People have long sought solace in support groups, finding comfort in shared experiences during challenging times. Despite this, research on the nonverbal behaviors that signify connection between participants, especially in virtual settings, remains limited. A recent study conducted by researchers from the University of Kansas and the University of Southern California sheds light on the behavioral markers of alliance in these groups, highlighting the potential for machine learning to enhance mental health support.
The research, published in the Proceedings of the 27th International Conference on Multimodal Interaction, analyzed data from 18 support groups comprising 96 participants. By evaluating the concept of dyadic alliance, which refers to the connection between two individuals, the study surveyed participants on their feelings of connection post-session. Using computational algorithms, researchers examined verbal and nonverbal communication—spanning language, audio, and visual elements—to explore the feasibility of machine learning in assessing these connections.
Demand for mental health services has surged in recent years, particularly in the wake of the COVID-19 pandemic. Yunwen Wang, assistant professor of journalism and mass communications at the University of Kansas and a lead author of the study, noted, “This project emerged from our focus on burnout among mental health professionals, especially given the increasing need for support during and after the pandemic.” Wang emphasized the importance of leveraging artificial intelligence ethically, suggesting that it could enhance access to mental health resources without replacing human therapists.
Participants in the study were engaged in online video conferencing sessions focused on general anxiety. Their mental health and emotional states were assessed before and after the sessions, during which they reported their perceived connection with one another. A virtual conversational agent, designed to facilitate these discussions, was operated by a human who could intervene if necessary.
The researchers recorded and analyzed participants’ verbal communications and nonverbal gestures—such as head nodding, smiling, and eyebrow movements—using advanced computational tools. This allowed for a detailed examination of communication dynamics, including the frequency of gestures and variations in pitch.
Strong predictors of dyadic alliance were identified, revealing that listeners who displayed more frequent head nods and fewer frowns fostered a greater sense of connection. Similarly, speakers who exhibited pitch variation and expressive facial gestures also reported stronger feelings of alliance. The findings underscore the significance of both verbal and nonverbal communication in measuring participant connections within group settings.
While the study suggests a promising role for machine learning in identifying behavioral markers of dyadic alliance, the researchers caution against unregulated use of AI in mental health contexts. Wang clarified, “The goal was not to replace human facilitators, but to compare how machine learning can estimate human perceptions of relationships within groups.”
The ongoing research also delves into ethical considerations surrounding AI in mental health care. Questions about privacy, the appropriateness of AI therapists, and the potential for AI technologies to be responsibly integrated into treatment are central to their work. “As AI chatbots gain popularity, we are engaging in critical discussions about their use in therapy,” Wang stated.
The research team continues to explore user trust in AI agents and how varying levels of AI involvement impact group dynamics. Ultimately, Wang believes that the insights gained from this study can lead to improved mental health services, particularly in understanding how genuine connections are fostered within support groups.
“The aim is to assess whether AI-assisted systems can be accepted by users, especially given the growing demand for mental health support,” Wang explained. She highlighted that the human-to-human dynamic remains essential, with AI serving primarily as a facilitator for meaningful conversations.
As this research progresses, it may pave the way for innovative approaches to mental health support, ensuring that technology complements rather than replaces the vital human connections that underpin effective therapy.
