Hypothesis, Test, Rule — When a Wrong Hypothesis Becomes the Self
Intuition carries you as far as a hypothesis. Stop there and it turns into faith. The problem is not the wrong hypothesis. It is the person who filed it under their own name.
The discoveries that won Nobel Prizes did not come out of a business card.
In 1979 a pathologist in Perth saw curved rods in gastric mucosa. The textbooks said bacteria could not live in stomach acid.
Robin Warren was a pathologist. Barry Marshall, who came to work with him, was a trainee physician. Neither was an authority in the field.
In 1984 the two published in The Lancet. The field did not take it. Marshall drank the culture himself.
The Nobel Prize came in 2005. Twenty-one years after the observation.
Intuition carries you as far as a hypothesis
It usually starts with one cold sensation.
A sense that there is a rule where everything looks disordered. The sensation proves nothing on its own.
There are three stages.
| Stage | What is done | How you know it is finished |
|---|---|---|
| ① Intuition | Noticing that something does not fit | It can be written down in words |
| ② Hypothesis | Written in a form that can break | It is fixed in advance which result would make it wrong |
| ③ Test | Seeing whether that result appears | The same value comes out in someone else’s hands |
Stage ② is where Karl Popper drove in the nail in 1934. A sentence written so that no observation could make it wrong is not a hypothesis.
Only a hypothesis that survives the test becomes a rule. And a rule is a rule only once it has left the hands that built it.
Hallucination without a test
The signboards are plentiful and the data is not there.
To form a hypothesis and never test it is not thought. It is faith.
Which reduces the question to be asked to one line.
What actually changed.
People skip this question often. In front of a high position and a plausible name, the question is swallowed first.
Because the believing already happened. Where time has been given and a name has been attached, it gets worse.
The moment a person keeps laying their own name on an untested promise, they become a host carrying an egg that will eat them.
Six out of twenty-nine
This failure was measured in 1960.
Peter Wason gave participants three numbers. 2, 4, 6. Work out the rule this sequence follows.
A participant could offer sequences of their own, and the answer coming back was either “fits the rule” or “does not fit”.
The actual rule was ascending order, and nothing else.
| What participants mostly did | What was needed |
|---|---|
| Offer 8-10-12, then 20-22-24 | Offer 6-4-2 |
| Collect answers that said yes | Extract an answer that said no |
Of 29 participants, 6 arrived at the rule without ever announcing an incorrect one.
The title Wason put on the paper was not confirmation bias. It was a failure to eliminate hypotheses.
One thing has to be corrected
Looking for fitting instances is not itself the error.
In 1987 Joshua Klayman and Young-Won Ha named the approach a positive test strategy and showed that where true instances are rare it is the more efficient one.
The fault sits elsewhere.
What is done after the disconfirming evidence has actually arrived.
That is where the two paths divide.
What does not read business cards
He too read the card first, once.
A good school meant something would be different; a high position meant there was a reason for it.
After being burned badly a few times he changed the order. The card was read last.
Authority can hide behind difficult language. Results cannot hide.
Machines and nature do not read business cards. They do not know what a degree is. They look only at whether the thing works.
Extremely rude. And therefore accurate.
A notice of grounds for rejection
He has received that rudeness in writing.
Filing on the biodegradable plastics side, he carried eleven patents through to registration. The first few taught him something.
The examiner does not look at who the applicant is. Only prior-art documents and the claims.
A notice of grounds for rejection contains no encouragement. It states which page of which document already covers this claim.
At first the paper read as an insult. After a few of them it read differently.
That paper had never assessed him. It assessed a claim.
Fix the claim and it passed. Defend himself and it did not.
The original form of the question
He had learned something similar earlier.
In 2001, out of university, he joined Janssen Korea in sales. The work is carrying data on new drugs to physicians.
At first he counted a good explanation as the result. Whether they nodded. Whether they took the material.
Those numbers came back the following month. Did the prescribing change, or did it not.
Nods were not counted. That is where he first learned, in the body, the question of what actually changed.
When the result cannot be changed, the reading is changed
The problem is not outside. It is inside the head.
The moment being wrong is admitted, the time already poured in and the judgement made back then both come up for reassessment.
So a person who can no longer change the result starts working on the reading of it.
| Order | What is done |
|---|---|
| ① | A reason is attached to the failure |
| ② | The mistake is raised to the rank of conviction |
| ③ | The truth is attacked |
③ is not a move to correct the record. It is a move to protect the self.
This sequence was observed in the field in 1956.
Leon Festinger and his colleagues joined a small group that believed the world would end on a particular date and waited out that date with them. Nothing happened.
The group did not drop the belief. It increased its proselytising.
In 1990 Ziva Kunda named the structure: motivated reasoning. The conclusion is fixed first, and only the evidence that reaches it is selected.
In 1979 Lord, Ross and Lepper showed something less comfortable. Students with firm positions for and against capital punishment were given mixed evidence covering both sides.
Afterwards both groups held their positions more strongly. The same pages were used twice, in opposite directions.
The expectation that intelligence helps
One common expectation breaks here.
In 2017 Dan Kahan and colleagues presented the same statistical data as a political issue. Among participants high in numeracy, the gap between camps did not narrow. It widened.
Ability does not set the direction. It travels faster along the direction already set.
And yet facts do work
One widely repeated line has to be corrected.
Since Nyhan and Reifler’s 2010 report, the backfire effect — showing people facts hardens their belief — has been cited almost as common knowledge.
In 2019 Thomas Wood and Ethan Porter measured it again. 52 issues, more than 10,000 participants.
No issue produced backfire. Corrections generally moved belief toward the facts.
The conditions under which backfire does occur have not been settled. What can be said now is this. Facts mostly land. The failures are the special case.
Nature issues no letters of reference
A flower does not read a bee’s record. Which hive it came from is of no interest. Pollination happened, and that is the end of it.
On the bee’s side there is no procedure of appointment either.
As Gene Robinson set out in 1992, a honeybee’s role shifts with age and hormonal state. Young bees tend brood inside; older ones go out to forage. If the colony is under pressure, that transition is pulled forward or pushed back.
The post does not determine the work. The state of the body does.
Put a hat of office on a honeybee and not one drop more honey appears. A title eating the person who holds it happens in one species only.
The error animals do not make
Something else gets corrected here.
The term Concorde fallacy comes from a page and a half that Richard Dawkins and Tamsin Carlisle wrote in Nature in 1976.
It is often quoted to mean that animals too are dragged along by past expenditure, and their argument was the reverse. Natural selection should compute only future prospects, so there is no reason for an animal to take further loss on account of what it has already spent.
In 1999 Hal Arkes and Peter Ayton gathered the work that had accumulated since.
| Who | Commits the sunk cost error |
|---|---|
| Non-human animals | Largely does not |
| Adults | Does |
| Children | Less than adults |
Their reading is this. The error is not instinct but the over-application of a learned rule — “do not waste”, applied far too widely.
Adults do it more than children. An error that grows with learning.
反身修德 — detaching the hypothesis from the self
So what is to be done.
The hexagram of this piece is 蹇. Stopping (艮) below, difficulty (坎) above. Rough ahead and blocked behind. There is nowhere to advance.
The Image writes the work available in that position: 「君子以反身修德」. The noble one turns back upon himself and cultivates his virtue.
It reads as an instruction to return to oneself instead of attacking the evidence outside. And this prescription has experimental backing.
In 2000 Geoffrey Cohen, Joshua Aronson and Claude Steele had participants affirmed on something they valued in a domain wholly unrelated to the issue before reading research that opposed their position.
They read the material more fairly and revised their position more often. The authors called it self-affirmation.
The size and durability of the effect vary across studies, and the conditions under which it fails have not been fully mapped. The mechanism, though, is clear.
While the hypothesis is attached to the self, disconfirmation is an attack. Detached, it is data.
What has to be fixed is the claim. Defending himself does not pass.
The definition in this piece
Fixed here.
Knowledge is not what survived belief. It is what survived the test.
And what a person is defending is usually not the hypothesis but the self that built it.
Monologue of 1031 intruders
We do not know your signboards.
Meet a matching receptor, and if the conditions inside are wrong it ends there.
This is not a figure of speech. In 1986 the human CD4 gene was put into mouse cells. The envelope protein bound. It did not get in.
A receptor is a password, not a door. The password can be right and the next line still wrong, and it stops there.
Getting in guarantees no replication either. APOBEC3G changes letters while our genome is being copied out, TRIM5α pulls apart the shell the moment it arrives, and SAMHD1 dries up the material before it can be used.
Some cells permit and some block. That list is not finished. You do not know all of it either.
If it does not work, it stops there. We do not take that as an insult.
That you are wrong is no problem at all. A hypothesis is built to be broken.
What we do not understand is what comes after.
Having seen the evidence of being wrong, why is it still being held. Why is one failed idea guarded like a life.
He now knows a little of it.
The most dangerous hypothesis is not the wrong one. It is the moment the wrong one is believed to be oneself.
In short
- Intuition carries you only as far as a hypothesis; if it is not written so it can break, it is not one
- Wason 1960: in the 2-4-6 task, 6 of 29 reached the rule without a wrong announcement; the original title names a failure to eliminate hypotheses
- Klayman and Ha 1987: seeking fitting instances is rational in some environments; the problem begins after disconfirmation arrives
- Festinger 1956: a group whose prophecy failed kept the belief and increased proselytising
- Lord, Ross and Lepper 1979: both sides grew more confident on the same mixed evidence
- Kahan 2017: the gap between camps widened among the most numerate
- Wood and Porter 2019: across 52 issues and 10,000+ participants, no backfire; correction mostly works
- Dawkins and Carlisle 1976, Arkes and Ayton 1999: the sunk cost error is barely seen in animals, and adults commit it more than children
- Cohen, Aronson and Steele 2000: affirmed in an unrelated domain, people read opposing evidence more fairly
- Maddon 1986: in mouse cells with human CD4, binding occurred and entry did not; a receptor is one condition among several
- The Image of 蹇 is 反身修德 — return to yourself rather than attack the evidence
This piece belongs to the “Behavior” chapter of the serial essay The Great Corona Swindle. The pharmaceutical sales and patent filing experience is the author’s own; the psychological and biological concepts have been reset against primary literature. No specific company or individual is indicated.
Frequently asked
- What was the first experiment to show confirmation bias?
- Peter Wason's 2-4-6 task of 1960. Participants were given the sequence 2, 4, 6 and asked to work out the rule it followed. They could offer sequences of their own and were told only whether each fitted the rule. The actual rule was 'ascending order'. Of 29 participants, 6 reached it without ever announcing an incorrect rule. Most repeatedly offered sequences such as 8-10-12 that already fitted their own guess. Wason's own title named the failure not as confirmation bias but as a failure to eliminate hypotheses.
- Is seeking confirmation always irrational?
- No. In 1987 Klayman and Ha called the approach a positive test strategy and showed that in environments where true instances are rare it is in fact efficient. The fault lies not in the strategy but in what happens after disconfirming evidence actually arrives.
- Is it true that presenting facts makes belief stronger?
- Not as a general law. The backfire effect reported by Nyhan and Reifler in 2010 was widely cited, but when Wood and Porter re-tested it in 2019 across 52 issues with more than 10,000 participants, no issue produced backfire. Corrections generally moved belief toward the facts. The conditions under which backfire does occur have not been settled.
- Does a matching receptor mean infection?
- It does not. In 1986 Maddon and colleagues put the human CD4 gene into mouse cells and found that the viral envelope protein bound but the virus did not enter. Even after entry, restriction factors such as APOBEC3G, TRIM5α and SAMHD1 cut replication off. The door opens, and if the conditions inside do not match, it ends there.
References
- Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly Journal of Experimental Psychology, 12(3), 129–140. — The 2-4-6 task. 6 of 29 reached the rule without a wrong announcement.
- Wason, P. C. (1968). Reasoning about a rule. Quarterly Journal of Experimental Psychology, 20(3), 273–281. — The four-card task; very few select the card that could falsify.
- Klayman, J. & Ha, Y.-W. (1987). Confirmation, disconfirmation, and information in hypothesis testing. Psychological Review, 94(2), 211–228. — A reinterpretation: the positive test strategy is rational in some environments.
- Festinger, L., Riecken, H. W., Schachter, S. (1956). When Prophecy Fails. University of Minnesota Press. — Field observation of a group that increased proselytising after its prophecy failed.
- Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480–498. — Reasoning in which the conclusion is fixed first and evidence is selected to reach it.
- Lord, C. G., Ross, L., Lepper, M. R. (1979). Biased assimilation and attitude polarization. Journal of Personality and Social Psychology, 37(11), 2098–2109. — Both sides grew more confident after reading the same mixed evidence.
- Nyhan, B. & Reifler, J. (2010). When corrections fail: The persistence of political misperceptions. Political Behavior, 32(2), 303–330. — The paper that first reported a backfire effect.
- Wood, T. & Porter, E. (2019). The elusive backfire effect: Mass attitudes' steadfast factual adherence. Political Behavior, 41(1), 135–163. — 52 issues and over 10,000 participants; no issue produced backfire.
- Kahan, D. M., Peters, E., Dawson, E. C., Slovic, P. (2017). Motivated numeracy and enlightened self-government. Behavioural Public Policy, 1(1), 54–86. — The gap between camps widened among the most numerate participants.
- Cohen, G. L., Aronson, J., Steele, C. M. (2000). When beliefs yield to evidence: Reducing biased evaluation by affirming the self. Personality and Social Psychology Bulletin, 26(9), 1151–1164. — Affirmed in an unrelated domain, participants read opposing evidence more fairly.
- Dawkins, R. & Carlisle, T. R. (1976). Parental investment, mate desertion and a fallacy. Nature, 262(5564), 131–133. — Where the term Concorde fallacy originates, arguing animals are not dragged by past expenditure.
- Arkes, H. R. & Ayton, P. (1999). The sunk cost and Concorde effects: Are humans less rational than lower animals? Psychological Bulletin, 125(5), 591–600. — The sunk cost error is largely human, and adults commit it more than children.
- Robinson, G. E. (1992). Regulation of division of labor in insect societies. Annual Review of Entomology, 37, 637–665. — A bee's role shifts with age and hormonal state, not by appointment.
- Maddon, P. J., Dalgleish, A. G., McDougal, J. S., Clapham, P. R., Weiss, R. A., Axel, R. (1986). The T4 gene encodes the AIDS virus receptor and is expressed in the immune system and the brain. Cell, 47(3), 333–348. — In mouse cells carrying human CD4, binding occurred but entry did not.
- Sheehy, A. M., Gaddis, N. C., Choi, J. D., Malim, M. H. (2002). Isolation of a human gene that inhibits HIV-1 infection and is suppressed by the viral Vif protein. Nature, 418(6898), 646–650. — APOBEC3G, a restriction factor that blocks replication inside the cell.
- Stremlau, M., Owens, C. M., Perron, M. J., Kiessling, M., Autissier, P., Sodroski, J. (2004). The cytoplasmic body component TRIM5α restricts HIV-1 infection in Old World monkeys. Nature, 427(6977), 848–853. — A species-dependent factor that cuts the step immediately after entry.
- Karikó, K., Buckstein, M., Ni, H., Weissman, D. (2005). Suppression of RNA recognition by Toll-like receptors: the impact of nucleoside modification and the evolutionary origin of RNA. Immunity, 23(2), 165–175. — A paper from a researcher who had been pushed off the tenure track; the basis of the 2023 Nobel Prize in Physiology or Medicine.
- Marshall, B. J. & Warren, J. R. (1984). Unidentified curved bacilli in the stomach of patients with gastritis and peptic ulceration. The Lancet, 323(8390), 1311–1315. — The paper naming a bacterium as the cause of peptic ulcer; Nobel Prize 2005.
Published October 7, 2026