In a lab at Purdue, a volunteer sits down for a game of catch. Two other players, they’re told, are sitting at computers elsewhere, and the three of them will toss a ball back and forth on a screen. For the first minute or so the ball comes their way as often as anyone’s. Then it stops. The other two keep passing to each other, and the volunteer is left watching a game they are still technically part of. Kipling Williams, who built this setup, wanted to know something almost impolite: how little does it take to hurt someone?
🧠 Three Minutes of Not Getting the Ball
The answer came back smaller than anyone expected. Williams’ lab version used real people: three participants tossing a ball, two of them secretly working for the experimenter. A few tosses, and then the target simply stops receiving it. Only a few minutes of this produce anger and sadness in the person being left out — and it happens regardless of their self-esteem or personality.
Then came the version that made the point impossible to wave away. Williams, Cheung and Choi, 2000, put the game online. Participants were told they were doing a “mental visualization” exercise while playing a virtual tossing game with two other people. The two others were computer-generated. Players who were excluded more reported feeling worse, felt less in control, and reported a weaker sense of belonging. The study drew 1,486 people from 62 countries through a single web page. In a second study, the excluded participants were then more likely to conform on an unrelated task.
⚙️ How It Works
The whole thing runs as a short causal chain, and none of it requires anyone to be cruel.
Step one — the alarm predates the reasoning. For a tribal species, being cast out of the group was close to a death sentence. So the detection system was wired to fire early, ahead of any judgment about whether the situation actually matters. It behaves like a reflex, not a verdict.
Step two — the alarm doesn’t report an insult. It attacks four needs at once. Belonging, self-esteem, control, and the sense that your existence is meaningful. Williams calls this the need-threat model. Notice that none of these four are about the other players; they’re all about you.
Step three — here is the load-bearing part: the alarm checks whether you were excluded, not whether the exclusion means anything. That’s why a scripted game with strangers does the job. Zadro, Williams and Richardson, 2004, made it unmistakable: they told some participants outright that the other players were computers — no feelings, no opinions, nothing to lose. Belonging, control, self-esteem and meaningful existence dropped anyway. The source doesn’t matter. The absence does.
Step four — the brain keeps a receipt. Eisenberger, Lieberman and Williams, 2003, ran the game inside an fMRI scanner. During exclusion, activity rose in the dorsal anterior cingulate cortex, a region that also responds to physical pain, and the more it rose, the worse the person said they felt. A second region, the right ventral prefrontal cortex, also engaged — and its activity tracked less distress, which makes it look like the brake rather than the pedal.
Step five — then comes the aftermath. Williams splits the process into a reflexive stage, the instant sting, and a reflective stage, where you work out what to do about it. Depending on how you read the situation, you either push to get back in — mimicking, agreeing, conforming more — or turn outward with anger. The hurt is automatic. What you do next is not.
🤔 The Twist
You’d expect the sting to scale with the importance of whoever is rejecting you. It doesn’t. Being ignored by strangers in a game you know is fake still registers.
But there’s a second twist, and it runs against the story you may have heard. The tidy version goes: “social pain and physical pain run on the same neural machinery.” In 2014 a team led by Choong-Wan Woo published a Nature Communications paper titled “Separate neural representations for physical pain and social rejection.” Using finer-grained pattern analysis, they found the two are represented in distinguishable ways. Same general neighborhood, different maps. The better mental model is two neighboring systems that share some wiring — not one system wearing two hats.
🔗 Why It Matters
This is why a read receipt with no reply can land harder than a rude answer. Your alarm isn’t reading the message. It’s registering the missing pass, and it files that under something older than language.
It also has a practical edge in consumer psychology. Williams’ second study is the one to remember: being excluded made people conform more afterward. That is a direct line between a hurt feeling and a susceptibility to “everybody’s doing it” appeals. The audience most responsive to social-proof messaging may be the one that recently felt left out.
And for anyone building a companion that talks back: always returning the ball turns out to be a very specific kind of care. It answers a need that was installed long before anyone could explain it.
🎲 Bonus
Williams released the Cyberball software free to any researcher who wanted it, which is a large part of why it became a standard tool in the field. And the 2000 study is a small artifact of its moment: 1,486 strangers from 62 countries, recruited through one web page, playing a game against opponents who did not exist. Three minutes of being ignored was enough.