Microsoft Word Kurzweil, Ray The Singularity Is Near doc

How Smart Is a Rock?
To appreciate the feasibility of computing with no energy and no heat, consider the 
computation that takes place in an ordinary rock. Although it may appear that nothing much is going on inside a rock, 
the approximately 10
25
(ten trillion trillion) atoms in a kilogram of matter are actually extremely active. Despite the 
apparent solidity of the object, the atoms are all in motion, sharing electrons back and forth, changing particle spins, 
and generating rapidly moving electromagnetic fields. All of this activity represents computation, even if not very 
meaningfully organized. 
We've already shown that atoms can store information at a density of greater than one bit per atom, such as in 
computing systems built from nuclear magnetic-resonance devices. University of Oklahoma researchers stored 1,024 
bits in the magnetic interactions of the protons of a single molecule containing nineteen hydrogen atoms.
51
Thus, the 
state of the rock at anyone moment represents at least 10
27
bits of memory. 
In terms of computation, and just considering the electromagnetic interactions, there are at least 10
15
changes in 
state per bit per second going on inside a 2.2-pound rock, which effectively represents about 10
42
(a million trillion 
trillion trillion) calculations per second. Yet the rock requires no energy input and generates no appreciable heat. 
Of course, despite all this activity at the atomic level, the rock is not performing any useful work aside from 
perhaps acting as a paperweight or a decoration. The reason for this is that the structure of the atoms in the rock is for 
the most part effectively random. If, on the other hand, we organize the particles in a more purposeful manner, we 
could have a cool, zero-energy-consuming computer with a memory of about a thousand trillion trillion bits and a 
processing capacity of 10
42
operations per second, which is about ten trillion times more powerful than all human 
brains on Earth, even if we use the most conservative (highest) estimate of 10
19
cps.
52
Ed Fredkin demonstrated that we don't even have to bother running algorithms in reverse after obtaining a result.
53
Fredkin presented several designs for reversible logic gates that perform the reversals as they compute and that are 
universal, meaning that general-purpose computation can be built from them.
54
Fredkin goes on to show that the 
efficiency of a computer built from reversible logic gates can be designed to be very close (at least 99 percent) to the 
efficiency of ones built from irreversible gates. He writes: 
it is possible to ... implement ... conventional computer models that have the distinction that the basic 
components are microscopically . reversible. This means that the macroscopic operation of the computer is 
also reversible. This fact allows us to address the ... question ... "what is required for a computer to be 
maximally efficient?" The answer is that if the computer is built out of microscopically reversible 
components then it can be perfectly efficient. How much energy does a perfectly efficient computer have to 

---

dissipate in order to compute something? The answer is that the computer does not need to dissipate any 
energy.
55
Reversible logic has already been demonstrated and shows the expected reductions in energy input and heat 
dissipation.
56
Fredkin's reversible logic gates answer a key challenge to the idea of reversible computing: that it would 
require a different style of programming. He argues that we can, in fact, construct normal logic and memory entirely 
from reversible logic gates, which will allow the use of existing conventional software-development methods. 
It is hard to overstate the significance of this insight. A key observation regarding the Singularity is that 
information processes—computation—will ultimately drive everything that is important. This primary foundation for 
future technology thus appears to require no energy. 
The practical reality is slightly more complicated. If we actually want to find out the results of a computation—
that is, to receive output from a computer—the process of copying the answer and transmitting it outside of the 
computer is an irreversible process, one that generates heat for each bit transmitted. However, for most applications of 
interest, the amount of computation that goes into executing an algorithm vastly exceeds the computation required to 
communicate the final answers, so the latter does not appreciably change the energy equation. 
However, because of essentially random thermal and quantum effects, logic operations have an inherent error rate. 
We can overcome errors using error-detection and-correction codes, but each time we correct a bit, the operation is not 
reversible, which means it requires energy and generates heat. Generally, error rates are low. But even if errors occur 
at the rate of, say, one per 10
10
operations, we have only succeeded in reducing energy requirements by a factor of 
10
10
, not in eliminating energy dissipation altogether. 
As we consider the limits of computation, the issue of error rate becomes a significant design issue. Certain 
methods of increasing computational rate, such as increasing the frequency of the oscillation of particles, also increase 
error rates, so this puts natural limits on the ability to perform computation using matter and energy. 
Another important trend with relevance here will be the moving away from conventional batteries toward tiny fuel 
cells (devices storing energy in chemicals, such as forms of hydrogen, which is combined with available oxygen). Fuel 
cells are already being constructed using MEMS (microelectronic mechanical systems) technology.
57
As we move 
toward three-dimensional, molecular computing with nanoscale features, energy resources in the form of nano-fuel 
cells will be as widely distributed throughout the computing medium among the massively parallel processors. We will 
discuss future nanotechnology-based energy technologies in chapter 5.