Well, first of all, this guy is echoing ideas first voiced by people like kurzweil. You may want to take a look to the original if you want to have a clearer idea of what he is talking about.
And now please keep in mind this are the conservative estimations. They think that, according to Moore's Law we must be able to have enough computer power to equal to the MAXIMUN ESTIMATED computer power of the human brain. But we all talking of a very conservative stimation here, and we may be for a surprise in the sort future. Let's take a look at how this estimations of the human brain computer power are performed:
- Average number of Neurons in the human brain (excluding the cerebellum):
20.000.0000.0000
- Average number of connections per neuron:
1.000
Each neuron can perform about 200 calculations per second, per connexion.
So, we have 20.000.0000.0000 X 1.000 X 200 = 4.000 TeraOps
Now, 4.000 TeraOps is about 100 times faster than the Earth-Simulator, the faster computer system in existence, and according to Moore's law, is going to take a while before we have a Data Center-wide cluster that powerful, not to mention a desktop system light enough so we could propel it around with two mechanical legs.
This is the logic after those "no AI before 2020# arguments we hear now so often. But us I said, this is the conservative estimation, and the conservative estimation is not the most likely scenario at all. Well, let me tell you something, and I know what I'm talking about, we will have a few nice surprises in the next few month. Let me give you a hint, there is a obvious flaw in that logic:
- Number of transistors in transistors in the AMD "Hammer" processor:
100.0000.0000
Each transistor can perform at 2.000.000.000 calculations per second.
So, we have 100.0000.0000 X 2.000.0000.0000 = 200.000 TeraOps
Acording to that logic, we may need a 200.000 TeraOps computer to emulate a AMD "Hammer" processor, what is oviuly untrue: 2Ghz Hammer can perform at only 4 TeraOps, and we just need, say, 2 1.8 Ghz Atlons to get to that speed.
The "peak" performance needed to contemplate all the possible states of the system is enormous, yes, but that is not realeted to the true capacity of the system. Not every single transistor in the system flops every cicle, that's not a realistic assumption, just a few of them do. Consecuently, the amount of information and operations you need to perform in order to emulate is orders of magnitude below the conservative estimation of the peak number of states you need to emulate. Now extrapolate to the H Brain. Is it more efficient than the hammer? No doubt. How much efficient is it? 10 Times? 100 times? 1000 times ? 10.000 times?
Even if the human brain happens to be 100.000 times more efficient than your tipical Pentium/Atlon, you'll need only a 2.000 nodes computer cluster to outperform it. And that is something we have at hand right now. The rest is just software.
Slashdot Mirror
That's a mistake, of course. A 2Ghz Hammer performs at about 4 GIGAops
Well, first of all, this guy is echoing ideas first voiced by people like kurzweil. You may want to take a look to the original if you want to have a clearer idea of what he is talking about. And now please keep in mind this are the conservative estimations. They think that, according to Moore's Law we must be able to have enough computer power to equal to the MAXIMUN ESTIMATED computer power of the human brain. But we all talking of a very conservative stimation here, and we may be for a surprise in the sort future. Let's take a look at how this estimations of the human brain computer power are performed:
- Average number of Neurons in the human brain (excluding the cerebellum): 20.000.0000.0000
- Average number of connections per neuron: 1.000
Each neuron can perform about 200 calculations per second, per connexion.
So, we have 20.000.0000.0000 X 1.000 X 200 = 4.000 TeraOps
Now, 4.000 TeraOps is about 100 times faster than the Earth-Simulator, the faster computer system in existence, and according to Moore's law, is going to take a while before we have a Data Center-wide cluster that powerful, not to mention a desktop system light enough so we could propel it around with two mechanical legs.
This is the logic after those "no AI before 2020# arguments we hear now so often. But us I said, this is the conservative estimation, and the conservative estimation is not the most likely scenario at all. Well, let me tell you something, and I know what I'm talking about, we will have a few nice surprises in the next few month. Let me give you a hint, there is a obvious flaw in that logic:
- Number of transistors in transistors in the AMD "Hammer" processor: 100.0000.0000
Each transistor can perform at 2.000.000.000 calculations per second.
So, we have 100.0000.0000 X 2.000.0000.0000 = 200.000 TeraOps
Acording to that logic, we may need a 200.000 TeraOps computer to emulate a AMD "Hammer" processor, what is oviuly untrue: 2Ghz Hammer can perform at only 4 TeraOps, and we just need, say, 2 1.8 Ghz Atlons to get to that speed.
The "peak" performance needed to contemplate all the possible states of the system is enormous, yes, but that is not realeted to the true capacity of the system. Not every single transistor in the system flops every cicle, that's not a realistic assumption, just a few of them do. Consecuently, the amount of information and operations you need to perform in order to emulate is orders of magnitude below the conservative estimation of the peak number of states you need to emulate. Now extrapolate to the H Brain. Is it more efficient than the hammer? No doubt. How much efficient is it? 10 Times? 100 times? 1000 times ? 10.000 times?
Even if the human brain happens to be 100.000 times more efficient than your tipical Pentium/Atlon, you'll need only a 2.000 nodes computer cluster to outperform it. And that is something we have at hand right now. The rest is just software.