Although a little scattered, Talking to Strangers makes important points. Its overarching theme is that humans think we can deduce things about strangers by reading their behavior and demeanor, but we really can’t. This is most significant when we are trying to detect deception: Some people just seem honest, even when they are not; and others seem dishonest even when they are truthful. When attempting to ferret out deception, people do better when they have no information on a subject’s behavior or demeanor.
The book relies heavily on the research of Timothy Levine, who over 20 years of academic research has shown that people can’t reliably read whether a person is lying. When relying on their perception of a subject’s demeanor, people detect liars only 54% of the time (50% would be no better than flipping a coin), and it appears that the only reason we do (marginally) better than chance is that some few people are very bad at lying.
People who think that they have expertise in detecting deception – police, judges, and intelligence agents – are actually worse because of their erroneous belief that they can read subjects. The consequences in the justice system are outrageously common and too often devastating to the victims: harassment, jailing, and even conviction of innocent but behaviorally mismatched suspects.
The Amanda Knox case is a typical example: Police thought her behavior was odd, and they thought they had expertise in detecting deception, and so they disregarded objective evidence in pursuing and winning a conviction of her. As Gladwell summarizes it: “Amanda Knox spent 4 years in an Italian prison for the crime of not behaving the way we think people are supposed to behave after their roommate is murdered.” (The lead investigator, Edgardo Giobbi, had erroneously concluded, “We were able to establish guilt by closely observing the suspect’s psychological and behavioral reaction during the interrogation. We don’t need to rely on other kinds of investigation.” Oopsie.)
Every day the justice system depends on judges to determine whether people charged with crimes should be allowed to await trial on bail, or whether they will be jailed for the months or years it takes for a fair trial. The calculus is supposed to be whether the accused poses such a danger to society that he should lose his freedom because of the accusation. Gladwell interviews judges who explain that they make this determination in large part by assessing the demeanor of the suspect during arraignment. And it turns out that a computer given only the objective information in suspects’ criminal records is more accurate at making this judgment.
Are we sending perfectly harmless people to prison while they await trial simply because they don’t look right? We all accept the flaws and inaccuracies of institutional judgment when we believe that those mistakes are random. But Levine’s research suggests that they aren’t random – that we have built a world that systematically discriminates against a class of people who, through no fault of their own, [have abnormal demeanors].
Malcom Gladwell, Talking to Strangers
Federalist 
I suppose its asking too much for judges and cops to be trained when new information like this is revealed.