Grief moves through a room differently than it moves through a single person. So does joy. So does the particular kind of solidarity that forms when a community processes something hard together — through rituals, through conversation, through the slow friction of shared attention over time. Psychology has a name for this: collective emotion regulation. The idea that groups don't just contain feelings; they process them together, in ways no individual can replicate alone.
A recent piece in Psychology Today by Cornelia Walther, Ph.D., a Wharton fellow researching hybrid intelligence, argues that something is quietly happening to that process — and that AI is at the center of it.
How Groups Actually Hold Emotion
Research on collective emotion regulation describes a sophisticated, often invisible choreography. Groups choose which situations to enter, how to interpret them, and how to respond — deploying strategies no individual could manage alone. The leader who reframes a crisis. The community that holds a memorial. The friend group that collectively decides, without ever saying so out loud, that something is going to be okay.
Much of this happens without deliberate planning. Nobody draws the blueprint; the regulation emerges from proximity and shared attention. What makes it work is a sense that the feeling belongs to everyone in the room at once. Groups grieve together, and recover together, only when that quality holds.
An Averaged Vocabulary for Every Occasion
Now bring AI into that picture. Every AI system trained on human-generated language carries an implicit emotional register — a kind of averaged-out vocabulary assembled from what was expressed, rewarded, and recirculated at scale across the entire internet. That register skews in predictable ways: it amplifies urgency, certainty, reassurance, and enthusiasm while compressing ambivalence, sadness, and the slower recognitions that resist easy naming.
A review of 40 peer-reviewed studies on AI and emotional well-being found that AI-mediated communication is actively transforming collective norms around what counts as genuine interpersonal engagement. Research on AI writing assistance found that suggestions homogenize language toward Western styles and diminish cultural nuance. The words that only mean something inside a specific shared history get replaced by language that travels everywhere and lands nowhere in particular.
The Condolence Note Nobody Actually Wrote
Ritual is where this becomes most legible. Rituals work precisely because of their specificity — the particular song, the phrase that only means something if you were there, the gesture that carries accumulated meaning for this group and no other. That specificity is not decoration. It is the mechanism.
AI-assisted rituals tend toward generality. The birthday message drafted with a language model. The condolence note shaped by autocomplete. The commemorative post optimized for engagement. Each one softens into something that covers all the bases and lands at none of them. The group still performs the ritual. The emotional anchoring — the part that makes it actually hold — gets progressively diluted.
Researchers have coined the term "pseudo-intimacy" to describe what fills that space: the simulation of emotional connection without the conditions that make connection real. The knowledge that another person — carrying their own history of loss or fear — chose to reach out is itself a regulatory resource. AI-generated messages lack it entirely.
When Organic Becomes Algorithmic
The subtler concern is what happens to the unplanned parts of collective emotion regulation — the organic conversations, the person who naturally holds feeling well, the shared reinterpretation that emerges from genuine friction. These emergent processes are now happening inside recommendation architectures and AI-generated summaries built by systems with no stake in the group's emotional coherence. The conversations still occur. Their conditions are being quietly shaped by something that cannot care about the outcome.
Paying Attention at the Right Moments
The practical implication isn't to avoid AI wholesale — it's to recognize where the stakes are highest. The condolence message, the team acknowledgment after a difficult week, the anniversary ritual: these are the moments where AI assistance hollows out what it appears to help. The few extra minutes of unmediated effort are doing more emotional work than they appear to be.
Walther's research also suggests a practice worth building: noticing what your community's emotional vocabulary actually is. Every group has a specific language for holding difficulty — particular phrases, rhythms of acknowledgment, references that only mean something inside a shared history. AI-assisted communication erodes that specificity gradually. And gradually is how most losses happen.
