Recently, Hertfordshire Constabulary completed a pilot scheme that involved conducting polygraph tests of a sample group of convicted sex offenders.
Deployed more frequently in the US, the polygraph is what most people think of as the traditional ‘lie detector’. In polygraph-based tests, the suspect is connected to a variety of measurement devices in order to determine whether they’re lying or telling the truth.
Apparently, the success of the Hertfordshire Constabulary pilot led to the reclassification of a number of offenders as being a higher risk than originally thought.
As is the case with law enforcement agencies, corporate investigators also find themselves trying to ascertain the truth behind suspects’ stories and any tool or device that can help distinguish liars from truth tellers is extremely powerful (and, therefore, very attractive).
In fact, the panacea of accurate lie detection has long been instilled in the public consciousness, but how valid are these tools? How strong is the scientific evidence supporting them?
After all, if they really are as accurate as is claimed, surely their use would extend beyond TV talk shows such as The Jeremy Kyle Show here in the UK and Maury over in the States?
Or is the technology so complex that it’s hidden behind the concrete walls of intelligence agencies? For the forensic technologist, how easy is it to apply those tools to the digital domain and catch online wrongdoers or fraudsters?
The subject of online deceit and lie detection is close to my heart. It was actually the subject of my doctoral thesis, and I spent many years researching the subject. What I discovered was a fascinating research area whereby the psychology of deceit was being applied to a new communications medium: a virtual world of chat rooms, e-mail and virtual reality.
In my experiments, I attempted to apply existing verbal, non-verbal and physiological lie detection techniques to text-based communication systems. What I found yielded even more questions than answers.
Diverse area of applied research
The field of lie detection is a diverse area of applied research. The reason for this lies in the broad range of psychological behaviours that underpin our common notions of ‘lies and deceit’.
For example, exaggeration, faked emotions, white lies and ‘acting’ are all forms of deceit with a very different psychological profile. Individuals and cultures can vary in their interpretation of these behaviours and, sometimes, it can be difficult to disentangle the truth from a falsehood.
This is where many lie detection techniques fall down. They fail to properly characterise the specific type of lie or deceit they’re trying to detect. For example, a fabricated witness statement that’s based (in part) on true events will require very different mental processes than faking shock at being informed of the crime in the first place. The former may require a focus on verbal indicators of deceit and the latter on non-verbal indicators.
Rather than dwell too much on these failings, I thought it might be interesting to cover some of the more successful techniques and suggest how they might be applied to ‘the digital world’.
Verbal deceit: the lie
Lies are a verbal form of deceit. However, many lie detection techniques don’t examine the contents of people’s lies in depth, preferring instead other leakage channels (such as non-verbal behaviour). The work undertaken on verbal lie detection would seem to have great applicability to text-based forums whereby many of these traditional non-verbal signals are hidden.
Many online groups are founded on principles of anonymous interaction, supposedly free from the shackles of stereotypes and marginalisation. However, the freedoms of this form of communication come hand-in-hand with the dangers of deceit. Anonymous voices from an ethereal domain may have grave real world consequences.
Textual conversation through technologies such as e-mail, Twitter and web chats tends to focus our minds on the content of what is being said, and it remains unclear as to whether this makes us better lie detectors or not. For the deceiver, verbal behaviours would seem to be easier to control than non-verbal behaviours. However, research has found that these controls can break down over time.
The most successful verbal lie detection techniques actually focus on ‘truth detection’ rather than lie detection. Techniques such as Statement Validity Assessment and Reality Monitoring are founded on the principle that, when we make up a story, we use a different part of our brain to when we remember a story. This can lead to behavioural changes that might affect the temporal order in which a story is told, the range of vocabulary, the willingness to admit mistakes or the use of subjective references (for example: “I then saw the van drive through the shop window” rather than “the van then drove into the shop”).
Generally, these techniques have been found to work best when used with children as part of a structured process of investigative interviewing and expert coding.
So, what would be required for such a technique to be used in a natural setting, say within an Internet chat room?
An obvious application would be identifying adults posing as children, and making up stories based around popular references that would not have been relevant during their youth.
I was interested in this very question, so I interviewed a number of individuals (in a chat room) using this technique, attempting to mirror the offline version as closely as possible. In addition, I monitored keyboard and mouse behaviours, such as deletions and typing speed.
Generally speaking, it takes more effort to lie than tell the truth, and I was looking to see if this extra effort interfered with motor control at the keyboard. Similarly, I was also keen to see if liars deleted more. I did notice some small effects along these lines, although nothing was particularly conclusive.
One of the interesting observations that altered how the technique could be applied concerned how our communication changes when we’re online. For example, the structure of conversational ‘turn taking’ can become confused and the text-based conversation may become either more formal or sloppier depending on a variety of factors (such as typing skills, familiarity with the medium and the need to respond quickly in acknowledging receipt of a message).
I still believe that further research in this area is needed, perhaps focusing on cues such as stereotypical language or using vocabulary and concepts that don’t fit with the purported identity of your chat partner.
“Look into my eyes”
Non-verbal techniques are more closely tied to our emotions, in the main because our emotions seem to be directly linked to our physiological systems of non-verbal communication.
Often, signs of stress or nervousness are used to indicate deceptive behaviour, such as playing with your hands, looking away or an increase in voice pitch.
Naturally, it’s important to be clear that a sign of stress is not a sign of lying per se. For example, someone may be exhibiting signs of stress because you’re accusing them of lying even if they’re innocent! Insurance companies have been known to use voice pitch as an indicator of potentially fraudulent claims, and there’s some discussion around using thermal imaging cameras in airports.
Some interesting work has been carried out around the facial muscles that can or cannot be controlled when expressing certain emotions. For example, the muscles that produce ‘crows feet’ when you’re genuinely smiling or laughing cannot be moved if you’re forcibly creating those expressions.
The scientist Professor Paul Ekman, consultant for the US TV show ‘Lie to Me’, has also conducted work on micro-expressions. These are emotional facial expressions that occur for very short periods of time (less than a second) before they’re moderated and covered up by the subject. These expressions are very short and difficult to observe in day-to-day conversation.
In the online domain, of course, we can perceive this non-verbal language through video conferencing software. This software can allow you to easily record footage that may be played back at a slower speed!
What about the polygraph?
The polygraph receives a good beating in academic literature. Of course, polygraph researchers are not oblivious to these critiques and have come up with a number of ingenious ways of trying to enhance the accuracy of the tool as well as make it less susceptible to defences and false hits.
Perhaps the most successful technique that uses a polygraph (or an EEG brain scanner) is the guilty knowledge test. This test entices the suspect to learn to recognise a number of images (a picture of some tools, for example). The suspect is then shown a succession of similar images and told to press one button when they see an image they’ve remembered and another when they don’t recognise the image.
Of course, interspersed within the images will be something that only a guilty suspect would recognise (in this instance, let’s say a particular wrench that was used to attack a victim).
It’s not expected that a suspect will hit the ‘wrong’ button when they see the wrench (although it can happen). However, when we recognise something our brains immediately attune to the stimulus for about 400 ms. During that time we cannot do anything else, including button pressing. Subsequently, the suspect will be statistically slower at pressing the button for the wrench than they will be for the other images.
Of course, if the suspect is aware of the technique then they can try to manipulate their responses. However, a consistent 400 ms time lag is not an easy thing to fake.
Trying to control behaviour
Often, the best lie detection techniques do not look at ‘lying’ behaviours at all. Instead, they examine behaviours indicating that someone’s trying to control their behaviour.
Theories such as interpersonal deception theory suggest that deceit is a highly reciprocal process. It follows that if I lie to you I might change my behaviour in accordance with what I perceive to be indicators of truthfulness (such as staring you straight in the eye).
That said, if someone does feel that they’re being lied to then they subsequently change their behaviour as well. If the liar notices this, then they will also adapt accordingly… and so on.
Many of these perceived indicators are often inaccurate, and research has shown that humans are pretty bad lie detectors. Perhaps this is because communication is fundamentally based on an assumption of truth.
In everyday life, the ‘perceived’ clues to deceit are used more than real cues to deceit, so if I want to deceive you then I should pay more attention to what you think are the cues to deceit rather than try to control the real ones!
Ultimately, it’s believed that everyday lie detection gets better in more interactive scenarios. Experienced liars will aim to reduce the interactivity of the conversation.
Exercising a degree of caution
Lie detection tools and technologies need to be used with caution. Anyone can buy a polygraph and a book on testing and get to work. These tools still suffer considerable validity challenges: the ones that are accurate are usually only effective under carefully controlled conditions.
In some situations these tools can lead to false confessions. Indeed there have been cases of people admitting to crimes they haven’t committed because they feel that no-one would believe them after failing a polygraph and that they were better off cutting a deal with the prosecution.
As investigators, we’re all attracted to the idea of a lie detector but my advice is to proceed with extreme caution, do your homework and never rely on this evidence in isolation.
There’s no highly accurate online ‘lie detection’ technique available at present. This is not an easy thing to achieve, but we are starting to see more technologies appear on the market (such as the Fraud Triangle Analytics technique developed by Ernst & Young and the Association of Certified Fraud Examiners) that analyse e-mail for a variety of psychological phenomena.
I feel that this area of research will only continue to mature as corporate investigators rely more heavily on e-mail-based evidence and intelligence.
Simon Placks leads the Ernst & Young IT Forensics team
Selected bibliography
