How to use this page
As in Lessons 1 and 2, keep a notebook beside you. At each Pause and answer box, write your answer before you open the fold. The fold is one line of reasoning to compare with yours, not the official answer.
Lessons 1 and 2 were about answers. This lesson moves one step earlier, to the question that sends you looking. The historical cases are real; sources are listed at the end.
What we assume you already know
- From Lesson 1: an observation is not a belief, and “I don’t know — here is how we’d find out” is a strong position.
- From Lesson 2: a claim is testable when it forbids some outcome; a test is honest only if it could have come out against you.
- From First principles, Lesson 2: a problem stated as a solution (“ship more ice”) can only improve that solution; stated as a function (“remove heat from a drink”), it opens other answers.
- Work can be turned into heat, but no machine gives out more energy than it takes in. We will see how people learnt this.
- You have spent time on a question that went nowhere.
If every answer starts with a question, how do I choose the question?
School trains us to answer questions others set. Outside school, nobody sets them. Questions follow:
- Can a question be wrong before anyone has answered it?
- Why do some questions absorb centuries and get nowhere, while others are settled in years?
- What happens when a question’s hidden assumption is false?
- If a question succeeds, who benefits, and who pays?
- Write down one question you are trying to answer at the moment — about work, study, health, family, or the world — exactly as it sits in your head. You will check it before the end.
Two questions after a bad result
A student fails a mathematics exam she had studied hard for. That night two questions go round in her head.
Question A. “Why me? Why does this always happen to me?”
Question B. “Which questions did I lose marks on — and what, exactly, went wrong on each one?”
She spends the evening on A. Next morning a friend helps her work through B with the marked paper.
- For each question, what would an answer look like? How would she know she had found it?
- Which question can be finished? Which can be asked for ever?
- What does Question A take for granted? Look at the word “always”.
After you have written your answers — compare your reasoning
Question B has an answer of a known shape: a list. Marks lost on careless algebra, on a topic she skipped, on running out of time. Each item can be checked against the paper, and each points to something she can do. When the list is complete, the question is finished.
Question A has no answer of any shape. “Why me?” asks for the reason the world picked her, and no observation would count as finding it. Every answer on offer — bad luck, fate, “I’m not a maths person” — is uncheckable, and the last would stop her trying. And “always” is an assumption her past results could refute.
The hurt behind Question A is real and deserves kindness. But as a road, it runs in a circle: the further she walks, the worse she feels.
A question is answerable when you can say in advance what an answer would look like and how you would recognise it. Unanswerable questions are not harmless: they use up time and invite uncheckable answers. The first check: “What would count as an answer — and could I ever get it?”
Where does the heat come from?
In the late eighteenth century most scientists explained heat as caloric: an invisible, weightless fluid that flowed from hot bodies to cold. Research questions followed: how much caloric does a body hold, and how does friction squeeze it out? (A related fluid, phlogiston, was said to escape in burning — yet metals heated in air get heavier, and in 1774 Lavoisier showed the gain was air taken up.)
Around 1797 Count Rumford, overseeing cannon boring in Munich, turned a blunt borer in a barrel under water and boiled the water in under three hours; the heat kept coming as long as the boring did. “Squeezing out” needed the chips to hold less heat than solid metal. He measured: no difference.
For centuries inventors had also asked: how do I build a machine that, once started, keeps doing useful work for ever? In 1775 the Paris Academy of Sciences resolved to stop examining such proposals.
In the 1840s James Prescott Joule asked: how much work makes how much heat? Falling weights turned a paddle wheel in water; he measured the fall and the warming, then repeated it with mercury and with iron rubbing on iron. In 1850 he reported heat always proportional to work: 772 pounds falling one foot warms a pound of water by one degree Fahrenheit, within about one per cent of today’s value.
- “How much caloric does this iron hold?” What does it take for granted? If that is false, could any measurement give a true answer?
- What result would have shown Joule’s idea wrong?
- What does the perpetual-motion question assume? Once Joule’s answer was known, what happened to it?
After you have written your answers — compare your reasoning
The caloric question carries a passenger: that caloric exists. If it does not, careful measurements are poured into a wrong story. This is a false presupposition, and a question built on one is a loaded question — logic’s standard example is “Have you stopped cheating in exams?”, where yes and no both admit the cheating. A loaded question cannot be answered correctly, only answered.
Rumford’s question — in effect, “where does this heat come from?” — assumed far less, and his chip test aimed where caloric would break. Joule’s names two measurable things and asks how they are linked. Had water, mercury and iron given very different ratios, a fixed exchange rate would have been dead: in Lesson 2’s terms, his answer could fail.
That answer, with work by Mayer, Helmholtz and others, became the conservation of energy, which settled the perpetual-motion question: you can’t get more work out than you put in. The US patent office, which does not ordinarily ask for working models, still demands one for perpetual motion.
Before investing in a question, ask: “What must be true for this question to have an answer at all?” If one is false, every answer is wrong, however careful. Prefer questions that ask for a measurable relationship — how much of this goes with how much of that — because their answers can fail, and so can teach.
Two clinics, one hospital
In the 1840s the Vienna General Hospital had two maternity clinics side by side. In the First, deliveries were done by doctors and medical students; in the Second, by midwives. Mothers in both died of childbed fever, but not equally: in 1846 the First Clinic recorded 459 deaths in 4,010 births (11.4%), the Second 105 in 3,754 (2.8%). The usual explanation was “epidemic influences” — changes in the atmosphere that came and went.
Ignaz Semmelweis, an assistant in the First Clinic, asked a narrower question: what is different about the First Clinic? Not crowding: the Second was more crowded. Not climate: the clinics were neighbours. Women who gave birth in the street on the way in rarely caught the fever. In 1847 a colleague, Jakob Kolletschka, died after a student’s scalpel cut him during an autopsy, with the same signs as the dying mothers. Doctors and students came to the First Clinic from the autopsy room. The midwives did not.
From mid-May 1847 Semmelweis required hand-washing in chlorinated lime before examinations. The First Clinic’s monthly death rate had been 18.3% in April. It was 2.2% in June, 1.2% in July and 1.9% in August.
- What observation could ever show “epidemic influences” wrong?
- Why is “what is different about the First Clinic?” easier to answer, and to test, than “what causes childbed fever?”
- What did the hand-washing idea predict before it was tried? Could it have failed?
- Who did the answer point at?
After you have written your answers — compare your reasoning
“Epidemic influences” fitted every rise and fall. Like Lesson 2’s horoscope, it forbade nothing — and it could not explain why two clinics under one sky differed by a factor of four.
Semmelweis swapped a question about an ultimate cause nobody understood for one about the difference between two groups alike in most ways. A comparison has a short list of candidates, each checkable. It also carried a function: not “explain the fever” but “stop mothers dying in this ward”. Hand-washing predicted that the First Clinic’s rate would fall towards the Second’s. It could have failed; it did not.
The answer pointed at the people asking. Many doctors rejected it, some offended that their hands could be unclean; it won wide acceptance only after germ theory explained it. When an answer blames the asker, expect resistance — including your own.
A good reframe turns a vague question about an ultimate cause into a sharp one about a difference you can observe and change: “What is different between where this happens and where it doesn’t?” Frame by function — name what must change, not the theory you hope to confirm.
A cheap cure for knock
So far, bad questions led nowhere. This one led exactly where it aimed.
Early car engines “knocked”: fuel exploded unevenly, wasting power and damaging engines. At General Motors’ research laboratory under Charles Kettering, Thomas Midgley Jr. and his team spent years seeking an additive that would stop knock effectively and cheaply. On 9 December 1921 they found that tetraethyl lead did it at about one part in 1,300 of petrol. Ethanol also worked, but only in large proportions, and it offered no patent; historians conclude commercial reasons weighed heavily in choosing lead.
In October 1924 five workers died at a New Jersey plant making the additive, and many more were poisoned. Public health experts, among them Alice Hamilton and Yandell Henderson, had already warned against spreading lead into the air. After a pause for review, a 1926 US committee found no firm evidence of harm, and sales resumed.
Decades later the geochemist Clair Patterson, measuring the age of the Earth, found lead contamination everywhere; in 1965 he published evidence that human exposure was far above natural levels. Lead harms the developing brain. The United States banned leaded petrol for road vehicles from 1996, India by 2000; the last country, Algeria, stopped in July 2021. The UN Environment Programme estimates the phase-out prevents more than 1.2 million premature deaths a year.
- Write the team’s question in one line. Answerable? Testable? True assumptions?
- What did success optimise? What was missing from the target?
- Who gained? Who paid — and when did anyone find out?
After you have written your answers — compare your reasoning
By the checks so far, the question was excellent: answerable, testable in real engines, and right that knock could be stopped. That is the uncomfortable lesson — the question succeeded.
The trouble was what it optimised. Effectiveness, cost and a patent were inside the target; the health of workers and of everyone breathing the exhaust was outside it. A question optimises exactly what it names, and treats whatever it leaves out as free. Workers paid within three years; the wider bill, paid by children over decades, went unmeasured until someone aimed a question at it.
One rewrite: “How can we stop knock without adding anything that harms the people who make, handle or breathe the fuel?” That rules lead out and points to answers that existed, at a higher price. Whether to pay it is a judgement of values — but the rewrite puts it on the table at the start, not fifty years later.
A question can be answerable, testable and built on truth, and still lead somewhere bad, because success optimises what the question names and ignores what it leaves out. Two more checks: “What does success optimise — and what is missing?” and “If we succeed, who pays, and do they get a say?”
Occam: the question that assumes least
Return to Vienna. Two questions were on offer about the same deaths.
Question 1. “Which epidemic influence in the air is causing childbed fever this season — and why does it strike the First Clinic harder?”
Question 2. “What happens to women in the First Clinic that does not happen in the Second?”
- List what each question takes for granted before any answer arrives.
- If Question 1’s first assumption is false, what happens to the work spent answering it?
Question 1 builds in an unseen agent in the air, seasonal change, and a reason to favour one clinic over its neighbour — three assumptions before the first observation. Question 2 assumes only a difference, which the death figures already showed.
Occam’s razor usually trims answers. It trims questions too: when two questions aim at the same goal, prefer the one that builds in fewer untested assumptions. It risks less, and fails in plain sight. The same rule picks Joule’s question over the caloric one.
A working model: five checks on a question
- Before opening ours, write your own checklist for a question in five lines or fewer.
- Which check would have caught leaded petrol? Which would not?
After you have written your checklist — compare it with ours
| Check | What it means | The question to ask |
|---|---|---|
| Answerable | An answer of a known shape exists, and you could recognise it. | “What would count as an answer — could I ever get it?” |
| Testable | Some result could show the answer wrong (Lesson 2). | “What would I see if the answer were wrong?” |
| Assumptions | What the question takes for granted is true, or has been checked. | “What must be true for this to have an answer?” |
| Target | Framed by function; success optimises what you actually care about. | “If this succeeds, what gets bigger — and what is left out?” |
| Who pays | The costs of success, and who bears them, are inside the question. | “Who pays if we succeed — and do they have a say?” |
The first three checks ask whether a road leads anywhere: they catch “why me?”, caloric and perpetual motion. The last two ask where it leads: they catch questions that succeed and harm. This is a model, not a proven theorem; we will push on it.
Exercise: check your own question
Return to the question you wrote at the start. Use one notebook page.
- Answerable: what would an answer look like — a number, a list, a yes or no by a date? If you cannot say, rewrite until you can.
- Testable: name one result that would show your answer wrong.
- Assumptions: list what the question takes for granted; mark each true, false or unchecked.
- Target: restate it as a function — a verb and an object. What would success make bigger? What is left out?
- Who pays: who is affected if the answer is found and acted on? Would they agree with your question?
- Rewrite it in one sentence that passes all five checks.
An example: “Why can’t I ever stick to studying?” becomes “Over the next two weeks, when and where do I manage half an hour of study without stopping — and what is different about the times I don’t?” That is Semmelweis’s comparison, applied to a timetable.
- What did your first wording assume or leave out? Give old and new versions to a friend, or a machine: which gets more useful answers?
After you have done the exercise — compare your reasoning
Most first questions fail the first or third check: an unanswerable “why”, or an unchecked “always”, “the best way to…”, “how do I get them to…”. Rewrites tend to become comparisons or measurements. The fifth check matters most when the answer will be multiplied — by a business, a government, or a machine.
Pushing on the model
- In 1961 a team of physicists answered “how large an explosion can we build and set off?” Which of the five checks does that question pass?
- Fritz Haber asked how to make ammonia from the nitrogen in air. If he succeeded, who would gain, and who would pay?
- Alchemists asked how to turn base metals into gold. Bad question, or early one?
After you have written your answers — one way to refine the model
A question can pass four checks and fail the fifth. On 30 October 1961 the Soviet Union exploded the largest bomb ever tested, later called the Tsar Bomba, over Novaya Zemlya in the Arctic: about 50 megatons, from a design for about twice that, scaled down to limit fallout. It was too large to be a practical weapon. The question was answerable, testable and true to physics; only the last check speaks to it. One of its physicists, Andrei Sakharov, became a leading voice against nuclear testing and received the 1975 Nobel Peace Prize.
The same road can fork. Haber answered his question (patent filed 1908), and Carl Bosch at BASF made it industrial. Fertiliser made this way fed about half the world’s people in 2008, by one widely cited estimate. The same ammonia makes explosives: when a British blockade cut Germany off from Chilean nitrates in the First World War, the process kept its munitions supplied. Here the who-pays check gives no verdict. It shows that the answer feeds and kills — and that someone must govern its use after the answer arrives.
A question can be wrong at one level and right at another. Gold by chemistry rests on a false assumption: chemical reactions rearrange atoms but cannot change elements. Ernest Rutherford asked instead what happens when alpha particles strike light atoms, and in 1919 reported nitrogen nuclei breaking apart — the first experimental change of one element into another. (In 1941 physicists made traces of radioactive gold from mercury: real transmutation, worthless as treasure.)
Honest limits of this model
- The checks inform; they do not decide or veto. Whether a cost is worth paying is a question of values, settled with other people. Refusing every question that could be misused would rule out most of science.
- Hindsight is cheap. The check finds only costs someone could foresee or measure, so ask “who pays?” again when new evidence arrives, as Patterson’s did.
- Don’t use them to kill curiosity. Rutherford’s “what happens if…?” needed no checklist. Use the checks before committing years, money, or other people’s safety.
- Some questions are not for evidence. “What makes a life good?” is not testable and still matters; as Lesson 2 showed, values are argued, not measured.
Why this matters now
Machines now answer questions in seconds, for anyone. When answers are cheap, the scarce skill is choosing what to ask. A machine will answer “why me?” with a kind paragraph, answer a loaded question without objecting to its assumption, and optimise exactly the target you hand it. It will rarely ask who pays. That check stays with you — and when paid work no longer sets your questions, you will be choosing your own.
Review & discussion questions
Answer these by explaining, in full sentences, as if teaching someone who missed the lesson. Try them alone first, then discuss with a friend or study group.
- Using the student’s two questions, explain what makes a question answerable, and why an unanswerable one is not harmless.
- Explain how the caloric question could be pursued carefully and lead nowhere. What is a loaded question?
- Why was Joule’s question a better road? Use Lesson 2’s idea of a result that could fail.
- How did Semmelweis reframe the question about childbed fever, and why was his answer resisted?
- The search for a cheap antiknock passed the first three checks. What went wrong? Rewrite the question so its costs are inside it.
- Run the five checks on the Tsar Bomba question and on Haber’s. What can the who-pays check do, and what can it not?
- Take a question you might ask a machine this week. Check it, rewrite it, and explain what changed.
…you can check any question — yours, a colleague’s or a machine’s — for whether it can be answered, whether its answer can be tested, what it assumes, what its success would optimise, and who would pay; and you have rewritten one of your own so that it passes.
Where the facts come from. Every historical case is a documented record; numbers are as reported in the sources below.
- Rumford’s cannon boring (Munich, c. 1797); water boiled in under three hours; heat capacity of chips unchanged: B. Thompson (Count Rumford), “An Inquiry concerning the Source of the Heat which is excited by Friction”, Philosophical Transactions of the Royal Society 88 (1798), pp. 80–102.
- Lavoisier, tin calcined in sealed vessels, 1774: A.-L. Lavoisier, “Mémoire sur la calcination de l’étain dans les vaisseaux fermés, et sur la cause de l’augmentation de poids qu’acquiert ce métal pendant cette opération” (read to the Académie royale des sciences, 1774); Lavoisier archive, CNRS.
- Joule’s paddle-wheel experiments; heat proportional to work; 772 foot-pounds per pound of water per °F (1850): J. P. Joule, “On the Mechanical Equivalent of Heat”, Philosophical Transactions of the Royal Society 140 (1850), pp. 61–82. Modern value about 778 foot-pounds per Btu; Mayer, Helmholtz and the conservation of energy: J. Young, “Heat, work and subtle fluids: a commentary on Joule (1850)”, Phil. Trans. R. Soc. A 373 (2015), 20140348.
- Paris Academy’s 1775 resolution on perpetual motion: Histoire de l’Académie Royale des Sciences, année 1775 (Paris, 1778), pp. 61–66; R. Hahn, The Anatomy of a Scientific Institution: The Paris Academy of Sciences, 1666–1803 (1971), p. 145. US patent practice: USPTO, Manual of Patent Examining Procedure, § 608.03 (“With the exception of cases involving perpetual motion, a model is not ordinarily required…”).
- Semmelweis — clinic figures for 1846 (459 of 4,010; 105 of 3,754); crowding and climate ruled out; street births; Kolletschka’s death; chlorinated lime from mid-May 1847; monthly rates April–August 1847; resistance and later acceptance: I. P. Semmelweis, Die Ätiologie, der Begriff und die Prophylaxis des Kindbettfiebers (1861), translated by K. Codell Carter as Etiology, Concept and Prophylaxis of Childbed Fever (University of Wisconsin Press, 1983); K. C. Carter and B. R. Carter, Childbed Fever: A Scientific Biography of Ignaz Semmelweis (2005).
- Tetraethyl lead — discovery on 9 December 1921; about 1 part in 1,300; ethanol as a known alternative; 1924 deaths at the New Jersey plant; warnings by Hamilton and Henderson; 1925–26 suspension and review: W. Kovarik, “Ethyl-leaded gasoline: how a classic occupational disease became an international public health disaster”, International Journal of Occupational and Environmental Health 11 (4), 2005, pp. 384–397; D. Seyferth, “The Rise and Fall of Tetraethyllead. 2”, Organometallics 22 (2003), pp. 5154–5178. Patterson: C. C. Patterson, “Contaminated and natural lead environments of man”, Archives of Environmental Health 11 (1965), pp. 344–360. US ban for on-road vehicles from 1 January 1996: US EPA, “EPA Takes Final Step in Phaseout of Leaded Gasoline” (29 January 1996). Algeria, July 2021; more than 1.2 million premature deaths prevented a year: UNEP, “Era of leaded petrol over, eliminating a major threat to human and planetary health” (30 August 2021). India by 2000: UNEP Partnership for Clean Fuels and Vehicles, leaded petrol phase-out status reports.
- Tsar Bomba — 30 October 1961, Novaya Zemlya; about 50 megatons; design for about 100 megatons reduced to limit fallout; impractical as a weapon: CTBTO Preparatory Commission, “30 October 1961 – The Tsar Bomba”; V. Adamsky and Yu. Smirnov, “Moscow’s Biggest Bomb: the 50-Megaton Test of October 1961”, Cold War International History Project Bulletin 4 (1994); Yu. Khariton and Yu. Smirnov, “The Khariton Version”, Bulletin of the Atomic Scientists 49 (4), 1993. Sakharov: Nobel Peace Prize 1975, NobelPrize.org.
- Haber–Bosch — 1908 patent; Bosch at BASF; munitions after the blockade of Chilean nitrates: Science History Institute, “Fritz Haber”; NobelPrize.org. About half of humanity fed (48% in 2008): J. W. Erisman et al., “How a century of ammonia synthesis changed the world”, Nature Geoscience 1 (2008), pp. 636–639.
- Rutherford, nitrogen disintegrated by alpha particles (1919): E. Rutherford, “Collision of α particles with light atoms. IV. An anomalous effect in nitrogen”, Philosophical Magazine, 6th series, 37 (1919), pp. 581–587. Oxygen identified (1925): P. M. S. Blackett, “The ejection of protons from nitrogen nuclei, photographed by the Wilson method”, Proceedings of the Royal Society A 107 (1925), pp. 349–360. Radioactive gold from mercury (1941): R. Sherr, K. T. Bainbridge and H. H. Anderson, “Transmutation of mercury by fast neutrons”, Physical Review 60 (1941), pp. 473–479.
- The ice trade and framing by function: see the sources of First principles, Lesson 2.
Attribution. Lesson structure — Foundation, questions, observations and deductions, refined models — adapted from John S. Hutchinson, Concept Development Studies in Chemistry (Connexions / Rice University), licensed under Creative Commons Attribution 2.0 (CC BY 2.0). The topic, examples, and text of this lesson are otherwise original to learn.curiosta.com.