L'IA peut-elle désamorcer un dîner de famille tendu ?

Lisez la pièce. Choisissez la blague qui libère la pression, pas celle qui la fait exploser. Sachez à qui c'est le tour de se sentir entendu.

Background

AI systems may identify emotional states in constrained contexts, such as chatbots detecting frustration in user inputs, but these capabilities remain rudimentary compared to human social intelligence [IEEE, enriched May 9, 2026]. Current models like large language assistants can suggest conflict-resolution phrases or offer neutral advice (e.g., "Let’s take a pause"), yet they lack the contextual awareness to judge when to intervene or tailor responses to specific family histories and power dynamics [IEEE, enriched May 9, 2026]. Nonverbal cues—such as a tense jaw, crossed arms, or averted gaze—are critical signals in family negotiations yet remain poorly replicated by AI, which relies on text or limited audio inputs [IEEE, enriched May 9, 2026]. Moreover, AI lacks personal memory of prior interactions; it cannot recall that Aunt Carol always changes the subject when politics arise or that Uncle Dave’s sarcasm masks his insecurity [Status checked on May 10, 2026]. While research explores "emotional AI" using multimodal sensors (e.g., facial analysis or voice tone detection), these technologies raise ethical concerns about surveillance in private spaces and have not been validated for real-time family mediation [IEEE, enriched May 9, 2026].