CogBot An Integrative Cognitive Architecture Aimed at Emulating Early Childhood Intelligence in a Humanoid Robot

Might it be possible, using a combination of current technologies, to create a software program enabling a robot to think, learn and converse roughly at the level of a young human child? While achievements like this have been elusive throughout the history of AI, we suspect that due to advances in AI, hardware and cognitive science, the answer is now: Yes, if the cognitive architecture of the program is right, and the combination of technologies is carried out according to the right theoretical principles. The goal of the proposed research is to explore the hypothesis that the cognitive synergy ensuing from integrating multiple symbolic and subsymbolic learning and memory components in an appropriate cognitive architecture and environment, can yield robust childlike intelligence. Toward this end, a three year software R&D project is proposed, aimed at creating a cognitive architecture called CogBot enabling a humanoid robot to carry out a set of mentally challenging tasks in a robot lab outfitted as a " virtual preschool. " The robotic platform for the project will be the Nao humanoid robot, connected via wifi with a compute cluster running the CogBot software, including standard multiprocessor servers and NVidia Tesla vector processing servers. Human child behavior will be used not as a target to be closely imitated, but rather as a guide to formulating appropriate tasks and metrics for instructing and assessing the robot. The goal is for the robot to learn, reason, create and converse at the qualitative level of a three year old human child. The core of the effort will be the CogBot integrative cognitive architecture, a fusion of of Itamar Arel's DeSTIN [1] deep learning system for perception and actuation, with Ben Goertzel's and Cassio Pennachin's OpenCog architecture for cognition and language [2]. Collectively, these systems involve sophisticated methods for visual and auditory pattern inference, language comprehension and generation, action planning, commonsense uncertain reasoning, and concept creation; and most of these capabilities have been evaluated in prior applications of narrower scope. In CogBot these methods will be integrated in a novel way, based on the principle of cognitive synergy, resulting in an overall cognitive architecture in which practical goals are achieved via multiple learning processes associated with multiple memory types cooperatively manipulating a common network-based knowledge store. While most of the components of CogBot are moderately similar to those used in prior AI systems, our hypothesis is that the proper integration …