Reading Terrorists' Minds About Imminent Attack

An anonymous reader writes "Imagine technology that allows you to get inside the mind of a terrorist to know how, when, and where the next attack will occur. In the Northwestern study, when researchers knew in advance specifics of the planned attacks by the make-believe 'terrorists,' they were able to correlate P300 brain waves to guilty knowledge with 100 percent accuracy in the lab, said J. Peter Rosenfeld, professor of psychology in Northwestern's Weinberg College of Arts and Sciences."

Comments

Re:Lab Accuracy != Real World Accuracy

[retchdog]

The 10/12 is a new study "without advanced knowledge of the `terrorist' plans," whereas the 100% was with this knowledge. The presence of this distinction should further set the stage for skepticism about their experimental design.

Further, they have achieved only 100% (resp. 83%) sensitivity (=true positives) with an unknown (or unreported) specificity (=true negatives) since they had no controls. What if I'm having an affair or high-stakes slightly-shady business deal in New York?

This study contradicts Rosenfeld's own research

[Cycon]

This is really interesting as Rosenfeld himself has previously railed against other neuroscientists for commercializing P300 based lie detectors with claims of 100% accuracy:
Simple, effective countermeasures to P300-based tests of detection of concealed information - J. PETER ROSENFELD,a MATTHEW SOSKINS,a GREGORY BOSH,a and ANDREW RYAN

"It seemed timely to investigate countermeasures to ERP-based tests also because although there have been many laboratory studies claiming 85-95% accuracy, only one field study has been published, but it reported approximately chance accuracy (Miyake, Mizutani, & Yamahura, 1993). Nevertheless, one user of these methods claims 100% accuracy and is presently attempting to commercialize them (see http://www.brainwavescience.com/). Finally, the ERP approach has now surfaced in popular novels, for example, Coonts (2003), as a foolproof method."

...

"It is noted that the subjects used by Farwell and Donchin were paid volunteers, including associates of the experimenters. Our presently reported study uses introductory psychology students as subjects, more like the subjects one might find in the field in the sense of relative lack of motivation to cooperate with operators, and perhaps lower intelligence."

The above is the original peer-reviewed paper, this review (also by Rosenfeld) below is more recent and concise:
http://www.srmhp.org/0401/brain-fingerprinting.html

I'm very skeptical

[Anonymous Coward]

Some people in my research team are working on P300 detection - here's how it works.

Basically, a P300 is a peak of cortical activity recorded approximately 300ms after perceiving something you expect to perceive - it has nothing to do with emotion, as the paper says. It's about attention and expectation. A simple example is the P300 speller: letters are blinking on a screen, and you focus on the letter you want. when your letter blinks, your brain generates a P300.

When detecting P300 with external electrode, there are several problems:

- some people are not able to generate P300 peaks (approx 5-10% if I remember correctly)
- the 300ms delay can vary from one person to another, even for the same person depending on the situation
- the P300 is drowned in noise, so you have to reproduce the experiment several times to cancel out the noise
- if you blink your eyes or contract your jaws muscles, you generate artifacts in the signal that are several orders of magnitude stronger than a P300
- to make it work properly you have to be relaxed, in a quiet environment - that's why we generally use visual stimuli. i'm not sure where the state of the art is with auditive stimuli
- if you drug the guy so he is calm and doesn't move, you are very likely to also affect his brainwaves, thus defeating the purpose.

Long story short: from what I know of the subject, P300 detection on a non-willing subject sounds unrealistic to me. It's all about researchers getting fundings by putting the word "terrorist" in their research proposal, which is very sad.

From Someone Who Works In This Lab

[alexsoko]

Hello, As both a) someone who works in this lab and b) someone who reads this site pretty religiously, I think I can address some of your guys concerns. 1) Specificity of questions - unlike a standard polygraph test, in a P300 CIT (concealed information test) subjects aren't asked questions as muscle movements or auditory stimuli may disrupt the electrodes ability to record P300. Instead, stimuli are presented silently on the screen and thus, if the subject 'recognizes' the stimulus he will generate a P300 whenever that stimulus is presented. However, in doing so, the list must initially be vetted with the subject who says if any of the items have specific relevance to him. (This would be like in an investigation if a police detective showed someone a list of people and asked if a POI knew any of them). 2. This isn't a 1 recognition stimulus identifies everything sort of thing. The same stimuli are shown to people literally hundreds of times and it takes a pattern of recognition to correctly identify someone as guilty. Also, there are levels of recognition. All of the responses are compared to one another to get a standard base, per each participant, of brain activity. Then each recognition pattern is compared to the pattern as a whole to determine guilty knowledge. 3. For critical information a more strict test can be performed which compares the strongest P300 to the second strongest P300. If that patters is statistically bigger then you can be certain that they have guilty knowledge of that item. 4. Several of the studies we have conducted have actually incentivized (given money) to people for trying any strategy possible to BEAT the test. 5. There ARE countermeasures for this test that you can do to try to hide your P300 responses - however this specific protocol is a COUNTERMEASURE RESISTANT TEST. Believe me, if you've thought of it - we've thought of it. 6. Yes, when using just a pure P300 analysis we don't get people with 100% accuracy. But after we adjust for countermeasure use, and analyze other behavioral and EEG data that is collected concurrently with the P300 we can get 100% accurate identification. 7. We do so without getting false positives. Like any tool in law enforcement (the polygraph, fingerprinting, etc...) it's not necessarily as important that any individual thing works as it is that of the array of tools used ONE of them catches the person. And you don't want to wrongly accuse anyone. Why our P300 research is special is because we get an extremely high detection rate with no false positives. 8. If you have more questions please respond to this comment and I will try to respond. --Alex Soko