A Cognitive Neuroscience-inspired Codelet-based Cognitive Architecture for the Control of Artificial Creatures with Incremental Levels of Machine Consciousness Klaus Raizer, André L. O. Paraense and Ricardo R. Gudwin 1 Abstract. The advantages given by machine consciousness to the control of software agents were reported to be very appealing. The main goal of this work is to develop artificial creatures, controlled by cognitive architectures, with different levels of machine consciousness. To fulfil this goal, we propose the application of cognitive neuroscience concepts to incrementally develop a cognitive architecture following the evolutionary steps taken by the animal brain. The triune brain WKHRU\ SURSRVHG E\ 0DF/HDQ DQG DOVR $UUDEDOH¶V &RQV6FDOH ZLOO serve as roadmaps to achieve each developmental stage, while iCub - a humanoid robot and its simulator - will serve as a platform for the experiments. A completely codelet-based V\VWHP ‡&RUH· KDV EHHQ LPSOHPHQWHG VHUYLQJ WKH ZKROH architecture. 1 INTRODUCTION 1.1 Motivation In this work, we are particularly interested in studying the cognitive architectures which were proposed to deal with the issue of consciousness [10, 24, 37]. Our main goal is to develop artificial creatures with different levels of machine consciousness, controlled by such architectures. To fulfil this goal, we propose the application of cognitive neuroscience concepts to incrementally develop a cognitive architecture following the evolutionary steps taken by the animal brain. Looking for inspiration in nature has been a successful way of discovering new solutions for problems in the fields of control, optimization, classification and artificial intelligence. Machine learning techniques such as genetic algorithms, ant colony optimization and neural networks are some examples of the remarkable muse nature can be [27, 25, 36, 16]. The advantages given by machine consciousness have been reported to be very appealing [20, 10, 9]. Nevertheless, the cognitive architectures which are able to benefit from it are not so many, and still under heavy development. So, the motivation to propose and implement yet another cognitive architecture, when there are so many of them already available, lies in the need for an architecture coherent with our hypothesis of a conscious codelet-based artificial mind, able to implement the animal brain in its different evolutionary 1 School of Electrical and Computer Engineering (FEEC), University of Campinas (UNICAMP), Brazil, email: gudwin@dca.fee.unicamp.br steps, and in the search for the sufficient feature set in the architecture for each of those steps that matches the results of natural selection along history. 1.2 Statement and Background of Research An artificial creature is an autonomous agent, a system embedded in an environment, sensing and acting on it, over time, in pursuit of its own agenda [21]. It can be controlled by a cognitive architecture, which includes aspects of the creature such as memory and functional processes [29], providing a framework to support mechanisms for perception, action, adaptation and motivation [39]. Cognitive architectures are control systems architectures inspired by scientific theories developed to explain cognition in animals and in men. These architectures are typically organized in layers [2, 13], with each layer representing a different level of control and specialized modules [15]. The most famous general cognitive architectures are SOAR [28] and ACT-R [1]. More recently, many specialized cognitive architectures have been proposed, emphasizing different aspects of cognition, e.g. emotions, attention, memory, consciousness and language. Each one has advantages and shortcomings, when compared to each other. Looking for inspiration in cognitive neuroscience, current research on artificial creatures has focused on the implementation of machine consciousness, with one of its major functions being to recruit relevant resources for solving new or difficult problems [33]. Recently, the study of machine consciousness in cognitive architectures applied to artificial creatures has particularly been exploited [4, 15, 33, 8]. Even though there is not a consensus on what exactly is PHDQW E\ ‡PDFKLQH FRQVFLRXVQHVV· DV GLIIHUHQW DXWKRUV LQGHHG have different perspectives on what they mean by ‡FRQVFLRXVQHVV· LQ D SUHYLRXV ZRUN IURP RXU JURXS >35], we investigated one interesting proposal, called the Baars-Franklin architecture. During this investigation, we evaluated the possible EHQHILWV WKDW VXFK ‡FRQVFLRXVQHVV· WHFKQRORJ\ ZKHQ DSSOLHd to the control of autonomous agents, could bring to such systems. Our main findings were that the main benefits brought by consciousness (as defined in Baars-Franklin architecture), are:  Executive Summary of Perception  The Possibility of Behaviour Automatization These two advantages arose from the main perspective on what is consciousness after all in the Baars-Franklin architecture, following Dennett [14]. According to this perspective, 43