Eight, ten, twelve. The person writes that down, hears that it fits, and writes another of the same kind. The sentence they then announce is the pattern they have been repeating. A different set of three, one that would have broken the pattern, never gets a turn. The yes felt like proof. It was only a yes.
P. C. Wason, then with the Medical Research Council's Industrial Psychology Research Group at University College London, set the task out in On the failure to eliminate hypotheses in a conceptual task, Quarterly Journal of Experimental Psychology, 1960. He sat 29 psychology undergraduates down one at a time, 17 men and 12 women, and gave them the numbers 2, 4, 6. Those three conformed to a simple rule about the relation between any three numbers, not about their size. The job was to discover the rule by writing further triples, with a reason for each, and to hear after each one whether it conformed. They were told to find it with as few triples as they could, and that the aim was not simply to find numbers that conform but to discover the rule itself. They were to announce it only when they felt highly confident.
The rule was wider than the pattern
Wason said yes when the three numbers were in increasing order of magnitude, and no when they were not. Rising by two is enough to earn a yes. It is not the rule. Any increasing triple earns a yes, including 1, 2, 3 and 2, 3, 7. A person who only offers 8, 10, 12 and 14, 16, 18 will hear yes every time and can announce "increasing intervals of two" with a straight face. That announcement is sufficient. It is not necessary. Every triple that fits it also fits the real rule, so the yes never corrects them.
Six of the 29 announced the correct rule on the first try. Thirteen announced one incorrect rule. Nine announced two or more incorrect rules. One announced none. The incorrect rules were, in nearly every case, sufficient ones of that kind. The people who got it right on the first announcement had generated a mean of 8.0 triples before they spoke. The people who announced a wrong rule first had generated a mean of 3.68. Speed was not a sign that the rule was in hand. It was a sign that a narrower rule had been allowed to stand.
What they refused to write down
Wason scored the thinking, not the charm of the announcement. For each reason a person gave, he looked at whether the later triples were compatible with that reason or incompatible with it. Compatible triples keep the reason alive. Incompatible ones put it at risk. He divided the incompatible count by the compatible count. The six who announced the right rule first had a mean ratio of 1.79. The 22 who announced a wrong rule first had a mean of 0.24. The paper reports that difference as significant. The people who were right had been trying to break their own reason. The people who were wrong had been collecting yeses.
He also counted negative instances, triples the real rule would reject. Up to the first announcement, those were 0.21 of the triples from the people who got the rule right, and 0.04 of the triples from the people who announced a wrong rule. The wrong-rule group almost never left the region where the real rule says yes. Inside that region, "rising by two" and "any three that increase" give the same answer. You cannot tell them apart by staying there. In 18 of those 22 wrong first announcements, the reason given for the very first triple became at least part of the rule they later announced. The first guess did not stay a guess. It hardened.
The warning was already in the instructions
Nobody needed a poster about open-mindedness. The instructions already said the aim was not to collect conforming numbers, and that an announcement should wait for high confidence. Twenty-two people announced a wrong rule anyway, and they did it on fewer triples than the six who were right. A slide that repeats that aim is the sentence they had already been given. It did not make them write 1, 2, 3.
A buyer who asks only whether the room can state its rule will pay for the fast announcement. That is the 3.68 result. The useful record is a case that still fits a wider reading and breaks the sentence they announced. If the wall is a column of examples that all fit, you bought enumerative induction. Wason's name for the other kind, the one that uses a case the current reason cannot survive, is eliminative.
What to ask for before anyone announces
Bring one rule the team is about to act on. It might be a claim about why a pilot is working, why a customer is leaving or why a draft is good enough to send. Write the rule in a sentence. Then write two triples, in the ordinary sense of two cases. One case should fit the rule and would still be true if a wider rule were the real one. One case should fit the wider rule and break the sentence on the board. If the second case is missing, the room is still on 8, 10, 12.
The standard I would take into a proposal is small. Did they write the reason for the first example before they wrote the rule, because that reason is what hardened in 18 of 22 wrong announcements. Did someone produce a case the announced rule cannot survive, and did the room keep the rule only after that case failed. Did more than one person do it. Six of 29 is not a room. A highlight of the one person who got there will not tell you what the next rule will do to everyone else.
Be willing to hear a yes that hurts the current sentence. A team that needs every example to confirm the rule it walked in with has booked the condition that produced the wrong announcement. Say that before the session, not after the wall is covered in triples that were always going to fit.
Where this sits in the week
Collective Campus has done capability building for teams since 2015. The critical thinking workshop is for a team that can bring a live rule and is allowed to try a case that would kill it. How a label fails to travel into later work is in how to judge a critical thinking workshop for transfer. The practised method that hides a shorter one is in the short solution a room can miss. Where judgement has to sit once a model is in the draft is in why critical thinking beat AI literacy.
What to bring
Bring one rule the team is using, the first example that made it feel true, and permission to test a case that would force a rewrite. If every case has to confirm the sentence on the board, the triple stays fitting and the rule stays wrong.
