Cortical architecture and self-referential control for brain-like computation

Discusses a new approach to understanding how the brain organizes computation. Progress in understanding the brain function under constant interactions with the sensory environment is hampered by inadequate models and theories. Obviously, current models and theories of brain computing still appear irrelevant when they are confronted with real-world problems. We argue that architecture in the brain does not reflect the result of thinking, the ready-made algorithm for solving a problem. Rather it should reflect the control that generates the constraints to select a proper algorithm for a specific problem that is posed by the input-or to create a new one if the application of the previously acquired ones does not provide a sufficient solution. We propose that a value system (based on a genetically imprinted a priori knowledge on coarse behavioral evaluation of sensory input) and neocortical columnar architecture are crucial elements of future artificial neural systems that are expected to emulate the performance of the brain. This should be the case especially for those cognitive tasks that appear easy for animals in their everyday life but turn out to be hopelessly tricky for the current generation of computers. In order to advance beyond the well known paradigms of current computational theory, we need a more functional understanding of brain-type computation.