Monday, August 22, 2005 • 12:15 PM • Robert N. Noyce Conference Room
Virgil Griffith Indiana University
On the viability of self-reproducing classical machines
In the early 1950's John von Neuman was investigating self-reproduction of analog machines from a "sea of parts." For simplicity, he assumed an idealized error-free environment and eventually designed self-reproducing Cellular Automata. John von Neumann's 29 stage 200,000 cell self-reproducing machine (as well as other self-reproducing CA's) are complex and extremely brittle to any noise in their environment. Given its extreme sensitivity to noise, Von Neuman's model is unrealistic for any physical machine, and is also less interesting as it makes evolution impossible.
This leads to the question of what are the necessary conditions to achieve self-reproduction in a stochastic environment in which there are errors in reproduction. We developed a simplified model to study when it is possible for a machine to copy itself with enough fidelity to be viable. One worries that if a machine is too complicated, and the noise is too high, there will be an error catastrophe in which the machine cannot reproduce sufficiently accurate to sustain itself. We make an estimate of the critical error threshold under a given set of assumptions about the fitness function characterizing the machines and their environment.
