Transistors: Pound for Pound
Many of the assorted nuts at the recent Singularity Summit apparently believe that the entire universe (or at least much of our planet) will be turned into a giant computer.
After all, if you extrapolate Moore's Law then very soon computers will be so powerful, that they will start to think for themselves and then start expanding over the face of the Earth like bacteria did billions of years ago. There is at least one problem with this: transistors have been getting more expensive over time.
I have already shown that small is expensive, but here I want to point out that the long term trend for transistors (at least) is to increase in cost when normalized with their size. In 1965, transistors were roughly 1 mm square and cost $1 each. Today, for 1 dollar you can buy 50 million transistors which makes it seem like they are much cheaper. However, they are much MUCH smaller today with features roughly 100 nm in size. The shrinkage in size greatly outweighs the cost reduction which means that pound for pound (or liter per liter) transistors are 4 to 5 orders of magnitude more expensive today then they were 40 years ago.
The fundamental reason for this is that the universe (through the 2nd Law of Thermodynamics) wants things to be spread out and hates it when structures are packed in tiny spaces. So, this trend will inevitably increase for Si/SiO2 integrated circuits and for any other potential replacement for them (i.e., molecular, quantum, DNA computing, etc.)
Even if we manage to invent an A.I., which is not inevitable, it will still be constrained by the laws of physics and economics which are driving the pound-for-pound increase in the cost of computing substrates.
This is yet another reason why the Singularity is physically impossible.
After all, if you extrapolate Moore's Law then very soon computers will be so powerful, that they will start to think for themselves and then start expanding over the face of the Earth like bacteria did billions of years ago. There is at least one problem with this: transistors have been getting more expensive over time.
I have already shown that small is expensive, but here I want to point out that the long term trend for transistors (at least) is to increase in cost when normalized with their size. In 1965, transistors were roughly 1 mm square and cost $1 each. Today, for 1 dollar you can buy 50 million transistors which makes it seem like they are much cheaper. However, they are much MUCH smaller today with features roughly 100 nm in size. The shrinkage in size greatly outweighs the cost reduction which means that pound for pound (or liter per liter) transistors are 4 to 5 orders of magnitude more expensive today then they were 40 years ago.
The fundamental reason for this is that the universe (through the 2nd Law of Thermodynamics) wants things to be spread out and hates it when structures are packed in tiny spaces. So, this trend will inevitably increase for Si/SiO2 integrated circuits and for any other potential replacement for them (i.e., molecular, quantum, DNA computing, etc.)
Even if we manage to invent an A.I., which is not inevitable, it will still be constrained by the laws of physics and economics which are driving the pound-for-pound increase in the cost of computing substrates.
This is yet another reason why the Singularity is physically impossible.
HP LaLancette said: "Today, for 1 dollar you can buy 50 million transistors which makes it seem like they are much cheaper. However, they are much MUCH smaller today with features roughly 100 nm in size. The shrinkage in size greatly outweighs the cost reduction which means that pound for pound (or liter per liter) transistors are 4 to 5 orders of magnitude more expensive today then they were 40 years ago. "
That is the dumbest thing I've ever heard. So, like, do you think that the processing power of a computer is proportional to the MASS of transistors in it rather than the number of transistors? ROFL
Did you hear about the AMD Athlon 2800 gram+ processor? Athlon have a special ultra-dense transistor design (the secret ingredient is depleted uranium), so that an AMD processor which weighs only 2000 grams is actually as good as an intel processor which weighs 2800 grams?
customer: "hey, I want to buy THE HEAVIEST COMPUTER IN THE SHOP"
shop assistant: "Ok, sir, we've got this very nice intel leaden-core system with a hard drive made of granite, that weighs, um half a ton, and it's only $599"
customer: "what, do you think I'm an idiot? that computer has far too many transistors in it. I want THAT ONE!"
shop assistant: "no, sir, that's a bronze statue, not a computer. It's only there for decoration"
customer: "and how much does it weigh?"
shop assistant: "about 10 tons"
customer: "how many transistors does it have?"
shop assistant: "none"
customer: "GREAT! I'll take it. Do you do home delivery?"
Reply to this
I don't know who arouses me more ... LaLancette or Roko...
Reply to this
I do not understand your purpose other than to show that today you can get more transistors for the same cost per pound. Could you show the actual math?
Reply to this
I think you may be missing the whole point of nanotechnology, my good man. There is no universal constant that designates matter and energy just to spread out, that happens as a by-product of entropy. You are right, things do in fact spread out (to quote a cartoon in American Scientist, "I've got the condensed matter blues,"), but you seem to be suggesting that we are living in a closed system. We are all nanotechnology mate - atoms, molecules; you know those mind-bogglingly small interconnected functional components we are made of. When you talk of expense what you mean is that something is unlikely - literally so. Anything unlikely requires (arbitrarily defined) effort to make it happen. Life on earth, for example, is extremely unlikely, the probability of a human coming together is infinitesimal, but here we are, not because a bunch of rocks collided and somehow we popped out (the twister through the scrap heap argument for ID), but because the right chemical conditions arose for simple, stable homeostats to become (relatively) feasible. Moreover, simple chemical homeostats are very different from humans, another infinitesimally improbable leap, and yet, again, here we are. Evolution works through indirection (not to sound too much like Kurzweil, but he was right on this), evolution does not get from 1 to 1014 by going 1+1+1+1+1+1.... etc., evolution gets from 1 to 1014 by going (1+1)+2)+4)+8)+16)+32)+64).... etc. It would have been astronomically expensive to ask Australopithecus to build you a laptop, but here we are, typing at each other on our laptops. When you say small is expensive, you mean that it is expensive for big things to do small things and small things to do big things - small things do small things very well, despite entropy, because we are not a closed system. In fact, that is what biology is in the first place, an open system that just keeps going by exporting entropy and importing free energy (and recycling, a great invention by any standard). The very point of nanotechnology is that, in this way, it mimics biology - it should be self-replicating and homeostatic, or, in a word, self-organising. Nobody wants to construct little nanocircuits with their fingers, we are not going to do it with a chisel or an axe, let alone by colliding a bunch of asteroids together - we are going to use complicate microfabrication techniques to build the next-generation microfabrication tools and so on. There are, of course, limits to this, like how much power we have available versus a specific problem with specific, realisable results (cost-effectiveness), but I think you are drastically underestimating how far we can and will go. I believe it would be a mistake to say that we will not at least best biology.
And as to these assorted nuts? Does that include Rodney Brooks and Douglas Hofstadter (who was at last year's), or Cory Doctorow? Which ones, exactly? Better not become a dogmatic ideologue while attempting to fight against them.
Reply to this
Post Script:
Just to clarify, so I do not sound like I am propounding Kurzweilian, "mystical-style" evolution (I agree, he never seems to explain what he thinks drives all this acceleration or what we can look for to test his hypothesis), when I say evolution gets from a simple chemical homeostat to us by going (1+1)+2)+4)+8)+16).... etc., what I do mean is that once you have got a human, you do not need to repeat the process of evolving a human to get a second one - evolution saves its progress informationally at each step, as do we (culturally). So in summary, while I do not think we are headed for a Kurzweilian Singularity (probably), I think we are probably headed somewhere, even if it it is not anywhere specific that we know of (basically, the future probabily will not look the same as the past).
Reply to this
I did similar calculations and got figures ranging from $1M to $150M/kg depending on the exact circuit and it's depth.
You are absolutely correct and absolutely wrong.
Yes, the stuff in your computer is horrendously complex and therefore horrendously expensive, but that is about to change. Hint: Potatoes are horrendously complex, but they're only $1/kg.
I suggest you familiarize yourself with the DNA Origami work done by Paul Rothemund, and the bis-amino polypeptide work done by Christian Shafmeister (not to mention the top-down approach being pursued by Zyvex).
Yes, Transhumanism has it's major fundamental flaws (which I've pointed out elsewhere; e.g. "Transhumanism: The New Master Race?" http://www.islandone.org/MMSG/99jan.htm#_Toc456110960, but it's dependence on nanotech is not one of them.
And there's one nice thing about Transhumanists- they're not a bunch of thumb-sucking, forrest gumpy, girly men afraid of change. They are willing to face the unknown future and shape it. Unlike the rest of American and European society, they're willing to challenge the universe, instead of just rolling over to die in some sustainable way. See Howard Bloom's "Screw Sustainability." http://lifeboat.com/ex/screw.sustainability
Reply to this
R u not at all familiar with Bucky Fuller's concept of *ephemeralization*?---that is, of doing ever-more with ever-less in terms of grams of mass, ergs of energy, and units of time *per* any particular technological function? Nanotech is (or at least will be) simply a continuation (albeit a rather dramatic and far-reaching one) of this long-standing techno-trend...
And, as for what is driving all this, check-out Jim Gardner's and John Smart's stuff. As the lamentably-late Robert Anton Wilson remarked: "**Something's** going on!!"
Reply to this