X Detectors

This is an interesting story on the BBC. It appears that as part of their probation, a pilot sample of convicted sex offenders are to be interviewed under a polygraph in an attempt to catch them sliding back into abusive behaviour.
I don't think any official body in the UK, certainly not the courts, police or the probation service are prepared to say that lie detectors "work" -- in the sense that they reliably detect when an interrogation subject is lying. The problems seem to be:

• Unconscious physiological arousal is not solely caused by lying (should this get a "duh"?),
• Some very dangerous people lie without turning a hair,
• Guilty subjects are disproportionately motivated to inform themselves about the devices and learn to overwhelm their measured responses with willed arousals,
• The innocent are undone by the free-floating guilt that afflicts so many of us (sometimes seriously), or by "false" positive rates that the American Polygraph Association seems to believe range up to 15%.

So this has been an obstacle to adoption of lie detectors in the UK. They don't work, and even if they did sort-of work the false positive rate would be oppressive in an population where even a small proportion of  people are guiltless. But investigators and enforcers love the idea of the polygraph: it's just so sciencey and promises an amazing shortcut. What polygraph enthusiasts want is a group which no-one will defend, which is universally assumed to be permanently guilty, and it looks like sex offenders are chosen.


The bit that interested me is the quote from Professor Don Grubin, the man behind the tests:

"Disclosures made during polygraph examinations, as well as conclusions drawn from passed or failed examinations, allow probation officers and the police to intervene to reduce risk ... Just as important, it is also aimed at enhancing the co-operation of offenders with supervision, helping them to focus on, and avoid, the sorts of behaviours that make re-offending more likely."

That is a very careful statement indeed, and I hope the Beeb haven't picked out something unrepresentative. Grubin is a proper academic at a proper university -- Newcastle -- where the university profile identifies his current approaches to sex-offenders as being polygraphy and Prozac. And on the strength of this quote, it seems that he finds the chief value of a lie detector is that it's called a "lie detector". He does mention passes and failures, but his focus is on the interview itself. It appears that the purpose of the "lie detector" is not to spot lies, but to persuade the subject that telling the truth is the best plan.

Now I don't think this necessarily a bad thing. We needn't worry about intelligent psychopaths who can fool the machine -- because this isn't about the interviewer believing the results. There's no objection to interviewing probationers -- it beats prison, and interviews in these particular cases might actually be helpful.

A little bit of stagy flim-flam in the form of lie detectors doesn't really make a moral difference -- it's on the same level as good cop/bad cop or Reid. I do worry that the idea of polygraphs as a worthwhile tools of investigation will acquire an spurious respectibility -- we mustn't reach a situation where a spoken denial plus a "lie" response is treated as a confession. I worry that if this goes beyond the pilot, it'll create a constituency of "skilled polygraph operators" which will tend to expand its area of operations regardless of value. But overall, when many of these subjects -- people convicted of nasty crimes with a huge recidivism problem -- believe that the impressive device can read minds, that's good, provided no-one, er, lies about it. And that's the rub.

Professor Grubin is treading a careful line. Somewhere on the continuum from

• "this machine has no real function, but we hope you will believe, mistakenly, that it is a lie detector", through
• "this machine records your physiological arousal and correlates it with your answers to the questions I ask", and
• "this is a polygraph, more commonly called a lie detector", right up to
• "this machine will tell me if you lie"

there is a moral limit. Grubin knows it's there. He's going to spend the next three years wondering whether he's gone over it.

And if we want to avoid dancing around with truth and falsehood we need a better name than "Lie Detector". The machines may have a use, but detecting lies isn't it.


[Updated 2012-07-20 when the pilot completed. Para after the bullets expanded to identify the appeal of  sex offenders as a target for this.]