Sunday, October 14, 2012

The Polygrapher's Dilemma: Ethics and Lie Detection

The Polygrapher’s Dilemma” is a short (probably shorter than this blog post in going to be) article by the psychophysiologist John Furedy. The article, which dates from 1993, presents a logico-ethical objection to classic lie detection test protocols. Specifically, it presents an objection to the Control Question Test (CQT) which is a type of lie detection test that relies on comparing a subject’s physiological responses to control and relevant questions. This is to be contrasted with something known as a Guilty Knowledge Test (GKT) which relies on a slightly different test protocol (more on this later). This post will discuss Furedy’s logico-ethical objection.

I have been interested in the merits (or de-merits) of lie detection tests for a long time. As part of my PhD research, I developed a side interest in whether there was anything to brain-based lie detector tests (which come in many forms, incidentally). This side interest blossomed into two research papers, which I published a couple of years back. The interest, however, was never purely technological or purely practical. As you might expect from someone writing a philosophy blog, brain-based lie detection interested me in part because it provided a testing ground for ideas in social epistemology and the philosophy of evidence. Some of these ideas have been discussed elsewhere on this blog.

Although I’ve been distracted by many other research projects over the past year or so, I am still interested in brain-based lie detection, and in order to prevent that interest from going stale, I am currently preparing a research paper that evaluates one such test using a number of concepts from epistemic systems theory, social epistemology and theories of democratic legitimacy. As part of my research for this paper, I’ve been collecting objections to lie detection tests that don’t focus purely on their epistemic or technological failings (important though they are). In this respect, Furedy’s paper caught my eye as it was an attempt by a leading psychophysiologist — and vociferous opponent of CQT lie detection — to present a largely ethical objection to the technology.

In the remainder of this post, I shall look at three things. First, I’ll explain, in rather cursory terms, how a CQT lie detection test works. Second, I’ll outline Furedy’s logico-ethical objection to the CQT. And third, I’ll try to evaluate Furedy’s objection before looking at the GKT and explaining why it might be a superior testing paradigm. As it turns out, it’s the GKT that most interests me, but it’s best introduced by way of contrast with the CQT.

1. What is a CQT Polygraph Test?
In some ways it is odd that the word “polygraph” has become synonymous with “lie detection”. Etymologically speaking, “polygraph” simply means “much writing”, which is an apt description of the output one receives from a lie-detector but which could equally well describe any device that records multiple data streams and displays them in a visual form. Nevertheless, the two have become synonymous and we must work with whatever linguistic quirks we have inherited.

The classic polygraph lie detector measures response profiles in the autonomic nervous system of a particular test subject. Thus, it will measure things like heart rate, galvanic skin response, systolic blood pressure and so on. Like any part of the nervous system (or body more generally) the autonomic nervous system is constantly active. So when one measures responses in that system, the visual output one receives will reveal a constantly variable level of activity, which can be displayed on a scrolling graph (as in the image below).

None of this gets us anywhere close to lie detection. For that, we need to ask test subjects questions, compare the response profiles associated with the answers to these questions, and we also need to make an assumption. The assumption is that when people lie their autonomic response profile will be markedly different than it is when they are telling the truth. There is some supporting theory here. The autonomic nervous system regulates the body’s fight-or-flight response: so if someone is distressed (as they might be when lying) we are likely to see increased activity in their autonomic system.
But other forms of distress, not indicative of lying, are likely to be associated with increased autonomic activity. This raises the disturbing possibility of a polygraph test resulting in many false positives, i.e. cases in which a person is identified as being deceptive even when they are not. A careful testing paradigm is needed to ensure that this does not happen. One of the key questions is whether the CQT is such a testing paradigm. Clearly, Furedy (and many others) think it is not. But before we prejudge the issue, let’s see what it is.

The CQT, in essence, consists in a test subject being asked two kinds of question: control questions and relevant question. The relevant questions cover whatever subject matter the tester is interested in. For example, if the tester is investigating a murder, the relevant question might be “did you shoot Mr X with a shotgun?”; if the tester is doing a background check for employment in the secret services, they might ask “are you, or have you ever been, a member of the communist party?”. The control question is supposed to be a similarly valenced question, covering subject matter that innocent subjects are likely to lie about (hence, it is also sometimes called a “probable lie” question). An example might be “have you ever lied to get out of trouble?”.

The belief is that an innocent (non-deceptive subject) is likely to respond more strongly to the control question, because that’s the one they are likely to be lying about. By way of contrast, the belief is that the deceptive subject is likely to respond more strongly to the relevant question, since that, for them, is the more important lie. (Note: multiple pairs of control and relevant questions may be asked in a single test. For simplicity’s sake I assume the test involves one question only for the remainder of this post)

2. The Polygrapher’s Dilemma
With this background detail on the test out of the way, we can move on to consider Furedy’s dilemma proper. An ethical dilemma arises whenever we have two choices to make, each of which is morally problematic (for consequentialist or deontological reasons) and which cannot be avoided by choosing some third (intermediate) option. Furedy’s contention is that the person administering a CQT lie detection test faces two unpalatable choices, with no alternative (palatable) ones. What are these two choices?

The answer comes from the structure of the CQT. As I just noted above, the CQT turns on the comparison of a subject’s responses to relevant R) and control C questions. If the response to the relevant question is greater than the response to the control question (R > C), then the subject is deemed “deceptive”. Contrariwise, if the response to the control question is greater than the response to the relevant question (C > R), then the subject is deemed “non-deceptive”.

One interesting feature of the CQT is that, in order for it to be effective, the tester has to formulate control questions that are emotionally similar to the relevant questions. The reason has to do with the danger of false positives. If the emotional valence of the relevant question is far higher than the emotional valence of the control question, then it’s likely to be highly disturbing to an “innocent” test subject and thus likely to elicit a higher physiological response. This will result in them being labelled deceptive despite the fact that they are not.

Furedy gives the example of the following control-relevant pairing from a case he was involved with:

Control Question: Have you ever done anything of which you were ashamed?
Relevant Question: Did you lick X’s vagina? (Where X was a four year old girl)

In this case, the emotional disparity between the control question and the relevant question is startling. The control question concerns something relatively innocuous and which is likely to garner assent; the relevant question concerns a serious criminal offence, and even an innocent person is likely to be placed under great stress when responding to it.

The thing is, the person administering the test is the one who formulates the control questions (usually after a pre-test interview) and so they have control over the emotional valence of the control questions. Because they have this choice, we get the first horn of our dilemma: if the tester formulates a control question which is markedly less emotionally disturbing than the relevant question, then they run the risk of false positive results. This is ethically undesirable because it could result in an innocent person being denied a job or arrested for a crime.

The second horn of the dilemma comes from the other choice that the polygrapher could make: trying to match the emotional valence of the two questions. Thus the polygrapher would ask a control question test that is either very similar to the relevant question or, more commonly, exploit the pre-test interview by convincing the test subject that certain minor offences or moral lapses have been proved (scientifically) to be linked to others (usually the subject of the relevant question). At this stage, the polygrapher will bandy about phony scientific claims that are designed to confuse the naive test subject in such a way that they can’t distinguish between the emotional valence of the control and relevant question.

Matching the emotional valence of the two questions in this manner should help to lower the rate of false positives (though, obviously, not eliminate it) but it does so at a cost. According to Furedy, the exploitative nature of the pre-test interview, coupled with the accusatory and high-pressure test environment, are likely to make the case of the matched-questions CQT traumatic to the innocent person. This is particularly so because the test subjects are not normally psychologically debriefed about the nature of the test. This is for good reason too (from the polygrapher’s perspective): explaining why they were questioned in that manner runs contrary to the spirit of the test, which is deliberately designed to encourage confusion about the nature and purpose of the questions being asked. If the test subject knows why they are questioned in a particular manner in advance, then they are likely to be able to cheat the test in various ways (by adopting “counter-measures”). Thus, maintaining a mystique about the test is important.

What’s more, there is no intermediate path that the polygrapher can take. If they adopt medium-strength questions, they are likely to get inconclusive test results, which are of no value. That gives us the Polygrapher’s dilemma, which can be formally stated as follows:

  • (1) If a person is faced with a choice between two courses of action (A, B) each of which is morally undesirable, then they are confronted with an ethical dilemma. 
  • (2) In administering a CQT, a polygrapher can either: (a) ask control questions that are markedly less disturbing than the relevant questions; or (b) ask control questions that are emotionally equivalent to the relevant questions. 
  • (3) If they ask control questions that are markedly less disturbing than the relevant questions, the test will result in a high number of false positives, which is undesirable for a variety of reasons.
  • (4) If they ask control questions that are emotionally equivalent to the relevant questions, the test is likely to be traumatic to the innocent test taker, which is undesirable for a variety of reasons.
  • (5) Therefore, in administering the CQT, the polygrapher faces an ethical dilemma.

Furedy is keen to point out that the polygraph test faces a number of scientific and empirical objections too, in addition to this ethical one, but he offers this argument nonetheless as it may persuade those who have, so far, been unperturbed by the scientific objections.

3. Evaluation and Way Forward
What I want to consider here is whether Furedy’s dilemma is a serious one. To evaluate this, I simply need to look at the premises of the preceding argument and work out whether there are good reasons to accept them. As regards the first horn of the dilemma — which we might call the false-positive horn — I am in agreement with Furedy. Creating that kind of emotional disparity between the two questions is likely to result in false positives and these are indeed morally undesirable for a variety of reasons (inculpation of the innocent, wasted resources and so on).

I’m rather less sure about the second horn of the dilemma — which we might call the traumatising horn. Is it really true that innocent test subjects are likely to be seriously emotionally harmed by the test? That looks like an arm-chair hypothesising to me. It does sound somewhat plausible, but one learns to be wary of what sounds plausible when it comes to the idiosyncrasies of human psychology. Any short-term trauma might be erased in the long-run. Also, one has to wonder whether all contexts in which a lie detection test are employed are likely to involve deeply disturbing questions of the sort alluded to above.

By itself, that mightn’t look like much of an objection to Furedy’s argument, but when you combine it with the fact that the CQT might be administered under certain extreme conditions — conditions in which a horrendous outcome might arise if the deceptive person is not located in a timely fashion — you might get a reason for favouring the test despite its traumatising effect. I have in mind here a kind of ticking bomb scenario, where a lie-detection test is used to identify those who are likely to have the information necessary to avert some disaster. The lie detection test would not actually provide you with the information needed to avert the disaster, but in separating out those who are likely to have it from those who are not likely to have, it could be a valuable first step in that direction. There are also other benefits that might result from weeding out the deceptive, and these might also compensate for the traumatising effect of the test.

Of course, whether or not a CQT lie detector could perform this function would depend on its scientific and empirical credibility, which is a topic for another day (quick answer: it seems to have little credibility). In this respect, it’s worth noting that there is an alternative testing paradigm, that is neither subject to Furedy’s logico-ethical dilemma, nor as scientifically dubious as the CQT. I refer to the Guilty Knowledge Test.

The GKT does not involve the comparison of responses to control and relevant question tests. Instead, it involves the comparison of physiological responses to multiple choice questions or multi-part stimuli. The goal of the test is to identify those who have “guilty” (or “relevant”) knowledge; not to identify those who are being deceptive (though that may be a side effect of the test). To give an example, suppose the police are investigating a murder, which took place in the victim’s kitchen. They might ask a suspect the following question as part of a GKT:

GKT Question: Where was the victim’s body found: (a) in the kitchen; (b) in the living room; or c) in the bedroom?

To the innocent person (provide these details are not widely known) each option would seem equally likely, and so when presented with the three options in sequence, the physiological response profile is likely to be the same. By way of contrast, the guilty suspect, who knows where the victim’s body was located, is likely to exhibit a greater response to option (a). This protocol does not involve the generation of potentially traumatic control questions, nor the exploitative and deceptive pre-test interview. Hence, it can avoid the worst excesses of the CQT.

To work effectively, the GKT requires certain information to be concealed from the innocent public, and also, most importantly, requires that there be some reliably detected physiological signal that encodes the distinction between “known” and “unknown” information. There are many such candidate signals, ranging from detectable brainwaves to autonomic responses but whether they are reliable or effective measures is a topic beyond the scope of this simple blog post.

What’s most interesting about the GKT is that many psychophysiologists, who are often quite dismissive of the classic polygraph lie detector, are quite ebullient about its potential uses in forensic and counter-terrorist contexts. For a collection of articles on this topic, I recommend Memory Detection: Theory and Application of the Concealed Information Test (Cambridge University Press, 2011).

No comments:

Post a Comment