RE: Central Control

[starlogo-users@media.mit.edu]

Yes, one can have visibility without control.

The issue, however, is control without visibility.

In unbounded systems such as the Internet there is neither central control
nor global visibility.  My concern is for security, survivability,
reliability and other similar properties that may be desired for
broad-based networks.  The desire is that these properties characterize the
network as a whole, while not necessarily characterizing any of the
individual nodes or components of the network.  Thus we might desire that
the network survive under certain classes of incidents or attacks, but
without expectation that any particular given node be able to survive such
an attack.  Similarly, how does one build reliable systems from unreliable
components?  Or, how can a species survive without any individuals of that
species surviving?

These are examples of emergent properties for which we seek solutions in
the form of emergent algorithms that have neither central control nor
global visibility.

My earlier point was that without global visibility it is very difficult to
exert meaningful central control.  Thus, if one is attempting to determine
whether an algorithm involves central control a reasonable place to begin
is the degree to which it depends on global visibility and if so how that
global information is used.

In the work I am doing we allow neither central control nor global
visibility.  Thus, at least in principle, our algorithms depend only on the
protocols of interactions between near neighboring entities (i.e., nodes,
processes, turtles, etc.).  A confusion that arises in the use of languages
such as Star Logo is that near neighbor information often can often only be
obtained through global search.  Although the process being simulated does
not inherently depend on or require global information, the only (or only
practical) way to implement it within the language is through the use of
global information.  If one uses global information to implement a process
that could in theory be implemented without such information, should that
process or its implementation be considered to depend on global
information? We would say no; that it is only a property of the
implementation and not of the abstract process being implemented.

It may that something similar is going when one person sees central control
where another does not.  It may be that the abstract process or algorithm
being implemented does not depend on central control, but that central
control mechanisms of the language are being used in the implementation.
If a turtle alters all turtles within three units distance by first marking
those at distance greater than three, altering all unmarked turtles, and
then resetting the marks, does the process involve central control? The
implementation clearly does.  The process being implemented may not
(assuming that the marks are not used for other purposes).


>I would not have thought that knowledge should equal control.
>
>We have a variety of institutions that aggregate widely-held information
>and make it generally available (newspapers, stock markets, etc.).
>Depending on how good a job the summary statistics do in capturing the
>distributed information, this may be akin to global visibility, but this
>doesn't say anything obvious about central control.
>
>Randy
>
>
>Prof. Randal C. Picker