The following article was rejected by David Kinny, the moderator of the
usenet newsgroup, comp.ai.


-----------------------------------------------------------------------
In article <abum4l$c0t$1@mulga.cs.mu.OZ.AU>, erayo@bilkent.edu.tr
(Eray Ozkural  exa) wrote:

>I would be rather hesitant to associate the availability of
>human-equivalent systems with processing power. As other posts
>indicate, we don't have a single theory that scales up in the sense of
>intelligence :)

Indeed. There are those who claim that the lack of processing power is
what is preventing us from building a human-level intelligence. They
are mistaken for several reasons:

1. Unless we have a comprehensive theory of the operation of the
brain, the processing power of our computers is irrelevant.

2. The number of neurons in the human brain is said to be about 100
billions. However, what most people seem to overlook is that only a
tiny fraction of the brain's neurons are active (firing) at any one
time. Catscans reveal that brain activity is severely localized during
most focused tasks. (My own experiments with spiking networks suggest
that the number of active neurons is three to four orders of magnitude
less than the total number of available neurons.) The idea here is
that CPUs need to attend to only a small subset of an intelligent
network at any given time. This drastically cuts down on the computing
power requirements.

3. A large percentage of the human brain is dedicated to maintaining
the proper functioning of the body and to things that our intelligent
machines will not need to concern themselves with (at least
initially): sex, thirst, etc...

4. If we knew how the brain worked, there would be nothing to stop us
from building a human-level AI right now. Given recent advances in
computer clustering and optical networking technologies, enough CPU,
FPGA and memory boards can be linked into a parallel supercluster to
simulate 100 billion neurons, even with the (incorrect) assumption
that they will all be active simultaneously.

5. There is no need to have a machine with 100 billion neurons to
achieve highly useful intelligence. A bee's brain has about 1 or two
million neurons (five orders of magnitude less than human brains) and
yet bees display amazingly complex and robust behavior. Current
machines can certainly match the neural processing power of a bee.
What accounts for the laughably poor behavioral performance of modern
robots/agents compared to that of a bee?

Surely the AI community cannot blame the lack of available computer
power for their failure to deliver a HAL-like intelligence. Something
is obviously amiss.

Temporal Intelligence:
http://home1.gte.net/res02khr/AI/Temporal_Intelligence.htm