You are blindfolded, holding a rope. When the signal comes, you pull as hard as you can, and a gauge records the force. On some turns you are told you are pulling alone. On others you are told that other people are on the rope with you, pulling too.
The researchers, Ingham and his colleagues, were chasing a puzzle more than sixty years old. Ringelmann’s measurements had shown that groups pull less than the sum of their members. In the figures most often repeated, eight workers together managed less than four times what one pulled alone.
The obvious explanation is timing. Eight people never heave at exactly the same instant. If that is the whole story, a person pulling alone has nothing to lose.
Nobody else was really pulling. The “others” were confederates who only pretended to. Every bit of force on the gauge came from you.
And still, when you believed one other person was on the rope, you pulled at about 90% of your solo effort. With two to six others, about 85%.
Add more hands to the rope and each one pulls a little less.
It was enough to believe the other hands were there.
Two ways to lose a pull
There are two separate leaks in a group, and the blindfold pulled them apart.
The first is coordination loss. People push at slightly different moments, get in each other’s way, do the same piece twice. This was Ringelmann’s own explanation. He was a French agricultural engineer who measured workers pushing and pulling loads in the 1880s and reported it in 1913, and he put the group shortfall down mainly to poor coordination, though he allowed that motivation might play a part. The social psychologist Ivan Steiner later called the gap between what a group could produce and what it does process loss, and found that in Ringelmann’s data it grew in step with the number of coordination links between the pullers.
The second is motivation loss: each person simply tries a little less. That is what the blindfold isolated. With no one else actually pulling, there was nothing to fall out of time with, and the effort still dropped.
In 1979 Bibb Latané, Kipling Williams and Stephen Harkins at Ohio State repeated the trick with clapping and shouting, using blindfolds and headphones so nobody could hear anyone else. For shouting, they estimated that about half of the shortfall came from poor coordination and half from people easing off. They named the second half social loafing. Latané had studied bystanders before this, and saw the same arithmetic in both: the more people share a job, the smaller the slice any one of them feels.
One explanation they offered is that in a group, nobody’s output can be told apart from anyone else’s, so nobody gets precise credit or blame. A reference summary of the research since then says people ease off less when their own effort can be identified, and when the task matters to them personally.
The two leaks need different fixes. Better timing does nothing for someone who has quietly decided the others will carry it.
Eight readers, one contract
Say eight people are asked to review a forty-page supplier contract by Friday. Everyone gets the same link and the same instruction: read it and flag anything wrong.
One of them knows exactly how this works. Seven colleagues have the file too, and two of them are lawyers. He reads the first ten pages properly, skims the rest, and closes the tab before lunch. With eight sets of eyes on it, he figures, anything serious in the back half will get caught.
Nothing about the task was unclear to him. He knew it was shared, and that knowledge is what made him read less.
If a few of the others reason the same way, the contract ends up with plenty of comments on page three and few on page thirty-one. Nobody shirked in a way anyone could point to.
Give each reviewer five named pages and every share becomes visible. In the research, that is one of the conditions under which people ease off less.
Agents on the rope
Anthropic’s 2025 write-up of its multi-agent research system is a useful test of which leak survives when the pullers are software. It doesn’t describe agents easing off. The failure it does describe is the first kind.
When the lead agent handed out short instructions like “research the semiconductor shortage,” subagents sometimes misread the task or ran the exact same searches as one another. In one case, one subagent looked into the 2021 automotive chip crisis while two others both investigated 2025 supply chains.
Anthropic’s fix was a better brief for each subagent, including clear task boundaries, so that no two of them pulled on the same stretch of rope.
Two subagents on the same 2025 supply chains, doing the same searches, and the work paid for twice.