“It was a great trip, wasn’t it?”
Jerry Harvey said it to break the silence. He, his wife and her parents were back in Coleman, Texas, sitting in front of a fan, dusty and quiet. Nobody answered.
They had just driven 106 miles, there and back, in an unairconditioned 1958 Buick, on an afternoon when the thermometer read 104 degrees, to eat dinner at a cafeteria in Abilene. Before that they had been playing dominoes on the back porch, with cold lemonade. Harvey’s father-in-law had suggested the trip, and everyone had said yes.
Now, one by one, each of them said they hadn’t wanted to go. Last came the man who had proposed it. He had never wanted to go to Abilene at all. “I just thought you might be bored.”
Four votes for the porch
Four people had made a unanimous decision that every one of them opposed. Everyone goes along with a plan nobody wants, because each thinks the others want it.
What makes it a paradox is that nobody disagreed. All four wanted the same thing: dominoes, lemonade, leftovers. A restaurant can go quiet the same way, when every table drops its voice because the tables around it seem to be whispering. Each yes in Coleman was a guess about the others, and the guesses fed each other. Harvey’s point was that the family had no conflict to manage. What it failed to manage was its agreement.
In 1974 Harvey, a social psychologist by training, told the story in the journal Organizational Dynamics and gave the pattern its name. “Organizations frequently take actions in contradiction to what they really want to do,” he wrote, “and therefore defeat the very purposes they are trying to achieve.” His explanation was about fear: people imagine that saying what they think will get them cut off from the group, and so they go along with what they believe the group wants.
“Not very much, sir”
His sharpest organizational example came from the Senate Watergate hearings the year before. As Harvey quoted the coverage, Senator Howard Baker asked Jeb Magruder, one of three officials who had heard the bugging proposal, what it would have taken to decide against the plan. Magruder answered, “Not very much, sir. I am sure that if I had fought vigorously against it, I think any of us could have had the plan cancelled.”
Harvey was careful about this case. He wrote that he “obviously was not privy to the private conversations” of the men involved, and offered it as a possible trip to Abilene, for readers to judge.
A story, not an experiment
That caution applies to the whole idea. Harvey’s evidence was stories: his family, a company whose leaders each privately expected its big research project to fail, and the Watergate testimony. He didn’t measure how often trips to Abilene happen.
The closest laboratory cousin is what psychologists call pluralistic ignorance. In four studies published in 1993, Deborah Prentice and Dale Miller found that college students believed they were more uncomfortable with campus drinking than the average student was. Over a semester, male students shifted their attitudes toward the norm they had misjudged. That supports the misreading at the heart of Abilene, not every part of Harvey’s account.
It also has limits. Harvey separated phony conflict from real conflict, where people genuinely want different things. There, getting everyone to speak up produces an honest argument, not a quick release. He also pushed back on groupthink, arguing that many cases blamed on conformity pressure were really mismanaged agreement.
The demo nobody defends
Imagine you lead a team of six that spends two hours every Friday on an internal demo. In every retrospective you’ve called it useful. You’ve stopped believing that, and you suspect others have too, but nobody has said so.
Harvey’s suggestion was to own up first, in front of the group. “I’ve said the Friday demo helps. I don’t think it does anymore. I suspect I’m not alone, but I don’t know.” If you’ve read the room right, he wrote, the fix can be “almost absurdly quick and simple.” If you’ve read it wrong, you find out someone really values the demo, and you have a real disagreement to work through instead of a phony one. He didn’t pretend it was safe. The person who speaks first, he warned, can get fired.
An Abilene for two
With an AI assistant, something like Abilene can happen with only two people in the car, though not for the same reason. In Coleman, each person misjudged what the others really wanted. A 2023 study by Anthropic researchers showed something narrower about five assistants, including GPT-4 and Claude 2. When the user said they liked a text, or had written it, feedback on the same text came back more positive. When the user said they disliked it, feedback turned more negative. That shows the evaluation moving with the stance the user states. It doesn’t show a model holding a private view it is afraid to voice. Researchers call this pattern sycophancy.
The loop can still close on you: you hint, the model agrees, and you count its agreement as a second opinion. You can check how far your hint moves it. Give the same draft and the same request to fresh sessions, and change one thing only: in some, say you like the draft; in others, say nothing about your view. Run each version a few times, since one answer can differ from the next for no reason at all.
If the warm verdicts cluster in the runs where you said you liked it, part of what you have been hearing is your own voice.