Microsoft Word Kurzweil, Ray The Singularity Is Near doc

Peeling the Onion.
The brain is not a single information-processing organ but rather an intricate and intertwined 
collection of hundreds of specialized regions. The process of "peeling the onion" to understand the functions of these 
interleaved regions is well under way. As the requisite neuron descriptions and brain-interconnection data become 
available, detailed and implementable replicas such as the simulation of the auditory regions described below (see 
"Another Example: Watts's Model of the Auditory Regions” on p. 183) will be developed for all brain regions. 
Most brain-modeling algorithms are not the sequential, logical methods that are commonly used in digital 
computing today. The brain tends to use self-organizing, chaotic, holographic processes (that is, information not 
located in one place but distributed throughout a region). It is also massively parallel and utilizes hybrid digital-
controlled analog techniques. However, a wide range of projects has demonstrated our ability to understand these 
techniques and to extract them from our rapidly escalating knowledge of the brain and its organization. 
After the algorithms of a particular region are understood, they can be refined and extended before being 
implemented in synthetic neural equivalents. They can be run on a computational substrate that is already far faster 
than neural circuitry. (Current computers perform computations in billionths of a second, compared to thousandths of a 
second for interneuronal transactions.) And we can also make use of the methods for building intelligent machines that 
we already understand. 
Is the Human Brain Different from a Computer? 

---

The answer to this question depends on what we mean by the word "computer." Most computers today are all digital 
and perform one (or perhaps a few) computations at a time at extremely high speed. In contrast, the human brain 
combines digital and analog methods but performs most computations in the analog (continuous) domain, using 
neurotransmitters and related mechanisms. Although these neurons execute calculations at extremely slow speeds 
(typically two hundred transactions per second), the brain as a whole is massively parallel: most of its neurons work at 
the same time, resulting in up to one hundred trillion computations being carried out simultaneously. 
The massive parallelism of the human brain is the key to its pattern-recognition ability, which is one of the pillars 
of our species' thinking. Mammalian neurons engage in a chaotic dance (that is, with many apparently random 
interactions), and if the neural network has learned its lessons well, a stable pattern will emerge, reflecting the 
network's decision. At the present, parallel designs for computers are somewhat limited. But there is no reason why 
functionally equivalent nonbiological re-creations of biological neural networks cannot be built using these principles. 
Indeed, dozens of efforts around the world have already succeeded in doing so. My own technical field is pattern 
recognition, and the projects that I have been involved in for about forty years use this form of trainable and 
nondeterministic computing. 
Many of the brain's characteristic methods of organization can also be effectively simulated using conventional 
computing of sufficient power. Duplicating the design paradigms of nature will, I believe, be a key trend in future 
computing. We should keep in mind, as well, that digital computing can be functionally equivalent to analog 
computing—that is, we can perform all of the functions of a hybrid digital-analog network with an all-digital 
computer. The reverse is not true: we can't simulate all of the functions of a digital computer with an analog one. 
However, analog computing does have an engineering advantage: it is potentially thousands of times more 
efficient. An analog computation can be performed by a few transistors or, in the case of mammalian neurons, specific 
electrochemical processes. A digital computation, in contrast, requires thousands or tens of thousands of transistors. 
On the other hand, this advantage can be offset by the ease of programming (and modifying) digital computer-based 
simulations. 
There are a number of other key ways in which the brain differs from a conventional computer: 
•