Minimal Agency Detection of Embodied Agents Hiroyuki Iizuka 1,2 and Ezequiel Di Paolo 1 1 Centre for Computational Neuroscience and Robotics, Department of Informatics, University of Sussex Brighton, BN1 9QH, UK 2 Department of Media Architecture, Future University-Hakodate 116-2 Kamedanakano-cho, Hakodate, Hokkaido, 041-8655, Japan ezca@sacral.c.u-tokyo.ac.jp, ezequiel@sussex.ac.uk Abstract. Agency detection is studied in a simple simulated model with embodied agents. Psychological experiments such as double TV-monitor experiments and perceptual crossing show the central role of dynamic mutuality and contingency in social interactions. This paper explores the ongoing dynamical aspects of minimal agency detection in terms of the mutuality and contingency. It is investigated how the embodied agents can establish a live interaction and discriminate this from inter- actions from recorded motions that are identical to the live interaction but cannot react contingently. Our results suggest that the recognition of the presence of another’s agency need not lie on complex cognitive individual mechanisms able to integrate past information, but rather on the situated ongoingness of the interaction process itself, on its dynamic properties, and its robustness to noise. 1 Introduction Social interaction may lead to coordinated behaviours when mutual anticipations are formed dynamically. This mutuality of influences is a key property of the in- teraction process but its dynamical characteristics have not been sufficiently investigated from a theoretical perspective. In contrast, important empirical ev- idence points to the central role of dynamic mutuality, or contingency, in sustain- ing and forming several aspects of an ongoing interaction. This is clearly shown in Trevarthen’s double TV-monitor experiments with infants [6,9], Nadel’s ex- tensions to these experiments [7], and in the perceptual crossing experiments by the GSP group at UTC Compi` egne [1]. In Trevarthen’s experiment, a mother and her baby are placed in separate rooms and allowed to interact only through video screens that display their faces to each other. During ‘live’ interaction, mother and infant engage in coordinated utterances and affective expressions. However, if a delayed video recording of the mother is displayed to the baby, the baby becomes withdrawn and depressed. This shows that it is not sufficient for the baby to sustain interaction that the mother’s expressive actions be dis- played on the monitor, but the mother is required to react ‘live’ to the baby’s own motions in order for the interaction to continue. It can be assumed that