Abstract

Embodied cognition is a radically new approach to the study of artificial intelligence . But because embodied cognition has only recently emerged, it does not have a well-defined set of central tenets that serve to unify its proponents. Classical AI, arguably the oldest approach to AI, employs proposition-sized representations and is based on the Physical Symbol System Hypothesis. Connectionism, by contrast, employs sub-symbolic distributed representations. Both classical AI and connectionism abstract intelligence from embodiment and attempt to build systems based on this abstraction. Embodied cognition on the other hand, examines intelligence in the context of embodiment and the environment. Merleau-Ponty's phenomenology is used to organize the criticisms that embodied cognition levels against classical AI and connectionism, organize the positive claims embodied cognition has made, and provide a set of central tenets for embodied cognition. In addition, recent evidence from biology and neuroscience are used to provide empirical support for the claims made by Merleau-Ponty and embodied cognition. Specifically, Marcel Kinsbourne's work in unilateral neglect, Barlow's studies of the horseshoe crab and Rizzolatti's research into cognitive maps is examined. Finally, a solution to the symbol-grounding problem is proposed. Classical AI and connectionism rely on the programmer or user to make the semantic connection between the AI system's internal symbols and the external objects they represent. By combining Merleau-Ponty and embodied cognition, however, it is possible to construct AI systems that solve the symbol-grounding problem and do not require a human being in order to make the semantic connection