Deconstructing and reconstructing turn-taking in caregiver-infant interactions: a platform for embodied models of early cooperation.

Despite recent advances in AI, building systems that can engage in natural and realistic cooperative interactions with human partners remains a challenge. In this article, we argue that as a precursor to modelling sophisticated cooperation in adults, it's useful to build a model of how cooperation develops in infants. We present a simple initial model of infant cooperation, embodied in BabyX - a hyperrealistic virtual simulation of an 18-month-old infant that can interact with human partners in real time. Our focus in this paper is on cooperative behaviours in the nonverbal domain. The framework for our model of these behaviours is a cognitive model of events and event processing. We detail how cognitive and motor mechanisms in BabyX lead to rudimentary cooperation manifested in nonverbal turn-taking. We introduce a novel empirical paradigm for testing BabyX's model of cooperation, by comparing her interactions with users with 'real' interactions between human infants and their caregivers, which we observed in a detailed empirical study. In this study, we find nonverbal turn-taking in human infant-caregiver interactions, consistent with our BabyX model. This initial model provides the foundation for a comprehensive, and developmentally consistent, model of human cooperation.