Adaptive behaviour through morphological plasticity in natural and artificial systems

Our concept of intelligence is changing. Embodiment has led to the rise of morphologies in Artificial Intelligence (AI) research. This thesis focuses on two research questions: 1) How can system morpholo­ gies, well-adapted to changing environments, be designed? 2) How can adaptive behaviour be generated through morphology? It is the fundamental argument of this thesis that morphological plasticity (MP), the environmentally induced variation in growth or development, can provide a solution to both ques­ tions. Specifically, this thesis is based around a detailed study of diatom valve morphogenesis. Diatoms, a unicellular organism, construct intricate siliceous structures (valves) around themselves which exhibit high plasticity to the environment. Diatom valve morphogenesis is a good example of how morphologies can be well-adapted to changing environments, an open problem in AI, and how adaptive behaviour can be generated through morphological processes alone. Through a constructivist approach this thesis contributes to both understanding of MP in natural systems and the design of MP algorithms for artificial adaptive systems. Several original models and frameworks are defined within this thesis: the Nature’s Batik Model of basic diatom valve morphogenesis; the Cellanimat, a ‘Dynamic Morphology’ based on the unicell, capa­ ble of MP driven adaptive behaviour through its unique ‘Artificial Cytoskeleton’ model of cytoskeletal dynamics; the Environment-Phenotype Map framework; and the Cellanimat Colony Model, which com­ bines all previous models for the investigation of MP mechanisms during diatom colony formation. Cellanimat dynamics and optimization are thoroughly investigated and the model is shown to be multi­ functional, evolvable, scalable and reasonably robust.