Rupert Cornwell: Out of America - The obsession with polygraphs ignores the problem of human fallibility

The Pentagon, the mightiest source of military power on the planet, is also its mightiest source of acronyms. Who else could have turned Maids from household helps into Multipurpose Automatic Inspection and Diagnostic Systems? In terms of creepiness, however, few come close to PCASS: Preliminary Credibility Assessment Screening System for acronym buffs, but in plain English, the hand-held instant lie detector.

Lie detector machines, or polygraphs as they are properly known, have been used here for criminal investigation and personnel screening for a century or more. The gut American conviction that there is no task that a machine cannot perform undoubtedly explains the number of inventions that have originated or been perfected here. Surely none, however, is more ambitious than the polygraph.

In sceptical old Europe, most countries treat polygraphs as a curiosity at best. In criminal cases, they have next to no weight.

Not so in the US. In almost half the 50 states, such tests are admissible in court as evidence. In 1992, a convicted Virginia murderer who protested his innocence until the end even underwent a polygraph test the night before his scheduled execution. Not surprisingly, he failed, and justice took its due course. Some two-thirds of police departments use such tests routinely, although no suspect can be forced to take one. The federal government administers an estimated 40,000 polygraphs a year, mostly for job screening.

Hitherto you've had the consolation of knowing what you were in for. There was a preliminary interview from the examiner, then you sat down in a large chair and got wired up to the machine. It registered the reaction with the oscillations of a pen point on rolling paper. The PCASS, on the other hand, miniaturises and speeds up the process. The device is linked to three sensors attached to the hand of the person undergoing "preliminary credibility assessment". Two sensors measure stress, as indicated by the electrical conductivity of the skin, while the third is attached to a fingertip to check the heart rate.

The gizmo is programmed for the examiner's questions. He punches in the answers, and within a minute or two the device delivers its verdict. If it flashes green, you're in the clear. Yellow means "don't know", while red signifies you're lying: in polygraph-ese, "showing deception".

Mercifully, the PCASS is not on general release. The Pentagon says the devices, costing $7,500 (£3,800) each, will be be supplied only to soldiers in hot spots such as Iraq and Afghanistan, where they have to make vital judgements about a someone they've arrested or who has come forward as a witness. The PCASS is at least an initial tool.

All of which is perfectly reasonable – except it begs the question of whether lie detectors work in the first place. The Pentagon claims the PCASS was more than 80 per cent accurate in tests. But it achieved that figure by excluding yellow "don't knows". Factor those in, and the accuracy rate drops to under 65 per cent, the same as the larger, old-fashioned polygraph. You'd be almost as well off tossing a coin.

The truth, however, is more nuanced. Lie detection is not so much a science as an art. In an interview on a Virginia radio station, John Sullivan, a former polygraph operator for the CIA, argued: "There is no such thing as a lie detector machine. When you have an X-ray, it doesn't make much difference who carries it out. But with the polygraph, the examiner plays a primary role. The process relies on my ability to get you to talk to me."

It is a prop for a skilful interrogator, but no more. The CIA relies heavily on lie detector tests, but its record is mixed. Aldrich Ames, the most damaging mole in the agency's history, denied to polygraph operators that he was working for a foreign intelligence service, and the machine didn't bat an eyelid. Ames was not nervous. "What makes it work is fear of failing," Sullivan said. "If you don't have that, the test is meaningless."

The Ames experience underlines the biggest weakness of the system. A "false positive", when a person is adjudged to be deceitful although he is telling the truth, can be brought on by nerves and can be corrected later. The real danger is a "false negative", clearing a person who is guilty. The remedy is simple. Don't put not your trust in acronyms such as PCASS – nor in any other mechanical child of man.