Why You Can't Tell When Someone Is Lying

It's Monday, October nineteenth. I saw the Nathan Fielder documentary You Can See Everything on Friday. It follows Elizabeth Holmes through the thirty-four days before she reported to prison, filmed inside her home with her partner and their two children.

What I keep thinking about is how she talks. In the film she'll describe something one way, then describe it differently later, and sound equally certain both times. Fielder said something afterward that matched what I noticed. He said she seems so genuine at times, and then he catches himself and remembers what she did.

That raised a question I wanted an actual answer to. If you sat her down with a polygraph, would it catch anything? And more broadly, can anybody tell when someone is lying?

Start with the rest of us. In 2006, two researchers published a meta-analysis of every study they could find on lie detection. Two hundred and six documents, and nearly twenty-five thousand people who were asked to spot the lies.

People got it right fifty-four percent of the time. Half the time is a coin flip. They correctly identified forty-seven percent of lies and sixty-one percent of truths.

Experience didn't help. People whose whole job involves this scored no better. Police officers came in at fifty-four and a half percent. Courtroom judges and psychiatrists landed slightly lower, at fifty-three percent. The same as someone with no training at all.

A follow-up analysis found the explanation. There aren't reliable behavioral differences between people who are lying and people who aren't. The things everyone believes in, the eye contact, the fidgeting, the pause before an answer, don't predict anything.

And lies get harder to spot the more they're told. If someone has repeated the same story several times, you're worse at catching it, not better.

So what about the polygraph? The National Academies reviewed the entire body of research on it in 2003, and the report is not flattering.

The first problem is what the machine measures. A polygraph records pulse, blood pressure, breathing, and how much you sweat. It doesn't measure lying. It measures arousal. The standard method assumes that lying produces more physiological stress than telling the truth, and that assumption has never been validated.

The second problem is arithmetic. The report walks through a hypothetical screening of ten thousand people, ten of whom are actually guilty. The test catches two of those ten. It also fails thirty-nine innocent people. So it misses eighty percent of the people it's looking for, and produces about twenty false accusations for every real catch.

The third problem is that you can beat it. Their conclusion: all of the physiological indicators the polygraph measures can be altered by conscious effort, physical or mental, and there's enough research to justify concern that effective countermeasures can be learned.

There's also a popular idea that psychopaths slip past the machine because they don't feel stress about lying. In a prison study that tested exactly that, psychopaths were detected just as easily as everyone else. They showed stronger stress responses, not weaker. A later review of the research calls the psychopath loophole a myth.

So the machine misses most of what it's looking for, flags innocent people at a higher rate than guilty ones, and can be defeated by effort.

Now the part that speaks to the film. In 2016, researchers at University College London ran an experiment. Volunteers looked at pictures of a glass jar full of pennies and sent estimates to a partner they never met. In some versions of the task, overestimating the number would earn them more money at the partner's expense.

People started with small exaggerations. Then the exaggerations grew.

Twenty-five of them did the task inside an fMRI scanner. The amygdala, which handles emotional processing, was most active the first time someone lied for their own gain. With each repetition that response declined while the lies got bigger. And the size of the decline predicted how much larger the next lie would be.

The researchers describe it as emotional adaptation, the same thing that makes a smell fade after a few minutes. The first time you cheat on your taxes, it feels terrible, and that bad feeling is doing real work. It's what stops you from going further. Once you've adapted, there's less resistance holding you back.

One evolutionary argument goes further than that. Self-deception may exist because it makes you a more convincing liar. If you genuinely believe the story, you don't leak the signals that give away someone who's consciously lying.

And memory works in the same direction. Memories aren't recordings. Each time you retrieve one, you rebuild it, and rebuilding introduces small errors. A story told often enough, with small edits along the way, stops being distinguishable from what happened. That's just as true for the person telling it.

Elizabeth Holmes is in a prison in Texas, about three years into a ten-year sentence. I don't know what she believes. A film can't tell me either.

What the research says is that the machine can't tell me. It measures stress, and stress is something a person can control or simply stop feeling. Inside somebody's head, a version of events they've told enough times becomes the version they remember.

Years from now she'll be out, and whatever she says about Theranos, there won't be a device in the room that can tell you whether she means it.

Stay informed, stay curious, and we'll see you tomorrow.