The perception- ... -action cycle cognitive architecture and autonomy: a view from the brain

Abstract

Autonomous systems with reasoning capabilities are systems able to perceive their environment and act on it by performing complex tasks automatically.Autonomous systems are also able to adapt to unforeseen operating conditions or errors in a robust and predictable manner without the need for human guidance, instructions or programming.To accomplish such complex feats they must master the powers of perception, recognition, attention, learning and memory, cognitive control, reward and motivation, decision making, affordance extraction, action planning and action execution (step 1).Once these powers are successfully mastered, then these systems may be embodied into a robot able to act in the real world (step 2).Their embodiment, however, cannot guarantee that these systems will be able to operate autonomously in the environment as they will still need to solve the issues of the real-time system operation, resource management and meta-learning (step 2).In their article "Cognitive architectures and autonomy: a comparative review" Thrisson and Helgasson reviewed a number of "autonomous" systems and architectures with general "cognitive" capabilities and compared and contrasted their performance in a hypothetical example of autonomous exploration of an environment by a robot.Instead of their criteria focusing on how the powers of perception, recognition, attention, memory, cognition, decision making and action planning and execution are achieved by these systems (step 1), the authors ignored these powers, and compared and contrasted the systems based on step 2's real-time processing, resource management, learning and meta-learning issues.The authors argued that the former functions (e.g.perception, recognition, attention, memory, etc.) are less important.I believe that dealing with the issues of real-time processing, resource management, learning and meta-learning first and comparing and contrasting the reasoning capabilities of systems based on them is similar to building a house from the roof down.The systems are forced to solve the real-time system operations of functionalities which they have not deciphered yet, so they will inevitably be dumb, as they will be empty shells not possessing any reasoning powers that will enable them to go beyond the information provided.Furthermore, though some of the reviewed systems are "biologically inspired" in that they depend on behavioral studies and test themselves by the replication of experimental behavioral data, none of these systems attempt reverse engineering of the brain circuitry that supports these behaviors.Reasoning is the highest faculty of the human brain and it depends on the majority of the brain components (perception, attention, learning and memory, decision making, action, etc.).The brain is a system that has evolved over a million or so years, so it is expected to provide a reasonably optimized solution to many of the cognitive tasks under consideration.