Silent s and the embodied side of collective intelligence. #
Posted July 24, 2026 [ Reviewed by Monica Vilhauer Ph.D.
](/us/docs/editorial-process)
Key points
- Social sensitivity, not just individual ability, is a key predictor of collective intelligence.
- AI-driven analytics can now code non-verbal behaviour like silence at millisecond-level precision.
- The challenge ahead is not teaching AI to notice behaviour, but to interpret it in context.
*Co-authored by Ceolla Dillon O’Rourke, Aoife Ní Chíobháin, and Michael Hogan. *
In education, collaborative learning is valued as a way to build the key collective intelligence skills that societies depend on: coordinating, communicating, sharing knowledge, perspective taking, and solving problems together. However, collaborative learning is not just about putting students in a group and asking them to complete a task. A group also has to manage the process of thinking together: taking turns, noticing confusion, regulating frustration, repairing misunderstandings, and deciding when to speak and when to wait. These social and embodied processes matter because collective intelligence (CI) depends not only on individual ability, but also on how well group members notice and respond to one another. Woolley et al. (2010) found that groups have higher CI when members show greater social sensitivity and share conversational turns more evenly. Successful collaboration depends partly on reading the room.
Researchers are increasingly using AI and multimodal learning analytics to make these subtle processes visible (Järvelä et al., 2020). Systems can detect speech patterns, s, facial expressions, and possible moments of difficulty. But noticing that something has happened is not the same as understanding what it means. A quiet moment might reflect confusion, frustration, individual concentration, or shared reflection. If social sensitivity matters for collective intelligence, can AI recognise the context behind these subtle signals, or can it only identify the pattern?
When the room goes quiet #
Dang et al. (2024) explored this question by focusing on something that is easy to overlook in groupwork: the moments when nobody speaks. Their study involved 30 Finnish secondary students working in ten triads on a face-to-face design task. Each student had a laptop, and together the group developed a food product on a shared Google document, working out ingredients and nutritional targets such as calories, protein, and fat. Halfway through, the researchers introduced a cognitive challenge: a customer’s voice message about a product allergy, which forced the groups to revisit their design. Three further messages followed at three-minute intervals, escalating from impatience to urgency to annoyance. The groups had to continue with their design work while also managing increasing emotional pressure.
Video and audio captured the group interactions in detail. AI-enabled tools then automatically segmented and transcribed the students’ spoken turns, which the researchers checked and refined. This made it possible to add silent s at millisecond-level timestamps and analyse them alongside the speech. Each spoken turn was also coded according to what the group was doing, such as defining the problem, educating one another, monitoring progress, agreeing on a next step, or regulating emotion. The researchers then used lag sequential analysis, a method for testing which coded events followed one another more often than chance, to examine how silence fitted into the discussion.
Not all silence is the same #
The analysis identified three deliberation tactics. In elaborated deliberation, groups engaged in extended back-and-forth reasoning. Silent s made up about 18% of interactions and tended to be brief, often appearing while students were educating one another. The authors suggest that these s may reflect introspection or cognitive effort, although the data cannot show exactly what students were thinking. In coordinated deliberation, silence was more frequent and often linked different stages of the task, such as generating an idea, checking progress, and agreeing what to do next. Here, silence appeared to function as a kind of junction between forms of group activity.
In solitary deliberation, silence became the dominant feature, accounting for about 36% of interactions. Groups showed fewer back-and-forth exchanges, and s were more frequent and longer than in the other tactics. The authors suggest that this could reflect individual reflection or internal cognitive processing, but it could also indicate a more fragmented or reserved group dynamic. The cluster also contained two moments coded as negative socioemotional interaction, moments of discouragement or low motivation rather than direct task discussion, and each was followed by silence. Whether that silence reflected emotional withdrawal, private reflection, or an effort to regain control of the task remains open to interpretation.
AI-driven analysis is what makes these findings promising: hand-coding silence across thousands of turns would be almost impossible, yet AI-driven analytics annotated these s at millisecond precision and aligned them with the talk — a level of fine-grained tracing that opens up questions researchers could not previously ask. But the findings are also preliminary. The study involved only 30 students, ten groups, and one task completed in a controlled laboratory setting. The emotional pressure was also deliberately introduced through scripted customer messages, so the students’ responses may not reflect tensions that develop naturally in a classroom or other real-world collective intelligence design scenarios. More importantly, the analysis identified patterns around silence without directly revealing students’ thoughts or feelings. AI made the s visible, but human interpretation was still needed to decide what those s might mean.
When should a teacher step in? #
In education, understanding a silence is only half the problem. The harder question is whether a teacher should step in, and when. Cohn et al. (2025) asked an experienced teacher to interpret a multimodal dashboard of two pairs of high-school students completing a collaborative programming task. The dashboard combined video, speech, screen activity, emotional cues, and participation patterns, so s could be read in context. The researchers distinguished a difficulty threshold, when students first appeared to struggle, from an intervention point, when support might actually help. In one scenario, s in discussion accompanied conceptual misunderstandings and ineffective debugging. In another, the students were still testing ideas and working meaningfully towards a solution. The teacher intervened in one case and waited in the other. The same outward signal—a —can call for different responses, which is why supporting collaborative learning takes such subtle, contextual understanding and skill.
Can AI know when to wait? #
Lag sequential analysis can show what tends to happen before and after a silent , but it is still looking back across recorded patterns. It cannot yet determine, in real time, what silence means or what should happen next. An AI system might detect repeated silences and interpret them as a sign that a group is stuck, but it could just as easily interrupt students during productive collective reflection. Making worthwhile predictions would require the system to consider much more than the itself: who had been speaking, what the group was doing beforehand, how long the task had been running, whether the emotional tone had changed, and whether progress was still being made. The difficult part is not teaching AI to notice a . It is preventing the system from treating every as the same kind of problem.
Artificial IntelligenceEssential Reads In the short term, dashboards like the one explored by Cohn et al. could help teachers notice groups that may need attention while leaving interpretation and intervention with the teacher. The longer-term challenge is whether AI can develop the social sensitivity that collective intelligence depends on: recognising both when a carefully chosen question or alternative perspective might move a discussion forward and when a group needs space to think for itself. The most socially sensitive AI may not be the one that always knows what to say, but the one that also knows when to stay quiet.