36 Essentials

Humanity went thousands of years without answers not because people were unintelligent, but because there was no method for asking. While disputes are settled by who holds more authority or which belief is older, wrong ideas survive just as long as right ones.

The scientific method did one simple thing. It turned claims into the form "if this were true, what would we have to observe?", and agreed to drop the claim when the observation fails to appear. Experiments, control groups, replication and peer review are all machinery for keeping that one promise.

Science is therefore not a collection of certainties but a way of handling doubt. An idea becomes testable only at the moment you can say in advance what result would make you abandon it.

Some explanations fit every outcome. If things go well, the theory gets the credit; if they go badly, conditions were not right. It sounds powerful, but nothing has actually been staked on it.

Useful theories take the opposite risk. A theory that says when and where an eclipse will be visible collapses if the sun stays put that day. Being able to collapse is not a weakness; it is precisely the measure of how much the theory says.

So when you hear an explanation, the question is not how plausible it sounds but what it forbids. The more outcomes it rules out, the more it is actually telling you.

Ice cream sales and drowning accidents climb together. Ice cream is not dangerous; heat drives both. Something sitting behind two variables and moving them at once is called a confounder.

Direction is a problem too. Whether exercise makes people healthy or health makes exercise possible cannot be settled by correlation. Reverse the order and the graph looks exactly the same.

To speak about causes you need intervention: split similar people at random, change the condition for one group only, and compare. Random assignment spreads even the confounders you have not thought of evenly across both sides.

"Will it rain?" has no yes-or-no answer. The answer is "in conditions like these, about three times in ten." Almost every real question has this shape, yet we keep demanding certainty and end up with no answer at all.

Thinking in probabilities lets you update. Something you put at 30% can move to 60% when evidence arrives. You are nudging a dial rather than replacing a belief, which hurts far less and can therefore be done far more often.

The other essential point is how rare events deceive. A test that is 99% accurate, applied to a disease that affects one person in ten thousand, will produce far more false positives than true ones. When the base rate is low, even an accurate test is usually wrong.

Learn a little of a field and it feels as if the whole of it is in view, because you cannot yet see what is missing. Go deeper and the list of things you do not know grows first; only then does confidence start to fall.

Confidence therefore does not track knowledge. In every field you will find the person who has studied for decades hedging while the newcomer states things flatly. Reading certainty as a signal of competence will mislead you often.

The fix is not only to learn more. Attach a confidence level to your judgements and check them against outcomes later. Recording how sure you were, rather than just whether you were right, is where self-correction begins.

Ask someone how many hair salons there are in their country, with no data to hand, and most give up. Yet take the population, guess how often a person goes, estimate how many customers one salon serves a day, then multiply and divide, and you land within the right order of magnitude.

The value of such estimates is not the answer. It is being able to notice when someone else's number is absurd. Told that a business earns a trillion a year, a crude multiplication of customers by average spend usually settles whether that is even possible.

All you need is a handful of anchors: the population where you live, a typical monthly income, the fact that a day holds 86,400 seconds, and that a human life is roughly 30,000 days. With those you can sanity-check most numbers you meet.

If we both defect we both lose, but if you cooperate while I defect, I come out ahead. So in a single meeting everyone defects and everyone ends up worse. This structure has been one of humanity's oldest problems.

Meet the same person repeatedly and the answer inverts, because what you stand to lose from tomorrow onwards outweighs today's gain. In tournaments, the strategy that endured best was simple: cooperate first, answer a betrayal once, and forgive immediately when the other side returns.

Cooperation is therefore a question of structure, not of virtue. People help each other where reputation persists, where they expect to meet again, and where betrayal becomes visible. The same people behave differently when interactions are anonymous, one-off and unobserved.

On pasture open to all, adding one more cow is a gain for me, because the cost of thinner grass is shared by everyone. When everyone runs the same calculation the pasture disappears. Fisheries, groundwater and clean air have the same shape.

For a long time the problem was thought to have only two solutions: state regulation, or carving the commons into private property. Research on actual villages found a third. The people who use the resource write their own rules, watch each other, and sanction those who break them.

Those communities shared features: clear boundaries, rules made by the users themselves, violations that are visible, and penalties that start light and grow with repetition.

Suppose you live with someone better than you at both cooking and cleaning. It still pays to divide the work, because every hour they spend cleaning is an hour not spent cooking. The measure is not what someone is good at but what they give up during that time.

This is comparative advantage. When each person takes the task whose sacrifice is smallest and trades for the rest, both end up with more than before, even when ability is lopsided. Trade between countries and division of labour inside a team run on the same principle.

But a bigger total does not mean everyone gains. If the gains are not shared, whoever lost their position is worse off unless they are compensated. Most arguments about trade are about distribution, not about the size of the pie.

If people only trade with those they know, markets never outgrow a village. Being able to send money to someone you met today and wait for goods is not a matter of trusting them. It works because machinery exists that makes breaking the deal costly.

Contracts, registries, guarantees, refund rules, dispute resolution and credit records are that machinery. None of it makes anyone virtuous. It makes cheating the more expensive option, so exchange can happen without trust.

Where institutions are weak, people retreat into kinship and old acquaintance. That is safe, but it shrinks the pool of possible partners and with it the range of opportunities. The value of institutions is that they let cooperation extend to strangers.

Light a lighthouse and every passing ship benefits. You cannot switch the beam off for the ships that did not pay. Nobody wants to carry the cost of such a thing first, because if someone else builds it you can simply use it.

Basic research, disease control, clean air, public safety, standards and weights all behave this way. Left to a market, less is produced than is needed. This is why taxation and shared contribution exist at all.

Treating free riding purely as a moral failing leads nowhere, because not paying really is the rational individual choice. What has worked in practice is collecting the cost from everyone automatically and making it visible where the money went.

Framed as a problem of choosing good leaders, it has no solution. Good people change as power accumulates, and you cannot choose their successor. Societies that lasted solved a different problem: how to limit the damage a bad ruler can do, and how to remove one without bloodshed.

The answer was to split power, to fix terms, and to write the rules before they are needed. The essential part is that whoever made the rules is also bound by them, and that changing the rules is deliberately made difficult.

This is not a system built for efficiency. Checks and procedures are slow. What they buy is that a wrong decision does not become irreversible. It is a trade of speed for recoverability.

During a cholera outbreak in 19th-century London, a doctor plotted the victims' homes on a map and saw them clustered around one water pump. This was before germs were known, yet removing the pump handle broke the outbreak. Cutting the route of transmission was enough without understanding the cause.

The near doubling of average life expectancy in the 20th century owes more to piped water, sewers, handwashing, disinfection and better nutrition than to dramatic surgery or new drugs.

Above all, fewer infants and children died, which pulled the average up sharply. Even when life expectancy was 35, plenty of people who reached adulthood lived into their sixties. What was short was not adult life but a child's odds of surviving.

A vaccine shows the immune system the markings of a pathogen in a form that cannot cause the disease. The body remembers, and responds immediately when the real thing arrives. You acquire the memory without paying for it with the illness.

There is a second effect. When most people around a carrier are immune, the pathogen cannot find its next host and transmission stops. Newborns and people in treatment who cannot be vaccinated are protected too. It is a rare case where an individual choice manufactures a public good.

Smallpox is the only human disease erased from the planet this way. It had killed more than 300 million people in the 20th century alone.

Only three things are required: variation that is passed to offspring, variation that changes the odds of surviving and reproducing, and enough time. Given those, forms fitted to the environment accumulate without anyone planning them.

The common misreading is that evolution moves toward perfection. It only ever picks what is slightly better right now. That is why the human eye has a blind spot where the optic nerve crosses in front of the retina, and why one nerve in the neck takes a needlessly long detour. These are the marks of repair, not design.

Evolution also has no purpose. Nothing happens "in order to improve"; what remained simply remained. Once you see the structure, antibiotic resistance and pesticide-resistant insects become the same phenomenon.

It feels as though memory is a file pulled from a drawer, but each recollection is reassembled from fragments. Things you heard afterwards or photographs you saw later get mixed into those fragments, and you cannot feel the seam.

Vividness is not accuracy either. Memories of shocking events feel especially sharp, yet when checked against records years later the details have often shifted considerably. Confidence and correctness move independently.

Convictions based on eyewitness testimony that were later overturned exposed this. As a result the way witnesses are questioned has changed: avoid questions that suggest an answer, and take the account as early as possible, before it is contaminated.

If you take something for a cold and feel better three days later, you cannot tell whether it helped or whether you were due to recover. People start treatments when symptoms peak, and things usually drift back toward average afterwards. Whatever you did will appear to have worked.

Expectation adds to this. Taking something with no active ingredient, believing it is medicine, genuinely reduces reported pain and improves symptoms. That is why comparisons are run so that neither the patient nor the doctor knows which is which.

So personal testimony, however honest, cannot establish that something works. People who did not improve rarely write reviews, so the reviews you see are always tilted toward those who did.

Set the Earth's formation at midnight and now at the following midnight. Life shows up around four in the morning, then stays single-celled for almost the entire day. Organisms large enough to see only proliferate after nine in the evening.

Dinosaurs arrive around 10:40 pm and are gone by 11:40. The human lineage splits off about 80 seconds before midnight. Farming begins in the final second and a half; written records fall inside the last second.

Seeing this scale once leaves two impressions at the same time: how brief our stretch of time is, and how unusual what was built inside it turns out to be.

Fold a sheet of paper 42 times and it reaches the Moon. You cannot actually fold it that often, but the arithmetic holds. Our minds cannot track something that doubles each step, because only straight-line growth feels natural.

So exponential processes look like nothing is happening for a long time and then seem to arrive all at once. Epidemics behave this way, so does compound interest, so does the performance of technology. In reality the rate never changed; it simply took time to reach a noticeable size.

The reverse is identical. Something shrinking a few percent a year shows no visible sign for a while and then passes the point of recovery. Recognising which regime you are in changes when you should act.

A person working hard all day produces about enough energy to run a few light bulbs. Yet we live in a world where one person accomplishes the work of dozens. The difference is not ability but a chain of energy sources: fire, draught animals, watermills, coal, electricity.

Every time energy became cheaper, the shape of society changed with it. Freeing people from grinding grain let them do something else, and that surplus produced technology and scholarship. Most of the reduction in working hours came from here too.

Energy is never created, only converted. Whatever you use is taken from somewhere, and after use it mostly disperses as heat that is hard to gather again. So the questions of where it comes from and what it costs always travel together.

Some things do nothing on almost every day and then, very rarely, decide everything: earthquakes, financial crises, pandemics, serious accidents. Judged from typical data alone, such events look as if they do not exist.

The difficulty is that these losses cannot be undone. Winning ninety-nine times and losing everything once makes the ninety-nine irrelevant. However good the average return, multiplying by zero even once leaves zero.

So the criterion changes. The question is not whether the expected value is high but whether you can continue after the worst case. Reserves, diversification, insurance, spare parts and contingency plans are all trades of a little return for survival.

When knowledge travels only by memory, what someone knew dies with them. However brilliant the next person is, they must start again from roughly the same place. What writing did was stop that reset.

Text crosses distance as well as time. You can convey a thought precisely to someone you never met, and unlike speech it can be reread and examined. Law, contracts and scientific papers, everything that needs precision, was built on top of writing.

Writing is also not merely transcription of thought. The moment you try to form sentences, the gaps show. What felt coherent in your head frequently collapses on the page.

When books were copied by hand, one cost as much as a house. Knowledge belonged to a tiny few, and each copy drifted further from the original as errors accumulated. Printing cut the price by a factor of hundreds and produced many identical copies.

Once many people could read the same text at once, they could compare and contradict each other. Verification became possible. The Reformation and the scientific revolution both stood on that condition. The change in the cost of copying came before the change in content.

Today copying costs essentially nothing, so good and bad spread at the same speed. The bottleneck is no longer access but filtering.

New knowledge does not settle in a vacuum; it attaches to what is already there. The broader your base, the more places there are for it to stick, so the same hour of study leaves more behind. This is why the beginning is always the slowest and hardest part.

Early slowness should therefore not be read as lack of talent. That stretch is where the attachment points are being built. The gap between people who get past it and people who do not widens over time, and it comes from persistence far more than from ability.

The same holds for a society. Money spent on education does not stop with that generation; it raises the point at which the next one starts. Humanity advanced quickly not because individuals became cleverer but because the starting line kept moving up.

A free ticket does not make the evening free. The price is the best thing you could have done with those hours. No money changed hands, so it is invisible, but the cost was paid all the same.

This view turns "should I?" into "this or that?". Taken alone, almost any option looks acceptable. Ranking only appears once an alternative sits beside it. Deciding whether to spend two hours in a meeting changes once you write down what those two hours would otherwise buy.

The mismatch is worst with time. Money can be earned again and time cannot, yet we attach a price to money and treat hours as free.

Judge a call centre on calls handled and staff end calls quickly. Judge a hospital on mortality and it admits fewer critical patients. None of this requires anyone to cheat. What is measured simply becomes the goal.

So once a number is the target, it no longer measures the thing it used to. A mortality rate that fell because care was good can now also fall by selecting patients. The figure improves while reality worsens.

The remedy is not only finer metrics. It is watching several measures that pull against each other, and having people look directly at whatever the numbers cannot describe.

Flight instructors long said that praising a good landing produced a worse one next time, while criticising a bad landing produced a better one. It genuinely looks that way. But this is not the effect of praise or criticism.

An exceptionally good day combines skill with luck. Next time the luck returns to normal and the score drops. An exceptionally bad day rises for the same reason. It happens whether or not anyone intervenes. This is regression to the mean.

So any improvement after intervening at an extreme mixes the real effect with this natural drift back. Separating them requires comparing against a similar group where nothing was done.

Finding a note and losing the same note are not equal in weight. Across many experiments, the loss registers close to twice as strongly. As a result people take larger risks in order to avoid booking a loss.

This tendency also inflates the value of whatever you already hold. The price at which you would sell an object jumps well above the price you would have paid for it moments earlier. It is why clutter accumulates, why losing positions are held, and why outdated methods survive.

Wording alone shifts decisions. "90% survive" and "10% die" state the same fact and produce different choices. Rewriting something framed as a gain into the language of loss exposes the standard you were actually using.

Offered 100 today or 110 in a month, many people take today. Offered 100 in a year or 110 in a year and a month, most take the later one. The waiting period is identical and the answer flips.

We discount value steeply in the near term. That is why the plan made last night is broken by this morning's self, and today's plan is broken by tomorrow's. It is not weak will; the arithmetic genuinely differs depending on when you run it.

So what works is not firmer resolve but removing options from your future self. Automatic transfers, rules set in advance and commitments that are hard to undo are all devices of this kind.

A map as detailed as the land would be as large as the land and therefore useless. Maps work because they discard. For someone who knows what was discarded they are tools; for someone who does not, they are traps.

Economic models, personality types, org charts and statistical summaries are all maps. Sorting people into four letters makes conversation easier, while most of a person is whatever those four letters left out. We use summaries because they are tractable, not because they are accurate.

So one question travels with every model: what is it deliberately ignoring? While you can answer that, the model assists your thinking. The moment you cannot, it starts replacing it.

A bounty on venomous snakes, the story goes, led people to breed them. When the bounty ended the snakes were released and the population ended up larger than before. Whether or not the anecdote is true, the structure is common.

Unintended consequences usually arise in three ways: people find a route around the rule, people feel safer and therefore act more riskily, and blocking a problem in one place moves it to another.

So the first question when designing a rule is how it will be exploited. Not because people are malicious, but because choosing what benefits you within a set of rules is simply what everyone does.

In wartime, someone proposed armouring the parts of returning aircraft that had the most bullet holes. A statistician objected: those were the places a plane could be hit and still come back. The armour belonged where there were no holes, because planes hit there did not return.

Collecting the traits of successful people to distil a formula is the same trap. Everyone who did the same things and failed is absent from the interviews. Looking only at survivors, any trait at all looks like a cause of success.

Surveys behave this way too. Only those who respond are counted, so the interested and the aggrieved are over-represented. If the sample does not represent the population, no quantity of responses makes it accurate.

Cut inventory and consolidate to one supplier and costs fall. Then that one supplier stops and everything stops. Plant a single crop variety and yields rise, until a disease suited to it arrives and the whole field fails at once.

Efficiency and resilience largely trade against each other. Spare stock, a second supplier, unused backup equipment and overlapping staff all look like pure cost on an ordinary day. What that cost was buying only becomes visible during a failure.

So judging by current efficiency alone is not enough. The companion question is how far the stoppage spreads when any single part stops.

When people share a background, conversation moves fast and conclusions arrive quickly. They also take the same things for granted and therefore share the same blind spots. Nobody objects, not because nothing is wrong, but because nobody can see it.

Seat belts and airbags were tested for years against an adult male body model, leaving other builds at higher risk. Drug dosages set from data on one sex produced the same kind of failure. Both were blind spots, not malice.

Groups with mixed perspectives decide more slowly and hold less comfortable meetings. That discomfort is the price. What it buys is a lower chance of a large error passing through untouched.

A hundred kept promises are a hundred small pieces of evidence. One broken promise makes all hundred get reinterpreted, because "perhaps it was like that then too" becomes possible. That is why recovery takes longer than the original building did.

Concealment that comes to light is the worst case. The hiding, more than the fact itself, signals that hiding will continue. Bad news does the least damage when you deliver it yourself, first.

Organisations follow the same rule. A single cover-up puts a decade of previous statements in doubt. This is why durable institutions choose not to punish whoever raises bad news early.

Screw threads, plug shapes, paper sizes, the order of a date, the way letters map to numbers. These exist not because someone invented them but because people agreed to match. Without that, every part would need its own factory.

The value of a standard does not lie in it being the best option. Even when something better exists, whatever everyone already uses wins. Keyboard layouts and railway gauges are familiar cases where an early accident hardened into something hard to change.

So when choosing a standard, reversibility matters as much as performance. Once set, society often cannot afford the cost of switching, even to something clearly better.