SPHERE: A Franco-Canadian Collaboration at the Intersection of Social Neuroscience and AI

The SENSUM community is excited to spotlight a new international research initiative that brings together two leading voices in social cognition and artificial intelligence: Justine Cassell (Inria Paris) and Guillaume Dumas (Université de Montréal, Mila, and CHU Sainte-Justine). Their joint project, SPHERE — Social Physiology and Human-like Embodied Response Engineering — marks a significant step forward in understanding and engineering human-like interaction in artificial agents.

Launched under the Inria Associate Team program (2025–2027), SPHERE aims to explore how social, embodied communication unfolds in real time — and how this understanding can inform the development of more adaptive, responsive conversational agents.

Two Leaders, One Shared Vision

Justine Cassell is a senior researcher at Inria Paris and a pioneer in the field of Embodied Conversational Agents (ECAs) — AI systems capable of combining speech with nonverbal behaviors such as gestures, gaze, and posture. Her work has been instrumental in understanding how machines can participate in natural human dialogue, particularly in educational and therapeutic contexts.

Guillaume Dumas, junior professor at the Université de Montréal and core researcher at Mila – Quebec AI Institute and CHU Sainte-Justine, is known for his contributions to social neuroscience, including the study of inter-brain synchrony — how the neural activity of two individuals aligns during real-time interaction. His lab develops computational and machine learning models that capture the dynamic, mutual nature of human communication.

Together, Cassell and Dumas are combining their expertise in computational social interaction and neurophysiological modeling to investigate how human brains and bodies coordinate during conversation — and how that synchrony can inform the next generation of interactive AI systems.

Scientific Ambition: From Neural Synchrony to Conversational Agents

SPHERE will focus on two main research axes:

1. Neuroscience Hyperscanning: Using EEG and fNIRS, the teams will study inter-brain synchrony during structured conversations, capturing both neural and behavioral data (e.g., gaze, gestures, prosody). This will provide insights into how alignment in brain activity relates to successful communication.

2. Multimodal Conversational AI: The project will also develop new architectures for conversational agents that can process and respond to multimodal inputs — from speech to neural signals — in real time. These systems will be tested in education and neurodiversity-related settings to evaluate their impact on learning and communication.

Throughout the project, researchers from Paris and Montreal will engage in reciprocal visits, workshops, and bi-weekly meetings to ensure cross-training and shared development. The collaboration will also train doctoral students in techniques at the frontier of social neuroscience and AI.

Shared Values: Ethics, Inclusivity, and Bilingualism

In keeping with both labs’ commitments to responsible research, SPHERE will uphold the highest standards in ethical data collection and algorithm design. Special care will be taken in working with children and neurodivergent individuals, and all agents developed will avoid reinforcing stereotypes or misleading users about their artificial nature.

The team will also explore bilingual conversational agents — capable of fluid code-switching between French and English — to reflect the linguistic and cultural diversity of Canada and France, and to enhance inclusivity in educational and therapeutic applications.

Toward Socially Intelligent Machines

SPHERE represents more than a scientific endeavor — it is a vision for how neuroscience and AI can come together to build more human-centered technologies. By grounding artificial agents in the dynamics of real-world human interaction, Cassell and Dumas aim to push the boundaries of what AI can achieve in terms of collaboration, empathy, and learning.

We look forward to following the outcomes of this ambitious project — and to seeing how it enriches both the SENSUM network and the broader fields of social neuroscience, AI, and human-computer interaction.

This content has been updated on 4 April 2025 at 12 h 46 min.