Implementing machine consciousness

Paulo Santos, University of Leeds Continued on p. 8 The last decade has seen the emergence of the new discipline of consciousness studies, which is now well established with journals (e.g. Cognition and Consciousness, Journal of Consciousness Studies), societies (ASSC, the Association for the Scientific Study of Consciousness), and conferences (the biennial Toward a Science of Consciousness, and the annual ASSC meetings) of its own. Consciousness studies encompass a huge range of activities: these are centred around psychology, but range from philosophy to neuroscience. Most recently, a new strand—machine consciousness—has emerged.1 In the 1990s, a small number of pioneers, including Stan Franklin in AI, Igor Aleksander in electrical engineering, and Gerald Edelman in biology, had begun to examine the possibility of creating consciousness in machines or software. Growing interest in these activities led to the 2001 Banbury Workshop Can a machine be conscious? At the end of this highly interdisciplinary event, the twenty-odd participants were asked to vote on whether they thought machine consciousness was in fact a possibility, and all except one agreed that it was. The outcome both surprised the participants, and gave the topic an impetus which is reflected in the level of interest in the symposium Next generation approaches to machine consciousness at AISB 2005. We set up the Machine Consciousness Group at Essex to explore the possibility of achieving machine consciousness in a robot through the formation and exploitation of internal models of itself and the world. This enterprise has been facilitated by the University’s support, via a £4.5m SRIF (Science Research Investment Fund) grant, for the building of a new robotic arena and workshop complex. Current work is directly funded by an EPSRC (Engineering and Physical Sciences Research Council) Adventure Fund award of £493,000, shared between myself and the visual neuropsychologists Tom Troscianko and Iain Gilchrist of the University of Bristol. This project, which runs until April 2007, involves the design and construction of a humanoid robot, and the study of the nature and development of its consciousness-related processes based on internal modelling. Why a humanoid robot? There are several reasons for this. We do not know very much about consciousness, but there has been a growing consensus that the origins of at least the lower levels of conscious phenomena are very strongly rooted in the body. If we are to produce consciousness in a machine, then its embodiment will be a critical determinant of the nature of that consciousness, and of its intelligibility and relevance to human consciousness. The robot (Cronos) is therefore being given a gross physical structure that is, as far as possible, qualitatively similar to the human body. Figure 1 shows the prototype (modelled on the body of its designer, Rob Knight): the basic humanoid skeletal structure of the (headless) upper torso is clear. The articulated skeleton models many of the constraints and degrees of freedom of our own body. In addition, the musculature uses a mixture of passive compliance and series-elastic actuators, ensuring that the motor programs used by the robot will be similar in important ways to those used in our own brains. When complete, the torso will be mounted on a wheeled base. Most autonomous mobile robots merely move through environments, but Cronos is being designed to be able to operate on the environment in ways comparable to those used by humans. At full extension, the arm and hand will be able to