"Cybernetics or Control and Communications in the Animal and the Machine," by Norbert Weiner was first published in 1949.  It is bit dated now, but still worth reading.  It provides an historical perspective on our understanding of this subject.  There are some speculations that have not proven correct, but the basic mathematics and science is as valid today as when it was written.  One of the problems discussed in Weiner's book comes from a military problem.  After World War II the military was trying to improve the targeting accuracy of its anti-aircraft weapons with automation.  The technology base of automation at that time was very different than the digital computer of today.  The first attempt, which failed, was based on calculating the trajectories of a projectile and the target to determine where they would intercept.  Since the physics and mathematics of doing this was an exact science, they expected extreme accuracy via automation.  They were wrong.  To find out why read Weiner's book.  For my purpose we only need to know that feedback was the central element of the solution.  They fired a first shot based on trajectory calculations, noted the miss distance and made an adjustment.  Since everything was moving very fast the adjustment had to be fast, thus the need for automation.  It worked. 

Stated in today's terminology, what Weiner recognized was that the problem of hitting a target is a goal seeking behavior problem.  He also showed why error correcting via feedback is easier than trying to calculate the path from fundamental principles.  The problem is in error control.  In a complex system errors can grow and a very small error at the beginning can be amplified so as to completely miss the goal.  Rather than trying to improve the accuracy at the beginning, it is often better to rely on feedback to bring us to our goal. 

In order for feedback to work your first shot must be reasonably close.  Feedback will close the gap much faster if the first miss is at least in the right area.  In some situations it is possible for the targeting system to oscillate wildly with a first bad shot.  When we discuss evolutionary search we will also learn that from where we begin the search matters.  There is also a matter of getting the right amount of delay in the feedback.  Get it wrong and feedback can cause wild variations without ever hitting the goal. 

ç  Prior Page of Text     Next Page of Text è
(C) 2005-2014 Wayne M. Angel.  All rights reserved.