Microsoft Word Kurzweil, Ray The Singularity Is Near doc

The brain uses emergent properties.
Intelligent behavior is an emergent property of the brain's chaotic and 
complex activity. Consider the analogy to the apparently intelligent design of termite and ant colonies, with their 
delicately constructed interconnecting tunnels and ventilation systems. Despite their clever and intricate design, 
ant and termite hills have no master architects; the architecture emerges from the unpredictable interactions of all 
the colony members, each following relatively simple rules. 
•
The brain is imperfect.
It is the nature of complex adaptive systems that the emergent intelligence of its decisions 
is suboptimal. (That is, it reflects a lower level of intelligence than would be represented by an optimal 
arrangement of its elements.} It needs only to be good enough, which in the case of our species meant a level of 
intelligence sufficient to enable us to outwit the competitors in our ecological niche (for example, primates who 
also combine a cognitive function with an opposable appendage but whose brains are not as developed as 
humans and whose hands do not work as well). 
•
We contradict ourselves.
A variety of ideas and approaches, including conflicting ones, leads to superior 
outcomes. Our brains are quite capable of holding contradictory views. In fact, we thrive on this internal 
diversity. Consider the analogy to a human society, particularly a democratic one, with its constructive ways of 
resolving multiple viewpoints. 
•
The brain uses evolution.
The basic learning paradigm used by the brain is an evolutionary one: the patterns of 
connections that are most successful in making sense of the world and contributing to recognitions and decisions 
survive. A newborn's brain contains mostly randomly linked interneuronal connections, and only a portion of 
those survive in the two-year-old brain.
17
•
The patterns are important.
Certain details of these chaotic self-organizing methods, expressed as model 
constraints (rules defining the initial conditions and the means for self-organization), are crucial, whereas many 
details within the constraints are initially set randomly. The system then self-organizes and gradually represents 
the invariant features of the information that has been presented to the system. The resulting information is not 
found in specific nodes or connections but rather is a distributed pattern. 
•
The brain is holographic.
There is an analogy between distributed information in a hologram and the method of 
information representation in brain networks. We find this also in the self-organizing methods used in 
computerized pattern recognition, such as neural nets, Markov models, and genetic algorithms.
18

---

•
The brain is deeply connected.
The brain gets its resilience from being a deeply connected network in which 
information has many ways of navigating from one point to another. Consider the analogy to the Internet, which 
has become increasingly stable as the number of its constituent nodes has increased. Nodes, even entire hubs of 
the Internet, can become inoperative without ever bringing down the entire network. Similarly, we continually 
lose neurons without affecting the integrity of the entire brain. 
•
The brain does have an architecture of regions.
Although the details of connections within a region are initially 
random within constraints and self-organizing, there is an architecture of several hundred regions that perform 
specific functions, with specific patterns of connections between regions. 
•
The design of a brain region is simpler than the design of a neuron.
Models often get simpler at a higher level, 
not more complex. Consider an analogy with a computer. We do need to understand the detailed physics of 
semiconductors to model a transistor, and the equations underlying a single real transistor are complex. 
However, a digital circuit that multiplies two numbers, although involving hundreds of transistors, can be 
modeled far more simply, with only a few formulas. An entire computer with billions of transistors can be 
modeled through its instruction set and register description, which can be described on a handful of written pages 
of text and mathematical transformations. 
The software programs for an operating system, language compilers, and assemblers are reasonably complex, but 
modeling a particular program—for example, a speech-recognition program based on Markov modeling—may be 
described in only a few pages of equations. Nowhere in such a description would be found the details of 
semiconductor physics. A similar observation also holds true for the brain. A particular neural arrangement that detects 
a particular invariant visual feature (such as a face) or that performs a band-pass filtering (restricting input to a specific 
frequency range) operation on auditory information or that evaluates the temporal proximity of two events can be 
described with far greater simplicity than the actual physics and chemical relations controlling the neurotransmitters 
and other synaptic and dendritic variables involved in the respective processes. Although all of this neural complexity 
will have to be carefully considered before advancing to the next higher level (modeling the brain), much of it can be 
simplified once the operating principles of the brain are understood. 

---