Physiology of Social Interactions with Natural and Artificial Agent
Topics: social interactions, artificial agents, MRI, brain, computational models
Disciplinary fields: neurosciences, cognition, language, informatics, automatism
Summary: The primary objective of the project PhysSocial is to propose an experimental approach to describe the physiological underpinnings of natural social interaction, and use these results to improve the social competence of artificial agents. As artificial anthropomorphic agents, including humanoid robots and computer-animated conversational agents, are expected to play increasingly important roles in contact with human – agents for e-commerce, tutors in e-learning or support in therapies – the question of their social competence is crucial. One hypothesis is that only when interacting with fellows do humans adopt an intentional stance (ie attribute mental states) towards the other agent, evidenced by activity in the medial prefrontal cortex. Achieving this objective requires the development of a neuroimaging experimental paradigm derived from social cognitive neuroscience. In the current project, a participant’s brain activity is scanned with functional magnetic resonance imaging (fMRI) while he discusses naturally with a human – intentional agent – or with an artificial agent – a humanoid robot or computer-animated agent – in the same setup. Data from multiple modalities, in particular behavioral and neurophysiological, is acquired during the discussions with human and artificial interlocutors. A statistical learning approach is proposed to analyze the complex, multidimensional, dataset in order to identify correlations between behavioral and neurophysiological events. It is predicted that a similar behavior leads to different physiological responses depending on the nature of the interlocutor. A computational model identifies single or complex behaviors from the human, or dynamic events from the human pair that elicit mentalization in the scanned participant. Identifying which aspects of the artificial agent social behaviors significantly affect the human interlocutor contributes to measuring objectively its social competence. It also identifies which behaviors should be in the longer term implemented in the artificial agent to its social competence.
Project duration: 03/01/2018 – 02/28/2021
Call for application: Interdisciplinarité 2016
Video: De l’humain : Comment rendre sociaux les agents artificiels ?, Collection “Expérimenter l’Avenir”by A*Midex