A new machine learning algorithm wants to tap into digital interactions that reveal when you're bluffing.
Researchers are finding ways to turn your phone into a lie detector.
A new machine learning algorithm developed by computer scientists at the University of Copenhagen can identify honesty -- and dishonesty -- by analyzing how you swipe or tap a smartphone. Dishonest interactions often take longer and involve more hand movement than honest ones, according to a research paper (PDF) to be published Friday.
The algorithm, called Veritaps, flashes a green check mark when truthful statements are entered into a smartphone and notes doubtful information with a red question mark. It gives recipients of questionable messages the option of asking the sender for more information. The experimental app runs on Android phones and isn't available to the public.
Aske Mottelson, one of the paper's authors, says the algorithm's ability to detect lies is comparable to that of a polygraph. However, it has its limits and shouldn't be used in courtrooms or other high-stakes environments, he said.
Humans have long been fascinated with the idea of detecting lies and have spent decades trying to create a device that can unequivocally tell when someone's bluffing. They're still working on it. The polygraph is the closest we've come to creating a gold standard in truth evaluation, but it's faced a fair share of criticism over accuracy since it was created in 1921.
Technology has tried to offer up 21st-century solutions, all of which rely to some degree on the body's response to lying. Converus, a startup, created a test that analyzes truthfulness by measuring subtle changes in a person's eyes, such as pupil dilation and blink rate. Its EyeDetect system is 86 percent accurate, the company says. The technology isn't meant to replace polygraphs, but rather to serve as another lie detection tool.
Toronto-based NuraLogix uses a conventional video camera -- like the one found in your phone or laptop -- to capture changes in blood flow in a person's face. The technology, called Transdermal Optical Imaging, then applies machine learning algorithms that can detect things like heart rate and blood pressure. People can't control these physiological responses, which makes it easier to tell when they're lying, says Kang Lee, the company's chief science officer.
Lee says they're using this technology to develop an app that can monitor a person's physical and psychological health. The app, which is in beta testing for iPhone, can measure your stress, heart rate and blood pressure in about 30 seconds by having you look at your phone. These same capabilities can easily be carried over into a lie detection app, he says.
"Machine learning and deep learning allow us to learn about patterns of lying versus truth telling in a much better way," Lee said. "That's a game-changer."
The researchers behind Veritaps conducted three studies to gauge how dishonesty affected a user's interactions with a mobile device. What they found is that liars paused before responding.
In one study, participants were asked to either lie or tell the truth about a color shown on the screen of their phones. The liars took longer on average to respond.
In another test, a person received money and was told to split it with a second person. When given the option to lie about the amount received, however, people generally took more time before entering the amount to give away.
In the last study, participants played a mobile dice game. Players weren't encouraged to lie but were rewarded based on their reported score, making it profitable for them to do so. The researchers found that truthful entries were tapped closer to the center of the screen and with more pressure. Dishonest players used more hand movement than honest players.
Veritaps might have implications for things such as tax returns, insurance claims and online marketplaces, the researchers say. For example, it could be used to evaluate the probability that the description of a car's condition in a Craigslist for-sale ad is accurate, or to flag fraudulent tax reports and insurance claims. The researchers also see Veritaps aiding in self-improvement; it could tell users when they're fibbing to themselves about skipping an appointment with their personal trainer, for instance.
Of course, mobile lie detection raises ethical questions. NuraLogix's Lee worries about invasion of privacy, given that personal information, such as heart rate and blood pressure, is collected. Mottelson says Veritaps or a similar algorithm could cause rifts between friends if they become suspicious of a questionable message. It could also lead to a rise in unwarranted suspicion in regard to online transactions. So, the Veritaps study cautions that its results shouldn't be seen as lie detection, but rather as a flag for information that should be verified.
"It's easy to foresee how this can be used in nonoptimal ways," Mottelson said. "What we suggest is that it doesn't report on lie classification, but instead it reports on truths, because that means that you can verify an interaction."
For better or worse, it seems smartphones will soon be smart enough to detect our unspoken truths, and our not-so-subtle lies.
First published April 19, 5:00 a.m. PT.
Update, April 20 at 9:24 a.m.: Adds more details on NuraLogix.
Tech Enabled: CNET chronicles tech's role in providing new kinds of accessibility.
Blockchain Decoded: CNET looks at the tech powering bitcoin -- and soon, too, a myriad of services that will change your life.