It would be nice, although a little scary, if they could. And there have been several reports of succesful trials under laboratory conditions. However, a new paper in Neuroimage reveals an easy way of tricking the technology: Lying In The Scanner.The authors used a variant of the "guilty knowledge test" which was originally developed for use with EEG. Essentially, you show the subject a series of pictures or other stimui, one of which is somehow special; maybe it's a picture of the murder weapon or something else which a guilty person would recognise, but the innocent would not.
You then try to work out whether the subject's brain responds differently to the special target stimulus as opposed to all the other irrelevant ones. In this study, the stimuli were dates, and for the "guilty" volunteers, the "murder weapon" was their own birthday, a date which obviously has a lot of significance for them. For the "innocent" people, all the dates were random.
What happened? The scans were extremely good at telling the "guilty" from the "innocent" people - it managed a 100% accuracy with no false positive or false negatives. The image above shows the activation associated with the target stimulus (birthdays) over and above the control stimuli. In two seperate groups of volunteers, the blobs were extremely similar. So the technique does work in principle, which is nice.
But the countermeasures fooled it entirely, reducing accuracy to well below random chance. And the countermeasures were very simple: before the scan, subjects were taught to associate an action, a tiny movement of one of their fingers or toes, with some of the "irrelevant" dates. This, of course, made these dates personally relevant, just like the really relevant stimuli, so there was no difference between them, making the "guilty" appear "innocent".
Maybe it'll be possible in the future to tell the difference between brain responses to really significant stimuli as opposed to artifical ones, or at least, to work out whether or not someone is using this trick. Presumably, if there's a neural signiture for guilty knowledge, there's also one for trying to game the system. But as it stands, this is yet more evidence that lie detection using fMRI is by no means ready for use in the real world just yet...
Ganis G, Rosenfeld JP, Meixner J, Kievit RA, & Schendan HE (2010). Lying in the scanner: Covert countermeasures disrupt deception detection by functional magnetic resonance imaging. NeuroImage PMID: 21111834
15 comments:
1) In a real test, how is a guilty person supposed to know with what "innocent" information he or she will be confronted?
2) Aren't there ways to detect minute muscle movements? Or if that's too complicated, aren't there drugs that will (temporarily) paralyze the body, leaving only a mind that involuntarily responds to the "guilty" or "innocent" stimuli?
Yes, human rights are in no way being violated if they have to be temporarily paralysed to prove their innocence. Wtf, mate.
Anonymous #1: 1) You wouldn't, but you could base it on the order of the stimuli or something. "I'll move my little finger for the first one, my toe for the second one" etc.
2) I think it's not the movements per se so much as the mental association. You wouldn't have to actually move, so long as when you see one of the control stimuli, you think "this means I have to move".
i Believe this article is not to teach how to fool the lie detector as much as to show how we might get false results... this is part of the experiment and they are showing the results.
Great post, on an interesting and slightly worrying field! Still can't quite believe there are a couple of companies already flogging lie detection by fMRI ("No Lie fMRI", etc! Shudder.)
Elie D: Right, it's an experiment - but it shows a way of fooling this type of test, that ought to work "in real life".
Put it this way, if I were, for some reason, faced with serious consequences if I failed this test in real life, I would use this method.
Okay, I haven't read the article yet. But I think the first commenter asked the right question. This flaw doesn't seem to compromise the real-world validity if you need to train up on the foils that will actually be used.
I'm also baffled by the idea that the accuracy could be "well below random chance." When making a binary decision, there's no such thing as a below chance decision rule.
Anonymous: With the countermeasures, about 70% of "guilty" people were wrongly classed as "innocent".
If you'd flipped a coin, you'd have got 50% right.
Although admittedly that does assume that you've got a prior probability of 50% guilty in your test.
and the funny thing is that in a real life scenario lot of the images would be actually relevant without them trying to fool the device.
Probably a stupid point, but could it be the case that the performance (FP or FN, whatever) goes down under stress conditions. Because that seems to be very likely...
finally, you agree that this has nothing to do with lie per se. don't you?
It's interesting that there were no false positives at all. I mean, I know the odds are fairly long, but at some point you are going to hit a date for someone in either group of individuals that has some kind of meaning for them, unless they collected all that relevant information up front.
Anniversaries, direct family members birthdays, etc. did they make the effort to collect all of that up front to ensure that none of their random dates would trigger a response?
"...unless they collected all that relevant information up front."
According to the Methods section, they did collect such information in advance for each subject and designed the stimuli accordingly.
of course we can fool this machine but only we need it when we don't want to fool it. All we know that mind is in our control to think which effect the result of machine.
- John Devis
Magento Themes
How many participants were in this study?
How many participants were in the study?
Surely great ways mentioned in order to go and fool a detector brain scan.... Well done keep posting stuff like this.....
Magento Themes
Post a Comment