Garold Stasser and William Titus ran an experiment that should be better known than it is. They published it in the Journal of Personality and Social Psychology in 1985.
Groups were asked to choose the best candidate for a role. In one condition, every member received the full set of facts about every candidate. In the other, the facts were split up, so that each member held some information nobody else had. Collectively the group still held everything it needed. The correct answer was reachable either way.
When everyone had the same information, 15 of 18 groups chose the best candidate. That is 83%.
When the information was distributed, 7 of 38 groups chose the best candidate. That is 18%.
The same task. The same total information. The same kind of people. The only difference was whether the facts arrived in one head each or in every head at once.
Stasser and Titus wrote that discussion tended to perpetuate, rather than to correct, the distorted picture each member had formed. That sentence is forty years old and it describes most meetings you have been in.
The finding held up, which is the part that matters
A single experiment is a curiosity. A replicated one is a fact about how groups work.
Lu Lu, Y. Connie Yuan and Poppy McLeod published a meta-analysis in Personality and Social Psychology Review in 2012, covering 65 studies, 101 independent effects and 3,189 groups. Two results stand out.
Groups discussed information that everyone already held at roughly twice the rate of information that only one person held.
And groups facing a hidden profile, where the answer required pooling the unique pieces, were eight times less likely to reach the correct solution than groups given the full information from the start.
Eight times. Not a bias to be aware of. A structural collapse in the group's ability to use what it actually contains.
Discussion tended to perpetuate, not to correct, the distorted picture each member arrived with.
Why unshared information loses, every time
The mechanism is not mysterious and it is not anybody's fault.
Shared information has an enormous advantage in a discussion. More people can raise it. When someone raises it, more people can confirm it, and confirmation feels like validation. It sounds like the group is converging on the truth, when in fact the group is converging on the intersection of what it already had.
Unique information has the opposite profile. Only one person can raise it. Nobody can immediately confirm it. It often arrives without the context that makes it make sense, because the context is also unique to that person. It carries social cost, because saying something nobody else can corroborate is exposing. And it competes for airtime with material that is easier to agree with.
So the room does not reject the unique contribution. It just does not pick it up. Ten seconds later the conversation has moved, and the person who spoke concludes it was not that important.
Three other well-established findings sit on top of this and make it worse. Irving Janis named groupthink in 1972, describing how the drive for concurrence suppresses dissent in cohesive groups under pressure. The groupthink manifests itself in the age of artificial intelligence as a pressure on dissent: if an AI model or autonomous agent makes a certain recommendation, it is difficult for an individual employee to oppose it. As Stasser and Titus showed, the group tends to reject the unique information possessed by an individual member. The employee conforms to the AI's "opinion" in order to maintain consensus.
Barry Staw's 1976 study in Organizational Behavior and Human Performance showed how commitment to a chosen course escalates precisely when the evidence turns against it. And Paul Nutt's 1999 study in Academy of Management Executive, tracking 356 decisions in medium and large organisations over two years, found that half failed in the sense that they were never actually used. Participation was used in 18%. Imposition was the tactic in around 63% of those cases. Nutt identified three key "management blunders" that lead companies to disaster: 1) "Idea Imposition" (rushing to judgement). 2) Limiting the search for remedies: leaders often make hasty decisions to avoid uncertainty, which leads to a radical narrowing of options and exploration. They often simply try to copy the "best practices" of others, believing that good decision-making processes take up too much time and money. In reality, correcting mistakes afterwards costs many times more. 3) Misusing power: leaders try to enforce their decisions alone.
Forty years of good advice has not fixed it
The remedies are familiar because they have been recommended continuously since the 1980s. Assign a devil's advocate. Ask the quiet people directly. Have the leader speak last. Build psychological safety. Go round the table.
They are all sensible. Some of them measurably help in a well-run session. And none of them has solved the problem, for a reason worth stating plainly.
Every one of those fixes is behavioural, which means it depends on a specific person doing a specific thing well, in real time, under time pressure, in front of their manager. It works when the facilitator is skilled and the meeting is calm and the group is small enough to go round.
None of it survives contact with an organisation at scale. Not across four hundred people. Not across time zones. Not across a decision that takes six weeks and involves nine meetings with different attendees. Not when the person holding the unique information is two levels below the person who would need to act on it, and was never in the room in the first place.
The advice asks people to compensate, by force of discipline, for something the container cannot hold. That was a reasonable ask when the container was a meeting and there was no alternative.
If there is a live question in your organisation where you suspect the answer is already inside the building, talk to us.
What a container would have to do instead
Take the mechanism seriously and the requirements fall out of it directly.
Unique contributions have to be visibly unique. In a conversation, a rare piece of knowledge and a widely held one sound exactly the same. They should not look the same. So let every contribution carry its ground: research, expert fact, lived experience, values or gut feel. Then a position held by one person on the basis of ten years in the field is legible as exactly that, rather than as one more comment.
Airtime has to stop being the constraint. The reason unique information loses is partly that speaking is serial and scarce. Contribution that is not competing for the floor removes most of the cost of saying the unusual thing.
Contributions have to be able to act on each other. The value is not in collecting the rare piece. It is in someone else building on it, or challenging it, and that relationship being preserved rather than flattened into a summary.
Thinness has to be visible. The most useful thing a group can know is where its picture rests on one person's word, or on nobody's. That is invisible in a discussion and it does not have to be.
Tension has to survive. Two well-grounded positions that conflict are the most informative object in the process. Resolving that early to look decisive throws away the only signal that the question is not yet settled.
And it has to accumulate. If understanding built in one round has to be rebuilt from scratch in the next, the group is paying the pooling cost over and over and losing more of it each time.
The problem was never the people
This is what Hunome is built to run. Each Spark carries characterisations, including its knowtype, the contributor's own account of how they know what they are contributing. A SparkMap holds the perspectives, their grounds and the connections between them, so the unusual position stays as analytically available as the common one. The Lens is how a SparkMap’s insights page reads what is there: where meaning is clustering, which chains of building have produced something new, where different ways of knowing are showing up across the map. What decision makers receive is deliberative intelligence: the full picture of what the group now understands, including where it is thin and where it is genuinely divided.
There is an old and slightly comfortable story about why organisations decide badly. It says the people were not smart enough, or not brave enough, or not honest enough with each other.
Stasser and Titus tested that story in 1985 and it did not survive. The same people, given the same total information in a different arrangement, went from 83% right to 18% right. Nothing about their intelligence or their courage changed between the two conditions.
What changed was what the room could hold. That is not a character problem, and it was never going to be fixed by asking people to try harder. It is a structural problem, it has been described precisely for forty years, and structure is a thing you can now build with for a difference in outcomes.