Microsoft Word Kurzweil, Ray The Singularity Is Near doc

The Software of the Brain.
The price-performance of computation and communication is doubling every year. As we 
saw earlier, the computational capacity needed to emulate human intelligence will be available in less than two 
decades.
5
A principal assumption underlying the expectation of the Singularity is that nonbiological mediums will be 
able to emulate the richness, subtlety, and depth of human thinking. But achieving the hardware computational 
capacity of a single human brain—or even of the collective intelligence of villages and nations—will not automatically 
produce human levels of capability. (By “human levels" I include all the diverse and subtle ways humans are 
intelligent, including musical and artistic aptitude, creativity, physical motion through the world, and understanding 
and responding appropriately to emotions.) The hardware computational capacity is necessary but not sufficient. 
Understanding the organization and content of these resources—the software of intelligence—is even more critical and 
is the objective of the brain reverse-engineering undertaking. 
Once a computer achieves a human level of intelligence, it will necessarily soar past it. A key advantage of 
nonbiological intelligence is that machines can easily share their knowledge. If you learn French or read 
War and 
Peace
, you can't readily download that learning to me, as I have to acquire that scholarship the same painstaking way 
that you did. I can't (yet) quickly access or transmit your knowledge, which is embedded in a vast pattern of 
T 

---

neurotransmitter concentrations (levels of chemicals in the synapses that allow one neuron to influence another) and 
interneuronal connections (portions of the neurons called axons and dendrites that connect neurons). 
But consider the case of a machine's intelligence. At one of my companies, we spent years teaching one research 
computer how to recognize continuous human speech, using pattern-recognition software.
6
We exposed it to thousands 
of hours of recorded speech, corrected its errors, and patiently improved its performance by training its "chaotic" self-
organizing algorithms (methods that modify their own rules, based on processes that use semirandom initial 
information, and with results that are not fully predictable). Finally, the computer became quite adept at recognizing 
speech. Now, if you want your own personal computer to recognize speech, you don't have to put it through the same 
painstaking learning process (as we do with each human child); you can simply download the already established 
patterns in seconds.