Embracing Embodied Social Cognition in AI: Moving Away from Computational Theory of Mind

Abstract

As artificial intelligence becomes more integral to daily life, the need to design AI systems capable of understanding human interactions is increasingly important. This paper delves into the integration of social cognition in AI, tracing back to its historical foundations and examining seminal theories like Newell's Bands of Cognition, Minsky's Society of Mind, etc., which have emphasized the importance of social cognition since AI's inception. We highlight the shortcomings of traditional computational theory of mind approaches, particularly in their failure to capture the embodied nature of social cognition. Advocating for including embodied socio-cognitive perspectives, we draw on theories such as Participatory Sensemaking and frameworks like Observable Creative Sensemaking. The paper further demonstrates the practical implementation of these concepts in AI through two case studies: one in co-creative dance AI and another in text-to-image generative AI systems.