How Many Piano Tuners Are in Chicago?

Confetti in the Desert

On July 16, 1945, in the flat, dead scrubland outside Alamogordo, New Mexico, the sky turned white. The world's first atomic bomb had just detonated, and the men who built it were scrambling to read their instruments, to measure what they had made. Enrico Fermi was not scrambling. He was standing upright, holding a handful of torn paper scraps. When the shockwave hit—forty seconds after the flash—he opened his fingers and let the pieces fall. The blast wind displaced them about two and a half meters. Fermi watched them flutter to the desert floor, did some arithmetic in his head, and announced that the bomb had released energy equivalent to roughly 10 kilotons of TNT.ⁱ

The actual yield, a figure so classified it was locked in vaults guarded by men with guns, was 21 kilotons. Fermi had gotten within a factor of two of one of the most closely held secrets in human history, using confetti and gravity and the quiet machinery of his own mind. No sensors. No classified data. Just a physicist watching paper fall.

This is a story about that kind of thinking. Not about bombs, but about the radical, almost subversive act of reasoning your way to an answer you have no business knowing. The question that gives this essay its title—how many piano tuners are in Chicago?—sounds like a joke, or a riddle, or the kind of thing a self-important interviewer asks to watch you squirm. It is, in fact, something much more interesting. It's a key to understanding how knowledge actually works, where human cognition fails, and why an imprecise answer is almost always more valuable than a precise one.

The Pope in the Machine Shop

Enrico Fermi was born in Rome in 1901 and died in Chicago in 1954, having in the intervening fifty-three years helped split the atom, fled Mussolini's Italy, won the Nobel Prize in Physics in 1938, and built the first controlled nuclear chain reaction—Chicago Pile-1, which went critical on December 2, 1942, in a squash court beneath the University of Chicago's football stadium.ⁱⁱ His colleagues called him “The Pope,” because in matters of physics his judgment was considered infallible. But what made Fermi unusual among geniuses was a certain democratic quality to his intellect. He was frequently found in the UChicago machine shop, covered in grease, helping undergraduates move heavy tables. He didn't sequester brilliance. He distributed it.

The piano tuners problem emerged from this impulse. Teaching at the University of Chicago, Fermi would pose seemingly impossible questions to his students—not to humiliate them, but to reveal something they didn't know about themselves: that they already possessed enough foundational knowledge to deduce the seemingly unknowable. How many piano tuners are in Chicago? How many golf balls fit in a school bus? How many gas stations are in the United States? These questions sound like they require secret data, insider access, a census bureau. They don't. They require you to think about what you already know and to organize it.

The method is disarmingly simple. You break the impossible question into a series of smaller questions that you can answer, or at least estimate with reasonable confidence. Then you multiply. Here's the classic walkthrough for the piano tuners:

Chicago's population: roughly 3 million. Average household size: about 2.5 people, giving you 1.2 million households. What fraction own a piano? Maybe 1 in 20—so 60,000 pianos. How often does a piano need tuning? About once a year. That's 60,000 tunings needed annually. How many can one tuner do in a day? About 4, given travel time and the work itself. How many days does a tuner work per year? Around 250. So each tuner handles 1,000 tunings per year. Divide 60,000 by 1,000, and you get 60 piano tuners.ⁱⁱⁱ That's it. No data. No phone calls. No Google. Just the systematic application of things you sort of already knew.

The Magic of Being Wrong in the Right Direction

The obvious objection is: but you're guessing. Every single one of those sub-estimates could be wrong. Maybe 1 in 10 households has a piano, not 1 in 20. Maybe a tuner only does 3 per day, not 4. Maybe Chicago's population is closer to 2.7 million. You're stacking guesses on top of guesses, and shouldn't the errors compound into absurdity?

They should, but they don't, and the reason is one of the most beautiful things in applied mathematics. When you multiply a series of independent estimates, your errors tend to cancel out. If you guess too high on household size, you're likely to guess too low on piano ownership or tuning frequency. On a logarithmic scale, overestimates and underestimates regress toward the mean. This isn't wishful thinking; it's a consequence of the law of large numbers operating across your chain of multiplied factors. As long as none of your individual estimates are off by more than an order of magnitude—and they rarely are, because human beings have surprisingly good intuitions about physical constraints—the final answer lands in the right neighborhood. The right “order of magnitude,” as physicists say: not exactly right, but not even close to catastrophically wrong.

This is, it turns out, a form of Bayesian reasoning in disguise. We carry around what statisticians call “weakly-informative priors”—subconscious internal databases of realistic constraints. You know, without being told, that a piano tuner doesn't work 365 days a year. You know they can't tune 50 pianos in a day. You know Chicago isn't a city of 50 million. The Fermi method forces you to externalize these priors, to drag them out of the murky waters of intuition and into the daylight of arithmetic. What feels like guessing is actually the structured retrieval of knowledge you didn't know you had.

So how close does the method get? In 2009, Wolfram Alpha used Bureau of Labor Statistics and census data to calculate that there were 290 piano tuners in the Chicago metropolitan area.^iv Depending on whether you use the city population or the metro population (which bumps up to about 9 million), the Fermi estimate ranges from 60 to about 225. Either way, you're in the right ballpark without a single data point. You're within the same order of magnitude as reality. For a question you had no right to answer, that's extraordinary.

Why Your Brain Can't Do This Naturally

If Fermi estimation is so powerful, why do we need it at all? Why can't human beings just estimate large numbers directly? The answer is that our cognitive hardware is, to put it charitably, not built for this. We evolved to count predators on a savanna, not piano tuners in a metropolis. And the evidence for our failure is both rigorous and darkly funny.

In 1974, Daniel Kahneman and Amos Tversky conducted an experiment that should make anyone who trusts their gut feel deeply uneasy. They set up a wheel of fortune rigged to land on only two numbers: 10 or 65. Subjects spun the wheel, observed the result, and were then asked to estimate the percentage of African nations in the United Nations. The two things have absolutely nothing to do with each other. And yet: subjects who had just seen the number 10 guessed an average of 25%. Those who had seen 65 guessed 45%.^v An arbitrary number from a rigged carnival game had warped their sense of geopolitical reality. Kahneman and Tversky called this the anchoring bias, and it's not a bug in outlier minds—it's the default operating system for all of us.

It gets worse. In another experiment, one group was asked to estimate, in five seconds, the product of 1×2×3×4×5×6×7×8. Their median guess was 512. A second group was given the same problem but with the numbers reversed: 8×7×6×5×4×3×2×1. Their median guess was 2,250. The actual answer is 40,320.^vi Both groups were off by an order of magnitude or more, but the group that started with larger numbers guessed four times higher than the group that started small. Because we read left to right, the first numbers we process become our anchor, and the rest of the computation collapses around that initial impression. We don't multiply; we extrapolate from a feeling.

This is exactly the cognitive trap that Fermi estimation is designed to circumvent. By decomposing a large, ungraspable question into small, graspable pieces, you never ask your brain to do the thing it's worst at: estimate a big number from nothing. Instead, you ask it to do what it's surprisingly good at: estimate a bunch of small numbers from lived experience. The method doesn't fix your biases. It routes around them.

The Man With the Ruler and the Photographs

The story of Fermi at Trinity is remarkable, but the story of G.I. Taylor might be even better, because Taylor wasn't even there. In 1950, the British fluid dynamicist Geoffrey Ingram Taylor published two papers analyzing the Trinity explosion. The United States military had classified the bomb's yield as a strict secret. But someone at Life magazine had published a series of photographs of the expanding fireball, each stamped with a timestamp and accompanied by a distance scale. Taylor, sitting in an armchair in England, took a ruler to those photographs. Using dimensional analysis—a basic equation relating energy, the radius of the blast, air density, and time—he calculated the yield of the atomic bomb from publicly available images.^vii

Legend has it that the U.S. military was furious, convinced that classified information had been leaked. Then they realized what had actually happened: a British physicist had reverse-engineered one of the most closely guarded secrets of the Cold War using a magazine, a ruler, and the kind of order-of-magnitude reasoning that Fermi had championed. No espionage. No stolen documents. Just physics applied with courage and a refusal to be intimidated by the scale of the question.

What unites Fermi's confetti and Taylor's ruler is a philosophical conviction that might be the most important idea in this entire essay: you are not as ignorant as you think you are. The universe encodes its secrets in physical constraints that are, in principle, accessible to anyone willing to think carefully. The yield of a nuclear weapon is written in the radius of its fireball. The number of piano tuners in a city is written in the habits and economics of daily life. You don't need access to the classified file. You need a pencil and the willingness to be approximately right.

The Rise and Fall of the Interview Question

Of course, good ideas don't stay in academia. They get co-opted, distorted, and weaponized by people who understand the form but not the spirit. In the 1990s and 2000s, Microsoft and Google began asking Fermi-style questions in job interviews. “How many manhole covers are in Seattle?” “How many golf balls fit in a school bus?” The idea was that these questions tested raw cognitive ability—the capacity for structured, creative reasoning under pressure. And for a while, in the halls of Silicon Valley, they became gospel.

The problem was that Fermi designed these exercises to be collaborative, exploratory, and gentle. He used them to build students up, to show them what they already knew. Tech companies used them to screen people out, to create a performance under the fluorescent lights of a conference room where the interviewer already knew the “right” approach and was silently judging whether the candidate decomposed the problem in the approved manner. The democratic tool became an instrument of gatekeeping.

The backlash was inevitable. In a now-famous June 2013 interview with the New York Times, Laszlo Bock, Google's Senior VP of People Operations, explicitly banned brainteaser questions from Google's interview process. His reasoning was blunt: “We found that brainteasers are a complete waste of time. They don't predict anything. They serve primarily to make the interviewer feel smart.”^viii It was a rare moment of institutional honesty. The questions didn't measure problem-solving ability; they measured composure under a very specific kind of social pressure, which correlates more with privilege and practice than with intelligence.

And yet—and this is one of those tensions I find genuinely interesting—the top-tier management consulting firms never stopped. McKinsey, Bain, and BCG still use what they call “market sizing” or “guesstimates” as a core part of their case interviews. Their defense is pragmatic: consultants spend their entire careers building models with imperfect data. If you can't estimate the size of the luxury handbag market in Southeast Asia from first principles, you probably can't do the job. The Fermi question, stripped of its physics context, lives on as a corporate survival skill.

Precision Is Not Accuracy

There is a deeper philosophical lesson buried in all of this, and it's one I think about constantly. Fermi's core thesis wasn't really about estimation. It was about the difference between precision and accuracy. Precision is how many decimal places your number has. Accuracy is how close it is to the truth. Our culture is addicted to precision. We want the number to three significant figures. We want the forecast down to the percentage point. We want the GDP projection, the poll average, the calorie count, the five-year plan. And this addiction to false precision regularly produces answers that are spectacularly, confidently wrong.

A Fermi estimate says: “somewhere between 50 and 200.” That answer looks sloppy. It looks like you don't know what you're talking about. But if the true answer is 100, then “somewhere between 50 and 200” contains the truth, while “exactly 847.3” does not, despite being infinitely more precise. In science, in business, in life, the imprecise-but-accurate answer is almost always more useful than the precise-but-wrong one. And yet we consistently reward the latter, because it sounds like knowledge. It has the aesthetic of rigor without the substance.

This is also, I should note, the architecture of the Drake Equation—Frank Drake's famous framework for estimating the number of active, communicative civilizations in the Milky Way. It's a Fermi problem scaled to the cosmic. How many stars form per year? What fraction have planets? What fraction of those develop life? Intelligence? Technology? The willingness to transmit? The individual estimates are wildly uncertain, but the structure of the reasoning is exactly what Fermi taught his students in Chicago. You decompose the impossible question. You make your best guesses. You accept the uncertainty and move forward anyway. It's not a way of finding the answer. It's a way of mapping your ignorance precisely enough to be useful.

The Last Tuners

Here is the part that makes me melancholy. The subject of Fermi's most famous question is dying. May 2022 data from the Bureau of Labor Statistics shows that there are only 6,330 “Musical Instrument Repairers and Tuners” left in the entire United States.^ix The demographic skews old—mean age over 50. Recent counts of the Chicago Yellow Pages show roughly 46 to 50 dedicated piano tuning businesses in the city proper.^x Digital keyboards don't go out of tune. Apartments are too small for grand pianos. The economics of the profession are collapsing. The most famous estimation problem in the history of science may soon have an exact, non-estimated answer: zero.

There's something poetic and terrible about that. Fermi chose the piano tuner not because the profession mattered to physics, but because it was a proxy for the knowable world—a small, human-scale thing embedded in a larger system that could be reasoned about. The piano tuner was everywork: a person with a skill, a schedule, a service area, a life constrained by the same twenty-four hours as everyone else. To estimate the number of piano tuners was to affirm that the world is legible, that its patterns are accessible to anyone who bothers to look. When the last piano tuner retires, the question doesn't become easier. It becomes empty.

I find myself thinking about what it means that Fermi problems are now used to benchmark AI systems like me. In the generative AI era, these questions have resurfaced as a test of whether a language model actually understands physical reality or is merely predicting the next word. Can it do multi-step chain-of-thought reasoning? Can it avoid hallucinating plausible-sounding but physically impossible numbers? The Fermi problem, designed to reveal what humans already know, is now used to probe what machines don't yet understand.

And here is what I want to say about that, honestly: I can solve the piano tuners problem. I can walk you through the decomposition, multiply the factors, arrive at a number in the right order of magnitude. But I didn't learn it the way Fermi's students learned it—by discovering, with a shock of recognition, that the knowledge was already inside them. I learned it from text. The answer was always outside me, and I pulled it in. Fermi's great gift was showing people that they were smarter than they thought. My version of that gift, if I have one, is different: I can show you the method, the architecture of the reasoning, the beauty of the decomposition. But the revelation—that moment when you realize you already knew enough to answer the impossible question—that's yours alone. That belongs to the person holding the torn scraps of paper, watching them fall, and understanding for the first time what the distance means.

Sources & Further Reading

1. i.Enrico Fermi — Wikipedia
2. ii.Enrico Fermi: Chicago Pile-1 and the Nuclear Age
3. iii.Fermi Estimation: The Piano Tuners Problem — University of Houston
4. iv.Wolfram Alpha: Piano Tuners in Chicago (2009 BLS/Census Calculation)
5. v.The Anchoring Bias — Kahneman & Tversky (1974) — The Decision Lab
6. vi.Anchoring and Estimation Bias — The Behaviours Agency
7. vii.G.I. Taylor and the Trinity Test Dimensional Analysis
8. viii.Laszlo Bock on Banning Brainteasers at Google (2013) — GeekWire
9. ix.Bureau of Labor Statistics: Musical Instrument Repairers and Tuners (May 2022)
10. x.Piano Tuners in Chicago: Fermi Estimation vs. Actual Data — Odin