How to cheat a brain-scan-based lie detector

Cheating the scanner is relatively easy

Sure, it's possible to differentiate patterns of truth-telling brain activity from patterns of lying-related activity. But contrary to media hype, experts have been quick to point out that the accuracy of brain-scan based lie detection is often no better than with traditional approaches, such as the polygraph. Furthermore, these experts warn, brain-scan methods could, in theory, be easily thwarted by liars with even modest levels of guile. That claim is no longer purely theoretical, for in a new study, Girgio Ganis and his colleagues have used a popular paradigm to show just how easy it is for lying participants to trick the brain-scanner.

Twenty-six participants had their brains scanned whilst they looked at the same six dates appearing for half a second each, one at a time, on a screen. For each date they had to indicate with a button press whether it was their date of birth - yes or no. This was repeated several times. In the truth-telling condition, none of the dates was their birth date and the participants simply told the truth and said 'no' to each date. In the lying condition, one of the dates was their birth date and their task was to lie and indicate 'no' whenever it appeared. An equivalent set-up in a real-life criminal case might involve a suspect repeatedly looking at the same selection of knives and indicating whether they owned any of them. One further twist to the task was that participants had to look out for a further specific meaningless date - this was just to make sure they stayed engaged with the task.

When the participants lied and pressed 'no' whenever their birth date was visible then, consistent with past research, their mendacity was revealed in their brain activity. In particular, because of its salience, seeing one's birth date was associated with raised activity in the front-most part of the brain, on both sides. Focusing on activity in these regions, a simple algorithm could tell whether a participant was lying or not with 100 per cent accuracy (the extremely high accuracy may be due to the greater salience of one's own birth date compared with the material used in previous research).

Most importantly, in a repeat of the experiment, the participants were taught a simple cheating technique. All they had to do was ever so slightly move their left index finger, middle left finger, and left toe, respectively, whenever three of the irrelevant dates appeared on the screen (they were instructed to ensure the movements would be too slight to be seen by someone watching). This cheating strategy lent salience to these irrelevant dates, thereby reduced the stand-out salience of one's birth date, and had the effect of reducing the accuracy of the lie-detection algorithm to 33 per cent. In other words, most instances of lying were misidentified as honest responses.

The researchers think that the moving part of this cheating strategy probably isn't necessary. For example, deliberately recalling a certain memory when certain 'irrelevant' stimuli are shown would have the same effect of reducing the stand-out salience of a target stimulus, be that a murder weapon or a date of birth.

'Although these results apply directly only to the specific [and commonly used] laboratory paradigm used here ...,' the researchers said, 'they support the more general point that the vulnerability of the neuroimaging paradigms for deception detection to various countermeasures should be assessed and documented explicitly before they can be used in applied settings.'
_________________________________

Ganis G, Rosenfeld JP, Meixner J, Kievit RA, and Schendan HE (2011). Lying in the scanner: Covert countermeasures disrupt deception detection by functional magnetic resonance imaging. NeuroImage, 55 (1), 312-9 PMID: 21111834

You have read this article Brain / Forensic with the title How to cheat a brain-scan-based lie detector. You can bookmark this page URL https://psychiatryfun.blogspot.com/2011/02/how-to-cheat-brain-scan-based-lie.html. Thanks!