Having written yesterday about false confessions, I sat up and took notice when (via Bruce Schneier) I saw this post from Ross Anderson at Light Blue Touch Paper, liveblogging from the Interdisciplinary Workshop on Security and Human Behavior at MIT last month on problems related to deception and computer security. (Here are photos from the event if you weren't on the exclusive invite list.) According to Anderson:
The first session, on deception, was fascinating. It emphasised the huge range of problems, from detecting deception in interpersonal contexts such as interrogation through the effects of context and misdirection to how we might provide better trust signals to computer users. ...That last observation identifies a key flaw in using behavioral cues for lie detection: Probably more people people deceive themselves into believing lies of all shapes, colors, sizes and motivations - particularly when justifying their own actions faced with accusations - than the rationalists among us would like to believe.
The first session brought home what a huge subject deception is. The first speaker, Paul Ekman, pretty well established the study of how people deceive each other in face-to-face contexts, and how we detect deception. We observe a social hot spot - a discrepancy or an implausible statement, say - and this prompts us to look for an emotional leakage, via something like a microfacial expression.
However the error rate can be high, as the hot spot can come from a context switch and the emotion being leaked could be (say) anger at being interrogated rather than fear of being caught in a lie. Othello’s error was to read Desdemona’s fear correctly but to misunderstand its cause. ...
In particular - and Paul Ekman agreed to this in the panel discussion - the best way to tell a lie is to deceive yourself into believing it. This should get a lot more research. Charlatans in particular - such as spirit mediums who pretend to talk to the dead - operate by inducing self-deception.
There's little doubt that detecting deception or mitigating the harm it causes (e.g., fraud) constitutes a major part of security of all types, not just computer security. The comments about deception, in particular, apply directly to day to day crime and punishment topics, not the least of which is the subject of police eliciting false confessions or identifying them after the fact.
A core method of modern American police interrogation relies on questioners judging verbal and nonverbal cues to determine if the interrogation subject is being deceptive, then developing an interrogation strategy grounded in that hypothesis. The most coercive tactics are reserved for offenders judged deceptive by police.
The problem is just as Anderson identified: The error rate is high, risking the same mistake committed by Othello when he falsely accused Desdemona. According to the (highly recommended) recent book by Richard Leo, Police Interrogations and American Justice (pp. 98-99):
Numerous controlled studies have shown that people are not good intuitive judges of truth and deception, typically performing at no better than chance levels of accuracy. Controlled studies have also shown that even investigators and other supposed experts who routinely evaluate deceptive behavior are highly prone to error. Moreover, Kassin and Fong have shown that police interrogators and others specifically trained in the [Behavioral Analysis Interview technique taught by John Reid and Associates] not only fail to discriminate accurately between true and false statements much of th time, but also that behavior analysis training actually lowers the ability of police interrogators to discriminate accurately between true and false denials. Further, such training inflates their confidence in their judgments. (citations omitted)Assuming police can accurately judge lie detection leads directly to false accusations and confessions. That realization led law enforcement in some European countries to switch to an information gathering rather than an inquisitorial approach to interrogations over the last couple of decades, in addition to recording the entire process.
If the total rate of wrongful convictions turns out to fall on the higher end of the range of estimates, it will be in large part because police are little better at judging deception or its causes than Othello when he murdered poor Desdemona, protesting her innocence to the end.
BLOGVERSATION: Scott Greenfield responds that it's not just police who can't accurately tell when someone is lying, but also judges and jurors.