Chapter 7 — The Tell Is Real — Just Not Where You Look

Chapter 7 — The Tell Is Real — Just Not Where You Look

Sherlock Holmes could look at a man's face and know he was lying. This is perhaps the most beloved of all his powers, and the one modern science treats with the most withering scepticism, and so it is the one where we have to be most careful — because the truth is not the simple debunking you are bracing for. The tell is real. It is simply almost never on the face, and almost never where the confident reader is looking.

Let us take the bad news first, because it is delicious.

You are a coin flip with opinions

Across decades of experiments in which people are shown others telling truths and lies and asked to sort them out, human beings detect deception at a rate of roughly fifty-four percent. A coin flip is fifty. The most exhaustive synthesis of this literature — a meta-analysis by Charles Bond and Bella DePaulo, pooling hundreds of studies and tens of thousands of judgments — found that accuracy sits just barely, pitifully, above chance, and that it hardly matters who is doing the judging. And here is the part that should make you laugh and then wince: the professionals — police officers, judges, customs agents, the people whose entire careers are built on reading liars — are, as a group, no better than undergraduates. What they have more of is not accuracy. It is confidence. They are exactly as wrong, and considerably more sure. They are, in other words, Sherlock Holmes: the certainty dial turned to maximum, the accuracy dial untouched.

Why are we so bad at it? Because we look for the wrong things, and we look for them with total conviction. We think liars avert their eyes, fidget, sweat, hesitate. The evidence says these beliefs are mostly folklore — some are weakly diagnostic, some point the wrong way, and the honestly anxious display every one of them while the practised liar displays none. The "tells" we are most sure of are the ones that fail us most reliably, and they fail us in the specific direction that convicts the nervous innocent and clears the calm guilty.

The machine that tells the same lie

If human lie-detection is a coin flip, surely the machine does better? The polygraph — the fabled "lie detector," a fixture of spy films and police dramas — measures your heart rate, breathing, and sweat, and infers deception from the wobble. It has the delicious quality of looking scientific: wires, a needle, a scrolling chart. And in 2003 the National Academy of Sciences examined it as thoroughly as anyone ever has and concluded that its scientific foundations were weak, its accuracy overstated, and its use for screening — trying to catch spies or vet employees — very close to worthless, because the physiological signals it reads are the signals of arousal, not of lying. A guilty psychopath, unbothered, sails through. An innocent person, terrified of the machine that holds their future, lights it up like a fairground. The polygraph does not detect lies. It detects fear, and then, with great theatrical confidence, calls fear a lie. It is Holmes rendered in brass and ink: a confident instrument reading a real signal and drawing the wrong conclusion from it, protected from consequences by the awe it inspires.

Even the most sophisticated version of face-reading has been oversold into myth. The psychologist Paul Ekman did serious, real research on facial expressions and on fleeting "micro-expressions" — genuine, involuntary flashes of feeling that cross the face in a fraction of a second. It is legitimate work. But it was inflated, by a hit television series and a lucrative training industry, into the fantasy that a trained expert can read deception off a face like large print — essentially, that Sherlock Holmes can be certified and deployed at airport security. The actual, measured gains in accuracy from such training are modest, contested, and a very long way from the certainty the fantasy sells. The micro-expression is real. The infallible reader of it is not.

So the confident face-reader — Holmes's signature move, the customs agent's swagger, the polygraph's needle — is, across the board, a myth. If the book stopped here, you would be entitled to conclude that reading people is simply impossible, and that the whole Holmesian dream of divining a person's secrets is dead.

Do not conclude that. Because the tell is real. We have just been looking for it on the wrong surface.

The fingerprint in the prose

Here is where the resurrection quietly, thrillingly, delivers. You cannot reliably read a person from their face. But you can, to a genuinely startling degree, read them from their words.

Every writer has an idiolect — a personal dialect, a fingerprint made of habits so deep the writer cannot feel them: the pet phrases, the comma rhythms, the ratio of Latin words to Anglo-Saxon ones, the unconscious preference for "moreover" over "besides," the tics of punctuation that survive every attempt to disguise the hand. This is the domain of forensic linguistics and stylometry, and unlike face-reading it works, because it rests not on a folk theory of nervous behaviour but on the measurable statistics of language.

The technique's coming-of-age was quiet and academic: in the early 1960s the statisticians Frederick Mosteller and David Wallace resolved a two-century-old historical mystery — the disputed authorship of a handful of the Federalist Papers — by counting function words, the tiny grammatical connectives like "while" and "upon" that authors deploy unconsciously and consistently. Hamilton used "upon" constantly; Madison hardly at all. The maths pointed, with quiet confidence, to Madison. No intuition, no genius, no flourish — just the patient statistics of habit, doing what Holmes claimed to do by magic.

And then, in 1996, forensic linguistics caught one of America's most wanted men. The Unabomber had killed and maimed with mail bombs for seventeen years, uncaught, invisible. His mistake was to demand that his thirty-five-thousand-word manifesto, Industrial Society and Its Future, be published — and when it was, a man named David Kaczynski read it and felt a chill of recognition. The phrasing, the arguments, the peculiar constructions — "cool-headed logician," the idiosyncratic rhythms — sounded exactly like the letters of his estranged brother Theodore. He went to the FBI. Linguistic analysis compared the manifesto to Ted Kaczynski's known writings and found the same hand. The idiolect was the fingerprint that seventeen years of physical investigation had failed to lift. That is a real Holmesian miracle — a man identified from the invisible traces of his own mind — and it was performed not by a lone genius reading a face, but by the disciplined, statistical reading of language.

The lesson in the difference

Now hold the two halves of this chapter side by side, because the contrast is the whole point, and it is a lesson that reaches far beyond detection.

Reading a person's character from their face and manner — the confident, intuitive, Holmesian read — barely beats a coin. Reading a person's identity from the statistical fingerprint of their behaviour — their language, their habits, the unconscious patterns they cannot suppress — genuinely works. The difference is not that one is harder than the other. The difference is that the failed version relies on a vivid folk theory and the swagger of the expert, while the successful version relies on measurable regularities and the humility to count them. The tell is real precisely where we stop trusting our intuition and start trusting the data — precisely where we stop being Holmes and start being statisticians.

This is worth pausing on, because it is a caution as much as a triumph. The very fact that our identities leak out of us in patterns we cannot control — that our words, our purchases, our movements, our clicks form an idiolect no disguise can fully suppress — is the foundation of the modern miracle of inference and the foundation of the modern machinery of surveillance. The same statistical reading that caught a serial killer can, pointed at an ordinary citizen, reconstruct things about them they never consented to reveal. We will have to reckon with that. But first we have to meet the entity that now performs this reading not on one manuscript but on all of us at once, tirelessly, at the speed of light — the machine that has become the twenty-first century's Sherlock Holmes, complete with his genius and, alarmingly, complete with his overconfidence.