Isolating the Black Box - Exploring the Effects of a Material Interface in Evolution in Materio

Abstract

Evolution in Materio (EiM) is the concept of exploiting intrinsic properties of various materials, or "black boxes", to perform computation. Motivation for EiM is multitudinous; new paradigms for computing, increased performance, and basic research on what computation is or can be. The work herein is part of the NASCENCE project, which aims at exploring nanoscale devices for computation and, further, how such information processing devices can emerge from bottom-up design processes, e.g. artificial evolution. In EiM research, instability and lack of determinism has been an ongoing problem regarding the evolution of useful computation. Meanwhile there has also been a lack of a clear, distinct border between the mechanism that exploits the material and the material itself. This raises questions like; where does computation arise? In the material, in the interface to the environment, or a combination?

In this thesis we have investigated the effect of drawing such a border by designing and implementing a material interface named optowall. This interface electrically isolates the material by using opto-isolators to transmit all signals by light instead of by an electrical connection. An experimental approach has been taken to develop methods to define properties in the material such as stability, functionality, and the nature of instability. By developing software that extracts the values of these properties, we have compared them in different configurations of the interface and found that the method of delivering the signal into the material has a significant effect on the computational properties. Hence, the impact also influence the search landscape for any evolutionary algorithm. We have also discovered that non-linear functionality can be exploited by the adjustment of current. Furthermore we have shown that the potential differences between concurrent input signals have a significant effect on stability and functionality.