Singularity

[blakehaydn]

I just listend to seti.org's "singularity" and wanted to know what you folks think about it.

"Every three years, the amount of computer processing you can buy for a thousand dollars doubles. By 2025, your desktop machine will have more computational power, and more memory, than you do. Is Homo sapiens about to replace himself? Guest Ray Kurzweil is suggesting that machines will soon outstrip human intelligence, leading to a discontinuity in our existence called The Singularity. Join us as we discuss this singular topic... and bring a machine with you."

http://seti.sitestream.com/AWA_05-11-27.mp3

 

[Roboramma]

So is drexlerian nanotechnology possible?

Maybe not in exactly the way he suggested, maybe even not quite as powerful as he suggested, but is it possible to build tiny machines that can manipulate matter on a molecular level? More efficiently than living things do?

I can't see why not, but all I know about it is what I read in his Drexler's book (it was a long time ago... Engines of Creation I think).

ETA, this is weird, I entered "Is Drexlerian Nanotechnology possible?" into yahoo, and it seems china doesn't want me to see any websites that have anything to do with nanotechnology. Odd, I wonder why.

 

[Iamme]

Ha. Up until my post...you only had a singular(ity) reply.

 

[Ziggurat]

I just listend to seti.org's "singularity" and wanted to know what you folks think about it.

"Every three years, the amount of computer processing you can buy for a thousand dollars doubles. By 2025, your desktop machine will have more computational power, and more memory, than you do. Is Homo sapiens about to replace himself? Guest Ray Kurzweil is suggesting that machines will soon outstrip human intelligence, leading to a discontinuity in our existence called The Singularity. Join us as we discuss this singular topic... and bring a machine with you."

http://seti.sitestream.com/AWA_05-11-27.mp3

The obsession with "computing power" is really misplaced. For example, you can make a chip that takes two bunches of numbers as input, pairs them up and multiplies each pair, then outputs those numbers as a new bunch. Do that fast enough, and you'll have more "computational power" than any computer that exists right now. But you won't have created anything more intelligent than your pocket calculator (in fact, less so). It takes a lot more than computational power to make intelligence, and on THAT front, we are not even approaching animals, let alone humans. To do that, we're going to need software much more complicated and sophisticated than anything we've got, and we don't know if we're even smart enough to ever do that. Hell, we still have problems making an operating system that doesn't crash.

 

[kuroyume0161]

The obsession with "computing power" is really misplaced. For example, you can make a chip that takes two bunches of numbers as input, pairs them up and multiplies each pair, then outputs those numbers as a new bunch. Do that fast enough, and you'll have more "computational power" than any computer that exists right now. But you won't have created anything more intelligent than your pocket calculator (in fact, less so). It takes a lot more than computational power to make intelligence, and on THAT front, we are not even approaching animals, let alone humans. To do that, we're going to need software much more complicated and sophisticated than anything we've got, and we don't know if we're even smart enough to ever do that. Hell, we still have problems making an operating system that doesn't crash.

I think that you spat on the nose of the donkey!

Whether we like it or not, computers are digital binary logic engines and animal brains are nothing of the sort (most likely analog, stimulus-based cascade receiver/transmitters). No matter how much 'computation power' you throw at a computer, it's still a dumb calculator - just faster. The difference is the underlying architecture (as noted in the first sentence). Brains don't (or don't appear) to work on binary logic, but rather on threshold stimuli response.

That noted, it will be interesting to see what kind of intelligence emerges from a complex digital binary system. You might not want to be around when it does!

 

[NeilC]

Surely software takes computers away from merely being binary calculators. It's possible that one might be able to design software models closely matching the function of "analog, stimulus-based cascade receiver/transmitters" or whatever it takes.

Clearly merely increasing h/w power isn't enough but is bound to be a factor.

 

[kuroyume0161]

As a developer, at the core, all computers operate on binary values (1's and 0's). There are macro operations, but nothing on the order of the coherence exhibited by organic nervous systems of any sophistication.

I will submit that given enough 'digital power', the semblence might emerge. But, from recent findings, we are problably discussing immediate-access storage on the order of quintobytes (TeraByte*MegaByte) and parallel processing on massive scales (not this dual-core junk). This would appear to be the minimum for a digital system to start emulating a complex organic analog system.

There are analog computers out there. They might be the only real worthwhile path to artificial intelligence/sentience - only time and experience wil tell.

 

[PixyMisa]

Whether we like it or not, computers are digital binary logic engines and animal brains are nothing of the sort (most likely analog, stimulus-based cascade receiver/transmitters).

Doesn't matter. A digital system can model an analog system to an arbitrary degree of accuracy.

There are reasons to think that the Singularity might not eventuate, but that's not one of them.

 

[PixyMisa]

I will submit that given enough 'digital power', the semblence might emerge. But, from recent findings, we are problably discussing immediate-access storage on the order of quintobytes (TeraByte*MegaByte)

If by that you mean millions of terabytes, then the term is exabyte.

and parallel processing on massive scales (not this dual-core junk).

Dual-core is for entry level servers and mid-range desktops today, not for serious work. Even the Xbox 360 has three cores.

There are analog computers out there. They might be the only real worthwhile path to artificial intelligence/sentience - only time and experience wil tell.

There's certainly no reason to think so.

 

[Soapy Sam]

Doesn't matter. A digital system can model an analog system to an arbitrary degree of accuracy.- PixyMisa.

Can an analogue system model a digital system to the same degree?

 

[hammegk]

And might we recall that "real time" actually is a factor?

 

[kuroyume0161]

If by that you mean millions of terabytes, then the term is exabyte.

Thanks for that. Wasn't concerned to go look up the prefix as I was tired and ready for bed and could not think of it offhand.

Dual-core is for entry level servers and mid-range desktops today, not for serious work. Even the Xbox 360 has three cores.

Yes, but I'm talking about super-super computer parallelism - the ability to process information with millions (in not hundreds of millions) of concurrent processes (neurons). No computer today (or in the near future) will be capable of this in real-time. Not SETI time 'share'. If you've seen 'neural cascades', you do realize that this is millions of neural firings. It used to be thought that a neuron could be modeled in a simplistic weighted stimulus system, but recent discoveries point to a much, much, much more complex system 'per neuron' (and there are billions of neurons in a human brain).

 

[kuroyume0161]

Doesn't matter. A digital system can model an analog system to an arbitrary degree of accuracy.- PixyMisa.

Can an analogue system model a digital system to the same degree?

Actually, all electronics are essentially analog. What we call digital binary is just the modification of voltages which are analog current and potential within components and traces. So, in the most fundamental sense, all digital systems are modeled from an analog one. It is just the interpretation of the signals that we deem digital.

I think that there are some limitations to the degree of accuracy that a digital system can be pushed to simulate an analog one (analog to digital sampling error - due to insufficient sampling rates and causing false harmonics (or data correlation) - comes to mind as a well-known issue).

 

[PixyMisa]

Yep. So-called "digital" systems are really analog systems doing their best to act digital. I know Hammy doesn't like me pointing this out, but computers aren't deterministic. Although designers do their best to minimise the frequence of hardware errors, they do occur, and are in many cases entirely random.

Now, let's go back to the question of computing power. Say we need, for the sake of argument, 1MFLOP (one million floating point calculations per second) to simulate one neuron. The Cell processor (which will show up next year in the Playstation 3) can perform, hang on, 3.2x10⁹x9x4x2=230.4x10⁹ floating point operations per second. 230,000 MFLOPS, so it can simulate 230,000 neurons. So, 4.3 million Playstation 3s networked together could simulate - simulate, mind you - a human brain in real time.

And by the end of 2006 we will have that many Playstation 3s networked together... Though they will mostly be used for playing Final Fantasy XII.

Wait five years and we will have processors four to eight times as powerful (assuming no advances that haven't already been proven in testing), and you would be able to build the thing for a couple of hundred million dollars - much less than many "bug science" projects like particle accelerators or space telescopes.

And that's to simulate the operation of the human brain, so it's a worst-case scenario. That computer will be many orders of magnitude more powerful than the human brain for any normal task.

 

[PixyMisa]

Hell, we still have problems making an operating system that doesn't crash.

We've had operating systems that don't crash for forty years. Just because you choose not to run them doesn't mean they don't exist.

 

[DanishDynamite]

Yep. So-called "digital" systems are really analog systems doing their best to act digital. I know Hammy doesn't like me pointing this out, but computers aren't deterministic. Although designers do their best to minimise the frequence of hardware errors, they do occur, and are in many cases entirely random.

Now, let's go back to the question of computing power. Say we need, for the sake of argument, 1MFLOP (one million floating point calculations per second) to simulate one neuron. The Cell processor (which will show up next year in the Playstation 3) can perform, hang on, 3.2x10⁹x9x4x2=230.4x10⁹ floating point operations per second. 230,000 MFLOPS, so it can simulate 230,000 neurons. So, 4.3 million Playstation 3s networked together could simulate - simulate, mind you - a human brain in real time.

And by the end of 2006 we will have that many Playstation 3s networked together... Though they will mostly be used for playing Final Fantasy XII.

Wait five years and we will have processors four to eight times as powerful (assuming no advances that haven't already been proven in testing), and you would be able to build the thing for a couple of hundred million dollars - much less than many "bug science" projects like particle accelerators or space telescopes.

And that's to simulate the operation of the human brain, so it's a worst-case scenario. That computer will be many orders of magnitude more powerful than the human brain for any normal task.

Very interesting stuff, Pixy.

But your extrapolation hinges on, at least, the idea of simulating a neuron by a 1 Mflop operation. I suppose it would be more than possible to simulate the neurons in an ant's brains with computers today, given an ant's size. Has such been done?

 

[hammegk]

I know Hammy doesn't like me pointing this out, but computers aren't deterministic.

I don't? Have I ever indicated that anything is error free?

Now, let's go back to the question of computing power. Say we need, for the sake of argument, 1MFLOP (one million floating point calculations per second) to simulate one neuron. The Cell processor (which will show up next year in the Playstation 3) can perform, hang on, 3.2x10⁹x9x4x2=230.4x10⁹ floating point operations per second. 230,000 MFLOPS, so it can simulate 230,000 neurons. So, 4.3 million Playstation 3s networked together could simulate - simulate, mind you - a human brain in real time.

If you could just show that a program could be written that utilized all that compute power, in real time, for something useful. And wow, how much heat will that sucker generate? And how big did you say it was? Last time I looked at something like that I came up with a cube 7 miles on each side.

And that's to simulate the operation of the human brain, so it's a worst-case scenario. That computer will be many orders of magnitude more powerful than the human brain for any normal task.

Unfortunately neither you nor anyone else has a clue how to program to "simulate a human brain", forget about "normal tasks" performed by a human brain.

 

[kuroyume0161]

But your extrapolation hinges on, at least, the idea of simulating a neuron by a 1 Mflop operation. I suppose it would be more than possible to simulate the neurons in an ant's brains with computers today, given an ant's size. Has such been done?

The current neural simulation capabilities are metered at about that of a cockroach. Not too far removed from an ant, I suppose.

But it's not just about speed of operation, it is also about data (storing neural states, connections, and simulating all of the variables). That's why I mentioned recent discoveries that put the neuron's complexity at levels closer to millions of bits (not the previous oversimplifications of several or hundreds of bits). Supposing just a million bits: (5*10^9) * (10^6) = 5*10^15 bits of data for a human brain (about 625 terabytes of storage - this could be permanent, but more than likely will need to be dynamic (i.e.: RAM)). Do you know of any system that can accomodate this much RAM? With the advent of 64-bit processors, we can assuredly address such vast memory spaces, but we'll need memory chips the size of current large hdds (200-400GB) to avoid filling a small office building with chips!

And I strongly agree with hammegk (yikes!) that simulation of neural responses over networks or internet would be hopelessly inadequate.

 

[PixyMisa]

I don't? Have I ever indicated that anything is error free?

That was your opinion last time the subject came up.

If you could just show that a program could be written that utilized all that compute power, in real time, for something useful.

How many examples do you need?

And those are all latency-tolerant examples, what is known as "embarassingly parallelisable" problems.

And wow, how much heat will that sucker generate?

A hundred megawatts or so. Which is rather a lot, I admit.

And how big did you say it was? Last time I looked at something like that I came up with a cube 7 miles on each side.

...

When did you do the calculation, 1922?

I'd estimate 50,000 square metres of floor space. That's 12 acres, assuming a single-story building.

Plus a few acres for the cooling pond for the air conditioning. Either that or build it in Anchorage and pipe the heat straight into the city.

Unfortunately neither you nor anyone else has a clue how to program to "simulate a human brain", forget about "normal tasks" performed by a human brain.

Actually, we do have a very good idea how to simulate a human brain. We're less clear about how the brain does what it does, but that doesn't matter, because we are talking about simulating the brain at the level of cell function, the level we do know.

 

[PixyMisa]

The current neural simulation capabilities are metered at about that of a cockroach. Not too far removed from an ant, I suppose.

Certainly the same order of magnitude, so no big deal.

But it's not just about speed of operation, it is also about data (storing neural states, connections, and simulating all of the variables). That's why I mentioned recent discoveries that put the neuron's complexity at levels closer to millions of bits (not the previous oversimplifications of several or hundreds of bits).

I'd be interested in a link, if you have one.

Supposing just a million bits: (5*10^9) * (10^6) = 5*10^15 bits of data for a human brain (about 625 terabytes of storage - this could be permanent, but more than likely will need to be dynamic (i.e.: RAM)). Do you know of any system that can accomodate this much RAM?

Um, yes. Any current server processor can handle that. No biggie.

With the advent of 64-bit processors, we can assuredly address such vast memory spaces, but we'll need memory chips the size of current large hdds (200-400GB) to avoid filling a small office building with chips!

The largest memory modules currently shipping are 4GB. We'd need 150,000 of them. Though for the rest of the calculations I was basing my numbers on the 45nm process, so divide that by 4.

And I strongly agree with hammegk (yikes!) that simulation of neural responses over networks or internet would be hopelessly inadequate.

Internet, yeah. I wasn't seriously proposing hooking up 4 million PS3s to do the job.

I don't see why you consider networks in general to be problematic. Inter-neuron communication is not exactly fast.