A while ago, an interesting paper on the promise and pitfalls of fMRI-based lie detection was published by Farah, Hutchinson, Phelps and Wagner (2014) in Nature Reviews Neuroscience. It is part of an ongoing article series by the journal examining the interplay between neuroscience and law. This installment discussed the reliability of observed associations between certain brain areas and deception, current limitations of fMRI-based lie detectors, how U. S. courts have treated appeal to fMRI data put forward as evidence as well as ethical and legal issues with the procedure. This post will also discuss ways of beating an fMRI-based lie detector.
Another article in that series that deals with common misconceptions about memory, memory distortions and the consequence of ignorance was covered here.
How does fMRI work?
An fMRI indirectly measure brain activity by measuring blood-oxygen-level dependent (BOLD) activity. This typically involve a lot of controls to make sure that researchers capture the neural correlates of what they want to study instead of irrelevant confounders. Typically, researchers compare BOLD activity during deception and truth-telling in an attempt to find the BOLD-signature of deception, which would give clues about the neural correlates of deception (i.e. patterns of brain activation associated with deception).
The theoretical rationale for fMRI-based deception is that there is probably a relationship between deception and cognition because deception is more demanding on memory and various executive functions than truth-telling.
What are the neural correlates of deception?
The paper performed a meta-analysis with the activation likelihood estimation (ALE) method. This is a way to measure overlap in neuroimaging data based on so-called “peak-voxel coordinate information” and thereby find out how reliable the association between deception and certain brain regions is. After applying their specific inclusion criteria, they identified 23 relevant studies. Their meta-analysis identified several areas as being associated with deception e. g. parts of the prefrontal cortex, the anterior insula and inferior parietal lobule. However, the between-study variation was enormous and no region was always identified.
Despite the apparent high identification rate of deception, fMRI-based lie detection has a long list of very important limitations that effectively undermine any confidence in this technique for legal purposes.
There are confounders that obscure the relationship between brain activity patterns and deception. For instance, studies on the efficacy of fMRI-based lie detection have involved different frequency of motor responses, different level of cognitive effort on memory, attention and emotion. Thus, it is not clear that the alleged neural correlates of deception are really due to deception and not these other factors.
There are research design limitations, such as the fact that most studies are group-level and not individual-level. In other words, data from different individuals have often been pooled to find associations between deception and brain activity. Little research is available for individual-level data i.e. trying to identify deception in individuals. Those few studies that have generated individual-level comparisons are plagued by the confounders discussed above. Information about specificity, sensitivity and population base rates are also often missing. This makes it difficult to impossible to estimate the true positive predictive value of a result indicating deception.
There is also many issues related to generalizable. Is the lab situation similar enough to real-world situations? Can results on undergraduates be extrapolated to career criminals? Are results generated from question of little personal importance informative about high-stakes questions? Does familiarity play a role? The method has also not been validated for individual differences: age, existence of psychiatric conditions and personality traits may influence results.
Individuals with antisocial personality disorder (colloquially known as “psychopaths”) typically do not show the same BOLD-signature to deception as does non-psychopaths. This is an issue affecting the classic polygraph as well, where psychopaths do not display the same level of autonomic reactions as average subjects. Research discussed in the paper has also shown that dishonest people tend to have better cognitive control, and thus had a reduced risk of getting their deception exposed.
Finally, there are a number of countermeasures that can be deployed. This includes small movements of fingers and toes, which dramatically reduced deception detection rate from 100% to 33%. Rehearsing or memorizing deceptive statements may also be effective, as the automation of lies reduces cognitive load.
To date, there have been three instances were fMRI-based lie detection results have been submitted as evidence to U. S. courts. In all three cases, the courts dismissed it based on considerations related to the Frye and Daubert standards. These standards attempt to safeguard the legal system from accepting pseudoscience as evidence and the common requirement in both standards is general acceptance and support in the scientific community (which does not exist at the present time). This highlight a need for scientists to serve as public experts in legal cases.
Privacy and the U. S. constitution
The possibility of fMRI-based lie detection raises important questions about privacy. Should the government be allowed to have access to our private, most inner-most thoughts? In the wake of NSA mass-surveillance, privacy is becoming increasingly important. Another issue is the U. S. constitution that prevents warrantless searchers (which includes these kinds of physical tests) and self-incriminating testimony. However, it has not been decided yet, legally, if fMRI-based lie detection qualifies as a test or testimony. This might mean that it could hang in limbo for quite some time before a decision is reached.
The researchers have three general recommendations for future strategies on the topic of fMRI-based lie detection:
(1) Different policies are required for different situations. It might be less important to impose restrictions on companies offering fMRIs for dating purposes than for its use in the criminal justice system.
(2) research is needed to remove confounders and validate fMRI methods for different populations under real-world situations.
(3) scientists need to be skeptical and provide science-based information and educate the public on this issue.
How to beat an fMRI-based lie detector
So based on the available scientific evidence and arguments presented in this review paper, how does one beat an fMRI-based lie detector?
: the easiest way to “beat” such a high-tech lie detector is not to consent to one. According the the U. S. constitution, the government may not perform warrantless searches and one cannot be forced to self-incriminating testimony.
: slightly wiggle your toes or fingers without anyone knowing. This drastically reduce the efficacy of detection. One might expect that this will be monitored in the future, but other countermeasures are likely to be developed in response.
: the more automatized the lie becomes, the less cognitive load and the less risk of the deception being identified. Criminals are probably already practicing their lies to make them seem more convincing to law enforcement personnel, so this is most likely a predictable extension of existing techniques.
: dishonest people display higher levels of cognitive control compared with honest people. This may indicate that people who tend towards dishonest behavior have a lower risk of having their deceptions caught by the detection system.
: psychopaths typically do not show the predicted BOLD signature in response to deception. In general, psychopaths rarely display the predicted responses to traditional polygraphs either. This contributes to making it very difficult to catch psychopath in a lie in this manner and many criminals could probably be diagnosed by antisocial personality disorder.