Skip to content

← Thinking Clearly

What facts can't settle

Chapter 25 · How science decides

In this chapter 4 sections
  1. The missing “should”
  2. Other ways of being right
  3. What a grandmother knows
  4. The claim, not the person

From the last day of 2025 to late April 2026, cameras strapped to trees in the forests of Nepal’s lowlands took pictures of whatever walked past them. The photographs told one animal from another, and at the end of July the count came out. Nepal had about 429 wild tigers, more than ever recorded. In 2010 it had about 121. Twelve years later it had reached 355, more than double, and so kept a promise it had made with twelve other countries to double its tigers by 2022.

Not everyone cheered. Tigers kill people. In the eight years to July 2026, tigers killed 85 people in Nepal, ten of them in the last of those years, in Banke, Parsa, Bardiya and Chitwan. People in the villages around the parks worry that more tigers will mean more attacks on them and their cattle. Some conservationists think the parks are close to holding all the tigers they can, and one wildlife expert reckons Nepal could keep 400 to 500. In Bardiya, 112 tigers share 968 square kilometers, seven or eight square kilometers each, far less than a tiger was supposed to need. “The reality we are witnessing today does not match what we learned from ecological theory,” said the park’s chief conservation officer.

So how many tigers should Nepal have?

Science has a great deal to say about that question, and it is good at saying it. It counted the tigers. It can count the deer and wild pigs they eat, map where the attacks happen, and find out whether more tigers really do mean more deaths. One researcher points out that in the past they haven’t always gone together. It can test what makes the forest safer: in one community forest beside a national park, people now go in only after nine in the morning, at fixed times, and the undergrowth is cleared so that a tiger can be seen. Each of these is a question about what is, or what will happen if. Each has an answer, even where nobody has found it yet.

Now imagine every one of them answered perfectly. Suppose you knew exactly how many tigers the forests could feed, and how many people would be killed each year with 300 tigers, or 400, or 500. You still wouldn’t know how many to have. That depends on how much a tiger is worth, set against a person’s safety, a farmer’s buffalo, the money tourists bring, and the plain fact that one of the world’s great animals still lives here. Two people can agree on every number and still answer differently, and neither has made a mistake about the facts.

The missing “should”

In 1740 the Scottish philosopher David Hume noticed something about the books on morality he had been reading. Each author would reason along for a while about what is and what is not, how people live, how the world goes. Then, all of a sudden, every sentence had an “ought” or an “ought not” in it. “This change is imperceptible,” Hume wrote, “but is, however, of the last consequence.” The facts had been about what is. The conclusion was about what ought to be, and nobody had explained how one led to the other.

Philosophers still argue about exactly what Hume meant, and that argument belongs to them. The everyday lesson is simpler. An argument that ends in “should” always has a “should” somewhere among its reasons, often one that nobody says out loud. “Attacks will rise, so Nepal should keep fewer tigers” leans on an unspoken step: keeping people safe matters more than having more tigers. “Tigers are endangered, so they should be protected whatever it costs” leans on another: we ought not to let a species die out, even at that price. Say the hidden step, and a shouting match about numbers turns into a disagreement about what matters, which is where it was all along.

None of this makes facts less important. They decide what a choice costs, and that can change people’s minds without changing anyone’s values. If rules about when to enter the forest cut the attacks sharply, more tigers cost fewer lives than people feared, and someone who cared about exactly the same things as before might now want more tigers. Facts also rule choices out. Wanting more tigers than the forests can feed is wanting something that can’t be had.

Most real quarrels are a tangle of the two, and the useful thing is to pull them apart. Which part of this could a measurement settle, and which part couldn’t? The two mistakes from sorting claims into kinds come back here. A quarrel goes wrong when someone presents a value as a fact (“science says we must”), because science says what will happen if, and the “must” always comes from somebody. It goes just as wrong the other way, when a fact is waved off as “just your opinion”.

Other ways of being right

Looking, testing and measuring is one way of being right about things. It isn’t the only one, and some of the oldest knowledge we have came by other roads.

Mathematics doesn’t use experiments at all. About 2,300 years ago Euclid showed that the prime numbers never run out, and the idea fits in a few lines. Take any list of primes, multiply them all together and add one. The new number leaves a remainder of one when divided by any prime on the list, so its prime factors, whatever they are, can’t be on the list. Whatever list you start with, there is always a prime missing from it. That isn’t evidence that makes the claim likely. It makes it certain, for every number, including numbers too big for anyone ever to write down.

Evidence, however much of it, isn’t enough for a mathematician. In the 1770s Leonhard Euler published the rule “a number squared, minus itself, plus 41”. Put in 1 and you get 41, a prime. Put in 2 and you get 43, then 47, 53, 61, and every answer is prime, forty times in a row. In a laboratory, forty out of forty would be a strong result. Put in 41 and you get 41 times 41, which is 1,681 and not a prime at all. There is a pattern about numbers, named after the mathematician George Pólya, that holds for every number anyone could check by hand, and for millions more, and then breaks for the first time at 906,150,257. So mathematics asks for a proof, and a proof, once it has been checked, doesn’t need to be revised when new evidence comes in, because there is no new evidence to come.

History has the opposite problem. Nothing in it can be run again. In May 1992, workers digging the foundation of a house in Maligaon, in eastern Kathmandu, found a life-size statue of pale sandstone lying face down about a meter under the ground, in four large pieces. On its base was a short inscription naming a great king, Jaya Varman, and a year: 107.

For about a century, Manadeva’s pillar at Changu Narayan, from 464 CE, had been taken as the oldest dated record of a king in the Valley, and by 1992 only a pillar inscription from Pashupati, five years older, had beaten it. If the year 107 counts from the same starting point as the Licchavi inscriptions that came later, the Shaka era, the statue was made in about 185 CE, and the written record of kings in the Valley goes back nearly three hundred years further. Most scholars read it that way, say the two who first published the statue. Not all. One Nepali historian argued for a later era, which would put it in the fourth or fifth century, two others read the number as 207, and another thinks the figure isn’t a king at all but a यक्ष (yakṣa), a nature spirit. A chronicle written down many centuries later, the गोपालराजवंशावली (Gopālarājavaṃśāvalī), lists a king named Jaya Varman. Historians are right the way a good detective is: by fitting together the script, the carving, the other records and the dates, saying how sure they are, and saying plainly where the evidence runs out.

Art is different again. No instrument measures whether a पौभा (paubha), a Newar religious painting, is beautiful, and two people who know a great deal can disagree. That doesn’t make every opinion equal. A good critic gives reasons you can check for yourself by looking: the line of a hand, how the colors balance, what the painter did with a tradition’s rules. You may still disagree, but you will see more.

And some knowledge is in the hands rather than in sentences. In Jumla, farmers grow rice at up to 3,050 meters, at Chhumchaur, among the highest rice fields in the world, where nights at sowing time still fall to about 5 °C. Around the twelfth of Chaitra, late March, they soak the seed in jute bags and keep it near the fire for four or five days until it sprouts. Then they scatter it on wet nursery beds, cover it with compost and spread the ash of burnt cow dung over the beds, which, a study of these fields explains, keeps them warm enough for the seedlings to come up and grow. Nobody needed a randomized trial to learn this. A farmer whose seed failed would know within weeks, and so would the neighbors. That is knowing how, and it is knowledge whether or not anyone ever wrote it down.

What a grandmother knows

On Simeulue, an island off the coast of Sumatra in Indonesia, the old people tell of 1907. That year a great earthquake was followed by the sea coming in over the land. Nobody knows how many died, but the stories say up to seven in ten of the islanders, some found later in the tops of coconut trees. The islanders have a word for it, smong, “the ocean coming onto the land”, and the stories came with a lesson that was handed down for almost a century: when the ground shakes hard, run to the hills.

On 26 December 2004 one of the largest earthquakes ever recorded struck about 40 kilometers from the island’s northern tip. At Langi, the village closest to it, the sea came ashore more than ten meters high less than ten minutes after the shaking stopped, and left nothing of the village but its concrete foundations. By then the people were on the high ground. Of Langi’s 800 people, nobody died. Of the island’s 78,128 people, the tsunami killed seven. A warning system would have been no help. There was no time for one.

On the mainland, at Jantang, 225 kilometers from the center of the earthquake, by some reports more than half of the 10,000 people died. The survivors the scientists spoke to hadn’t known that shaking could bring the sea. They heard sounds from the ocean like gunshots and took them for the fighting that had gone on there for years, so they stayed in their houses.

So the islanders’ grandmothers knew something true and important that the experts of their own government had failed to act on. It’s worth asking why their knowledge was so good, because the answer also shows where such knowledge goes wrong. In 2009 two psychologists from schools of research usually seen as opposed, Daniel Kahneman, who studied the mistakes in people’s judgment, and Gary Klein, who studied how experts decide in real situations, wrote a paper together about when experience can be trusted. Their answer had two parts. The world has to be regular enough to be predictable, and the person has to have had the chance to learn its regularities. The sea that follows an earthquake is about as regular as anything gets, and the lesson was written in the dead. The rice in Jumla gives its answer every single year. Where the world teaches clearly, a tradition can hold knowledge that took generations to gather, and nobody should sneer at it.

Cures are the hard case. Most people who take a remedy get better anyway, the ones it didn’t help rarely tell the story, and the answers come slowly and mixed up with everything else in a person’s life. Generations of careful watching can’t untangle that, which is how bloodletting lasted two thousand years among clever doctors. So a tradition is a fine place to start looking for medicines. That is how the malaria drug artemisinin was found, when an old Chinese handbook pointed Tu Youyou’s team to a plant and to a way of preparing it, and tests on mice and then on patients showed that it worked. But each specific claim that a thing cures something can still be put to a fair test, and the test can come out either way.

Honey for a child’s cough came out fairly well. In 2018 reviewers brought together six randomized trials with 899 children, all older than a year, and found that honey probably eases coughing better than nothing or a dummy syrup, and may work about as well as dextromethorphan, a common cough medicine. The evidence was moderate rather than strong, and most of the children had it for just one night. (Honey isn’t for babies under one. It can carry bacteria that make a baby seriously ill.)

Guggul came out less well. It is a resin from a shrub, Commiphora mukul, long used in Ayurveda, and widely taken in Asia to lower cholesterol. There was even a good reason to expect it to work: experiments had found that chemicals in it act on proteins the body uses to handle cholesterol. In 2000 and 2001, doctors in Philadelphia gave 103 adults with high cholesterol either guggul or a dummy pill, chosen by chance, for eight weeks, with nobody knowing who had which. The “bad” cholesterol of the dummy group fell by 5%. In the guggul groups it rose by 4 to 5%, and six people taking guggul broke out in a rash, against none on the dummy. Some earlier studies had found a benefit, so the evidence on guggul is mixed rather than closed, and one trial of Americans on an American diet doesn’t answer every question. But it answered the one it asked.

Neither test said anything about Ayurveda as a whole, or about anyone’s grandmother. Each tested one claim, about one remedy, for one problem. That is all a fair test ever does, and it is the reason it is fair.

The claim, not the person

The grandmother who swears by a remedy isn’t foolish. She watched carefully, often for a lifetime. What she lacked was a control group, and for most of history nobody had one, including the best-trained doctors in the world. The islanders of Simeulue and the doctors who bled their patients were using the same kind of knowledge, long watching passed down. The islanders were right and the doctors were wrong because of the question each was asking, not because one lot were cleverer.

That’s why sneering gets things wrong twice. It mixes up the claim with the person, so that a question about whether a remedy works turns into a question about who is stupid, which nobody wants to lose in front of their family. And it throws away too much. Someone who laughs at everything old would have laughed at smong too.

Testing a claim is a way of taking it seriously. It says that what this person believes matters enough to find out whether it’s true. A remedy that passes the test has won something no story could give it. One that fails has told us something, and the person who offered it has lost nothing but a mistaken idea, something every person carries now and then.

Back in the forests of the Tarai, the cameras will go up again, and the counts will get better, and so will the rules that keep people safe. Each new fact will change what the choices cost. None of them will say how many tigers are worth how much risk, or whose risk it should be. That part will be settled the way such things always are, by people weighing what they care about, and it goes best with all the facts in hand and an honest word for the part that facts can’t settle.

Found a mistake? Please tell us.