How Liars Give Themselves Away: New Methods for Detecting Deception

Lying is considered immoral and wrong, yet as humans, we do it over and over again. Deception is present in all areas of our lives, from social interactions to matters of national security. At the same time, existing methods for detecting lies are still not reliable enough. That’s why scientists are searching for new, objective ways to identify deception.

Analyzing Micro-Expressions to Detect Lies

Anastasia Shuster from Tel Aviv University and her colleagues have found a way to “read” tiny facial muscle movements to help determine if someone is lying. Their system correctly detected lies in an average of 73% of cases. “It’s not perfect, but it’s much better than any existing facial recognition technology,” notes one of the study’s authors, neuroscientist Dino J. Levy.

In their 2021 experiment, electrodes measured facial muscle movements in 40 volunteers who were either lying or telling the truth. A machine learning algorithm was then used to gradually learn patterns in their facial expressions.

“According to our data, deception is revealed through involuntary micro-expressions lasting 40-60 milliseconds, which don’t match the emotions the face is trying to convey with the lie,” the research team explained. People perceive these lightning-fast, involuntary muscle reactions unconsciously—and they can at least partially supplement our judgments about lying.

To capture these brief twitches, the team developed very small but highly sensitive electrodes that easily detect the electrical impulses of facial muscles.

• Liars show contractions in specific facial muscles.

During the study, two people connected to electrodes sat facing each other. One person wore headphones and had to either repeat what they heard (truth) or say a different word (lie). The other person had to guess whether their partner was lying. Meanwhile, electrodes on the participants’ foreheads and cheeks recorded their involuntary muscle reactions.

The participants themselves only recognized when their partner was lying about half the time. The electrodes, however, achieved much better results: by detecting muscle reactions in the face, scientists were able to identify lies in all 40 participants.

Researchers recorded activity in the corrugator supercilii muscle (between the eyebrows) and the zygomaticus major muscle (cheek area) as participants listened, spoke, and responded. They also tracked the connections between facial muscles in these areas. Additionally, the experiment found no link between how long a person thought about their answer and whether they were lying.

Levy and his colleagues wrote in their article that, using machine learning technology, they successfully detected lies in all participants. However, the experimental algorithm still needs refinement, and the study showed that people’s characteristic muscle patterns tend to change over time.

• (a) Each participant’s face was fitted with a matrix of eight electrodes: five recorded the zygomaticus major (cheek area), and three recorded the corrugator supercilii (eyebrow area).
• (b) Participants took turns as Sender and Receiver. The Sender heard a word through headphones (“KAV” or “ETZ”), then either repeated the word (Truth) or said a different word (Lie). The Receiver indicated by pressing a key whether they believed the Sender (Truth) or not (Lie). Two trial examples are shown: lie-truth (top panel) and truth-lie (bottom panel).
• (c) Participants lied in about half the trials, and the frequency of lying increased between the two stages of the experiment (top panel). Receivers’ ability to detect lies was random, and time or monetary incentives did not affect their performance (bottom panel).

“Interestingly, people who were able to successfully deceive their human partners were also poorly detected by the machine learning algorithm,” the researchers noted.

However, the pattern of involuntary muscle tension does not appear the same in everyone. Scientists observed two completely different patterns. In some participants, only the eyebrow muscle (corrugator supercilii) reacted during lying. In others, the forehead remained still, but the muscle that lifts the corner of the mouth (zygomaticus major) twitched.

Both muscles play important roles in emotional facial expressions. The zygomaticus major is responsible for smiling and is associated with positive feelings and laughter. The corrugator supercilii is more often linked to negative emotions and a displeased facial expression. Why these two opposite muscles react to lying remains unclear.

“Since this was an initial study, the lies were very simple,” explains Levy. “In our study, it was very easy to lie. But in real life, we tell much longer and more complicated stories that mix truth and fiction.” The team plans to investigate how muscle reactions manifest in such cases in future research.

Over time, scientists hope to teach artificial intelligence to analyze facial movements without electrodes. Special high-resolution cameras could capture micro-expressions.

Analyzing Words and Gestures

Previously, in 2015, specialists from the University of Michigan developed a program that could recognize lies in 75% of cases. Rada Mihalcea and her colleagues used 121 video recordings from public court hearings—61 with deceptive testimony and 60 with truthful testimony. The average video length was 28 seconds.

Court hearings are an ideal source of such data because facial expressions and gestures are more pronounced. Defendants are highly motivated to be believed, so they clearly display gestures and facial expressions.

In developing their program, the American researchers focused on analyzing words and gestures, rather than physiological parameters like those used in polygraphs. They were able to correctly identify lies in three out of four cases.

• During training, the algorithm was provided with video footage, textual descriptions of the defendants’ gestures and facial expressions, and transcripts of their words. For example, the following signs were observed in the videos: head movement forward (deceptive testimony), movement of both hands (deceptive), movement of one hand (deceptive), raised eyebrows (truthful), furrowed brows (deceptive), looking up (truthful).

The following behavioral traits were found in liars and truth-tellers. Liars tend to frown, gesture with both hands, and stare intently at the questioner. They also tend to use interjections. Truth-tellers often raise their eyebrows, close their eyes, and shake their heads. They also use first-person pronouns more frequently.

Future Prospects for Lie Detection

The methods described above could become alternatives to traditional polygraphs in the future.

The polygraph works by analyzing physiological reactions such as heart rate, blood pressure, breathing rate, temperature, and skin conductivity. These are functions people can, in principle, learn to control, allowing them to beat the polygraph.

Despite continued use by law enforcement, polygraphs are often inaccurate. The American Polygraph Association claims its specialists can detect lies and truth with up to 90% accuracy. Critics say the real figure is closer to 70% or even lower. There have been many errors in polygraph testing, where innocent people were accused of serious crimes and real criminals escaped punishment.

By using new methods of lie detection (analyzing facial muscle contractions, words, and gestures) together with physiological measurements, it may be possible to improve the effectiveness of lie detection.

In the future, electrodes may no longer be needed, as special high-resolution cameras could detect muscle twitches. In banks, police interrogations, or airports, such cameras with built-in artificial intelligence could distinguish truth from deception.