Detecting lies with fMRI

In 2006, a company called No Lie MRI began advertising their ability to detect "deception and other information stored in the brain" using functional magnetic resonance imaging (fMRI). They were not the first to make this claim. Two years prior, a company called Cephos had been founded on the same principle. Both companies were launched by entrepreneurs who hoped to one day replace the polygraph machine and its recognized shortcomings with a foolproof approach to lie detection.

Within several years after the establishment of these companies, their services were used on multiple occasions to obtain fMRI data to submit as evidence in court--in both criminal and civil cases. In each instance the court declined to admit the evidence, but the decision to exclude it was carefully deliberated. In one 39-page opinion written to recommend excluding the evidence in a Medicare fraud case, the judge indicated that fMRI evidence might one day be admissible even if it hadn't been determined to be an extraordinarily reliable method of lie detection. Thus, it seems that the use of neuroimaging for lie detection may be a real possibility at some point in the future. But how realistic is it right now?

Designing studies to test fMRI-based lie detection

To be able to evaluate the current state of the research regarding fMRI-based lie detection, it's important to have an understanding of how studies in this area have been designed and conducted. The goal of these studies has been to identify differences in the brain activity of participants when they are telling the truth vs. when they are lying. Thus, participants are usually instructed to lie when asked about certain topics and told to answer other questions truthfully. The brain activity observed during the truthful responses is compared to brain activity observed during deceptive responses in an attempt to isolate areas that are active only during prevarication.

For example, in one study participants were directed to "steal" one of two items: a watch or a ring. Then, the subjects were put into an fMRI machine and questioned about the mock thievery, but they were told to deny taking anything. The subjects were asked about both items, but because they had only taken one of the two, some of their responses naturally were lies and others were the truth. Researchers then compared the participants' brain activity during truthful and deceitful responses.

Other studies vary considerably in the details, but have a similar general approach. In one experiment, participants were given two playing cards and then told to deny having one of them when asked about it. In another example, participants picked a number between three and eight and then were told to deny picking their chosen number when it was shown to them. Thus, most of these experiments involve "directed deception," where a participant is told to lie about a particular experience while their brain activity is being observed.

What the studies have shown

The collection of studies investigating the neural correlates of deception has grown considerably over the years, leading to a number of candidate regions being singled out as potentially playing a role in deception. In a recent publication, Farah et al. conducted a meta-analysis of these studies to attempt to identify brain structures that had consistently been activated during lying. The investigators analyzed 23 studies which, as a whole, described a total of 321 foci of interest.

Farah et al. found significant variability among the results of the studies, and no one region was activated in all of them. However, a number of areas were generally more likely to be active during deception, including parts of the prefrontal cortex, inferior parietal lobule, anterior insula, and the medial superior frontal cortex.

These commonly-activated areas may provide us with some clues as to where we should be looking for the neural correlates of deception. To have confidence in fMRI for lie detection, however, it will be important to see consistent activation of a particular network of areas from study to study. Because the studies included in this meta-analysis all differed slightly in experimental design, it is not surprising they resulted in slightly different patterns of brain activity. Even if with further studies we see a more consistent pattern of activation, however, can we be confident it is representative of lying? Some would argue that we cannot.

Problems with fMRI-based lie detection research

Beyond the lack of uniform results in laboratory studies of deception, there are a number of other hurdles associated with using fMRI for lie detection. One major difficulty is in determining that the brain activity we see occurring during deception is specific only to lying. For example, if participants are instructed to steal an item and then are asked to lie about stealing it later, the increased activity observed when they are asked about the item could be associated with deception. But, it could also be that their unique personal experience with the item (e.g. having "chosen" to steal it as opposed to one of the other items) causes a different pattern of activity that is representative of familiarity. In other words, different memory systems may be accessed for items that one has more knowledge of, and the activation of these networks could be responsible for the differences in brain activity observed in deception vs. sincerity conditions.

A study conducted by Hakun et al. serves as a good example of this problem. The investigators in this experiment asked participants to choose a number out of a series of numbers. Then, the participants were put in an fMRI machine and questioned about the number chosen; half of the participants were told beforehand to lie about which number they picked while the other half were asked to remain silent. Some of the areas identified in the meta-analysis discussed above that seem to be activated during deception were activated in both groups of participants (half of whom were not lying--nor even talking). Thus, it may be that activation in these areas is not specific to deception, but involved with memory retrieval, attention, or some other aspect of higher-order processing.

Another issue with using fMRI for lie detection involves the real-world applicability of the studies in this area. For example, when you ask someone to lie about a number they chose from a list or even about a mock crime, the lie they tell will generally not have a great deal of emotional significance for them. In real-life, however, lying is often associated with high emotionality, stress, anxiety, etc. Thus, it is unclear if the brain regions identified in fMRI studies of lie detection are only likely to be activated during the more dispassionate type of deception that occurs under laboratory conditions. If so, it could mean that other areas of the brain are more likely to be activated during "real" lies, which could suggest that fMRI lie detection based on current data might overlook activity often associated with lies in the real world.

Additionally, most of these fMRI studies are conducted on healthy participants (often college undergraduate students). It seems likely that these individuals may display different brain responses to lying than, for example, a sample of participants with a criminal past (a population this technology might be expected to work with if it were to have validity in the legal realm). For instance, one fMRI study conducted with people who had a criminal history and a diagnosis related to psychopathy found a different pattern of brain activity during instructed lying than that seen in other populations. Thus, brain activity during deception may differ from individual to individual depending on other personality characteristics, personal background, etc. Until we can be more certain about which areas of the brain are activated in everyone when they lie, it is difficult to have much confidence in the use of fMRI to detect deception.

One other potential problem--which applies to using any method to attempt to detect lying--is that we must be aware of countermeasures the subject may use to evade detection. Countermeasures are actions an individual might be able to take to disrupt the lie detection device's ability to accurately identify dishonesty. For example, some have asserted that slightly painful actions like biting the tongue during control questions in polygraph tests can raise physiological responses during those responses, making baseline activity too high to be able to detect the deviations from it that occur during true deception. Countermeasures to disrupt fMRI-based lie detection are still relatively unknown, but studies suggest they may be fairly simple to implement. In one study experimenters were able to detect lying in participants with 100% accuracy. However, when they asked participants to wiggle their fingers and toes "imperceptibly" during the fMRI scanning the accuracy was reduced to only 33%.

A long way to go

Thus, despite the optimism that led some to invest in businesses devoted primarily to fMRI-based lie detection, it seems the method still has a long way to go before it can be considered valid and reliable. Consequently, it is not something we are likely to see gain widespread acceptance anytime soon. But that does not mean it is not a future possibility. As our methods of neuroimaging improve and we develop better ways of identifying specific structures that are necessarily activated during certain behaviors, highly accurate neuroimaging for lie detection may emerge, bringing with it a collection of ethical dilemmas about how it should be used. However, such developments are likely decades away; for now the knowledge about the lies we tell will remain safely sequestered in our own heads.

Farah, M., Hutchinson, J., Phelps, E., & Wagner, A. (2014). Functional MRI-based lie detection: scientific and societal challenges Nature Reviews Neuroscience DOI: 10.1038/nrn3589