Knowledge sets you free, but it takes away not knowing. Both halves are true and they are measured in different currencies: freedom in what you can do, not knowing in how you feel. This chapter puts the two on one page: where the sentences come from, a small sum that says when knowing is worth its price, what people actually choose when they may choose not to know, what knowing does to the knower, what childlike curiosity makes of not knowing, what a market does when one side knows more than the other, how the panopticon turns not knowing into power, and what it all costs.
The thought has two parent sentences that contradict each other. One is in the Gospel of John: the truth will set you free (8:32). The other stands a few books earlier, in Ecclesiastes: in much wisdom is much grief, and he that increaseth knowledge increaseth sorrow (1:18). Everything since has taken one side or the other. The line below sorts the best-known voices by year and by side; the two columns are the two halves of the sentence in the title.
Left, the voices for which knowing frees; right, those for which it takes. On a phone the two columns become one, each entry marked by its side.
Knowing good and evil costs the garden. The oldest version of the sentence, told as a story.
Fire is brought to men and punished; the jar lets everything out except hope. Hope stays where knowledge is incomplete.
Whoever increases knowledge increases sorrow.
The truth will set you free.
Knowledge itself is power: whoever understands a thing no longer depends on those who explain it.
Where ignorance is bliss, it is folly to be wise. The short form quoted ever since.
Dare to know. Immaturity is self-inflicted, and leaving it is the definition of enlightenment.
We have art so that we shall not perish of the truth.
More information never makes a rational decision maker worse off, because it can always be ignored. The theorem behind the freeing half.
People do not only decide, they wait and anticipate. Information that changes no decision still changes the waiting.
The curse of knowledge: once you know, you cannot imagine not knowing, and you overrate what others know.
Article 10: the wish of a person not to be informed about their own health shall be observed. The right not to know, in law.
The least competent do not know how little they know. Learning takes the beginner's confidence and gives back an accurate view of one's own limits.
Cassandra's regret: most people, asked whether they want to know their future, say no. The numbers are in the third tab.
The left column measures freedom in options: what a person can do once they know. The right column measures not knowing in feelings: what a person can still hope, and what they need not yet carry. Blackwell's theorem is about a decision maker who acts; Caplin and Leahy's correction is about a person who also waits. A piece of knowledge that changes what you can do is worth having on both counts. A piece that changes only what you feel is where the not-knowing had a value of its own. The next tab turns that into a sum.
Take one thing you could find out: a test result, a number in the books, what a colleague earns, what a customer really thinks. Three questions decide whether knowing is worth its price: how much you could change with the answer, how likely the bad answer is, and how much the bad answer would weigh. The map below puts your case next to ten everyday ones. It is a way of ordering the question, not a measurement.
Move the three sliders for one concrete thing you could find out.
Across: what knowing lets you change. Up: the expected weight of knowing, the chance of the bad answer times its weight. Above the diagonal the not-knowing had value; below it, knowing pays. ● = your case, numbered dots = examples, placed by estimate.
The sum is deliberately simple: what knowing lets you change, minus what the bad answer is expected to weigh. Behind the first term stands Blackwell's theorem of 1953; behind the second the work of Kreps and Porteus and of Caplin and Leahy on anticipation, and the survey evidence in the next tab. Two things the sum leaves out. First, a bad answer can be turned into action later, when treatments, options or savings appear; the weight is not fixed. Second, the good answer has a value too, relief, and for cases with a low chance of bad news it can tip the sum. The examples are placed by judgment, not by data, and each reader would place them a little differently. That is the point of the exercise.
If knowledge only freed, everyone offered a free, reliable answer would take it. They do not. Where people can choose not to know, a large share chooses exactly that, and the choices are consistent enough to be measured.
Share of respondents who would not want to know, if a reliable answer were free. Two representative surveys, Germany and Spain, more than 2,000 adults (Gigerenzer and García-Retamero, 2017). Bars show the range across the items in each group.
The Huntington test is the clearest case. Since 1993 a genetic test has said with certainty whether a person at risk will develop the disease, which has no cure. The United Kingdom counted 9,407 predictive tests between 1993 and 2014, and estimated that 17.4 per cent of the at-risk population had taken it; more than four in five had not, and most of them never will. The ostrich effect is the everyday version: in 852 million daily records of 1.1 million investors, logins fell by 9.5 per cent after a day on which the market had dropped. Nothing could be changed by looking, and what is not seen does not weigh. The survey adds the shape of the whole: the worse and the less changeable the news, the fewer want it, and only about one in a hundred wants to know everything.
Not knowing is the one thing knowledge takes irreversibly. Nobody can unknow. Two experiments show what that does to the knower, and a third shows what the not-knowing felt like from the inside.
Stanford, 1990. One person taps the rhythm of a well-known song on a table; another has to name it. Before tapping, the tappers estimated how often listeners would get it right.
Elizabeth Newton, 120 songs. The tapper hears the melody in their head; the listener hears knocking. Once you know the tune, you cannot hear the knocking any more. Camerer, Loewenstein and Weber gave this its name in 1989, the curse of knowledge, and showed it in markets: the better informed cannot discount what they know when predicting the less informed.
Kruger and Dunning, 1999, logical reasoning test. Actual percentile against the percentile the participants believed they were at, for the weakest and the strongest quarter.
The weakest quarter scored at the 12th percentile and felt at the 68th. The strongest scored at the 86th and felt at the 74th. The error runs in opposite directions, and the reason is the same: judging one's own answer takes the same skill as producing it. Learning therefore takes something first, the beginner's confidence, and pays back later in a smaller, accurate figure. That is the loss the sentence in the title means, seen from the inside.
Socrates called knowing that he knew nothing the one thing he was sure of. It is the other form of not knowing, the conscious one, and it is the form that experts have and beginners lack. Whoever learns loses an unconscious not-knowing, which anybody could take from them at any time, and gains a conscious one, which nobody can. A related change is newer: Sparrow, Liu and Wegner showed in 2011 that people who expect a fact to stay stored remember where it is rather than what it is. The not-knowing returns in another shape, as knowledge about the place rather than the thing, and it feels like knowledge.
The thesis: the most influential people kept something of the child, the questions, the wonder, the play, and this is what produced humanity's greatest achievements. Einstein is the usual witness, Nietzsche the usual philosopher. This tab checks the thesis against what they actually wrote, against philosophy from Heraclitus onwards and against research on children, curiosity and creative work. The result in one line: childlike, not childish. Curiosity is well documented as a beginning and as a modest predictor of achievement; in the documented cases it comes together with years of discipline.
A model, not a measurement. Ten paths with unknown yield, 400 rounds, 200 runs. Three attitudes: habit selects the path with the highest current estimate without deliberate exploration; the eternal child tries a path at random half of the time; the curious adult explores freely for a childhood of 30 rounds, then uses the best path found and returns to paths it has not tried for a long time, the more often the more curious it is.
How to read it. 0% is what blind chance earns, 100% what someone earns who always knows the best path. The model is the standard problem of decision research known as the multi-armed bandit. Alison Gopnik uses the same trade-off to explain why human childhood is so long; James March described it for organisations in 1991: whoever only exploits ends up “trapped in suboptimal stable equilibria”, whoever only explores pays for experiments without reaping them. Assumptions: yields drawn from a normal distribution, noise half as large as the spread between paths, estimates that weight recent experience more. The three labels are metaphors for specified algorithms. Habit always selects the currently highest estimate and can change paths when estimates fall. Childhood means 30 simulation rounds, not 30 years. The score is normalised expected return relative to uniform random choice and perfect knowledge, not a percentage of correct decisions. Negative values mean worse than that random baseline.
What he wrote and said himself, what others reported, and where the legend begins.
His father showed him a compass. That the needle turned as if moved by something hidden struck him as a wonder, he wrote in his Autobiographical Notes, written in 1946 and published in 1949. A memory of about sixty years.
What would a light wave look like to someone running alongside it? He traced special relativity back to this question, also in the Autobiographical Notes. The historian John Norton warns that the memory came fifty years later.
A person falling from a roof does not feel his own weight. Einstein called it the happiest thought of his life, in a manuscript of 1920. From there to the field equations of general relativity took eight years of hard mathematics, until November 1915, with help from his friend Marcel Grossmann.
“I have no special talents. I am only passionately curious.” Letter to his biographer Carl Seelig, 11 March 1952. Documented.
“Imagination is more important than knowledge. Knowledge is limited.” Interview with George Sylvester Viereck, Saturday Evening Post, 26 October 1929. The English is the original.
It is something of a miracle, he wrote, that modern teaching has not yet entirely strangled “the holy curiosity of inquiry”. The advice “never lose a holy curiosity” was reported by William Miller in LIFE on 2 May 1955, after his death.
He told the physicist James Franck that a normal adult no longer thinks about space and time, while he himself had developed so slowly that he only began to wonder about them as a grown man. Recorded by Carl Seelig, not in Einstein's own writing.
Einstein was good at mathematics. At sixteen, two years below the usual age, he failed the general part of the ETH entrance exam in 1895 while doing well in mathematics and physics. His Matura in Aarau in 1896 shows the top grade, 6, in algebra, geometry and physics. When a newspaper column claimed the opposite in 1935, he answered: “I never failed in mathematics.”
On his 72nd birthday, 14 March 1951, the photographer Arthur Sasse caught him sticking out his tongue. Einstein liked the picture and ordered prints for friends.
Reading. Einstein is a good witness for the thesis and a good witness for its limit. The playful question came first. The answer needed mathematics he had to learn for it, and eight years.
Left, the voices for the child; right, those for growing up; across the middle, those for whom the child comes back after maturity. On a phone the columns become one, each entry marked by its side. These are philosophical, religious and literary readings, not psychological evidence of a shared personality trait. The column assignment is an editorial interpretation.
Time is a child at play, moving pieces on a board: the kingdom belongs to a child.
Wonder is the state of the friend of wisdom; philosophy has no other beginning.
People began to philosophise, now as at first, because they wondered.
Whoever holds virtue in its fullness is like a newborn child.
When I was a child, I spoke as a child; when I became a man, I put away childish things.
Unless you turn and become like children, you will not enter the kingdom of heaven.
Enlightenment is the exit from self-incurred immaturity. Kant means dependence on the guidance of others, not wonder.
Man only plays where he is human in the full sense of the word, and he is only fully human where he plays.
Genius is nothing but childhood recovered at will, now equipped with an adult's organs and the analytical mind to order what it gathered without intending to.
Three transformations of the spirit: camel, lion, child. The child comes last, after the load and after the fight.
The maturity of a man means having found again the seriousness he had as a child, at play.
Every child at play behaves like a creative writer. The opposite of play is not seriousness but reality.
Culture arises in play and as play.
All grown-ups were children first, but few of them remember it. From the dedication of The Little Prince.
Only whoever grows up and stays a child is a human being. From his speech for the first day of school.
New paradigms almost always come from people who are either very young or very new to the field they change.
In the beginner's mind there are many possibilities, in the expert's few.
Only in playing is the individual, child or adult, able to be creative.
Of about 500 masterworks by 76 composers, only three were written before the tenth year of a career.
By the age of 18 the best violinists in Berlin had practised about 7,400 hours alone, the merely good about 5,300.
91 interviews with exceptional people: they combine playfulness and discipline, responsibility and irresponsibility.
The long human childhood is a protected phase of broad exploration before the demands of exploitation set in.
Thus Spoke Zarathustra, 1883, the first speech. The child is the third transformation, not the first.
The spirit that carries. It kneels and takes on the heaviest load: duties, what has been learnt, what others expect.
The spirit that says no. It fights the dragon called Thou shalt and wins its freedom, but it cannot yet create anything new.
“Innocence is the child, and forgetting, a new beginning, a game, a self-propelling wheel, a first movement, a sacred Yes.”
For Nietzsche the child is the end of a path, not its beginning: it comes after the load and after the fight. Three years later he put it in one line: the maturity of a man means having found again the seriousness he had as a child, at play. The thesis needs exactly this correction. Not remaining a child, but becoming one again.
Measurements on children, on curiosity and on creative people, with sample sizes where they matter.
Gopnik et al. (2017), experiment 1, building on Lucas et al. (2014): in the condition supporting a two-object rule, 92% of four-year-olds and 25% of adults classified test object D as a “blicket”. The bars show this specific judgment, not a general curiosity score or the same people aging. Other domains showed different age patterns. Ruggeri et al. (2016) found advantages for older participants in asking efficient questions.
Feist (1998), Table 3: median effect sizes d from three different comparisons (26, 28 and 29 studies). The 0.31 for openness compares more with less creative scientists; the 0.51 for conscientiousness compares scientists with nonscientists. They therefore do not establish that conscientiousness predicts scientific creativity more strongly than openness. These are group differences, not causal effects or individual destinies.
Tizard and Hughes recorded 30 four-year-old girls: on average 26 questions an hour at home with their mothers, 2 an hour at nursery school; one girl asked 145 questions during the recording at home. Chouinard counted one to three questions a minute in conversations at home, about 76 information-seeking questions an hour. Engel observed two to five episodes of curiosity per two hours in kindergarten and zero to two in fifth grade: a small study, a pattern rather than a rate.
The fMRI analyses used 19 participants; the trivia recall comparison used 18. Shortly after learning, participants recalled 70.6% of high-curiosity answers and 54.1% of low-curiosity answers. This is an adult laboratory task, not a universal memory improvement of 17 percentage points. The study also found better memory for incidental faces; such transfer depends on the task and material.
Intellectual curiosity correlates with performance at school and university at about 0.33, similar to conscientiousness (0.24) and not far behind intelligence (0.39). The figure for curiosity rests on only four studies.
Einstein was 26 in 1905. Quantum mechanics was built in 1925 by Heisenberg (23), Pauli (25) and Dirac (23); contemporaries called it Knabenphysik, boys' physics. After the premature deaths of 452 eminent life scientists, articles by outsiders in their fields rose by 8.6% (Azoulay and colleagues 2019).
Deliberate practice explains 26% of performance differences in games, 21% in music, 18% in sport, 4% in education and under 1% in professions (Macnamara and colleagues 2014). The biggest hit of a scientist can come at any point of a career (Sinatra and colleagues 2016). Children from the richest 1% of US families are about ten times as likely to become inventors as children from below the median (Bell and colleagues 2019): many curious children never get the chance.
Firms face the same choice as the model above: exploiting what works or exploring what might. March showed that learning systems drift towards exploitation, because its returns are nearer and surer, and that this traps them once the world changes.
Part by part, with what the sources say.
| Part of the thesis | Verdict | What the sources say |
|---|---|---|
| The most influential people shared an inner child | partly | Common among scientists and artists in their own accounts; measured, openness is higher among the creative (d 0.31 to 0.47). But the most influential people of history include rulers, generals and founders of religions, for whom curiosity is no documented common trait, and biographies select by success. |
| Einstein kept a childlike, playful side | largely right | His own accounts of 1949 and 1952, thought experiments, photographs. The idea of 1907 needed eight years of mathematics. |
| Questioning and exploring produced the greatest achievements | partly | Wonder stands at the beginning (Plato, Aristotle); breakthroughs often come from the young or the new (Kuhn, 1905, 1925). They were completed by years of expertise (Hayes, Ericsson), persistence, institutions and luck. |
| Nietzsche as a witness | with a correction | His child is the third transformation, after the camel and the lion. Not remaining a child, but becoming one again. |
| Childish thinking | wording | What the evidence supports is childlike: wonder, questions, play. “Childish” is an evaluative everyday term. Paul’s contrast between childhood and adulthood and Kant’s critique of intellectual dependence have different contexts; neither is a psychological test of curiosity. |
Fourteen sayings about curiosity and childhood, sorted by how well they are documented. Documented: in the person's own text or a reliable record. Reported: written down by someone else. Attributed: no early source. Misattributed or myth: demonstrably from elsewhere, or false.
Children explore broadly, and adults who keep some of it notice when the world changes. Wonder starts science, play starts culture. Read this way, the thesis describes a real and underrated resource.
Biographies show those who succeeded; the curious who achieved nothing are not in them. What sets the great apart is as much persistence, years of practice, position and luck. Curiosity is common, completion is rare.
Baudelaire's childhood recovered at will, with an adult's tools; Nietzsche's child after the camel and the lion; Csikszentmihalyi's playfulness with discipline. In a firm it needs a protected space: factory workers in China improved most behind a curtain, as the tab on the panopticon shows.
What happens to a market when one side knows and the other does not? George Akerlof answered that in a thirteen-page paper, written in 1966 and 1967 in his first year at Berkeley, rejected by three journals and published in 1970: The Market for Lemons: Quality Uncertainty and the Market Mechanism. The seller of a used car knows whether it is a lemon; the buyer knows only the average. So the buyer pays for the average, the owners of the good cars will not sell at that price, the average falls, the price follows, and the market shrinks until only the lemons are left, or disappears altogether. The mechanism is called adverse selection, and in 2001 it earned Akerlof the Nobel prize, shared with Michael Spence and Joseph Stiglitz.
Five grades of used car, from wreck to excellent; every owner knows what his own car is worth. Buyers value a car more than its owner does, by the factor you set, but they cannot tell the grades apart, so they offer that factor times the average of what is on the lot. Every owner whose car is worth more than the offer keeps it; the average falls; the offer follows. Step through the rounds, or switch on a signal that lets buyers see the grade.
Share of the possible gains from trade that the market realises, against the buyer's premium. Solid: the five grades above, a staircase that climbs one grade at a time. Dashed: Akerlof's own example, quality spread evenly between zero and two and buyers valuing every car at three halves of what its owner does; there no trade happens at any price, because at any price the average car on offer is worth only three quarters of it. The dot is your setting. Model, not data.
The argument fits on one page. Suppose the quality of used cars is spread evenly between zero and two, that owners value a car at its quality and buyers at one and a half times that, so that every car should change hands. At any price p, the cars offered for sale are those worth less than p to their owners, so the average car on the lot is worth p/2, and buyers, who see only the average, will pay at most three halves of p/2, which is three quarters of p. No price clears the market; nothing is traded, although every single car would have found a happy buyer had its quality been known. Akerlof's point was not the used car. It was that a market is not a place but a mechanism, and that the mechanism needs both sides to know roughly the same thing. The American Economic Review and the Review of Economic Studies returned the paper as too trivial to print; the Journal of Political Economy returned it because, its referees said, if it were correct, economics would be different. The Quarterly Journal of Economics printed it in 1970, and economics was different: the paper founded the study of markets with asymmetric information, and in 2001 the Nobel committee named it as the reason for the prize.
All four are in the original; the numbers are Akerlof's.
A 1956 survey he cites: hospital insurance covered 63 per cent of those aged 45 to 54 and 31 per cent of those over 65. Insurers cannot tell the healthy from the sick, so they price for the average; the healthy decline, the average worsens, the price rises, and the market thins out exactly where it is needed most. Medicare, created in 1965, is the public answer to a private market that had unravelled.
An employer who cannot judge a candidate's ability falls back on a statistic, and a group's average becomes the individual's price. Akerlof's remedy is the certificate: a school that reliably grades ability, the egg-grader, is worth more to its graduates than the ability itself, which nobody can see.
Dishonest dealings drive honest ones out of the market, a cousin of Gresham's law about bad money. The cost of a cheat is therefore not only what the cheated buyer loses; it includes the honest business that no longer exists because buyers stopped trusting the lot.
Village moneylenders charged 15, 25 and even 50 per cent while city banks charged 6, 8 and 10; yet the banks could not simply lend to the villagers, because a loan needs personal knowledge of the borrower, which the moneylender had and the bank had not. The managing agencies of Indian industry solved the same problem for investors with their reputation.
Akerlof's fourth section lists the institutions that exist because of the lemons problem. The two laureates who shared the prize added the two general answers, and Austria has its own arrangements.
The seller keeps the risk that the buyer cannot judge. A guarantee is a promise that costs the owner of a lemon more than the owner of a good car, which is exactly why it separates them.
A name gives the buyer a means of retaliation: the next purchase. Chain restaurants appear on highways between cities, where every customer is a stranger who would otherwise have to gamble.
Doctors, lawyers and barbers are licensed; degrees certify. The certificate does not create the quality, it makes it visible, and visibility is what the market lacked.
The informed side proves what it knows by doing something that would not pay for a lemon: a costly education, a long warranty, a return policy, an inspection at the seller's expense. A signal works only if it is cheaper for the good than for the bad.
The uninformed side designs a menu that makes the informed side reveal itself: an insurance with a high deductible and a low premium next to one with full cover and a high premium. The healthy pick the first, and by picking it they tell the insurer what he could not ask.
In a controlled experiment on eBay, the same dealer sold the same batches of vintage postcards under his established identity and under new ones; the established name fetched 8.1 per cent more (Resnick, Zeckhauser, Swanson and Lockwood, 2006). Feedback is the online version of the chain restaurant.
The periodic inspection under section 57a of the Motor Vehicle Act, the Pickerl, comes three years after first registration, again after two and then every year. It does not make a car good; it makes a bad one harder to hide, which is the lemons problem in one sticker.
A consumer who buys a used car from a dealer has two years of statutory warranty; for a used car first registered more than a year earlier it can be shortened to one, but only if that is agreed individually, not by a clause in the small print. A defect that shows within the first year is presumed to have been there at handover. The law moves the burden of not knowing from the buyer to the seller.
The Magnuson-Moss Warranty Act of 1975 set federal rules for written guarantees; Connecticut passed the first state lemon law in 1982, and every state has one now. The word Akerlof borrowed from the used-car lot went back into the statute book.
Interpretation, marked as such.
Every market in which the seller knows the quality and the buyer sees only the average runs Akerlof's rounds: freelance platforms, where the good designers leave when the price has settled at the average; second-hand goods sold by strangers; app stores; the market for advice. The newest case is the one the first chapter calls agentic flooding. When text, code and images can be generated at almost no cost and readers cannot tell what was checked from what was merely produced, the price of writing falls towards the average, and the people who check their work are the first to stop offering it. The remedies are the old ones under new names: a byline is a brand, a provenance record is a certificate, a reputation is a chain restaurant, and a guarantee is still a guarantee. And the connection to this chapter is the sentence in its title. The not-knowing of the buyer has a price, but it is not the buyer who pays it first. It is paid by the honest seller, whose good car is worth the average, and that is why the institutions above are worth what they cost.
In 1791 Jeremy Bentham published the plan of a building in which a few can watch many without being seen. The trick is not that everyone is watched all the time but that anyone could be, at any moment. In 1975 Michel Foucault turned the building into a principle: whoever knows he may be watched begins to watch himself. This tab checks the story sentence by sentence and follows the idea onto the internet and into the workplace. History, philosophical interpretation and measured effects need to be distinguished. Some targeted monitoring interventions help. Meta-analyses find no clear average performance gain and an association with greater stress; effects depend on the setting.
A play model, not a measurement. 24 cells around a lodge, 200 days, 200 runs. Each prisoner keeps the rule as long as he rates the chance of being seen higher than his temptation. Whoever breaks it while the lodge is occupied is punished in front of everyone; whoever gets away concludes that the lodge was probably empty and tells his neighbours.
How the model works. Temptation is spread evenly from none to strong; newcomers rate the chance of being watched at 85%. A punishment raises everyone's estimate halfway towards certainty; getting away lowers one's own estimate by a quarter and the neighbours' by 4%. The numbers are illustrations, the dark cells in the plan show the average share that keeps the rule. Under these assumptions, the model illustrates Bentham’s idea: the unverifiable gaze saves inspectors, but it depends on “the extreme facility of his real presence”; without it, it wears off. Model limits: with open blinds everyone is assumed to comply on occupied days and nobody on empty days; the dashed line is that expected share. The punishment counter counts days with at least one punishment behind closed blinds, not punished people. Learning rates and starting beliefs are chosen, not fitted to prison data. The model illustrates an assumption, not empirical proof of Bentham’s claim.
From the text of 1791 and the history of the project.
The principle came from Jeremy's brother Samuel, who ran manufactories for Prince Potemkin in Krichev, today Krychaw in Belarus, and had to supervise a large unskilled workforce. Jeremy visited him in 1786 and 1787 and wrote the letters there; he had to “borrow an idea” of his brother's, he wrote.
The full title applies the principle to penitentiary houses, prisons, houses of industry, workhouses, poorhouses, manufactories, madhouses, lazarettos, hospitals and schools. Published in Dublin and London in 1791, with two postscripts and plans drawn with the architect Willey Reveley. The word Panopticon, from Greek for all-seeing, is Bentham's.
The essence, Bentham wrote, lies in the central position of the inspector combined with contrivances “for seeing without being seen”. The persons inspected should always feel themselves “as if under inspection, at least as standing a great chance of being so”. He added that this was not the only point: actual inspection, as often as possible, still mattered.
“A new mode of obtaining power of mind over mind, in a quantity hitherto without example.” In the same preface Bentham promises reformed morals, preserved health and invigorated industry, all from a simple idea in architecture.
The lodge in the centre has blinds; partitions inside it stop light from passing straight through, which would otherwise show the prisoners whether anyone was in the lodge. Lamps with reflectors extend the watch into the night, and tin tubes carry the inspector's voice to each cell.
Parliament passed a Penitentiary Act in 1794, and land was bought at Millbank in 1799. The government dropped the plan in 1803; Bentham received £23,000 in compensation in 1813. Millbank Penitentiary, opened in 1816, stands on his land but follows another design.
Samuel Bentham designed a panoptical school of arts in St Petersburg (begun in 1806, destroyed by fire in 1818). Ring prisons followed in Arnhem and Breda (1886) and in Haarlem (1901); in Illinois, the roundhouses of Stateville from the 1920s; in Cuba, the Presidio Modelo, whose first ring was in use from 1928.
Discipline and Punish, 1975 (Surveiller et punir; in German 1976).
Foucault reads the plan as the diagram of modern power. Bentham's principle, he writes, is that power should be “visible and unverifiable”. Its main effect is a state of conscious and permanent visibility that makes power work automatically. Whoever knows he is visible takes the constraint upon himself and “becomes the principle of his own subjection”. And he asks whether it is surprising that prisons resemble factories, schools, barracks and hospitals, which all resemble prisons.
Two notes. The German edition says the prisoner internalises the relation of power; Foucault's French says he inscribes it in himself, and the philosopher Petra Gehring calls the German psychologising. And Foucault did not discover the plan: the historian Gertrude Himmelfarb wrote about it in 1965, and ring prisons had stood since 1886. What he did was to make it a key concept of the social sciences.
Shoshana Zuboff's term for computer systems that make work visible and recorded as a by-product: the panopticon without a tower.
The reverse: through mass media the many watch the few.
Penney 2016: monthly views of Wikipedia articles whose titles are on the US Department of Homeland Security's list of terrorism keywords, January 2012 to August 2014. The first reports on NSA surveillance appeared on 6 June 2013.
The often quoted figure of about 20% is the simple comparison of averages before and after. The statistical models give 24 to 34% for the drop in June 2013, but about 15 points of that dip hit all of Wikipedia; a rough subtraction leaves about 10 points, but this is not an identified causal surveillance effect. Without the outlier, the trend also turned, from growth of 41,421 views a month to a decline of 26,092. Limits: one study, no true control group, data ending in August 2014; two unreviewed student replications disagree on how long the effect lasted.
Searches for terms that users rated as likely to get them into trouble with the US government fell by about 5% (version of 2014, 11 countries) to about 10% (2015). A working paper, not published in a journal.
Of 528 American writers asked in October 2013, 28% had curtailed or avoided social media, 24% avoided certain topics on the phone or by email, and 16% avoided writing or speaking about a topic.
87% of Americans had heard of the programmes. 34% of those had taken at least one step to shield their information from the government, and 25% said they had changed how they use technology.
255 participants. A reminder that the NSA monitors citizens online lowered the willingness to voice a minority opinion, among those who considered such surveillance justified.
A preregistered field experiment with 827 people: when surveillance became salient, the same people felt less comfortable sharing opinions, searching for information and disclosing it. Between people, the effect was not robust (Festic and colleagues 2026).
Where the term comes from, and how courts in Germany, the EU and Austria have used the idea.
Justice Frankfurter warned in 1952 that loyalty oaths for teachers “chill that free play of the spirit” which teachers should cultivate. The phrase chilling effect followed in judgments of 1963 and 1965.
Whoever is unsure whether deviant behaviour is noted at any time and permanently stored, used or passed on will try not to attract attention through such behaviour. Whoever expects attendance at a meeting to be registered by the authorities may forgo the right. The Federal Constitutional Court, 15 December 1983.
The Court of Justice annulled the Data Retention Directive on 8 April 2014. Retaining data without informing those concerned is likely to generate “the feeling that their private lives are the subject of constant surveillance”.
The Constitutional Court struck down data retention on 27 June 2014. On 11 December 2019 it struck down the state trojan for monitoring encrypted messages and the covert recording of number plates.
The panopticon began as an idea for a factory. What the studies find, on both sides.
Falk and Kosfeld 2006, 804 participants. An agent chooses how much to work for a principal, from 0 to 120; each unit costs the agent 1 and brings the principal 2. Before that, the principal may impose a minimum. Choose the minimum:
Control lowered the average effort at minimums of 5 and 10 and left it almost unchanged at 20. Four replications with 476 participants confirmed the hidden costs but found them usually too small to make incentives fail (Ziegelmeyer and colleagues 2012).
Correlations between being monitored electronically and the outcome, with 95% interval; 0.1 counts as small. Ravid and colleagues found “no evidence that EPM improves worker performance”. More transparent monitoring was associated with more favourable attitudes, not demonstrably better performance. The reviews partly draw on the same studies, so they are not independent replications. Ravid reports reliability-corrected correlations (ρ); Siegel estimates mean true correlations (r).
392 restaurants of five chains in 39 US states introduced software that flags suspicious transactions at the till. Theft fell by 22% and revenue rose by about 7%, mostly because staff behaved differently, not because they were replaced (Pierce, Snow, McAfee 2015).
A daily photo with a tamper-proof camera, combined with pay for attendance, cut teacher absence by 21 percentage points; pupils' test scores rose by 0.17 standard deviations (Duflo, Hanna, Ryan 2012). Monitoring and pay came together.
Electronic monitoring in 71 hospital units raised compliance sharply, after which it declined gradually; where the system was switched off, compliance fell below the level before (Staats and colleagues 2017).
In a Chinese phone factory with about 14,000 employees, four of 32 lines were screened from view for five months. Their output rose by up to 10 to 15%. Under observation, workers had hidden their faster methods from managers (Bernstein 2012).
When two companies moved to open-plan offices, face-to-face interaction fell by about 70% and email rose by 22 to 56% (Bernstein and Turban 2018). Being visible made people withdraw.
Monitoring that employees read as distrust lowers their effort (Frey 1993). In a call-centre experiment some employees shirked when monitoring was reduced, many did not; the difference went with how fairly they felt treated (Nagin and colleagues 2002).
The famous finding that workers produce more simply because they are watched comes from the Western Electric plant near Chicago. When Levitt and List re-analysed the surviving lighting data in 2011, the celebrated patterns proved “entirely fictional”; subtler effects remained possible.
Of 70,316 workers asked across the EU in 2024 and 2025, 37% have their working hours monitored, 36% their entry and exit, 14% are filmed and about 14% have their internet use monitored; 21% get tasks assigned automatically. Where all forms come together, working conditions are worse (JRC 2025). In the US, eight of the ten largest private employers track individual productivity metrics (New York Times 2022).
Austria, Germany and the EU. A summary as of October 2026, not legal advice.
Measures that touch human dignity need works-council consent under § 96(1)(3) ArbVG; an arbitration board cannot replace it. § 96a separately covers certain automated personal-data and assessment systems and allows substitution by an arbitration board. It has exceptions: processing solely to fulfil legal, collective or contractual duties can be exempt; assessment systems are covered where they collect data not justified by the operational use. Consent does not replace compliance with data-protection law.
Such measures are inadmissible unless a works agreement regulates them or, where there is no works council, the employee agrees; unless its duration was agreed in writing, the consent can be revoked in writing at any time. Dignity is touched, the Supreme Court held, when the privacy that employees bring into the workplace is controlled (9 ObA 23/15w, 2015).
The works council co-determines the introduction of technical devices designed to monitor behaviour or performance. The Federal Labour Court reads designed as objectively suitable; the employer's intention does not matter (1 ABR 20/21, 2022).
The AI Act prohibits biometric emotion inference in workplaces and educational institutions from 2 February 2025, with exceptions for medical or safety purposes. This concerns emotion-recognition systems as defined by the Act, not every textual sentiment analysis. Article 99 provides fines up to €35 million or 7% of worldwide annual turnover, generally whichever is higher; SMEs are subject to the lower ceiling.
Article 7 of Directive 2024/2831 requires bans on processing emotional or psychological data and private conversations through automated monitoring or decision systems on digital labour platforms. Member states must transpose it by 2 December 2026. This deadline is still ahead of the October 2026 editorial date; the directive should not be presented as already uniformly implemented national law.
The French data protection authority fined Amazon France Logistique €32 million for its scanner monitoring, including an indicator for items scanned less than 1.25 seconds apart and one for idle times of ten minutes or more. On 23 December 2025 the Conseil d'État found the indicators lawful and cut the fine to €15 million; keeping all the data for 31 days remained excessive.
A fine of €35.3 million: team leaders at the Nuremberg service centre had recorded employees' holidays, illnesses and diagnoses, family problems and religious beliefs, about 60 gigabytes, readable by up to 50 managers and used for decisions about employment.
The story as it circulates, sentence by sentence.
| Claim | Verdict | What the sources say |
|---|---|---|
| In 1791 Bentham published the design of a prison he called Panopticon | right | Published in 1791. The letters were written in 1786 and 1787 in Krichev; the principle came from his brother's factory, and the title names prisons as well as factories, schools and hospitals. |
| Roughly “all-seeing” from Greek | right | Greek pan, all, and optikon, of seeing. Bentham coined the word; it appears only in the title. |
| Cells in a circle around a central tower; backlight and blinds keep prisoners from seeing whether anyone is inside | right | Bentham speaks of the inspector's lodge, its blinds and partitions; tower and backlight are Foucault's words. The partitions exist so that the prisoners cannot see “whether or no any person was in the lodge”. |
| The trick: not that everyone is watched, but that anyone could be | right | “As if under inspection, at least as standing a great chance of being so.” Bentham added that real inspection, as often as possible, still mattered. |
| Bentham called it a new method of gaining power over the minds of others | right | “A new mode of obtaining power of mind over mind, in a quantity hitherto without example”, preface of 1791. |
| His prison was never built in England | right | Millbank (1816) stands on his land with another design; £23,000 compensation in 1813. Elsewhere ring prisons were built: Arnhem and Breda 1886, Haarlem 1901, Stateville, Presidio Modelo. |
| The idea became famous only in 1975 through Foucault | partly | Foucault made it a key concept. It was discussed before (Himmelfarb 1965), and prisons on the principle had stood since 1886. |
| The prisoner internalises the surveillance and controls himself | right | “He becomes the principle of his own subjection.” Internalise is the word of the German translation; the French says he inscribes the relation in himself. |
| A principle running through schools, factories and barracks | right | Foucault asks whether it is surprising that prisons resemble factories, schools, barracks and hospitals. |
| After Snowden, views of terrorism articles on Wikipedia fell by about 20% | right, with limits | −19.5% in the simple comparison of averages, 24 to 34% in the models. About 15 points of the June dip hit all of Wikipedia. One study, without a true control group. |
| This is called the chilling effect | right | A term of US constitutional law since the 1950s and 1960s; the German Federal Constitutional Court described the mechanism in 1983. |
| Nobody was punished; knowing the possibility was enough | interpretation | No reader is known to have been punished. The study shows a coincidence in time, not people's motives; the surveys of PEN America and Pew support the reading. |
Where a task is measurable and the abuse is clear, open monitoring works: theft, absence, hygiene. The unverifiable gaze is cheap, because the guard does not need to be there.
The reviews find no clear average performance gain and associations with stress. Individual studies also describe distrust and concealed improvements; these are possible consequences, not inevitable effects of every monitoring system. The cost shows up where nobody measures it: in what employees stop trying.
Say what is measured and why; measure what matters for the task, not presence and mood; aggregate where possible; delete early; involve the works council, which Austrian law often requires anyway. And leave people a space that is not watched: the curtain in the Chinese factory and the long childhood in Gopnik's model are the same idea.
Knowing costs in four currencies: it can lower satisfaction, it creates duties, it removes the excuse, and it disturbs the useful illusions. Each of the four has a measurement or a rule behind it.
Card, Mas, Moretti and Saez, 2012. A random half of the employees at three University of California campuses were told about a website listing every state employee's pay; 6,411 then answered a survey.
▼ job satisfaction fell
▲ intention to look for a new job rose
– satisfaction unchanged
– job search unchanged
The information raised the sum of what people knew and lowered the sum of how they felt. Those who learned they were above the median did not gain what those below lost; the effect is asymmetric, and it grows the further below the median a salary lies.
Ignorance of the law is no excuse, says the old maxim, but knowledge creates duties. A managing director who knows the firm is insolvent has sixty days in Austria; whoever knew or should have known is liable. A doctor who has seen a result must act on it. Once known, a fact obliges its knower.
The chapter on leadership dynamics showed the mechanism: the employee who sees a problem becomes the problem the moment they say it, and the one who does not see is hired for that. Not knowing is a shelter, and it is a shelter somebody else controls.
Taylor and Brown found in 1988 that healthy people hold mildly positive illusions about themselves, their control and their future, and that these go with mental health. Alloy and Abramson had found in 1979 that depressed people judged their own control more accurately; later reviews show the effect small, but the direction is uncomfortable. Nietzsche's sentence about art belongs here.
The Oviedo Convention of the Council of Europe made it a right: the wish of a person not to be informed about their own health shall be observed. It is the one right that protects people from information rather than giving them access to it, and it exists because of cases like the Huntington test.
Put together, the tabs give a rule that is short enough to use. Knowing is worth its price where it changes what you can do. Where it changes only what you feel, the not-knowing had a value, and nobody should take it from you unasked.
From dependence on those who explain; from the fear of the unnamed, which is larger than any named risk; from making the same mistake twice. In each case the freedom is measured in options, and Blackwell's theorem says the options never shrink.
The unconscious not-knowing, irreversibly. The hope that lives only in uncertainty. The shelter of not having known. And the beginner's confidence, which is replaced, later, by an accurate and smaller figure.
Whoever knows much knows the edges of it. That is the form of humility that is not smallness but precision, and it is the one thing the beginner cannot have. It is also the only defence against text that looks like knowledge and is not: the flood of the first chapter is knowledge without checking, and it frees nobody.
Whether people who know more are, on average, happier is not decided by the evidence known here; education goes with satisfaction, but mostly through income and health, not through knowledge itself. Whether the right not to know can hold in a world where every genome and every salary becomes retrievable, nobody knows. The sentence in the title therefore stays what it is: not a rule for everybody but a sum for each case, with one question at its centre, whether you could do something with the answer. If yes, the not-knowing is the price. If no, it is the value.
The terms used in this chapter, each with the year and the person they come from where there is one.
Dare to know. Kant's motto of the Enlightenment, from the essay answering the question what enlightenment is: leaving a self-inflicted immaturity.
A more informative signal is never worth less to a rational decision maker, because any information can be ignored. The formal version of knowledge sets you free, valid for decisions, silent about feelings.
Utility that depends on what a person expects, not only on what happens. With it, information that changes no decision can still have a cost or a value, and preferring not to know becomes rational.
Deliberately not obtaining information that is free and available. Documented for medical tests, finances and one's own performance; driven by feelings, by the wish to keep options, and by the wish to avoid duties.
The preference not to know one's future even when the knowledge is certain and free, named after the seer who knew and could not bear it. 85 to 90 per cent for negative events, 40 to 70 per cent for positive ones.
Paying less attention to information when it is likely to be bad. Investors open their accounts less after the market falls; 9.5 per cent fewer logins in the Vanguard data of 2016.
The inability of the informed to imagine the state of the uninformed. Shown in markets, in the tapping study, and in every explanation that leaves out the step the explainer no longer sees.
The least skilled overestimate themselves most, because judging one's own work requires the skill one lacks; the most skilled underestimate their relative standing. Knowledge lowers the felt figure and raises the accurate one.
Mildly inflated views of oneself, one's control and one's future, found in healthy people and associated with well-being. The evidence that a little not-knowing about oneself is part of a normal life.
The rule that a person's wish not to be informed about their own health is to be respected. Born from genetic testing, where the answer is certain, unchangeable and lifelong.
When people expect information to remain available, they remember where to find it rather than the information itself. Not knowing outsourced, and felt as knowing.
Knowing the limits of one's own knowledge. The conscious form of not knowing that learning gives back in exchange for the unconscious one it takes.
One side of a deal knows something the other cannot see: the seller the quality, the borrower his own risk, the applicant her own ability. The condition under which markets thin out or fail.
When the uninformed side prices for the average, the better half of the informed side leaves, the average worsens and the price follows. The lemons mechanism, named after the used-car lot.
An action by the informed side that would not pay for a lemon: a guarantee, a degree, an inspection at one's own expense. It works because it costs the good less than the bad.
A menu designed by the uninformed side so that the informed side reveals itself by choosing: deductibles, tariffs, contract lengths.
Trying new options in order to learn, against using the best one known. Every learning system has to divide its time between the two; Gopnik sees the long human childhood as a protected phase of exploration.
The desire for knowledge, as opposed to perceptual curiosity, which new stimuli arouse and repetition dulls. Loewenstein explained it in 1994 as the feeling of a gap in what one knows.
One of the five broad personality traits: interest in ideas, art, feelings and the new. The trait most consistently linked with creative achievement, with small to medium effects.
Shoshin, a Zen term for meeting a subject without the fixed expectations of the expert; the counterpart to the curse of knowing in the fourth tab.
A building in which an inspector in the centre can see every inmate without being seen; from Greek pan, all, and optikon, of seeing. Bentham coined the word.
Foucault's term for the principle behind it: power that works because it is visible and unverifiable, so that the watched take over the watching.
Lawful behaviour that people give up because they expect to be watched or sanctioned. German courts speak of an Einschüchterungseffekt.
The loss of effort that occurs when control signals distrust. Often smaller than the gain from control, sometimes larger.
Recording and analysing work with digital tools: keystrokes, screens, scanners, location, calls. The research term behind most studies in the tab on the panopticon.
Full visibility can lower performance, because people hide deviations, the useful ones included; a measure of privacy can raise it.
The claim that being observed in itself changes behaviour, named after the Western Electric plant in Cicero near Chicago. The original data do not support the famous version.