Home / Other / The Painful Truth About Lie Detectors: There Are No Infallibles

The Painful Truth About Lie Detectors: There Are No Infallibles

We all want to know when someone is lying, and our entire culture is fascinated by the ability to detect liars. The flood of detective series cannot pass without a quality scene on a lie detector, and many series also show human polygraphers who uncover various frauds and lies. Records of attempts to detect lies, whether by technical means or by observing skilled observers, date back at least 3000 years, and forensic techniques for detecting lies have become increasingly popular since the invention of the polygraph in the early 20th century, with the latest methods involving advanced brain imaging. However, despite everything, an infallible method of lie detection has yet to be found. In fact, most methods of lie detection do not detect lies at all – instead, they register physiological or behavioral signs of stress or fear.

Old Methods

Methods for detecting or identifying lies have likely existed since humans have. A stick across the back seems like a good starting method for detecting whether a cave colleague is lying or if someone else really ate the rabbit meant for everyone. Of course, such a thing is not documented, but one of the first documented examples of lie detection dates back to around 1000 BC. The Chinese, in fact, had a method involving rice. A suspect believed to be lying had to fill their mouth with dry rice. After some time, the grains of rice would be checked, and if the rice remained dry in the mouth, the person would be punished. The theory behind this practice was that it was believed a person would stop salivating during emotional distress such as intense fear. Even after 2487 years, no better methods have been devised, so in the book ‘Malleus Maleficarum‘ (‘Hammer of Witches’), German theologian and inquisitor Heinrich Kramer writes about how to test whether witches are telling the truth or not. He mentions a test in which the accused would be thrown into water, and if they sank, they were considered innocent, while if they floated, they were surely guilty. This means that those who did not know how to swim would be innocent if they managed to wriggle out of the water after the judges decided they had spent enough time under it to prove their innocence. Swimmers seemed to be sentenced to death. Another excellent example from the Middle Ages is testing with hot coals. The suspect had to take hot coals in their hands, and if they burned themselves – they were guilty.

Are You – Not – Guilty

In addition to such recorded methods, there are many more examples rooted in mythology and legends. For instance, in Egypt, tears were considered a sign of truth, so if the accused shed any during the trial, they would be quickly acquitted of all charges. The ancient Egyptians were clever, and they are credited with the branding method. A suspect would be branded with a seal that had ‘special’ ink, and if the ink print remained on the skin, they were guilty. Historically, it was very difficult to escape one’s cursed fate because all methods were generally such that you were guilty.

Despite all the techniques and technologies, an infallible method of lie detection has yet to be found. In fact, most methods of lie detection do not detect lies at all – instead, they register physiological or behavioral signs of stress or fear.

Only in the early 20th century did the first machines for lie detection, the so-called polygraphs, emerge. The most well-known of these is the analog polygraph, which usually has three or four ink-filled needles dancing on a strip of moving paper. The suspect has sensors attached to their fingers, hands, and body, and the machine then measures breathing rate, pulse, blood pressure, and sweating while the suspect answers a series of questions. However, there are also problems with the accuracy of the machines and the possibility that a skilled person can deceive them, so such systems are constantly being improved, and recently new, artificial intelligence-based ones have emerged.

Training the Algorithm

Research shows that people are generally poor at distinguishing lies from truth, leading to the belief that artificial intelligence will help improve our chances and will be better at detecting truth than cunning old-fashioned techniques like polygraph tests. Today, it seems that all hopes are placed on AI, whatever the topic. AI-based lie detection systems could one day be used to help us separate facts from fake news, assess claims, and potentially even spot fabrications and exaggerations in job applications, so lies from CVs could disappear without a trace. Perhaps it should also poke around university diplomas in these areas, and we might find out sooner who studies, I don’t know, say in Tuzla. In a recent study, Alicia von Schenk and her colleagues developed a tool that is significantly better than humans at spotting lies. Von Schenk, an economist at the University of Würzburg in Germany, and her team then conducted some experiments to discover how people used it. In a way, the tool was helpful because those who applied it were better at spotting lies, but it led people to make many more accusations against others – the conclusions of the experiment. In their study published in the journal iScience, von Schenk and her colleagues asked volunteers to write statements about their weekend plans. Half the time, people were encouraged to lie; a credible but untrue statement was rewarded with a small cash payout. The team collected a total of 1536 statements from 768 people. They then used 80 percent of these statements to train the algorithm on lies and truths using Google’s AI language model BERT. When they tested the resulting tool on the last 20 percent of statements, they found that the AI tool could successfully determine whether a statement was true or false in 67 percent of cases. This is significantly better than the typical human, who is usually correct only half the time.

AI Better Than Humans?

The study revealed that respondents who used the tool genuinely believed the results of artificial intelligence and ‘heavily relied on its predictions.’ Von Schenk says that this heavy reliance on AI can shape our behavior because people generally assume that others are telling the truth. This was confirmed in the study. Namely, although respondents knew that half of the statements were lies, only 19 percent labeled them as such. But this changed when people decided to use the AI tool, as the accusation rate rose to 58 percent. – In a way, this is a good thing because these tools can help us spot more lies we encounter in our lives, such as misinformation we might come across on social media. However, it’s not all rosy. Such a tool could undermine trust, a fundamental aspect of human behavior that helps us build relationships. If the price of accurate judgments is the deterioration of social connections, is it worth it? – researchers of the study wonder.

Records of attempts to detect lies, whether by technical means or by observing skilled observers, date back at least 3000 years, and forensic techniques for detecting lies have become increasingly popular since the invention of the polygraph in the early 20th century, with the latest methods involving advanced brain imaging.

And there is also the question of accuracy. In her study, von Schenk and her colleagues were only interested in creating a tool that is better than humans at detecting lies. This is not too difficult, considering how terrible we are at it. But she also imagines that a tool like hers is used for routine assessments of the truthfulness of posts on social media or for searching for false details in resumes. In such cases, it is not enough for the technology to just be ‘better than humans’ if it will make more accusations.

Sinful Polygraphs

Von Schenk notes that we should remember the fallibility of historical lie detection techniques. The polygraph was designed to measure heart rate and other signs of ‘excitement’ because it was believed that some signs of stress are unique to liars, but they are not. And we have known this for a long time. For this reason, the results of lie detectors are generally not admissible in American court proceedings. Imperfect tools of artificial intelligence may have an even greater impact because they are so easy to scale, adds von Schenk. In one day, you can polygraph only so many people. The scope for AI lie detection is almost unlimited compared to previous methods.

According to the European Artificial Intelligence Act, high-risk artificial intelligence systems such as AI detectors that include emotion recognition in educational institutions and workplaces will be restricted and banned. However, restrictions for AI technology with ‘unacceptable levels of risk’ will only come into effect in six months, and the actual law will not be passed before 2025. It will then take another two years before all rules and obligations begin to be applied.

– Given that we have so much fake news and misinformation, these technologies can be useful. However, you really need to test them – you need to be sure that they are significantly better than humans. If the AI lie detector generates many accusations, it might be better not to use it at all – concludes von Schenk.

DIGITAL FRAUD DETECTION: An Application for a Broader Perspective

For just a hundred dollars, you too can become a real AI detective with this new tool/application for lie detection. LiarLiar.AI is an advanced lie detection tool powered by AI and computer vision technology and is a blend of technology and psychology. Unlike the polygraph, which relies on physiological reactions, LiarLiar.AI analyzes video sources in real-time to detect micro-expressions, fluctuations in heart rate, and subtle changes in body language. ‘This comprehensive approach encompasses more potential indicators of fraud, offering a broader perspective,’ says the company on its website. LiarLiar.AI uses advanced AI algorithms for analysis using a mix of models and records the facial and body movements of the person, including gaze orientation. Additionally, it can determine a person’s heart rate by tracking micro-movements on the forehead. This wealth of data is then processed to identify potential signs of fraud. LiarLiar.AI can be used remotely and works seamlessly with any video call software, whether it’s Zoom, Google Meet, Teams, Skype…