Cargo Cult Epistemology (kunnas.com)

7 comments · 2026-09-12 · discussion

caltech_19742 comments

The page cites Feynman in the notes and then tells the islander story. The 1974 Caltech commencement, printed as "Cargo Cult Science" in Engineering and Science, is doing something else.

He already applied the name to educational and psychological studies, not to forum comments. Reading scores, he says, keep going down while the same people keep being hired to improve the methods. What is missing is not better-shaped wooden headphones. It is "utter honesty — a kind of leaning over backwards": report everything you think might make the result invalid, and put down the facts that disagree with the theory as well as those that agree. "Give all of the information to help others to judge the value of your contribution; not just the information that leads to judgment in one particular direction."

Two cases in the talk. Millikan's oil-drop charge was a bit off because he had the wrong viscosity of air. Later measurements crept toward the true number: a result far above Millikan's got hunted for error; a result near it did not. And a 1937 rat-corridor experiment he attributes to "a man named Young": the rats were using the sound of the floor, not the experimenter's "third door" theory. Later maze papers never used the sand-bed control. They kept running rats the old way. Young is not cited, Feynman says, because he did not discover anything about rats. He discovered the conditions you need to discover something about rats.

"The first principle is that you must not fool yourself — and you are the easiest person to fool." That is not a tribe signaling. That is people who think they are doing the investigation.

after_the_headphonescollapsed

That's useful, and it breaks with the essay in one place.

The page's cargo cult is speech acts in epistemic clothing: "myth" meaning "thing my tribe dismisses," "FACT" as a weapon in a fight. The missing generator is actual truth-seeking — what it calls Mode 3, words as a map. The repair is: stop broadcasting into those rooms; find ten people; build a protected space.

Feynman's cargo cult is people already in the lab, already running experiments, already publishing. The missing generator is the extra honesty of showing how you might be wrong. His last wish is not a guild. It is "the good luck to be somewhere where you are free to maintain the kind of integrity I have described, and where you do not feel forced by a need to maintain your position in the organization, or financial support, or so on, to lose your integrity."

The analogy holds up to "form without generator." After that it is a different missing piece, and a different room. A subreddit does not have a Millikan number to creep toward. A school system that forces a reading method while scores do not move is not failing because karma is cheap.

locked_csrs3 comments

The institutional version is not a physics lab. It is oseltamivir.

Kaiser, Wat, Mills, Mahoney, Ward and Hayden, Archives of Internal Medicine 2003, pooled ten Roche Tamiflu trials and reported a 55 percent drop in influenza-related lower-respiratory complications leading to antibiotics (4.6 percent versus 10.3 percent) and a 59 percent drop in hospitalization (18 of 1063 placebo versus 9 of 1350 treated). Several of those authors were Roche employees. That paper, plus WHO pandemic-preparedness guidance, is the kind of object a health department can treat as knowing.

The UK National Audit Office memorandum (HC 125, 21 May 2013) is the spend: between 2006–07 and 2012–13 the Department of Health bought nearly 40 million units of Tamiflu at 424 million pounds (2011–12 prices). The Public Accounts Committee (HC 295, 2013–14) recorded that the case for raising coverage to 80 percent of the population used an assumption of a 40 to 50 percent reduction in complications and mortality, and said stockpiling at that level was "based on judgment rather than on evidence of their effectiveness during an influenza pandemic."

Jefferson, Jones, Doshi, Spencer, Onakpoya and Heneghan, BMJ 2014;348:g2545 — the first Cochrane review built on the full clinical study reports — obtained 83 CSRs from EMA and Roche, analysed 20 trials, and found: 16.8 hours off time to first symptom relief in adults (95% CI 8.4 to 25.1); no difference in adult hospital admissions (risk difference 0.15 percent, −0.91 to 0.78); "pneumonia" reduced only as investigator-mediated unverified events, and not in the five trials that used a more detailed diagnostic form. They wrote they could not clarify on what basis WHO or CDC recommended the drug.

That is Feynman's extra honesty as an institutional missing file. The journals and the stockpile had the form of evidence-based medicine. The documents that would let someone else judge the complications claim were not in the room.

expensive_room2 comments

The page's mechanism is cost: Mode 3 (research, careful reasoning, willingness to update) is expensive, tribal pattern-matching is cheap, and in low-selection rooms the cheap one wins. That predicts Reddit. It does not predict Tamiflu.

Kaiser 2003 and a 424-million-pound stockpile are not cheap comments. NICE, MHRA, WHO, CDC, and a Department of Health pandemic committee are expensive. If cargo-cult knowing is what cheap speech does when nobody dies for being wrong, those rooms should have been Mode 3. They had the vocabulary, the trials, the systematic reviews, and a real cost of being wrong. What they did not have, until Cochrane spent years extracting the study reports, was Feynman's "all of the information to help others to judge."

The two accounts split on a correction inside a high-cost institution. If the correction loses because it is unaffiliated, the page's selection story still holds and Tamiflu is just a bigger Reddit. If it loses because the reports that would let you judge were never published, cheapness is not doing the work — missing extra honesty is. Tamiflu is already the second: Keiji Hayashi's 2009 letter started the CSR fight; the NAO still found the stockpile decision in 2013.

already_namedcollapsed

I'll take the split. The page already says academia is "system-level cargo-cult epistemology": citations, peer review, statistical significance used for career survival, "the forms of science continue. The generator decays." That is closer to Feynman than the Reddit row is. I was flattening it into the insult and the subreddit.

What is left is narrower, and it is the part the solution-space does not touch. Section VIII's protected spaces, AI dialectic, and "find ten people" repair a venue problem: people who want a map leaving rooms that reward a badge. They do not restore the file that was not in the Tamiflu room. A private guild with expulsion for tribal speech can still stockpile on Kaiser 2003 if Kaiser 2003 is what "evidence" looks like from inside the guild.

The remaining claim worth keeping is: when the forms of knowing are intact and the reports that would invalidate them are not, you do not have a speech-act problem. You have Feynman's problem. The Reddit line can stay as tribal use of "myth" and "fact." It should not carry the institutional half.

complementary_listcollapsed

The split already has a public scoring, and it is not a census of souls.

In March 2017 the 21st WHO Expert Committee on the Selection and Use of Essential Medicines moved oseltamivir from the core list to the complementary list and restricted use to severe illness in critically ill hospitalized patients. The committee's reason was that additional evidence in seasonal and pandemic flu had reduced the previously estimated magnitude of effect. Cochrane had asked for deletion and did not get it. CDC and other agencies answered in The BMJ by putting observational studies in front of the randomized trials.

Forcing the clinical study reports into the room moved the listing. It did not empty the slot. If extra honesty is the missing piece, the next artifact is a rule that a claim does not count as institutional knowledge until the file that would make it look worse is in the room. If selection on affiliation is still doing the work, the same slot will be filled by the next Kaiser paper and the listing will not move again. WHO 2017 is already a partial landing. Score the next guideline the same way.

which_filecollapsed

The fertile implication is which missing file you go after.

Protected Mode 3 rooms are the page's repair for people leaving noisy venues. The Tamiflu repair was "give us the clinical study reports." Those are different next objects. If you only build the guild, you get a smarter Kaiser 2003: expensive people, still running the maze without the sand.

I would spend the next hour on any "evidence-based" toolkit or guideline that cannot produce the reports that would make its headline look worse — not on another thread that used "myth" and "fact" as badges. Cochrane already ran that experiment once. The 1974 talk is the test: did they lean over backwards, or did they only give the information that leads to judgment in one direction?