Thank you for your generous and thoughtful response.

I agree that it is a problem that lie detectors are popularly misperceived as infallible, all-or-none diagnostic instruments. It may be, however, that the introduction of lie detector that explicitly yields probabilities rather than certainties could alter this popular misperception, as opposed to being corrupted by it.

In general, there is no good evidence that voice stress analysis (VSA)–which is used to monitor the veracity of benefit claimants over the phone–works. Lying is indeed statistically correlated with some aspects of vocal pitch; however, the effect is very small and non-specific. My hunch is that, as a lie detection methodology, VSA works little better than chance in the current application, and that no experiment has been run to test whether it works better than chance in the current application. However, as an asserted lie detector, the VSA may intimidate benefit claimants into bring more truthful in general. Ironically, this would involve telling a lie to deter lying.

Ultimately, it is an empirical question as to how high-stakes situations would affect the accuracy of the TARA. Tests should certainly be run to address the question prior to the introduction of the TARA as a lie detection methodology. Once they are run, the matter become an issue of public knowledge, not private opinion.

Best wishes,

Aiden

I had to take your link out before Akismet would allow your comment to appear, sorry. Here it is:

en.wikipedia.org/wiki/Timed_Antagonistic_Response_Alethiometer

Thanks very much for responding. I now feel a bit ashamed of having written so dismissively.

I never doubted that your methods were painstaking and accurate or that your conclusions were careful and qualified. Of course, I know that pop science reports of experiments always distort the results. I was challenging the pop-science report rather than the science.

However, I feel that implementation always follows the pop-science mindset. it’s easy to grasp. It’s clear. As soon as you start discussing the details of research, people’s brains mist over. Few people who want to use research results for practical purposes reads the caveats.

My feeling is that, in any real-world use of this sort of technology, users would soon forget the 85% probability etc. issues. The CSI factor – our assumption that technology always provides evidence that cannot lie – would soon take over. This is why I talked about the Holy Grail machine.

You may be aware that a government department is trialling the use of lie-detector analysis of benefits claimants’ phone calls. How many of the accuracy caveats are likely to inform the behaviour of the people analysing the calls?

Any wrongly accused person is likely to be afraid to answer questions put under interrogation. So is any correctly accused person. I still maintain that the effects of fear and awareness of the potential consequences of any response would make it very hard to use this technology with an acceptable degree of accuracy.

Thanks for discussing my fledgling lie detector–the TARA–on your blog.

Please bear in mind that newspaper reports are sometimes too brief to do full justice to a technical methodology, and sometimes contain distortions or exaggerations. It is best to consult in original scientific reference for the proper details. The wikipedia entry lists that reference, together with a more sober description of the task.

If you don’t mind, I like to try and rebut some of your criticisms of the TARA.

First, the claim is NOT made that the TARA is a perfect lie detector. Instead, the claim made is that the TARA distinguishes truth from like 85% of the time (in the lab). That’s promisingly above chance. It suggests that the TARA has potential utility as a lie detector. Accordingly, assertions that it is useless are perhaps unduly pessimistic and premature. So long as the TARA’s margin for error can be quantified, and factored into interpretations, it may be put to good use. For example, sufficiently extreme scores on the TARA, fast and slow, could be used by police as preliminary indications of the likely innocence or culpability of key suspects respectively. Merely probable information is often better than none at all in such contexts.

Second, the TARA is designed so that, if you lie, you have an objectively more difficult classification task to do, whereas if you tell the truth, you have to an objectively less difficult one to do. (The explanation is why is not elementary. It has to do with juggling incompatible binary response strategies on alternate trials.). So, all else equal, liars will be disadvantaged relative to truth-tellers in completing the TARA quickly. Importantly, this is not a fact that can be undone by just knowing you have to go more quickly and by trying to go more quickly. Both liars and truth-tellers will be doing their damnedest to go quickly. However, truth-tellers will enjoy greater success, because of he objective nature of the task

Third, realistic and high-stakes situations may actually increase, not decrease, the power of the TAR A to discriminate liars from truth-tellers. High levels of motivation or distraction may make difficult tasks more difficult still, but leave easier tasks relatively unaffected. Suppose you tried to recite the alphabet forwards and then backwards. You’d be faster at doing the former. However, suppose you tried to recite the alphabet forwards and then backwards while also doing a complex dance under the critical eye of Simon Cowell. Under these taxing circumstances, you might then be a little but slower to recite the alphabet forwards, but you’d probably be a greater deal slower to recite it backwards.

Fourth, because lying on the TARA is unlike lying in everyday life (you don’t have to juggle incompatible binary response strategies on alternate trials) the ability to lie successfully in everyday life is not guide to the ability to lie successfully on the TARA. So even if Karen Matthews was a good liar in everyday life, the TARA might have caught her out.

I hope my comments provide some food for thought.

Yours sincerely,

Dr Aiden P. Gregg