Adaptive And Robust Artificial Organisms

Hava Siegelmann, Associate Professor of Computer Science, University of Massachusetts Amherst

Adaptive and self-monitoring systems are two desired properties of agents that are being tuned to their environment and can survive damages. Many biological systems are adaptive and self-monitoring to some extent, and can hint at possible ways of attaining these properties.

We first introduce an artificial multi-cellular organism that exhibits in addition to self-construction and -repair, also an adaptive sensing mechanism. The sensors have multiple sensing modes while being capable of having one mode at a time. We demonstrate how these sensors adapt to their environment in an optimal way to transfer the sensory data.

We then allow the cells (agents) the possibility of mutating their genes and study the self-monitoring property. The agents can repair genes to some extent but if they acquired un-repaired mutations and are no longer aiding the environment they are considered /aberrant./ If these injuries are not repairable, they may escalate so that it is beneficial that the agent kills itself. This gives rise to an artificial creature made of multiple agents in which local death prevents the unhealthy agents from destroying the system. This work may shed light also on tumor formation and suppression and we will explore some dynamical treatments.

Audio Slides