Information Technology Reference
In-Depth Information
The brain has multilevel hierarchical organization. The most important fact is
that no single-layer model is capable of reproducing the functions of the brain.
The brain is not a system with consecutive information processing.
In essence, Arbib's ideas underlie the interest in neural networks that emerged in
that time. Moreover, this was the interest in complex variants of semiconductor
computer architecture that would allow for a manifold increase in the parallelism of
computation.
In those same years, Michael Conrad, the leading US expert in the field of
molecular electronics, published a detailed comparative analysis of information
characteristics of the brain and the von Neumann computer. A remarkable feature
of Conrad's approach was that it was based on general information concepts,
without tying them to any physical implementation of the devices. Conrad com-
pared the fundamental differences in information processing by the brain and by
computer, based on the exclusion principle he had previously introduced:
“A system cannot at the same time be effectively programmable at the level of
structure, amendable to evolution by variation and selection, and computationally
efficient.”
These differences are summarized in Fig.
8.8
. They include the possibility of
structured programming of the systems, the parallel or serial nature of information
processing, the vertical or horizontal flows of information, etc. In essence, in this
analysis Conrad demonstrated that the brain and the von Neumann computer are
two extreme alternatives for information processing systems.
Since Conrad's approach is quite general, not tied to any physical implementa-
tion, it appeared attractive to use it to assess the place of distributed reaction-
diffusion devices in the general system of information processing devices. Consider
the basic information characteristics of reaction-diffusion devices. In contrast to
the von Neumann computer, reaction-diffusion devices are not externally program-
mable. Their dynamics is determined by the state (and structure) of the medium as
well as by control stimuli.
Even the simplest devices exhibit a very high degree of parallelism, mixed
continuous-discrete dynamics, and vertical fluxes of information transmission
and processing. Even in a simple system one can identify:
The level of macro-micro data transformation, i.e., the level of data input
Dynamics at the molecular (micro) level which implements the method of
information processing
The level of micro-macro transformation of information, i.e., physicochemical
readout of the solution to the problem
The degree of self-organization of chemical reaction-diffusion devices is high.
Moreover, they manifest gradualism, i.e., small changes in the state of the medium
(the concentrations of its components and temperature) lead, in a certain area of the
states, only to a relatively small quantitative rather than sharp qualitative change of
dynamic regimes. This feature, in essence, is the basis for constructing systems with
training.
