In 2014, two psychologists at the University of Chicago paid commuters to talk to strangers on the train. The commuters predicted they would hate it. They were wrong in a specific, measurable, and replicable way. Twelve years later that result is still the cleanest example of a thing the entire internet should know: humans are bad at predicting how good a conversation with a stranger will be, and we're bad in a direction that costs us happiness.
The Chicago train study
Nicholas Epley and Juliana Schroeder, then both at Chicago Booth, had a simple question. People on commuter trains have an obvious choice every morning: talk to the stranger sitting next to them, or sit in silence. They almost universally pick silence. Why?
Their hypothesis was that commuters expect the conversation to be unpleasant or to leave them feeling worse. So they ran the experiment. They paid commuters to either (a) talk to a stranger, (b) sit in solitude, or (c) do whatever they normally do. Then they measured mood after.
The result: commuters who talked to strangers reported a more positive commute than those who didn't. Significantly. Predictably. Across multiple replications. The commuters who predicted they'd hate it were wrong. Even the introverts. Especially the introverts.
Why we predict it'll go badly
Epley's follow-up papers explain the mechanism. We have what he calls a “pluralistic ignorance” problem: each of us silently assumes the stranger doesn't want to talk, even though the stranger is making the same assumption about us. Both parties stay silent because each one is, in good faith, being “considerate.” Both end up worse off.
Underneath that, there's a more primal mistake. We expect strangers to be more threatening than they are. We expect conversations to be more awkward than they are. We expect to come out feeling drained when we usually come out feeling lifted. The prediction error is robust across age, gender, country, and trait introversion. It is a genuine human bug.
The depth surprise
A 2022 paper from Kardas, Kumar, and Epley took the question further. What if you ask people to skip small talk and go straight to deep questions? Surely thatis where the badness lives. They had pairs of strangers swap one of two question sets:
- Shallow: “What's your favorite TV show? Tell me about the last time you went to a beach.”
- Deep: “Can you describe a time you cried in front of another person? When did you last sing to yourself?”
The participants predicted the deep conversation would be more awkward, more uncomfortable, less enjoyable. The deep conversation was rated more enjoyable and produced more connection. The shallow one was rated, predictably, mid.
We have a working theory of what makes conversations “safe” that actively prevents us from having the conversations we'd like best.
Why this matters more on the internet
The internet was, briefly, a stranger-conversation utopia. Early IRC, early Twitter, early Reddit. Strangers found each other through interest. The medium did the introducing. Then platforms optimized for retention by surfacing people you already knew, and the stranger-conversation utopia got buried under follower graphs.
The cost is real. Loneliness research from the post-2020 era keeps finding the same thing: the medium of communication matters less than the variety of conversations. A weekly chat with a stranger about something you both care about does more for well-being than a hundred messages with people you already know. The stranger conversation forces you to articulate yourself fresh. You can't coast on shared history. You have to actually say what you mean.
Video matters
There's a separate strand of research on the modality of stranger contact. Text-only stranger interactions help less than voice-only, which helps less than video. The mechanism is the “they're a real person” effect: text leaves your brain free to fill in the worst version of who they are; video replaces that with the actual version. Almost always the actual version is better than the imagined one. The same prediction error from the train shows up here too.
What to do about it
Three concrete moves, in order of effort:
- Stop predicting how the next stranger conversation will go. Your prediction is wrong by a known direction. Just have it.
- Start with a deep question. Counterintuitively, this works better than starting shallow. Ask “what have you changed your mind on lately?” before asking what they do. The depth is comforting.
- Use video when you can. Voice second. Text third. Modality is doing more work than you think.
The PodRandom angle
We built PodRandom because we kept reading these studies and reading our own group chats and noticing the disconnect. The internet had quietly become hostile to the single most reliable mood-improver in the social sciences literature. We wanted a product where the friction of finding a stranger was zero, where the medium was video, where the conversation was framed around something you both cared about, and where you could keep the recording if you wanted.
It's a pretty literal application of the research. If you've never tried the “talk to a stranger” thing online with video and a topic prompt ready, the prediction error described above is going to apply to you, too. The conversation will be better than you expect. Even if it isn't, the next one probably will be. That's the whole game.