Sustaining Evolution for Shallow Embodied Intelligence

Abstract Lenski’s experiments with E. Coli show Biology can sustain continual evolutionary improvement. However long term evolutionary experiments (LTEE) with evolutionary computing find that information theory’s failed disruption propagation (FDP) in deeply nested genetic programming (GP) hierarchies can greatly slow adaptation. We propose that researchers aiming at embodied artificial intelligence should control software robustness by using porous high surface area geometrical architectures, perhaps composed of many shallow mangrove like tree structures intimately linked to their data rich fitness environment.