I'm interested in computational techniques in which the information content
of a problem drives a description of the problem to evolve into a description of the
solution. On this view, information (usually mutual information) exerts a force by
virtue of the form and function of the symbols which encode it.
Essentially, this casts information as a potential and mutual information as a
potential difference. The force referred to is therefore just the gradient of this
potential. Then an algorithm is simply a computational model of the symbolic
dynamics induced on the semantic space of symbol configurations by a pattern of
such forces. And a solution, finally, is nothing other than a configuration which
minimizes this potential.
|