Can AI tell if you’re lying?

You might think you're a good liar but does AI beg to differ?

Lying - we’ve all done it. But how accurate are we really at telling if someone is lying to us? And could a machine be better?

Although you may think you can always tell when someone is lying to you, studies suggest that we are only about 54% accurate at detecting deception. Research has revealed that we are usually very bad at telling if strangers are lying, as we do not know them well enough to notice if things like their mannerisms or voice change when they are not telling the truth. Surprisingly, we are also usually very bad at telling when people we are very close to, like family, are lying, possibly because they know us well enough to be able to fool us when they want to. That means that we are best at noticing lying in friends, people we spend a lot of time with but do not know extremely well.

Contrary to popular belief, non-verbal communication (body language) is actually very unreliable when trying to work out if someone is lying. Scientists say we should focus on verbal communication; for example, the pitch of many people’s voice goes up when they lie. However, this is hard to detect unless you’re listening out for it. We also have to take into account how long someone has had to prepare a lie. When people are put on the spot, their answers tend to be short and general, but when they have had some time to create a lie, their answers are often longer and more detailed. But again, this is pretty broad as the way people lie differs greatly from person to person.

So, humans are hardly better than chance when detecting lies. But can a machine do any better? A new lie detector has been introduced at border control. It is called iBorderCtrl and has been implemented in several airports to assist border guards in their questioning, trying to eliminate unconscious bias. It is based on a system called Silent Talker, which combines information from 38-40 non-verbal channels from the face simultaneously. It tries to generalise about deceptive non-verbal behaviour. However, non-verbal communication is unreliable when detecting lies. So, as this may assist border guards, it is definitely not foolproof.

What about a traditional lie detector (polygraph) test? During a polygraph test, sensors are attached to the person’s body which usually measure breathing rate, pulse, blood pressure and perspiration. When the test starts, the questioner usually asks the person taking the test a few simple questions to establish the norms for their signals. The questioner looks at the graphs after the test to see if anything changed. A significant change e.g. higher blood pressure or increased perspiration usually means the person is lying. This is by far the most reliable way of detecting lies, but it is not perfect, and a good liar can fool the test.

So basically, although humans are pretty bad at telling if someone is lying, there is no foolproof way of detecting lies. And that essay that was due a week ago is totally in your locker...

Ruby V