Microsoft Word Kurzweil, Ray The Singularity Is Near doc

Analytic Versus Neuromorphic Modeling of the Brain.
A good example of the divergence between human 
intelligence and contemporary AI is how each undertakes the solution of a chess problem. Humans do so by 
recognizing patterns, while machines build huge logical "trees" of possible moves and countermoves. Most technology 
(of all kinds) to date has used this latter type of "top-down," analytic, engineering approach. Our flying machines, for 
example, do not attempt to re-create the physiology and mechanics of birds. But as our tools for reverse engineering 
the ways of nature are growing rapidly in sophistication, technology is moving toward emulating nature while 
implementing these techniques in far more capable substrates. 
The most compelling scenario for mastering the software of intelligence is to tap directly into the blueprint of the 
best example we can get our hands on of an intelligent process: the human brain. Although it took its original 
"designer" (evolution) several billion years to develop the brain, it's readily available to us, protected by a skull but 
with the right tools not hidden from our view. Its contents are not yet copyrighted or patented. (We can, however, 
expect that to change; patent applications have already been filed based on brain reverse engineering.)
7
We will apply 
the thousands of trillions of bytes of information derived from brain scans and neural models at many levels to design 
more intelligent parallel algorithms for our machines, particularly those based on self-organizing paradigms. 
With this self-organizing approach, we don't have to attempt to replicate every single neural connection. There is a 
great deal of repetition and redundancy within any particular brain region. We are discovering that higher-level models 
of brain regions are often simpler than the detailed models of their neuronal components. 
How Complex Is the Brain?
Although the information contained in a human brain would require on the order of one 
billion billion bits (see chapter 3), the initial design of the brain is based on the rather compact human genome. The 
entire genome consists of eight hundred million bytes, but most of it is redundant, leaving only about thirty to one 
hundred million bytes (less than 10
9
bits) of unique information (after compression), which is smaller than the program 
for Microsoft Word.
8
To be fair, we should also take into account "epigenetic" data, which is information stored in 
proteins that control gene expression (that is, that determine which genes are allowed to create proteins in each cell), as 
well as the entire protein-replication machinery, such as the ribosomes and a host of enzymes. However, such 
additional information does not significantly change the order of magnitude of this calculation.
9
Slightly more than 
half of the genetic and epigenetic information characterizes the initial state of the human brain. 
Of course, the complexity of our brains greatly increases as we interact with the world (by a factor of about one 
billion over the genome).
10
But highly repetitive patterns are found in each specific brain region, so it is not necessary 
to capture each particular detail to successfully reverse engineer the relevant algorithms, which combine digital and 
analog methods (for example, the firing of a neuron can be considered a digital event whereas neurotransmitter levels 
in the synapse can be considered analog values). The basic wiring pattern of the cerebellum, for example, is described 
in the genome only once but repeated billions of times. With the information from brain scanning and modeling 
studies, we can design simulated "neuromorphic" equivalent software (that is, algorithms functionally equivalent to the 
overall performance of a brain region). 

---

The pace of building working models and simulations is only slightly behind the availability of brain-scanning 
and neuron-structure information. There are more than fifty thousand neuroscientists in the world, writing articles for 
more than three hundred journals.
11
The field is broad and diverse, with scientists and engineers creating new scanning 
and sensing technologies and developing models and theories at many levels. So even people in the field are often not 
completely aware of the full dimensions of contemporary research.