PBS’ Frontline has a new show called The Real CSI that examines many of the most commonly used criminal forensics techniques and shows the problems and limits of how useful they are. In a segment just released from the show, they look at a study about cognitive biases in fingerprint analysis:
In 2004, cognitive neoroscientist Itiel Dror set out to examine whether the process of fingerprint analysis, long considered one of the most reliable forms of forensic science, can be biased by the knowledge examiners have when they attempt to find a match for prints from a crime scene.
In the clip above from tomorrow night’s film The Real CSI, FRONTLINE correspondent Lowell Bergman explains the way Dror constructed an experiment using the case of Brandon Mayfield. Mayfield, an Oregon lawyer, was at the center of international controversy in 2004 after the FBI and an independent analyst incorrectly matched his prints to a partial print found on a bag of detonators from the Madrid terrorist bombings.
Dror asked five fingerprint experts to examine what they were told were the erroneously matched prints of Mayfield. In fact, they were re-examining prints from their own past cases. Only one of the experts stuck by their previous judgments. Three reversed their previous decisions and one deemed them “inconclusive.”
Dror’s argument is that these competent and well-meaning experts were swayed by “cognitive bias”: what they knew (or thought they knew) about the case in front of them swayed their analysis.
Which is why the analyst should be entirely isolated from the case, should know nothing about it at all before comparing fingerprints, and should also have to compare the fingerprint to several different prints, not just the one the police think might be a match. That’s how you avoid such cognitive bias. It’s the same principle that should govern police lineups, the officer administering the lineup should not have any idea which person in the lineup is the one the police think committed the crime.
15 comments
Skip to comment form ↓
The Lorax
April 25, 2012 at 2:17 pm (UTC -4) Link to this comment
It’s the same principal that should be used in any scientific research: double-blind experiments. Not just for fingerprints.
Katherine Lorraine, Chaton de la Mort
April 25, 2012 at 2:20 pm (UTC -4) Link to this comment
Zoom and enhance?
gshelley
April 25, 2012 at 3:00 pm (UTC -4) Link to this comment
T he lack of blinding in lineups and other forms of police identification still amazes me. I’m not sure if the people who allow the police officer present ad the line up to be someone who thinks “number 4 is guilty” don’t realise how easy it is to sway the witness, or just don’t care.
d cwilson
April 25, 2012 at 3:34 pm (UTC -4) Link to this comment
It’s been know for a long time now that fingerprint identification is very subjective. I always find it amusing on cop shows when the technician announces that he’s found a “5 point match”. There is no universal standard for how many points of similarity constitutes a positive match. It’s entirely based on the expert’s judgment.
slc1
April 25, 2012 at 3:36 pm (UTC -4) Link to this comment
Re d cwilson @ #4
That’s not true in many, if not most states. The minimum number of points of identification required to declare a match is written into legislation. For instance, an 8 point match is required in Virginia, a 9 point match is required in Federal court, and a 12 point match is required in California.
Doug Little
April 25, 2012 at 3:55 pm (UTC -4) Link to this comment
I think this idea should be extended to juries as well. The defendant remains anonymous to the jurors and they are only presented the facts of the case. This could be done electronically so not even the location of the potential crime would be known.
Doug Little
April 25, 2012 at 3:56 pm (UTC -4) Link to this comment
Sorry, the location would NOT be known
David C Brayton
April 25, 2012 at 4:26 pm (UTC -4) Link to this comment
One thing about fingerprint analysis is that it is not nearly as black and white as one might expect. A lot of analysis goes into determine whether there is a match and there is substantial room for interpretation.
gvlgeologist
April 25, 2012 at 4:29 pm (UTC -4) Link to this comment
Don’t worry, Doug, you got it right the first time: “not even the location of the potential crime would be known” (my emphasis).
Doug Little
April 25, 2012 at 4:47 pm (UTC -4) Link to this comment
Oh yeah, my bad.
Lyle
April 25, 2012 at 9:23 pm (UTC -4) Link to this comment
Which is why the analyst should be entirely isolated from the case, should know nothing about it at all before comparing fingerprints, and should also have to compare the fingerprint to several different prints, not just the one the police think might be a match.
This. The protection from confirmation bias is of extreme importance, particularly when somebody’s rights and liberties are on the line. This country has such an obsession with criminal justice but not enough to be serious about applying due process.
dingojack
April 26, 2012 at 1:45 am (UTC -4) Link to this comment
gshelley (#3) – “The lack of blinding in lineups and other forms of police identification still amazes me“.
You wanna drag some poor shmuck in from the street for a line-up –
and then blind him!
Harsh.
:) Dingo
——–
More seriously, where are you going to get a pool of people who are approximately the same build, hair colur, height & etc. as the suspect, espescially so that even the cops don’t know which is which?
Perhaps you could do it virtually, using randomly generated characters similar to the digitised image of the suspect.
ischemgeek
April 26, 2012 at 6:07 am (UTC -4) Link to this comment
@dingojack: All you’d have to do is to isolate the cop(s) who are going to deal with the witness from the case. Don’t let them know which of the people in the lineup is the suspect.
And they don’t pick people who look so similar they could be same-zygote twins. They have to pick people who have similar build, hair color, eye color, and height so that they don’t bias the witness’ call. As a matter of fact, standard procedures advise against using non-suspects that resemble the suspect too closely. They want them close enough that the witness won’t identify the suspect just because xie stands out, but not so close that even hir friends might have to do a double-take to tell which is which.
dingojack
April 26, 2012 at 8:49 am (UTC -4) Link to this comment
So you’d do the ID parade in a different precinct or something?
Dingo
redpanda
April 27, 2012 at 2:15 am (UTC -4) Link to this comment
Is this the sort of thing that AIs (the predator drone kind, not the terminator movie kind) will be able to do well at in the near future? It seems like with an advanced enough algorithm and enough data we should be able to compare any two fingerprints or partial fingerprints and get an accurate assessment of the probability that they are both from the same person.
That way with a pair that’s fairly similar but not conclusively identical, you can get a neat number like 72% or something that a jury can weigh more objectively than some forensics analyst’s biased opinion.