The robot that knows when you're lying: Scientists create an AI that can detect deception in the courtroom (and it's already 'significantly better' than humans)

  • The system, called DARE, was trained by watching 15 videos of people in court
  • It was trained recognised five expressions that indicate someone is lying
  • These are frowning, raised eyebrows, lips turning up, lips protruded and head tilt
  • In a final test, the system performed with 92 per cent accuracy
  • The researchers describe this performance as 'significantly better' than humans 

From a raise of an eyebrow to a tilt of the head, we use several micro-movements when we're lying without even knowing it.

Now, scientists have developed an artificial intelligence system that can detect these micro-expressions and detect if you're lying – and it's already 'significantly better' than humans.

The researchers hope their system could soon be used in courtrooms to tell if people on the stand are telling the truth.

Scroll down for video 

The team from the University of Maryland and Dartmouth College trained the AI to recognise five micro-expressions known to indicate that someone is lying - frowning, eyebrows raising (pictured), lip corners turning up, lips protruded and head side turn

The team from the University of Maryland and Dartmouth College trained the AI to recognise five micro-expressions known to indicate that someone is lying - frowning, eyebrows raising (pictured), lip corners turning up, lips protruded and head side turn

HOW DOES IT WORK?

The researchers trained the AI to recognise five expressions known to indicate if someone is lying - frowning, eyebrows raising, lip corners turning up, lips protruded and head side turn.

After watching 15 videos from courtrooms, DARE was then tested on whether it could tell if someone was lying in a final video.

The researchers said that DARE managed to spot 92 per cent of the expressions, which the researchers describe as a 'good performance on the final deception detection task.' 

To compare how effective DARE was, the researchers gave the same task to human assessors.

Results showed that the AI was better than humans at spotting a liar.

Advertisement

The AI system, called Deception Analysis and Reasoning Engine (DARE), has been developed by researchers from the University of Maryland and Dartmouth College.

To develop DARE, the researchers trained the system using videos of people in the courtroom.

In their study, published in arXiv, the researchers, led by Dr Zhe Wu, said: 'On the vision side, our system uses classifiers trained on low level video features which predict human micro-expressions.'

The team trained the AI to recognise five micro-expressions known to indicate that someone is lying - frowning, eyebrows raising, lip corners turning up, lips protruded and head side turn.

After watching 15 videos from courtrooms, DARE was then tested on whether it could tell if someone was lying in a final video.

Results showed that DARE managed to spot 92 per cent of the micro-expressions, which the researchers describe as a 'good performance.' 

To develop the AI system, the researchers trained the system using videos of people in the courtroom. After watching 15 videos from courtrooms, DARE was then tested on whether it could tell if someone was lying in a final video

To develop the AI system, the researchers trained the system using videos of people in the courtroom. After watching 15 videos from courtrooms, DARE was then tested on whether it could tell if someone was lying in a final video

Lips protruding are known to signal that someone is lying
Frowning is a indicator that someone is lying

Micro-expressions known to indicate that someone is lying - frowning (pictured right), eyebrows raising, lip corners turning up, lips protruded (pictured left) and head side turn. The team used court videos (pictured) to train the AI 

The researchers then gave the same task to human assessors, who were only able to pick up 81 per cent of micro-expressions. 

Results showed that the AI was better than humans at spotting if someone was lying.

The researchers said: 'Our vision system, which uses both high-level and low level visual features, is significantly better at predicting deception compared to humans.'

Results showed that DARE managed to spot 92 per cent of the micro-expressions, which the researchers describe as a 'good performance'

Results showed that DARE managed to spot 92 per cent of the micro-expressions, which the researchers describe as a 'good performance'

The researchers suggest that the system could be even more effective if the AI was provided with further information.

They added: 'When complementary information from audio and transcripts is provided, deception prediction can be further improved.'

CAN AI JUDGE IF YOU HAVE THE FACE OF A CRIMINAL?

Last year, a controversial paper was released, which investigated whether a computer could detect if a human could be a criminal, by analysing their facial features.

The study involved 1,856 faces of Chinese men aged 18 to 55, which were 'controlled' to account for 'race, gender, age and facial expressions.'

730 of the photos belonged to criminals – although the images were not mugshots.

The images were fed into a machine learning algorithm, which used four different methods (classifiers) of analysing facial features, to infer criminality.

The researchers write: 'All four classifiers perform consistently well and produce evidence for the validity of automated face-induced inference on criminality, despite the historical controversy surrounding the topic.

'Also, we find some discriminating structural features for predicting criminality, such as lip curvature, eye inner corner distance, and the so-called nose-mouth angle.'

Advertisement

The comments below have not been moderated.

The views expressed in the contents above are those of our users and do not necessarily reflect the views of MailOnline.

We are no longer accepting comments on this article.