Through Reasoning to Cognitive Machines

Abstract

We approach the problem of creating a cognitive machine by initially analyzing nonlinguistic reasoning in chimpanzees and crows at the level of neural simulation. We present two principles for such reasoning as these animals possess: the presence of coupled triples of forward and inverse internal motor models coupled to buffer working memories, and of a mechanism to modify goal values in order to create sub-goals for guiding actions in a sequential order. A brief extension of these structures to begin the creation of low-level reasoning machines is then presented. How this can be extended to language is then considered in two steps: through the creation of a language learning machine, with semantic and syntactic structures up to phrase structure insertion mechanisms, and then how this system might be used for linguistic/logical reasoning. The importance of attention is then emphasized in order to handle the increased complexity of the encoded stimuli, and how an attention control system can begin to grant a minimal level of consciousness (through the use of a copy of the attention movement signal).