Tunable epigenetic catalysis: programming real-time cognitive machines
Rodrick Wallace · International Journal of Parallel Emergent and Distributed Systems · 2010
We adapt arguments from developmental biology to the design of real-time cognitive machines, via the homology between information source uncertainty and free energy density, in the context of the powerful asymptotic limit theorems of information theory. This leads to a model of epigenetic catalysis for programming such devices in which an ‘epigenetic farmer’ directs machine behaviour by lowering the relative effective energy of the desired developmental pathway, leading to a computed answer that may range from some simple fixed point to, in a coevolutionary machine, a particular highly structured dynamical behaviour pattern. Our approach may be of singular use for understanding the inevitable patterns of canonical and idiosyncratic failure that will afflict these machines.