MELISSA BLOCK, host:
A failure of the FBI's computerized system to match fingerprints allowed a wanted sex offender to walk free in Georgia. Authorities say after he was released, Jeremy Brian Jones went on to kill four women. Last year Jones was picked up on minor offenses and fingerprinted. But when those fingerprints were entered into the computer and sent to the FBI, there was no match. The check did not turn up the fact that he had previously been arrested for rape and had jumped bail. He's since been arrested and is in jail. Kenneth Moses is a forensics expert. He helped install automated fingerprint systems in California.
Mr. Moses, the FBI is saying this isn't human error; it's a rare case of this technology failing.
Mr. KENNETH MOSES (Director, Forensic Identification Services): That makes a lot of sense 'cause, in fact, these systems are what they call `lights out.' There is no human interaction in the identification process. The computer does the search and then decides whether or not there is the match or not.
BLOCK: And when the system fails, why does it fail?
Mr. MOSES: Well, none of these systems are 100 percent accurate. They only reach into the high 90s. When I started designing these systems, they were in the 70s, so they've improved quite a bit, but you're never going to get to 100 percent.
BLOCK: Now in those times that it does miss, what's going wrong? What's the computer not recognizing?
Mr. MOSES: The most common reason is the quality of the database; that is if it's searching for a particular person and that person's prints were taken, say, five or six years ago by some jailor in a local community, and he didn't do a careful job or he smeared the prints, that's what they receive at the FBI. If the database print on a person is very bad, the computer will miss it.
BLOCK: When the fingerprints are entered into the system and the computer starts doing its work of matching, what's it doing exactly? How is it making those matches?
Mr. MOSES: Well, what it does is the computer will scan each finger with the line--dot of light. And what it's looking for is where the fingerprint ridges, the little black lines that make up the fingerprint--what they do besides going on unbroken. So if they stop all of a sudden, like you have a dead end, it will mark that point. If one ridge forms into two, it'll mark that point. And in any fingerprint there's probably in excess of 50 to 100 of these points. And once it has those points, it joins the dots together like geometry, and it makes a geometric figure which becomes the fingerprint.
BLOCK: Now in this case, in the case of Mr. Jones, apparently the image was a digital image. It was taken fairly recently. Any way of knowing how good a print that might have been?
Mr. MOSES: Well, without looking at it, you don't know, but the fact that it's digital doesn't mean it's accurate. For instance, if you just roll your own finger on the desk, if you move that finger, slide it or push down too hard or not hard enough, even a digital image will be less than ideal. Now there's a wide range of tolerance that the computer has for poor-quality fingerprints. After all, there are 10 fingers, and all you need is a couple of them good to make a match. But if it's a very poor-quality set of prints, it will miss. And once it misses, it then assumes this is the first time this person's been arrested, and it will assign--the computer will assign it a new criminal identification number, which is what happened here.
BLOCK: Given the system the way it is now, what could be done to make it better? What's the thinking of ways to improve the matching system?
Mr. MOSES: Well, one way on the computer end is to make it more tolerant of poor fingerprints. And let's say if it searches a couple of fingers and they're bad, it keeps trying to find clear parts of any finger that might be good. But to tweak it that much would probably cost millions of dollars. You're probably better off focusing on the local level, to improve training and to improve the taking of fingerprints to begin with.
BLOCK: There are critics, of course, who, especially in the criminal justice system, say we've just become too reliant on these fingerprint matches, that they're not reliable.
Mr. MOSES: These fingerprint systems have increased the safety of everybody in society. When I was starting out, before we had these computers, we missed people all the time doing manual searches. People would be arrested, we'd take their fingerprints and manually search them. And by the time the searches came back from the state or federal government, it may be four to six weeks. Meanwhile, this guy's out and about. You compare that to today, where you get a fingerprint match in minutes. These people are still standing there when you find out who they really are and that they have warrants in various places.
BLOCK: Well, Mr. Moses, thanks for talking with us today.
Mr. MOSES: My pleasure.
BLOCK: Kenneth Moses is director of Forensic Identification Services in San Francisco.
(Soundbite of music)
BLOCK: You're listening to ALL THINGS CONSIDERED from NPR News. Transcript provided by NPR, Copyright NPR.
