Developmental Neurosimulation as a Route to Embodied Morphogenetic-inspired Intelligence

Abstract

The Developmental Neurosimulation approach allows us to approximate simple connectomes and their associated behaviors that emerge in an embodied context. By combining morphological and behavioral bio-inspiration with topological constructivism, the relationship between morphology and connectome has been demonstrated in three ways: providing a substrate for information acquisition, the role of pattern formation and developmental transformations in information processing, and the use of epigenetic functions to simulate critical periods and other forms of nonlinear acquisition. Embodied connectomes are constructed from innate instruction sets, which result in morphogenetic-inspired substrates for information acquisition. In particular, connecting a connectionist network with an embodied input/output network can act to structure input data with respect to orientation behaviors towards a stimulus. This results in a highly structured connectome topology exhibiting developmental freedom, or alternate forms of equivalent connectivity given plasticity and constraint. To grow topologically complex embodied connectionist networks, we utilize generative pattern formation as a mechanism for morphological transformation. This produces bodies of different shapes and internal networks of different sizes, and leads to acquisition that is grounded in how the agent actually utilizes the environment. Finally, successful developmental neurosimulation requires a variable model of temporal information processing that more closely resembles naturalistic interactions. To capture this in a computational model, we propose using epigenetic functions to enable stage-specific acquisition. Stage-specific acquisition, consistent with morphological development and enabling critical periods, structures association-building through increasing network capacity. Ultimately this contributes to an agent-based form of generative intelligence with a capacity for behavioral complexity. We will also demonstrate developmental neurosimulation with two examples: a sensory array influenced during a critical period, and the action of developmental freedom on a small, dense Braitenberg Vehicle connectome.