Chapter 13: How to Predict the Future

In the 1860s, chemists faced a mess. Sixty-three known elements, each with its own quirks, and no logic connecting them. Dmitri Mendeleev refused to accept that nature was chaotic. He arranged the elements by atomic mass and valence, and a structure emerged: the properties of each element repeated in a predictable rhythm. He called it the Periodic Law.

What Mendeleev did next revealed the true power of his discovery. He didn't just organize what was known—he used the pattern to see what wasn't. Where the table showed gaps, he predicted entire elements: their weight, their density, even how they would react with other substances. Critics mocked him. Anyone claiming to know the properties of elements no one had ever seen seemed absurd, even delusional. Mendeleev held his ground. In 1875, gallium was discovered. In 1886, germanium followed. Both slid into his empty slots as if built for them, matching his predictions almost exactly. He had seen a pattern invisible to everyone else, and trusted it enough to stake his reputation on elements that didn't yet exist.

But a pattern by itself is not enough, and this is where most admirers of Mendeleev misread him. He did not merely notice that properties repeated. He had a claim about why they repeated, and that claim—not the pattern—told him what belonged in the gaps. The distinction between seeing a pattern and understanding one is the distinction between a guess and a prediction worth staking a reputation on.

Explanatory Reasoning

In 1870, Jules Verne sent his characters in a submarine toward the South Pole, chasing an answer to one of the era's great mysteries: was Antarctica a landmass, or just a frozen sea like the Arctic? No one had been there. No one could check. And yet Verne's characters worked it out from their armchair, using nothing but reasoning.

Their argument ran like this: icebergs form only from glaciers, and glaciers form only on land. Near the North Pole, glaciers are scarce, confined to places like Greenland, so icebergs there are rare. But near the South Pole, icebergs are abundant. If icebergs require land to form, then somewhere beneath that ice must lie a continent, unseen and undiscovered. They were right. Beneath the ice lay a landmass, though decades of exploration would pass before anyone could prove it.

What makes this remarkable isn't that the guess turned out correct. It's how the conclusion was reached. There was no statistical pattern here, no "we've found continents everywhere else, so we should find one here too." That would be induction: extrapolating from past instances to predict a future one. This was something sharper. It was an explanation: a specific claim about what icebergs are, how they form, and why their abundance demands a hidden landmass. Change any detail of that explanation and it collapses. That rigidity is exactly what makes it powerful.

This distinction, between prediction by pattern and prediction by explanation, marks the difference between guessing and understanding. Probability works when you have large numbers of identical, repeatable events: coin flips, dice rolls, actuarial tables. But the economist Frank Knight built a career on a warning that almost nothing in real life fits that mold. "Any given instance is so entirely unique," he wrote, "that there are no others, or not a sufficient number, to make it possible to tabulate enough like it to form a basis for any inference of value about any real probability in the case we are interested in." A business decision, a marriage, a war: each occurs once, under conditions that will never recur in quite the same combination. Treating a forecast like "we are sixty percent certain profits will rise" as an objective probability is not merely imprecise. It is meaningless, and dangerously so if mistaken for genuine certainty. John Maynard Keynes, writing from a rival school of economics, arrived at the same place in his 1921 A Treatise on Probability: real uncertainty is too irreducible, too qualitative, to fit inside a formula.

Understanding, by contrast, doesn't require memorizing outcomes at all. No astronomer has ever memorized every recorded planetary position, yet many understand planetary motion completely, because understanding rests not on accumulating facts but on possessing the right concepts and theories. A single formula from general relativity can predict planetary positions more accurately than any archive of past observations, and it applies to cases that have never even been recorded. But knowing the formula is not the same as understanding gravity. The formula only becomes meaningful once it is explained: once someone describes what gravity actually is, a curvature in four-dimensional space and time, and why matter bends it that way. The predictions are a byproduct of that explanation, not its purpose.

Science rejects most bad ideas without ever testing them. Consider the claim that eating a kilogram of grass cures the common cold. It is technically falsifiable—you could test it and prove it wrong—but no one bothers, because it offers no explanation of how or why it would work. An infinite number of equally untested, equally baseless claims could be invented tomorrow. What separates science from noise is not merely the capacity to predict, but the demand for an explanation that cannot be casually altered without falling apart. As physicist David Deutsch puts it, a good explanation is hard to vary: tweak the details and the whole account breaks down. That is the quality Verne's characters stumbled onto with their icebergs.

Prediction, then, is not the goal. It is a consequence, one of many, that flows from understanding how something works. Mistaking prediction for the point is like mistaking a spaceship's fuel consumption for its purpose: burning fuel is necessary, but the mission is to deliver the payload.

The Limits of Inductivism

Suppose I told you I had just completed a rigorous statistical study of Barack Obama's mortality. Sixty-four years, more than 23,000 daily observations, and not a single death recorded. By any conventional measure of statistical significance, I could confidently declare the man immortal. The absurdity of that conclusion reveals something most people get backward about how knowledge works: more confirming evidence does not make a hypothesis truer. It just makes the eventual surprise more devastating.

This is the black swan problem, and it predates Nassim Taleb by centuries. David Hume raised it in A Treatise of Human Nature, and John Stuart Mill later sharpened it into its classic form: no number of white swans observed can prove that all swans are white, but a single black swan is enough to destroy the claim entirely. For most of European history, "all swans are white" looked like settled fact, until explorers reached Australia and encountered Cygnus atratus, jet black and unmistakably real. Thousands of confirming instances collapsed the instant one counterexample appeared.

None of this makes observation worthless. It works, but in only one direction. Consider two statements: "No swan is black, because I've examined four thousand swans and found none," versus "Not all swans are white." The first can never be proven, no matter how many white swans you catalog; you'd need to inspect every swan that has ever existed or ever will. The second requires only one specimen to confirm. As Taleb puts it in Fooled by Randomness: "I can use data to disprove a proposition, never to prove one. I can use history to refute a conjecture, never to affirm it."

This asymmetry is the foundation of real scientific reasoning, and it has nothing to do with piling up trials. Mix baking soda and vinegar and you'll watch carbon dioxide bubble to the surface. Repeating that reaction in a million kitchens worldwide adds no new knowledge; what matters is the chemistry explaining why it happens, an explanation strong enough to be falsified if wrong. Science advances by proposing bold explanations and hunting for the flaw that breaks them, not by accumulating instances that merely fail to contradict what we already believe.

Taleb, in Antifragile, extends this into a broader principle: knowledge grows more reliably by subtraction than by addition. "We know a lot more what is wrong than what is right," he writes, because "negative knowledge, what is wrong, what does not work, is more robust to error than positive knowledge." Disconfirmation is rigorous. Confirmation is a story we tell ourselves, one counterexample away from unraveling.

Fallibilism

Plato believed truth sat in a realm of perfect forms, waiting to be uncovered: fixed, knowable, available to anyone rigorous enough to grasp it. Karl Popper, and later David Deutsch, demolished that comfort. Everything, they argued, is fallible. Not just religious dogma or political ideology, but mathematics, physics, and every scientific theory ever proposed, including the ones we're most certain of today.

This doesn't mean truth is a fiction. It means truth is a horizon, not a destination: something we approach asymptotically, closing the distance without ever arriving. Two failed philosophies sit on either side of this idea. The relativist claims nothing is true, so why bother searching. The dogmatist claims to have already found it, so why keep searching. The fallibilist rejects both. We cannot possess objective truth, but we can make guesses, subject them to brutal criticism, and inch closer to something more useful than what we believed before. Knowledge, in this view, is the best explanation that has survived our attempts to break it.

Under this framework, every theory falls into one of two categories. It has either been tested and found wanting, which Popper called falsified, or it hasn't been proven wrong yet, meaning it remains exposed, waiting for the test that will eventually catch it. There is no third category called "proven." Popper took the point from Kant: the mind does not passively receive the world; it imposes its own structure on what it perceives, which means we never encounter reality raw enough to be certain we have seen all of it.

Because all ideas, including the laws of physics, are fallible, the work of learning never ends. Days after being celebrated worldwide for general relativity, Albert Einstein wrote to Max Born: "I have not yet eaten enough of the Tree of Knowledge, though in my profession I am obliged to feed on it regularly." Charlie Munger has said that "every year that goes by that you do not get rid of a long-held belief is a year wasted," and he urges updating mental models the moment a better one appears. Thomas Kuhn, in The Structure of Scientific Revolutions, showed why this is harder than it sounds: experts often resist new frameworks because their authority depends on the old one. Theories exist precisely so they can be replaced, which means everything you currently believe is a candidate for revision.

Reach

In 1943, a young statistician named Abraham Wald received a strange assignment from the U.S. military: study the bullet holes in planes returning from combat and recommend where to add armor. The obvious answer was to reinforce the areas riddled with the most damage. Wald saw the opposite. The planes that came back showed only where a plane could be hit and survive. The ones that mattered, the ones shot down, never made it to his sample. He told the military to armor the spots with no bullet holes at all, the engines, because those were the hits that brought a plane down.

This is what separates a merely useful idea from a powerful one. Wald's insight into survivorship bias reappears far beyond aircraft: in the false confidence of the investor who studies only successful funds, in the historian who reads only surviving texts. The measure of an explanation's power is not how well it solves the problem in front of you, but how many other problems, in other domains, it happens to solve as well. The physicist David Deutsch called this quality reach: "the ability of some explanations to solve problems beyond those that they were created to solve."

Most people chase information. The strategist chases reach. A fact about aircraft armor is a fact. A principle about hidden data is a lens you carry into finance, medicine, and history. This is why the sharpest minds in any field are often generalists in disguise: they've learned to recognize when a pattern from one domain is quietly governing another.

Claude Shannon lived by this instinct. While building the mathematical foundation of information theory at Bell Labs, he needed an equation to measure uncertainty, the average number of yes-or-no questions needed to identify a message. When he showed his formula to the mathematician John von Neumann, von Neumann pointed out that it was structurally identical to the entropy equation from thermodynamics, the same one describing the disorder of gas molecules in a room. Shannon adopted the name outright. He had stumbled onto a pattern that didn't just describe communication. It described disorder itself, wherever it appeared. That equation now sits at the foundation of computer science, cryptography, and even the analysis of financial markets, where the same mathematics that describes a particle drifting at random through a fluid turns out to describe the drift of a stock price. The math didn't know it was being borrowed. It simply worked, because the underlying pattern was real.

Francis Galton found a different kind of reach a century earlier, and it would prove just as durable. Studying the height of sweet peas, Galton noticed that the offspring of unusually tall parent plants tended to be shorter than their parents, drifting back toward the average. He called this regression to the mean, and at first it looked like a narrow botanical curiosity. It was not. Daniel Kahneman rediscovered the same pattern decades later while training fighter pilot instructors in the Israeli Air Force. The instructors swore that praise made pilots perform worse and criticism made them perform better. Kahneman saw what was actually happening: performance oscillates around an average, so an unusually good landing is likely to be followed by a worse one regardless of what the instructor says, and an unusually bad one by a better one. The instructors weren't shaping outcomes; they were mistaking a statistical rhythm for cause and effect.

That single insight, once you see it, is impossible to unsee. It explains why the team that wins spectacularly often stumbles the next season, and why the hot streak in the market rarely lasts. It is not decline. It is math, quietly returning to its center.

Reach is also what makes learning fast. Most beginners assume learning means memorizing thousands of isolated facts. It rarely does. As John Reed wrote in Succeeding, what matters is identifying the handful of core principles, usually no more than three to twelve, that govern a field. Once you see them, the flood of details stops looking random. What seemed like a million separate facts reveals itself as combinations of a few recurring patterns.

Experience and expertise are not the same thing. Palmer Luckey has described his own approach as deliberately broad rather than deliberately deep: enough exposure across fields to recognize a useful structure when it appears. Sam Hinkie defines expertise as a predictive model that works, regardless of how you arrived at it. Years in an industry can leave a person with stale assumptions; a sharper mind can arrive later and see the pattern faster.

Language learning makes this visible. If you hunt for patterns, progress speeds up immediately. English words ending in "al" often carry straight into Spanish: natural, fatal, liberal, ideal. A small insight removes a large amount of friction. The same is true of frequency. Learn the hundred most common words and you unlock a surprising share of the language. Then comes the next pattern: people do not speak in single words. They speak in phrases. So the faster route to fluency is not endless vocabulary lists but the most common phrases. This is how learning compounds. You stop collecting fragments and start seeing the machinery underneath.

The Trap of Domain Specificity

A statistician can spend all day teaching probability, then walk out into the street and make the same intuitive errors as everyone else. This is not a failure of intelligence, but a property of how the mind is built. Knowledge rarely transfers cleanly from one context to another, because our reactions are domain specific: what we think depends on the setting a problem arrives in, not its underlying logic.

This is why expertise so often fails to generalize. A person can be sharp in the boardroom and careless with their own finances, rigorous in a lab and superstitious in their personal life. They are not being inconsistent, only reacting to a different domain. The way forward is to notice the switch happening, to deliberately drag a lesson learned in one context into another where it does not naturally belong, and to keep testing an idea on multiple levels rather than trusting that understanding it once means understanding it everywhere.

Reasoning from First Principles

The prevailing wisdom among engineers building race cars in the early twentieth century held that heavier cars gripped the track better, and that grip meant speed. It was obvious. It was consensus. It was wrong. Ettore Bugatti looked at the same physics and reached the opposite conclusion: a lighter car would be faster. He built accordingly, and he was right. Bugatti did not guess better than his rivals. He refused to inherit an assumption he hadn't tested himself.

The difference is between reasoning by analogy and reasoning from first principles. Most people build on top of what others have already concluded, which is efficient, but it's also how entire industries end up chasing the same dead ends in lockstep, mistaking consensus for truth. Alan Turing took the alternative to its extreme: he disliked seeing what others had already accomplished, preferring to reinvent solutions from scratch. Mitchell Waldrop wrote of him in The Dream Machine, "he wasted a lot of time and reinvented the wheel, but he came to understand things deeply." The inefficiency was the point; borrowed knowledge is shallow knowledge.

Nassim Taleb frames the same principle in economic terms: any advantage that's obvious gets competed away instantly. "If it were known and obvious," he writes of a hypothetical restaurant secret, "then someone next door would have already come up with the idea and it would have become generic." The payoff of any venture is inversely proportional to how expected it is. None of this is a case for contrarianism. The market pays for correct disagreement, not disagreement itself.

Peter Thiel offers the clearest answer to how you find that disagreement: look for secrets. "Every correct answer is necessarily a secret," he writes in Zero to One, "something important and unknown, something hard to do but doable." Secrets about nature wait in the unexamined corners of the physical world. Secrets about people are things others don't know about themselves, or won't reveal. Building something valuable starts with asking both questions: what has nature not yet told you, and what are people not telling you?

The Billion-Dollar Pattern

Three hundred employees. That's the entire headcount of Renaissance Technologies, a hedge fund that in the mid-2010s pulled in more than $7 billion a year in trading gains, more than the annual revenues of Under Armour, Levi Strauss, Hasbro, and Hyatt Hotels combined. Those companies employ tens of thousands of people. Renaissance needed a few hundred mathematicians and a shared belief that the market was not random and could be solved. The average employee there holds nearly $50 million in the firm's own funds. Its founder, Jim Simons, was worth an estimated $23 billion by 2019. This is what happens when pattern recognition is taken to its logical extreme.

Simons didn't set out to become the richest mathematician in history. He set out to solve a puzzle. When he finished college, he told a friend that conquering the market's ancient riddle "would be remarkable," and he wanted to be the one to do it with math alone. He believed it was among the hardest unsolved problems he could find. He was right. It took him over a decade of failure to prove it. Two headstrong mathematicians he recruited as partners saw their collaborations collapse amid losses and bitterness. At one low point, results were so poor that Simons had to halt trading entirely, and one employee grew worried enough to fear Simons was contemplating suicide.

Robert Mercer, later the firm's co-chief executive, admitted that some of its signals make no intuitive sense, and that a few had been traded for fifteen years by people who could not say why they worked. No mechanism, no theory, and the best track record in the history of investing. By everything argued so far, that should not be possible.

Renaissance had no theory of why any individual signal worked, but it had a rigid theory of everything surrounding them: that faint structure decays once discovered, and that the binding constraints are data quality, transaction costs, and capacity rather than cleverness. They built a cleaned price archive reaching back decades, modeled their own execution costs, and retired any signal the moment it stopped surviving tests on unseen data. That is conjecture and refutation with a budget.

Decision Making

Theodore Roosevelt once confessed that if he could be right 75 percent of the time, he would reach the highest measure of his own expectations. He said this from the White House, having led a nation and negotiated the end of a war between empires. At Citadel, one of the largest quantitative funds in the world, founder Ken Griffin has said the firm's best stock pickers are right only 53 percent of the time, barely better than a coin flip. These are the numbers at the top.

Most decisions are not reasoned from first principles. They are predictions, quiet bets placed on how the future will unfold, built from pattern-matching against everything you have seen before. This holds as true in a cage as in a boardroom. The UFC fighter Dan Hooker says it without decoration: at the highest level, fighting is pattern recognition—studying an opponent's tendencies to exploit them while scrambling your own so no one can do the same to you. A fighter with simple, readable patterns is already halfway beaten.

A pattern does not need to be perfectly true to be worth using. It only needs to be useful. As the statistician George Box put it, "All models are wrong, but some are useful." Newton's laws are not fully correct—Einstein showed that—yet they were still good enough to help put humans on the moon. For most purposes, Newton's errors are so small they barely matter. Einstein's relativity is more precise, but it is also more complex, and in most situations that extra precision adds little. It becomes necessary only in extreme cases, such as GPS, where tiny errors compound into major inaccuracies. What matters is whether a model works for the problem at hand.

The danger is that not every compelling pattern is even that useful. Some are simply false, and false patterns are everywhere. John Locke drew the distinction sharply: idiots cannot associate ideas at all, but madmen associate them incorrectly, with total conviction. Confirmation bias is the polite version of that failure, hunting for evidence that flatters a pattern you've already, wrongly, decided is true. Consider the persistent belief that going outside without a coat causes colds. There's no real evidence for it, yet people cling to the pattern while remaining oblivious to the actual culprit, touching contaminated hands to the eyes and nose.

And not every decision can be reduced to data, objectivity, or mathematical certainty. Some decisions are matters of judgment, and judgment is inherently subjective, contested, and often controversial. Bezos's decision to keep lowering prices at Amazon was not a purely data-driven conclusion. It was a bet, a hunch that customers would value lower prices and keep coming back to the place that made buying easiest and cheapest. Many people urged him to raise prices and improve margins. He ignored them. That is what judgment looks like. It cannot be proven in advance, only defended after the fact.

None of this excuses paralysis. The standard isn't certainty; it's usefulness, tested against results and revised when the world pushes back. That is what disciplined decision-making comes down to: not the promise of being right, but the habit of building patterns useful enough to act on, discarding them the moment they stop earning their keep, and stacking the odds, decision after decision, in your favor.

Opportunity Cost

Charlie Munger stated the principle without qualification: "All intelligent people should think primarily in terms of opportunity cost. When deciding whether to do something, compare it with the best opportunity you have." Every choice carries a hidden price, the value of what you didn't choose. Most people never calculate this price. They ask whether an opportunity is good, not whether it's the best available to them.

The discipline lies in ranking opportunities before acting on any of them. Sort them into four categories: health, knowledge, relationships, and wealth. Health behaves differently from the other three. Its upside is bounded—sleep well, train hard, eat well, and the returns taper off—while its downside is not, since losing it caps everything else you might do. That makes health a floor to defend rather than a frontier to maximize.

Knowledge, relationships, and wealth have no such ceiling. A new skill compounds into new opportunities. A strong relationship opens doors no resume can. Capital, reinvested, grows on itself. This is why the disciplined thinker weighs opportunities in that order, knowledge first, then relationships, then wealth, built on top of a floor of health that is maintained rather than optimized.

The question is never simply "is this a good use of my time?" It's "is this the best use of my time, given everything else I could be doing instead?"

Speed in Decision Making

Barack Obama once said that big decisions don't require certainty, only conviction: "You don't have to get to 100 percent certainty on your big decisions, get to 51 percent, and when you get there, make the decision quickly and be at peace with the fact that you made the decision based on the information you had." That is not recklessness. Clarity rarely arrives before action does.

Hamlet understood the trap, even as he fell into it. He warned that overthinking corrodes resolve, that "the native hue of resolution is sicklied o'er with the pale cast of thought," and that hesitation causes "enterprises of great pith and moment" to "lose the name of action." His tragedy was awareness of his own indecision without the will to break it.

Elon Musk's approach names the line where thinking should stop. When two paths seem roughly equal, he does not chase the marginal edge: "rather than spend time trying to pick which one was slightly better, we would just pick one and go. Sometimes we were wrong and picked the suboptimal path, but at least we moved fast." Momentum beats precision when precision isn't available. Better to choose and adjust than to stall in search of a certainty that decisions of consequence rarely offer.