Computer-Assisted Witness Identification

Witnesses are much more accurate at identifying criminals when computers assist in the identification process, not police officers.

A major cause of miscarriages of justice could be avoided if computers, rather than detectives, guided witnesses through the identification of suspects. That's according to Brent Daugherty at the University of North Carolina in Charlotte and colleagues, who say that too often officers influence witnesses' choices.

The problem was highlighted in 2003 when the Innocence Project in New York analysed the case histories of 130 wrongly imprisoned people later freed by DNA evidence. Mistaken eyewitness identification was a factor in 77 per cent of the cases examined.

Makes sense to me.

Posted on October 7, 2009 at 7:12 AM • 19 Comments

Comments

gregOctober 7, 2009 7:58 AM

As a witness myself I can attest to the "bias" of police officers.

I had a guy come into my house while i was inside... the guy freaked out and ran away, but the police had a long series of rapes and burglars they believed was done by the same person.

Anyway about 3 months later they asked my to identify him out of about 20 different mug shots. He was most definatly not any of them.

But that wasn't good enough, I was told they know it was this guy (pointing to one of the photos) and just needed me to say so for the arrest warrant. They got quite abusive toward me when i refused, and assured then I would say it wasn't him in court too.

I don't really trust officers any more...

DinoOctober 7, 2009 8:02 AM

Influence whther known or not is always there. Picking up on sometions emotions can be overpowering, even when the individual is trying to be professional. The use of computers is a nice idea to avoid that.

Muff PotterOctober 7, 2009 8:53 AM

Richard: no, he wouldn't. The problem is not that the cop knows whodunnit, the problem is that the cop will be a person just like you and me, with both conscious and unconscious biases.

Have you ever wondered, for example, why African Americans are much more likely to be convicted of murder (and much more likely to receive a harsher sentence when convicted) than European Americans, even though hardly any jury member these days will be outright racist the same way that people were in the past? It's the same mechanism.

HJohnOctober 7, 2009 9:26 AM

@greg: "But that wasn't good enough, I was told they know it was this guy (pointing to one of the photos) and just needed me to say so for the arrest warrant. They got quite abusive toward me when i refused, and assured then I would say it wasn't him in court too."
__________

That was wrong of them to treat you that way.

I have quite a few police officer friends, and they are human like the rest of us. It isn't easy on them to have someone who quite obviously, to them, has committed multiple crimes yet, in each instance, there was just enough reasonable doubt to tie their hands. After a while, though in each individual case there is reasonable doubt, the reasonable doubt in the police officer's mind has faded with recurrence. I can see where they would be frustrated greatly, but that is certainly no excuse.

Before anyone thinks I'm justifying the police officers, I am not. What I discussed is precisely the reason a computer assisted identification process would be superior. It has no opinion or feelings, only facts.

My father was once mugged on vacation, right in front of my sister. My sister described the mugger, and the police knew who it was and found him walking a few blocks away. When he was brought in, my sister said she was 80% sure that he was the guy. The police told her that wasn't good enough. Fortunately, a camera, though not high enough quality for a positive ID, placed someone wearing the same clothes at the scene and he confessed. I respect the police in this case for letting the evidence speak for itself rather than force the issue. Though my dad was the victim, I would have chosen an 80% chance the mugger walks over a 20% chance an innocent person was wrongly convicted.

AKeckOctober 7, 2009 9:55 AM

I am optimistic about this technology, but I take issue with the claim "Witnesses are much more accurate".

From the New Scientist article, "They found comparable results both times, suggesting Officer Garcia performed as well as an impartial human."

This didn't sound like "much more accurate" to me, so I dug into the linked research paper.

"Although the
virtual officer condition produced slightly more correct identifications than did the
live officer condition (virtual = 31.1% vs. live = 19.2%), this difference was not
statistically significant. The percentage
of filler identifications was nearly identical across conditions (virtual = 31.1% vs.
live = 33.3%), and the difference in incorrect identifications was also not statistically
significant (virtual = 37.7% vs. live = 47.4%)."

This is promising research, but let's not get too sensational Bruce. Great info, regardless.

JeremyOctober 7, 2009 10:07 AM

The computer aided would be superior I think. The problem is when officials use the computer aided solution as the ONLY source. It would be pretty easy to frame someone right?

Also, what if some agency or gov decided it would be great to use the same mechanism as a fund raiser like the red light cameras?

I don't plan on breaking any laws anytime soon, so I'm not really worried about it, but in a police state this could be really scary. Beyond reasonable doubt would have no more meaning.

anonymousOctober 7, 2009 10:30 AM

"I don't plan on breaking any laws anytime soon, so I'm not really worried about it, but in a police state this could be really scary."

Really! I certainly don't want to be driving behind you.
Or did you mean that you aren't going to break any laws that are commonly enforced?

HJohnOctober 7, 2009 10:33 AM

@:Jeremy: "The computer aided would be superior I think. The problem is when officials use the computer aided solution as the ONLY source. It would be pretty easy to frame someone right?"
____________

I think that, by definition, it wouldn't be the only source. It would still need corroborated by the eye witness, there would have had to have been some reason the suspect was in the lineup, and hopefully addition evidence.

On surface it seems harder to frame someone, but, as with anything, someone may think of a way.

HJohnOctober 7, 2009 10:42 AM

@Jeremy: "I don't plan on breaking any laws anytime soon, so I'm not really worried about it"
_____

It's the innocent that have reason to worry. Which is why we have all the legal protections. If investigators and police never got the wrong guy, it wouldn't be necessary.

Petréa MitchellOctober 7, 2009 10:49 AM

I'm with AKeck. All this seems to say is that if you don't have a live human available who doesn't know who the suspect is, using a computer will work just as well.

But look at what "just as well" means! With the photo of the suspect available, 31.1% of the time, someone else is picked out. Without the suspect in the mix, there's a false accusation 50.8% of the time.

The problem as described at the beginning of the paper is that eyewitness identification is unreliable. The only reasonable conclusion here is that it should not be attempted at all.

BetaOctober 7, 2009 10:53 AM

@greg: "But that wasn't good enough, I was told they know it was this guy (pointing to one of the photos) and just needed me to say so for the arrest warrant. They got quite abusive toward me when i refused, and assured then I would say it wasn't him in court too."

That took courage. I'll remember this story the next time I'm on a jury.

Clive RobinsonOctober 7, 2009 10:59 AM

First thiings first,

I have seen a number of studies that show "eyewitness identification" on a "line up" is at best 30% within 4 hours of a full face close distance view of the "perp" and with them still wearing the same garments.

This drops to around 25% within 24 hours and less than 20% after three days.

Other studies have shown that general hight and build are more accuratly remembered as is distinctive garments, hair colour is usually incorrectly described.

One report I remember was one that investigated witness charecter and reliability of observation.

It turns out the people who are least "certain" are generaly more accurate, but more suceptable to prompting.

All in all I would most definatly not trust eye witness evidence as being more reliable than throwing dart blind folded at a wall full of photos.

Mark ROctober 7, 2009 11:51 AM

The article makes a few leaps of logic that don't seem justified by the facts.

First they bring up the statistic on eyewitness misidentification being involved in 77% of cases that were overturned by DNA evidence.

From this they assume that the police officers running identification process must be the problem... this seems like a good hypothesis.

They then create a robo-cop to run the identification process instead, and they test its accuracy against an unbiased person. Since robo-cop does equally well, they declare success.

But none of this demonstrates anything about the theory that the cops are biasing the results, or that the robo-cop will improve the situation... or did I miss that part?

Clive RobinsonOctober 7, 2009 11:59 AM

Opps,

In my above,

"30% with 4 hours"

Should be,

"30% within 4 hours"

That is if a witness is interviewed less than 4 hours after the incident they pick the correct person aproximatly 30% of the time.

JasonOctober 7, 2009 1:12 PM

I'm reminded of a scene in the movie Candyman that always seemed sort of odd.
The heroine is attacked by a thug who claims to be Candyman. The police produce a line-up and she correctly identifies the guy.
The police are jubilant and are all like, "Good girl!"
They knew this was the guy but just needed someone to say so. They had been trying to get "something" on him for a long time as he was terrorizing the neighborhood but folks were afraid to speak up.
It was a movie, and it wasn't beneficial to the plot to have the wrong guy arrested, but I wonder if, in real life, she would have been at least unconsciously coached as to which person would be the one they would like her to choose.

BF SkinnerOctober 7, 2009 1:30 PM

some studies indicate a good 30% of us will aggree with whatever the herd says when we know they are wrong. When there's an authority figure involved that number goes up.

I've seen people believe everything the computer tells them. So maybe not so good.

bobOctober 8, 2009 7:07 AM

@Beta, Greg: The most recent time I served on a jury, this one juror just kept saying over and over "if he is innocent, then why doesn't he take the stand and say so". I pointed out that there could be many reasons why, but it doesn't matter why, the consititution says he is INNOCENT until PROVEN guilty and he also has the RIGHT not to testify if he does not wish to.

To which his response was "yes, but if he was REALLY innocent he would take the stand and say so..." [rinse, repeat] I finally got him to stop SAYING it, but I believe that was the underlying reason he voted to convict. (I voted to convict also, but for a bunch of other reasons)

It scares me that people who think like that may be on MY jury someday. [and no, the fact that I dont commit crimes will not guarantee I am not tried for one; in fact the trend seems to be the other way]

Leave a comment

Allowed HTML: <a href="URL"> • <em> <cite> <i> • <strong> <b> • <sub> <sup> • <ul> <ol> <li> • <blockquote> <pre>

Photo of Bruce Schneier by Per Ervland.

Schneier on Security is a personal website. Opinions expressed are not necessarily those of Co3 Systems, Inc..