Microsoft Word Kurzweil, Ray The Singularity Is Near doc

A Virtually Unlimited Limit.
As I discussed in chapter 3 an optimally organized 2.2-pound computer using 
reversible logic gates has about 10
25
atoms and can store about 10
27
bits. Just considering electromagnetic interactions 
between the particles, there are at least 10
15
state changes per bit per second that can be harnessed for computation, 
resulting in about 10
42
calculations per second in the ultimate "cold" 2.2-pound computer. This is about 10
16
times 
more powerful than all biological brains today. If we allow our ultimate computer to get hot, we can increase this 
further by as much as 10
8
-fold. And we obviously won't restrict our computational resources to one kilogram of matter 
but will ultimately deploy a significant fraction of the matter and energy on the Earth and in the solar system and then 
spread out from there. 
Specific paradigms do reach limits. We expect that Moore's Law (concerning the shrinking of the size of 
transistors on a flat integrated circuit) will hit a limit over the next two decades. The date for the demise of Moore's 
Law keeps getting pushed back. The first estimates predicted 2002, but now Intel says it won't take place until 2022. 
But as I discussed in chapter 2, every time a specific computing paradigm was seen to approach its limit, research 
interest and pressure increased to create the next paradigm. This has already happened four times in the century-long 
history of exponential growth in computation (from electromagnetic calculators to relay-based computers to vacuum 
tubes to discrete transistors to integrated circuits). We have already achieved many important milestones toward the 
next (sixth) paradigm of computing: three-dimensional self-organizing circuits at the molecular level. So the 
impending end of a given paradigm does not represent a true limit. 
There are limits to the power of information technology, but these limits are vast. I estimated the capacity of the 
matter and energy in our solar system to support computation to be at least 10
70
cps (see chapter 6). Given that there 
are at least 10
20
stars in the universe, we get about 10
90
cps for it, which matches Seth Lloyd's independent analysis. So 
yes, there are limits, but they're not very limiting. 
The Criticism from Software 

---

A common challenge to the feasibility of strong AI, and therefore the Singularity, begins by distinguishing between 
quantitative and qualitative trends. This argument acknowledges, in essence, that certain brute-force capabilities such 
as memory capacity, processor speed, and communications bandwidths are expanding exponentially but maintains that 
the software (that is, the methods and algorithms) are not. 
This is the hardware-versus-software challenge, and it is a significant one. Virtual-reality pioneer Jaron Lanier, for 
example, characterizes my position and that of other so-called cybernetic totalists as, we'll just figure out the software 
in some unspecified way—a position he refers to as a software "deus ex machina,"
2
This ignores, however, the specific 
and detailed scenario that I've described by which the software of intelligence will be achieved. The reverse 
engineering of the human brain, an undertaking that is much further along than Lanier and many other observers 
realize, will expand our AI toolkit to include the self-organizing methods underlying human intelligence. I'll return to 
this topic in a moment, but first let's address some other basic misconceptions about the so-called lack of progress in 
software.