Slashdot Mirror
IBM Working on Brain-Rivaling Computer
Obdurate writes "The first supercomputers to approach and even surpass the processing power
of the human brain are to be built by IBM, under a $184M contract
announced by the US Government yesterday.
ASCI Purple and Blue Gene/L will be the fastest and most powerful machines built,
with a combined capacity equal to the 500 best of todays computers."
I once had an exercise in a business math class where half had calculators and the other had nothing. Calculator users *had* to use the calculator. The teacher then asked simple arithmetic questions - 2x2, 3 minus 1, etc. Of course, the people without calculators could answer first.
The fastest computer in the world will always be limited to how quickly data may be fed to it. One way or another, a human will have to direct this operation - if only for safety and security considerations.
I think with the interesting people, their lives can't possibly be wrapped up into a nice little package.
The processing power of a honeybee's brain in terms of the power needed for it to perform flight as it does, and find honey, and return to the hive, etc., has been estimated at 60 teraflops. The idea that 6 times as much processing power = the human brain seems reasonably foolish. I think ultimately, the problem is that people tend to think of brains as giant calculating machines, when they're not -- there's a great deal of hardwired logic controlling things like breathing and reflexes, that aren't so much mediated by calculation, as they are by simple input output "black-box" sort of processes. This is another reason attempting to equate a brain to a giant computer seems foolish.
Kargis Strong, MD
I think I read somewhere that brain fires bursts of neurotransmitters in the range of 40 Hz. That's right, ladies and gentlemen, you're conciousness is running on a processor that's slower than the chip in your GBA or your Palm Pilot.
I think that what most people don't get is that the brain is not that powerful a computer... It's just very, very good at what it's supposed to do.
Think of it this way. Instead of a computer and mobo combination, consider the brain as dozens and dozens of embedded micro-controllers that talk to eachother via a protocol. Each one is very specific. We have one that handles getting audio signals, one that handles getting video signals... and then completely different controllers for recognizing voice, music, speech, text, and images. There is one overlying controller-- the frontal lobe-- but most of what is does is pattern matching and random number generation. It's the combination or all these working together, not the raw ability of the brain to process information, that makes the magic of 'conciousness'.
The next Slashdot story will be ready soon, but subscribers can beat the rush and slashdot the links early!
It's not anything remotely like a human brain. They're making some rough analogy between storage size, processing speed, and the number and nature of neurons in the human skull. This is just a really really really fast/big version of existing machines.
Again, for those who haven't read Douglas Hofstadter's excellent books GEB and MMT - being human-like is a *really* tough thing for a computer, and we haven't even begun to figure out the basics of it on paper. Maybe in 100 years we'll understand the problem better, but I'll place my bets now that when we do we'll finally realize it's futile to try to mimic it.
11*43+456^2
Are we talking about the brain as we use it, or the brain, at it's full potential?
I don't think IBM know... Seems like some silly marketing ploy like Intel's "The Pentium III makes Internet faster".
I mean... Does even anyone know how quick the brain is at it's "full potential"? Do we even have a unit in which we measure brain "quickness"? I don't think brains go well with FLOP's. As someone in another thread said: "with easy numbers I can do 1 - 2 FLOP's". Still, we can do stuff we haven't even come close to with today's technology.
I wonder if there's a science that research the possibilities to adapt human behavior and thinking to computers? That's usually the major flaw with today's robots, etc. We have pretty much unimaginable power in the super computers of today, but the computer "minds" we've produced so far are still at a laughable stone age level. Why? Do we *still* need more power to make a computer be able to follow a natural conversation (without pre-made replies)? Or do we simply not have the theory to approach the problem and we're essentially just standing there saying "duh?" at the problem of having a computer to truly *know* grammatics and form sentences on its own?
Sure, we have neural networks, and that might be a nice *foundation* for simulating human minds, but how to do it in practice? How to write the actual code? Again, are there even a science for this?
Beware: In C++, your friends can see your privates!
While the human brain is usually not very good at such linear calculations, hence the popularity of a calculator, its true power lies in it's massively parallel processing.
/. expression, the brain functions very similar to a beowolf cluster. We can design computers (very expensive ones, though) that can simulate many of the simpler activities that humans are capable of (such as complex pattern recognition, primitive conversation skills, and rule-based systems of cause and effect,) but to do all of these at once is still well on the horizion.
To tie in an ever popular
-Space for rent
The Raw processing power of the brain is very high, but its actual effectiveness and speed is crap. The reason is the IO speeds, the network interface (spine) has poor throughput and requires lots of individual channels rather than being able to operate as a simple bus, this means loads of wasted space when a channel isn't doing anything.
/dev/random would be pretty useless, and yet the software in humans means that this is a greatest advantage.
The external interfaces are even worse, these make the brain totally useless for many tasks that computers can process in seconds. As an example try raytracing a rendering a scene using crayons and doing the maths in your head.
So the human brain totally and utterly is secondary to the computer already.
Apart from the fact that humans can be inspired. The solution may take a computer 100 years to attack by brute force and it will get there... but a smart person will do it in minutes because "its obvious".
Computers already outstrip us in terms of processing, but while they are just grown up calculators they miss the essence of human processing. A computer hardwired to mutate everything via
It will be generations before computers will have reached a stage they can start doing the obvious. The limited processing of the brain has produced the people on the Jerry Springer show and Isaac Newton, it ain't the hardware, its the software that counts.
An Eye for an Eye will make the whole world blind - Gandhi
But, to be fair, humans are at a disadvantage in terms of how numbers are represented in our brains and how we take input. For example, have a handwritten piece of paper scanned, OCRed, and then perform math. We handle the 'hard part' of scanning, recognizing, and transferring the information via a convenient avenue for the computer/calculator to handle. Even general purpose computers are quite restricted and highly optimized to handle a small subset of things, while human brains are extremely general purpose, but not optimized for intense mathematical operations.
XML is like violence. If it doesn't solve the problem, use more.
Lots of interesting things about this:
First, the real issue is not hardware or CPU cycles -- it is software. Tired of Seti@home? Let's build a distributed processing network that has as many CPU cycle equivalents as the human brain! Oh yeah, that's already been done. Ok, so why doesn't it "think" yet? Oh yeah...software.
The issue is how to integrate storage, processing, "RAM", etc. into a software package that can emulate a human brain's method of thinking (which may be a very bad, krufty method of developing consciousness -- why would anyone use meat for processors? What a kludgy hack!).
(OT: what if "thinking" software is _not_ GPL'ed? That could be really frightening. So could security issues for "thinking" machines.)
Second, the next issue is why should we compare digital thinking machines to biological ones? Maybe it is the only benchmark we can think of, but given the truly awkward way in which light-sensitive cells were adapted for inclusion a biological thinking machine (see Francis Crick's "Astonishing Hypothesis"), why can't a much more efficient independent decision making machine be developed from digital equipment (not DEC, btw) actually designed for the purpose?
The human brain/computer comparison is really a red herring. The only reason to create a human-like digital thinking machine/emulator (and you thought WINE was hard to use...) might be to pursue immortality. I think the more likely reason is that it would be the ultimate species-wide circle jerk. Humanity getting off on creating humanity. Bleh. Let's set our sights a little higher.
guac-foo
Lots of petrified grits
Don't forget that the human brain is capable of designing devices to do certain things it's bad at (math) at incredible speeds. Which makes determining the intelligence/brain power of humans a recursive problem.
Be wary of any facts that confirm your opinion.
Sure, this thing can do 100 teraflops, but does that mean that it has any intelligence? That it can learn? Those are the true qualities of the human brain, and without those ASCI Purple is just an incredibly large and expensive calculator.
Again, are there even a science for this?
Yes. It's called "Artificial Intelligence". People involved are often, but not always, also "Computer Scientists". They usually write programs in strange languages like "Common Lisp", "CLIPS" and "Prolog", with the occasional Neural network simulation like "SURF-HIPPO" (Don't ask...)
The thing is, computing SPEED is not what matters. In fact, the most "intelligent" seeming systems are often large networks of small computers with good interconnects. Making them faster makes the system faster, but not smarter. The computer is only the substrate. Highly parallel symbolic manipulation seems to matter much more. Some of the most convincingly "generally-intelligent" programs can barely count!. All the computational substrate is there for is to support the symbolic-computation level above.
Unfortunately, thanks to the "AI Winter" (where AI researchers overpromised and underdelivered in the 80s), there are relatively few people active in the field. It also doesn't help that each time an AI researcher makes a computer do something that "will never be done by a computer", the general public redefine AI to mean "not that"!
You might quite surprised at how far ahead of the "mainstream" AI research really is.
Read "Godel, Escher, Bach: An Eternal Golden Braid" some time, bearing in mind that it was written 20 years ago! It's really regarded as quite passé in AI circles, but ideas in it STILL haven't made it into the computing mainstream...
Look at it this way. Go outside, on a windy day (adding more variables to the mix) and have someone throw you a football/basketball/baseball/frisbee/whatever. It probably takes 3-4 seconds at most for the ball to reach you, and looooong before that, your brain completed a monstrous calculus problem. It figured in the position of the thrower, the wind velocity and direction, direction/speed of the ball, the ball's arc of travel, and in the next split second, sent signals to your legs and feet to move your body to the ball's expected landing spot.
But wait, it's the ball's landing spot minus about five feet, because your brain figures you want to be positioned to catch the ball when it's about 4-5 feet off the ground. It simultaneously sends signals to your hands and arms, positioning them to catch the ball, taking into account the ball's speed, size and mass.
A lot of calculations in an extremely short period of time! And, if you think that's impressive for a human brain, the brain in that dumb mutt of yours in the back yard can do the same thing when you toss him a tennis ball.
ASCI Purple will be built using 12,544 IBM Power5 microprocessors, the same chips that are used in Apple PCs and Nintendo games systems.
Umm, how about...NOT. Just because they're all PowerPC based doesn't make them the same. Based on that logic a 386 and a Pentium 4 are the same too, just beacuse they're both built on the x86 architecture.
Power 5 (can't find a link) is a generation of chips that are related, but further on the horizon than the chips Apple is buying (both are Power 4 spin-offs, but quite different). The chips used in the Nintendo GameCube are not even related -- they just happen to also be made by IBM -- not to mention they are several years old while the above chips are not even available yet.
Then again having a server class chip in a Nintendo might be interesting...
Actually, individual neurons in the brain fire up to 200 times a second (200 Hz).
However, a (biolgical) neural network isn't hardware or software. It's both. It's a network build up out of simple elements, which together form logic circuits.
In a computer the hardware (the logic components) form a circuit which can be used to run software on. A neural network IS the software.
For this reason it's wrong to compare a computer with a neural network.
A neural network doesn't even have strictly defined areas which process certain kinds of data. Due to its adaptive nature, an unused part of the network will be taken up by bordering parts and used for different purposes. A person who loses a hand will have the part of his brain which was previously used to control this hand 'absorbed' by sections of the brain with totally different functions.
So in short, a computer is hardware, a neural network is software in a physical form.
Site & blog: http://www.mayaposch.com
Digital computers and the human brain work on completely different computational principles. The people who run these meaningless calculations on the "processing power of the brain" take each synapse to be a bit. That's absolute bunk when you're talking about the nonlinear properties of even small networks of neurons, much less the massively complex architecture of the brain. Until we actually develop an understanding of how neural networks (real neural networks, not the stuff that drives touchpads) operate, we can't even begin to make realistic comparisons.
btw, I'm a ee who does neuroscience research, so I'm not talking out of my ass here.
Some men spend their entire lives trying to kill themselves for having been born. --Ross MacDonald
Even if they can compute as many instructions per second as the human brain is capable of, it won't matter unless they have a good enough understanding of AI to come up with software that can mimic human types of intelligence -- things like intuition, insight, creativity.
Otherwise it'll just be a very expensive, very fast data processor.
You see? You see? Your stupid minds! Stupid! Stupid!
>But here is the kicker: Will those 100 teraflops be flops that can use thousands of inputs?
...And, to go with that, will they find someone smart enough to actually implement some adaptive technique that emulates a human brain?
--
God is the only form of extraterrestrial life that we could ever possibly communicate with -- SETI is a joke, people
"I'll believe all this bullshit when I see someone actually create an organic, conscious brain from scratch. Until then, its all moot."
But, how would you know that the computer actually *is* conscious, and not just pretending to be conscious? Sure, you know that *you* are conscious, but how would you know for sure that anybody else in the world is? It's an interesting puzzle, and one that will probably never leave the domain of philosophy and religion...
>
> Considering we can blow up the surface of the world a couple of times(at least) over with our existing stockpiles, why are we spending ANY money on ANYTHING except REDUCING said stockpiles?
It's you :-)
Seriously - reducing the need for large nuclear stockpiles exactly why the money's being spent on simulations.
Nukes are complicated devices, composed of weird stuff (the fissionables and other what-not), and normal stuff (the explosives that trigger the weird stuff).
Over time, the weird stuff changes its properties. So does the normal stuff.
One of many issues with nukes is that if you're gonna throw one at someone, you want to be damn sure it goes off. Otherwise, you've probably just given your enemy enough weird stuff that they could build their own bomb. This, I think we can agree, is a Bad Thing.
If you're going after a guy in a hardened bunker, and your nuke blows up but doesn't blow as strongly you thought it would, you may have to lob another one at the same target. And that means you need to have more nukes in reserve.
And worse yet, if you're going after the same bunker, but your nuke works a little too well, you've just wiped out a city instead of just the few hundred feet around your target. This is inefficient at best, and barbarism at worst. (The early fusion bombs had this "problem", and some tests resulted in radiation exposures far greater than was expected, mainly because the bomb was "better" than it was supposed to be.)
If you want to cut down on the number of nukes in the arsenal, a good way is to make sure that you've got a few very good ones that always go off when they're supposed to, with the correct amount of "boom".
One way to make damn sure your nukes blow up when and how big they're supposed to is to test them regularly. I'll grant that mushroom clouds over the Nevada desert were probably very pretty to watch, but they were also pretty messy for those living downwind. Bad idea.
The second way is underground testing, which solves most of the "downwind" problem, but can still result in some leakage under some circumstances.
That really only leaves one other option - to run simulations. Lots of simulations. Using the best math your scientists can come up with, and the fastest computers your geeks can build. No radiation leaks, and what you learn while building the supercomputers can be used for building higher-performance computers for peaceful purposes in the future.
I dunno about you, but I'll take Door Number Three any day.
Computers were already able to do things, that brains couldn't do, 20 years ago - regarding speed and precise memory (that's what I call processing power).
... ...; someone must write programs for computer to make them work, so there is little chance that computers will ever be more 'intelligent' than the one who learned them how to become intelligent.
Brains will probably always be able to do things, that computers can't do, not even with EXTREMELY much processing power.
Why?
====
A Brain works in a different way. It's good at fuzzy-logic and at distinguishing important from less important information.
To let a computer's pure logical processing power act like a brain, you have to simulate all that "fuzzy-logic" with complicated mathematics.
Computers can do a lot of things, which not even thousands of brains could do correctly, or at least in an acceptable period of time. Weather/climate simulations, sound-processing,
There are also a lot of things, which can only be done by thinking, by being creative,
We do not even know exactly, what 'intelligent' means, from a technical point of view.
Brains are more powerful than computers, and Computers are more powerful than brains.
That's what you get if you compare apples with pears.
But is the brain calculating this or rather looking up the answer? I know as a toddler I couldn't catch squat, but as I got older I got better. Was the reason increased proceesing power, my brain got bigger. Or more experience, I'd caught a lot more balls by then.
I doubt very much the brain is clunking through calculus.
Sure it is. What do you think "more experience" means? It means that the neurons in your brain have reconnected in ways to tackle a task better each time. It doesn't necessarily mean your brain did it one way or another. Let's look at the two ways that a wetware computer could catch the ball:
A) Mathematics. [Input: (Here is the ball now. And here is where it is now. And this is roughtly how fast the wind is blowing and what direction it is coming from...) -> Process (Compare position of the ball at time A to that of time B, then to time C, the path is making an arc... Extrapolate that arc. Where will the ball be at time D? -> Output (Move those hands and catch!)]. That doesn't necessarily mean you used more neurons (your "bigger brain") to do it. It's like taking a chunk of mixed silicon and metal and turning it one step at a time into a 3GHz custom CPU. Reorganization made for faster processing.
B) Look up tables. Keep a log of past experiences, the solution to each experience and reference it each time a task is done. Certain things your brain probably only uses a lookup table for -- digit - by - digit multiplication for example. The brain recognizes a Platonistic "football-ish" object and throws it into the works. It thinks, what did I do the last time I had a football pitched it right at my noggin?
But you can't tell me that the circumstances are the same every time someone throws you the ball. If your brain was simply trying to catch by following previous experiences, it would fail to find a previous experience when the wind suddenly shifts and blows hard. Or you trip over a rock, stumble and still make the catch. Or the ball travels at a different speed. Do you just stand there, or improvise? If your brain isn't doing any actual number crunching to catch that ball, did you only catch it the last time by chance? And just think of how much storage space would be needed to hold every experience! Quite the cluttered mess. It makes much more sense in this situation to reply more upon the math than it does look up tables.
So the last poster was right. A brain does do math to catch that ball. And you're right, a brain does reference previous experiences when trying to catch that ball.
Since this math is done by specialized brain functions that were prepared to do just that, and are inseperably integrated with other brain connections -- it doesn't mean that you could take that calculus ability and use it for another task. But the math is being done.
Remember "Bring 'em on"? *sigh