This is Science Fictions with Stuart Ritchie, a subscriber-only newsletter from i. If you’d like to get this direct to your inbox, every single week, you can sign up here.
Here’s something you might’ve missed from this week’s King’s Speech: news about lie detectors. Amid various policies on fighting crime such as longer prison sentences and more powers for police to enter buildings where they’ve detected stolen mobile phones, the Government also announced that it was expanding the use of polygraph tests.
There seems to be some confusion over exactly what the change is. The Times on Wednesday reported that “lie detectors will be used in more cases than just those involving terrorism and domestic abuse”.
However, in the Government’s briefing notes that expand on the policies in the King’s Speech, it just states that the plans are to “give probation officers more powers … to use polygraph tests on serious terrorist or sexual offenders to better manage their risk”.
Maybe something got garbled along the way. But regardless of exactly how they’re being used, it’s surprising – and depressing – that polygraph tests are being used in our justice system at all – because polygraph tests don’t work.
Are polygraphs 80-90% accurate?
Here’s something reassuring: polygraphs – which take various physiological measurements, often including heart rate, blood pressure, breathing rate, and sweating rate, some or all of which are assumed to change when someone is lying – aren’t being used in court to help decide whether or not a suspect is guilty. Indeed, they aren’t admissible as evidence in any court in the UK (a fact that should stick with you: if courts won’t accept them, why should anyone else?).
Instead, they’re being used on convicted terrorists, sex offenders and domestic abusers to make sure they’re sticking to the rules of their licence after they’re released from prison. It’s not exactly clear, but one assumes they’re asked questions about who they’re associating with, where they’ve been, and what material they’ve accessed on the internet, and if caught in a lie they might be recalled to prison.
Are we confident that the polygraph can really tell when one of these criminals is lying? We have to rely on the evidence, and the UK Government states (without giving a source) that the polygraph test is “80-90 per cent accurate”.
Those kinds of numbers are often thrown around in discussion of the polygraph, and come from various scientific studies and reviews. They make the polygraph sound like a near-magical machine that can detect dissembling with astonishing accuracy. But there are some big problems with the kinds of studies that come up with such high accuracy rates for the polygraph test.
First, people in a laboratory study of deception, perhaps where they’ve been asked to pretend to have committed a crime as part of a game, are unlikely to react the same way during a polygraph test than those who really have been accused of committing a crime. Not only that, but in lab studies you know for sure who has committed the “crime” (because there’s no actual crime and it’s an artificial situation). That’s not the case in real life, where verifying whodunit is vastly harder. That means that what we want to see are real-world studies of the polygraph test, where genuine suspects are being tested.
Right off the bat, then, we rule out an awful lot of studies, which weren’t done in real-life situations where a crime really has occurred.
But the truly important issue is one to do with confessions.
The confession confounder
Often in a criminal case a confession is what separates the guilty from the innocent CUT, and there isn’t any outside evidence that clinches the case CUT. Imagine you’re testing a set of potential suspects with a polygraph. For one suspect, the machine buzzes: you’ve caught them in a lie! In most cases you’ll then ask that suspect to confess.
Let’s imagine they do: “It’s a fair cop. You got me. I did it.” In that case, it’s a win for the polygraph: at least according to the confession, it successfully identified the right person. You can add that to the record: the polygraph worked and found the person who committed the crime.
But imagine they don’t confess. Imagine none of the suspects confess. In that case, there’s no answer to the question of who committed the crime: there’s still just a big question mark. Crucially, you can’t say whether or not the polygraph worked, because – again, since the confession is often all you have to go on – you don’t have any evidence that any given suspect committed the crime.
Cases like that don’t get included in the results of polygraph studies, because there’s no “ground truth” of who actually committed the crime to go on.
You can see why this is a problem: the studies will be strongly biased to include only (or largely) the first kind of case, where the polygraph seemed to have “worked”, and not the second kind of case, where it didn’t. This amazingly circular process is a very plausible explanation for the very high (“80-90 per cent”) numbers you see bandied about for polygraph effectiveness.
As one 2019 study put it: “If polygraph testing actually had no better than chance accuracy, by basing validity studies on confession-verified [results] selected in this manner, a researcher could misleadingly conclude that the technique was virtually infallible”.
Oh dear.
The truth
Don’t take my word for it. Back in 2003, the US National Academy of Sciences put together a report on “The Polygraph and Lie Detection”, which concluded that “almost a century of research in scientific psychology and physiology provides little basis for the expectation that a polygraph test could have extremely high accuracy”. And it’s not like there’s been any breakthrough evidence since then.
The 80-90 per cent number quoted by the UK Government just isn’t supported by reliable science, and we have good reason to think it’s drastically inflated. Is the real number above 50 per cent? That is, is the polygraph better than just flipping a coin to work out whether someone is telling the truth or lying? Possibly. It’s not completely implausible that there’s some combination of measurable physiological clues to deception.
But “somewhere above 50 per cent” isn’t what we want in a criminal-justice context, especially if we’ve fooled ourselves into believing that the true accuracy is 90 per cent and thus where we might rely on the machine far more than we should. That’s not even to mention the fact that many people are aware of how the polygraph works and can use so-called “countermeasures”, like specific breathing techniques, to fool the test.
The 2003 report noted that: “Countermeasures pose a potentially serious threat to the performance of polygraph testing because all the physiological indicators measured by the polygraph can be altered by conscious efforts…”.
At the very least we want the people using the polygraph to be very aware of this and other limitations that reduce its accuracy – and in quoting the “80-90 per cent” figure, the Government has shown that’s not the case.
It seems likely that, despite the evidence, many people have been hoodwinked by the “science-y” nature of the polygraph. They think: “It measures all these different things, and spits out this very convincing-looking graph. I’ve seen police using it on TV shows dozens of times! How can it be wrong?”
But if we really want “evidence-based policy”, we’d need a lot more than the appearance of science. We’d need studies that convincingly showed it was an accurate tool. We simply don’t have that at present.
This isn’t just an “academic” discussion. An error on the polygraph could help a terrorist, sex offender, or domestic abuser to continue to live in the community when in fact they’re breaching their licence. Given the potential for re-offending, that could be a big mistake to make.
No word of a lie: the UK government should scrap the polygraph entirely.
Other stuff I’ve written this week
As well as the stuff about lie detectors, another one of the announcements in the King’s Speech was about self-driving vehicles. Here’s my article on how they’ve certainly been over-hyped, but shouldn’t be ruled out just yet.
Science link of the week
Do you remember the whole room-temperature superconductor drama from a few months ago? Well, there’s a whole entirely separate room-temperature superconductor drama that’s playing out in multiple scientific journals, including Nature. You can read the most recent update here.
This is Science Fictions with Stuart Ritchie, a subscriber-only newsletter from i. If you’d like to get this direct to your inbox, every single week, you can sign up here.
