of them in this fashion is just bad.
Capitalism is a superrealistic model of a big-buisiness environment in a rather fun game. It is probably more directly related than any of the above. They're making a new version now, Capitalism MBA, for use in an instructional environment, but it will still probably be a fun game.
However, on a chip where heat is not the main issue (as the Crusoe easily may be), it is possible to overclock to a point where, heatwise, the chip is fine, but in which it will never boot because signals are simply not moving fast enough for instructions to process. Cooling can help, but overheating is not the problem in such a case.
(Try this: take a 486DX33 and overclock it to 50. With cooling it will easily run, but it will not heat dangerously with just an ordinary heatsink fan. However, it will not run.)
Seriously, I use KDE on my Dual OC 550 Celeron A, but on my P75 laptop, nothing quite beats a well-configured copy of TWM. Has anybody out there actually USED this little wm? It's fast as sin.
LED's are just a P/N junction, today we can fit millions of these on a chip. The junction in an LED isn't really that much bigger, so theoretically these things could be built like integrated circuits, including the current-limiting resistors and the latch-and-multiplex circuits sitting at the base of the LED itself. We're only talking 20000 micron technology IC's here, which is hardly difficult by any stretch.
The current problem is the patent on the new high-intensity blue LED if I recall correctly.
Liquid crystal displays actually have to realign crystal cells, which is quite a lengthy process (which causes the ghosting). LED's are just a P/N junction. Turn power on -> light starts as soon as the junction triggers. Turn power off -> light dies.
OLED's should have comparable performance to LED's, which can have refresh-rates in the megahertz.
Somebody moderate this up (if I'm not the only one this deep in the thread)! I just about died when I read it, probably the funniest thing thusfar on this story!
Quake2 in text mode... Fun stuff ;)
First off, yes, I _HAVE_ worked with analog neural networks. They tend to produce some rather interesting results at times. But that's not what I'm referring to in particular.
Realtime and hardware are both VERY relevant. These lead to quite a bit of numerical error in the dataset, so the same network may not perform the same way even given the exact same input conditions.
Yes, I fully agree with one thing. "CONVENTIONAL" neural networks ARE statistical models, mostly because they're being done WRONG and for the WRONG reasons. Conventional neural networks have no place in a serious discussion about AI.
Lastly, we see why most people dismiss AI. The failure to understand the effect of true chaos on intelligence. No purely computational solution can ever be truly intelligent. Even when applied to a conventional neural network, the addition of a HARDWARE random number generator (read: one that generates REAL random numbers, not numerical approximations) can significantly improve response times. With something a little less conventional, the results are even more dramatic.
I don't think I'll ever succeed in convincing you that AI is possible, because you seem dead-set against it. I will point out, however, that when I said "totally unexpected conclusion", I meant to say "totally unexpected but more CORRECT conclusion". The fact of the matter is that the partially-hardware systems demonstrate the same emotional and learning capacities of the lesser biological lifeforms in behavioral tests. Fine, you can argue that this still does not make it intelligent, then you can also go and point out that for the same reasons, a ringworm is also totally unintelligent (is it?).
Anyways, I'm done with this thread. The only way I could convince you is by breaking NDA, and that's the last thing in the world I want to do. Now I'll get off the soapbox.
Such is Canada. We don't worry about our government, because the only thing it's good at is infighting and buying $20000 doorknobs. They waste lots of money, yes, but what government doesn't? If that's our biggest complaint, we have little to complain about.
Now for a list of why I say this: We have free crypto (afaik), you can use radio scanners to pick up cell-phone conversations legally (whether this is good or bad is your decision... I don't do it, but some do...), very very little is censored.
There is no 'internet decency act' equivalent up here. In fact, there was a recent ruling stating that the ban against downloading child porn was unconstitiutional. (Mind you, creating such stuff is still VERY illegal.)
We all grumble about the government here, but trust is a non-issue. How many people hear about CECIS (the canadian equivalent of the CIA, possibly spelled wrong)? Not many. Why? Because it is TINY, and has almost NO power inside of the country!
Oh, and the status-quo is protected with a passion here. Any attempt at censorship results in a terrible onslaught of outraged protests, and quickly dies a silent death.
Basically, we've proven that a non-government is better than a totalitarian democracy ;)
However, the other two, adaptive knowledge-based systems and neural networks, are not.
First, I should define an adaptive knowledge-based system. Basically, a KBS is a big database search program. An adaptive KBS is a KBS with the ability to self-modify, either through the use of built-in matching rules or by the use of other systems (like NN).
Onwards and upwards, about neural networks, you seem to be using the 1980's definition, straight out of a textbook.
When I said "dabbles in AI professionally", I mean that I've actually done a fair amount of stuff with it professionally, but I'm not the researcher, I just code it. I've seen NN systems that fly in the face of all current models.
In fact, it's been years since I've seen a backpropogation or Bayesian NN used inside these walls. We've been using realtime systems closer to CORE, which are not simple back-to-front-and-learn-backwards networks. They are latticed in all directions, a lot of them are at least partially hardware.
You seem to have the misconception that an NN-based system outputs a number. This is only the case for what exists in the public domain. I assure you, a lot more than this exists. Like a robot that demonstrates a fear of light for instance.
NN systems are not merely a statistical system, but an approximation of the brain. In some of our recent tests (I cannot go much into specifics, but I can say that they involve pulses over time) we've seen some interesting things happen, and quite often, the NN jumps to a totally unexpected conclusion. It's only a matter of time before the approximation becomes accurate enough.
And while I agree that AI has been having a rocky road, this is due to the overwhelming feeling that it must be 'programmed' in. In fact, it must NOT be. We've seen time and again hardware systems succeed where a software model of the same model fails. Why? Chaos. Computers are terrible at calculating chaos. Hardware, on the other hand, has lots of little transmission delays.
Anyways, I've already said more than I'm probably cleared to say (aint NDA's a pain?).... But I hope that clears up some of the common misconceptions.
First off, we have countries like Australia and the US slowly eradicating privacy and freedom. This is in itself a bad thing.
Here's the kicker though. When the Y2K bug hits (which will, in and of itself, be a mostly non-event now...), the ensuing paranoia, rioting, etc will cause considerable chaos. Don't forget that Clinton still has the ability to sign exectutive orders! Does anybody else see the US becoming a dictatorship soon?
Regardless of Y2K, the other side of the coin is large, multinational corporations and conglomerates like the RIAA and MPAA. With the legislation being passed, soon these companies will have absolutely NO liability to consumers, and, in fact, to the government itself!
Then again, I oculd just be paranoid and sleep-deprived from working quadruple-overtime. Heh, whatever. Glad I live in Canada.
What about if the robot was a neural system? For me, this isn't just idle debate; I've worked extensively with neural networks, even in the context of robotic intelligence. The one thing I can tell you right now is that you DO get robots with personality disorders. In one of my preliminary tests, I had a computer vision system go into "complete negative lockdown", which is the equivalent to suicidal depression.
Granted, at the 50-neuron level, it certainly was not self-aware. However, most of my decent tests of neural systems all point to the existance of true artificial intelligence in them. Yes, it is about as intellegent as a ringworm, and possesses all the intelectual and emotional development of one.
But the point is, from the second I switch on a neural network, I hardly have the foggiest notion what it'll be like. The same matrix that went negative-lockdown stayed "sane" for about a full minute when I tried the same test with the room lights darker.
While other experimental techniques like genetic algorithms, fuzzy logic and adaptive knowledge-based systems do still have the capability to mess up in ways the designer did not intend (your favorite search engine is proof of that), neural networks (which most experts believe is the most promising currently) have the capability for a MUCH larger range of reactions.
Somehow I doubt many of us would like to have "Marvin the Paranoid Android" driving our car. But with neural systems, you cannot be sure about the differences between one neural map and another. The equations for them become exponentially complex with netsize, so much so that for a 50-neuron system, the equations are often hundreds or even thousands of pages long.
Oh, and it doesn't necessarily take an AI system to design something patentable. A simple brute-force "try everything until it works" system works quite well for any problem domain with a well-defined method of simulation (mechanical design, architecture, etc). While the calculations would take months to perform, I have seen a few of these work wonders.
Just the opinion of somebody who dabbles in AI professionally.
Although, just the same, VNC in windows doesn't sound as powerful overall. Ah well.
Take them to task on this one. With a blatant disregard for the GPL like this, the bare mention that you might send lawyers their way is liable to stir up some attention. Of course, since they have no legal leg to stand on, I seriously doubt that they will call the bluff.
It was only a matter of time before I learned how to do cool stuff like flying and teleporting and creating objects and stuff (some of these are actually quite difficult!). The hardest one of course is changing universes...
Now I'm trying to figure out how to extend dreams... Oh well...
It always bugged me that movies depicted an inability to simply "unplug" (hence "Existenz" is more accurate than "The Matrix".) Sudden disconnection would be a very dizzying, disorienting experience but NOT a fatal one.
Ah well.
The trick that most people miss is that the brain already has a full VR facility built in. Most proposed direct neural interfaces override real muscle controls and send responses back to real sensor inputs. This has advantages and disadvantages. The main advantage is that it is usually easier.
However, feeding signals back into the sub-conscious brain would produce much more vivid simulations. This capability is still outside of reach, but it seems to be getting closer every day. The most major benefit is that the dream interface would be able to read "commands" as well as motions.
Everybody seems to be assuming that speed and timing of the controllers are particularly challenging. First of all, the controllers are already WAY faster than they have to be. Imagine a dedicated G4 or PIII or whatever chip to control nothing but your movements?
Second, there are already plenty of fast-acting powerful force-feedback actuators. 10-gauge nitinol wire (muscle wire) has an actuation time well under a tenth of a second for a standard controller, and a high-current pulse-modulated controller can actuate it even faster (the pulse-width must be kept small to avoid burning the wire). And wire of that gauge has a VERY significant pull. The only problem is the heat generated, but insulation is definately not an unsurmountable challenge.
While I'm not doing much in VR hardware, I do work with the software for VR systems. Most of this is still early development or prototype, but little of it remains in the pipe-dream stage.
I work with the software for VR systems. So far, this idea seems to have the most promise, although I have not yet personally had a chance to work with it. But short of full neural interface, this is as close to real as we will get.