Wednesday, April 16, 2008

Truth or ??

The U.S. military is deploying a handheld “lie detector” for battlefield use in Afghanistan (and then perhaps Iraq). It consists of a handheld computer that has three wires connected to two types of sensors that are attached to the hand of the person being questioned. One type of sensor measures changes in capillary dilation at the tips of the fingers, while the other measures the perspiration on the surface of the skin. The interrogator (most likely via an interpreter) asks a series of questions, then will enter the (yes/no) answers. After a delay for processing (via a proprietary algorithm developed), the handheld will display green (truth), red (false), or yellow. A yellow calls for a repeat, a case of double-yellow then signifies no opinion. 

The new device is the PCASS, or Preliminary Credibility Assessment Screening System. Following a series of tests the DOD says that the device has an accuracy rate of 63% to 79%, or, excluding the yellows and only considering green/red, 79% to 92% (trial numbers shown below): 

The intent is to use this as a screening tool (e.g. after a roadside bomb) to triage suspects to identify those that should get further investigation. The process and rules surrounding PCASS use include:

1. Operators are supposed to get 40 hours of training (24 hours classroom and 16 laboratory)
2. Only certified operators are to conduct PCASS examinations (certification requires an internship of 25 examinations after training)
3. The device is supposed to be used ONLY for initial screening – from the regulations, “PCASS examination results shall not be the sole means to validate information…”
4. Yellows are not to be taken as “false”
5. “The PCASS operator must obtain the examinees’ voluntary consent before conducting an examination.”
6. The operator has to document each examination, and the paperwork “shall be maintained for at least 20 years.”

OK, so how realistic are the rules governing PCASS use? Are the troops going to be asking permission to conduct PCASS examinations after being subject to an IED attack? Seems unlikely. For example in Haditha 24 civilians were killed by troops when a patrol suffered killed and wounded in such an attack. Although some members were put on trial the prosecutions fizzled, because even though investigations uncovered some issues/lapses (e.g. no positive IDs, etc.), the troops had been under attack and mostly followed the rules of engagement. If troops suspect someone are they going to let them go after a double-yellow? In real life in dangerous situations, will PCASS only be used for initial screening and be followed up by other methods of investigation? Seems unlikely.

There is also criticism of the device and its utility. Some contest the accuracy numbers. Others wonder how applicable/correct an algorithm developed in a lab setting and tested on Americans will work on Afghans and Iraqis in battlefield situations and in the field. And re the hardware, how good will a portable sensor attempting to measure slight perspiration changes be in an Iraqi summer when it is 120 degrees in the shade? Also, will this become a technological “crutch”? Will the false negatives (per the study the device will show green for 2.8% percent of those actually deceptive, and yellow for 5.6% i.e. 8.4% of those lying will “pass”) allow “bad guys” to get through? What about false positives? Even at the quoted accuracy 11.4% of those telling the truth would show up ‘red’ and 22.9% as ‘yellow’, so potentially 11 (or more likely 34) of every 100 people tested would then be suspect (and probably hauled off and warehoused in the prisons for follow up). At this point the base rate fallacy comes into play, so even if the accuracy rates of these devices was much higher than currently claimed they would still be troublesome.

The bottom line is that much as it would be nice to find a technological magic bullet (e.g. see OPED14 Science to the Rescue) it is unlikely that PCASS is it.

U.S. Soldiers to Receive Lie-Screening Devices
New anti-terror weapon: Hand-held Lie Detector
Operational Approval of the PCASS
Efficacy of Prototype Credibility Assessment Technologies PCASS Final Report

No comments:

Post a Comment