Microsoft Word Kurzweil, Ray The Singularity Is Near doc

 
Reversible Computing.
Ultimately, organizing computation with massive parallel processing, as is done in the human 
brain, will not by itself be sufficient to keep energy levels and resulting thermal dissipation at reasonable levels. The 
current computer paradigm relies on what is known as irreversible computing, meaning that we are unable in principle 
to run software programs backward. At each step in the progression of a program, the input data is discarded—
erased—and the results of the computation pass to the next step. Programs generally do not retain all intermediate 
results, as that would use up large amounts of memory unnecessarily. This selective erasure of input information is 
particularly true for pattern-recognition systems. Vision systems, for example, whether human or machine, receive 
very high rates of input (from the eyes or visual sensors) yet produce relatively compact outputs (such as identification 
of recognized patterns). This act of erasing data generates heat and therefore requires energy. When a bit of 

---

information is erased, that information has to go somewhere. According to the laws of thermodynamics, the erased bit 
is essentially released into the surrounding environment, thereby increasing its entropy, which can be viewed as a 
measure of information (including apparently disordered information) in an environment. This results in a higher 
temperature for the environment (because temperature is a measure of entropy). 
If, on the other hand, we don't erase each bit of information contained in the input to each step of an algorithm but 
instead just move it to another location, that bit stays in the computer, is not released into the environment, and 
therefore generates no heat and requires no energy from outside the computer. 
Rolf Landauer showed in 1961 that reversible logical operations such as NOT (turning a bit into its opposite) 
could be performed without putting energy in or taking heat out, but that irreversible logical operations such as AND 
(generating bit C, which is a 1 if and only if both inputs A and Bare 1) do require energy.
48
In 1973 Charles Bennett 
showed that any computation could be performed using only reversible logical operations.
49
A decade later, Ed 
Fredkin and Tommaso Toffoli presented a comprehensive review of the idea of reversible computing.
50
The 
fundamental concept is that if you keep all the intermediate results and then run the algorithm backward when you've 
finished your calculation, you end up where you started, have used no energy, and generated no heat. Along the way, 
however, you've calculated the result of the algorithm.