On March 11, 2004 a series of bombs exploded aboard four commuter trains in Madrid, Spain, killing 192 people and injuring 2050.
The Spanish police recovered a bag containing detonating devices which had a latent fingerprint that the Spanish shared with the United States Federal Bureau of Investigation (FBI).
The FBI apparently ran a check on the fingerprint using the FBI’s Automatic Fingerprint Identification System (AFIS). AFIS uses a pattern recognition algorithm to generate an ordered list of possibly matching fingerprints.
One of the fingerprints in this list matched Brandon Mayfield, a Muslim American attorney from the Portland, Oregon region. Expert latent fingerprint examiners from the FBI proceeded to positively identify the fingerprint from the Madrid train bombing as belonging to Brandon Mayfield; at least, this is what the FBI claimed at the time.
Mayfield was arrested as a material witness in the bombing and a great deal of information about him seems to have been leaked to the press. Meanwhile, the Spanish police matched the fingerprint to an Algerian man whom they arrested. The Spanish police directly challenged the FBI identification of Mayfield, leading to his eventual release.
Mayfield later successfully sued the FBI for his treatment.
Fingerprint identification has been in widespread use in the United States since the 1920′s where popular culture has, until recently, held that fingerprints are unique. According to some reports, people have even been executed based solely on a fingerprint identification.
This is striking since most human and automatic pattern recognition abilities and algorithms have significant false positive and false negative error rates. The author had the experience in 2002, prior to the Mayfield case, of trying to locate scientific studies confirming the accuracy of latent fingerprint identification without success. In fact, there have been a number of cases prior to the Brandon Mayfield case in which fingerprint identification was shown to have been wrong.
The suspect had an airtight alibi. DNA tests contradicted the fingerprint identification and cleared the suspect. These cases of incorrect fingerprint identification have always been blamed on fraud or error by the human latent fingerprint examiners rather than a case of identical fingerprints.
Some facts about fingerprints. Identical twins usually, perhaps always, do not have the same fingerprints. This means a fingerprint test can discriminate between the otherwise identical twins. Dramatic demonstrations of this remarkable fact helped convince juries in the 1920′s to accept fingerprint identification. However, these demonstrations do not constitute rigorous scientific statistical studies of the accuracy of latent fingerprint identification.
A small minority of people do not have fingerprints. Some people have fingers with very shallow ridges which, in practice, makes fingerprint identification more difficult. Contrary to some claims, fingerprints can be altered by scarring and wear and tear. Automatic fingerprint recognition algorithms have had substantial problems with people who work with their hands.
Discussions of the accuracy of fingerprint identification often confuse the accuracy for comparisons of all ten prints, all ten fingers, and the rates for a single or a few prints. For example, automatic fingerprint recognition algorithms were very accurate with all ten fingerprints in 2002 but much less accurate for a single print such as the forefinger or thumb. It is difficult to get all ten prints in the real world.
Latent fingerprint identification is performed by human examiners. There are automatic fingerprint recognition programs such as AFIS but these are probably not as accurate as human beings. This is not unusual. In general, human pattern recognition abilities are significantly better than automatic methods based on mathematical, statistical, or scientific methods: artificial intelligence, pattern recognition, machine learning, and other synonyms.
This is something to keep in mind when scientists, attorneys, or others denigrate eyewitness testimony. Nonetheless, human pattern recognition abilities are imperfect. There are false positive and false negative rates. Eyewitnesses do misidentify people and objects. Human fingerprint examiners almost certainly have non-zero false positive and false negative rates.
In the wake of the Mayfield case, an FBI Laboratory review committee evaluated the scientific basis of friction ridge examination (fingerprint identification) and recommended scientific research including a study of the accuracy of latent fingerprint examiners (!).
The National Research Council (NRC) also identified the need for evaluations of fingerprint examination decisions in a study in 2009. The FBI recently published a report on such a study in the Proceeding of the National Academy of Sciences (Accuracy and reliability of forensic latent fingerprint decisions, PNAS, April 25, 2011). This study found a 0.1% false positive rate and a 7.5% false negative rate.
It is worth considering this for a moment. Fingerprint identification is in widespread use in the United States. People are routinely convicted or cleared of crimes based solely or in part on fingerprint identification. Fingerprints have long been portrayed and perceived as unique.
Fingerprint identification is usually perceived as a highly scientific form of identification. Yet basic scientific studies of the accuracy of the technique appear to have been lacking until recently. This lack has only become apparent recently as unfavorable comparisons to the seemingly rigorous statistical basis of DNA profiling (formerly known as DNA fingerprinting) have been made as well as the extensive publicity received by the Mayfield case, much higher than previous misidentifications which lacked the post 9/11 terrorism angle.
The uniqueness of fingerprints seems to be one of those things that “everyone knows” that has a remarkably weak basis in fact. Indeed, this seems to happen from time to time in supposedly fact-based scientific and engineering fields. A significant number of scientific and technological breakthroughs have occurred when someone went back and questioned the underlying evidence or data behind something “everyone knew.”
In most crimes, only one or a few partial fingerprints are recovered, such as the thumb and forefinger used to hold an object, e.g. the bag in the Madrid train bombing. There are over six billion people on Earth. Suppose that one in a million people have the same or essentially the same partial prints; an examiner cannot tell the difference. This means that, in fact, there would be about six thousand (6,000) possible matches including the guilty party.
With automobiles, trains, and especially air travel, it is probable that a substantial proportion of these six thousand suspects live within traveling time of the crime and lack an alibi. There was a small possibility that Brandon Mayfield traveled secretly from Portland, Oregon in the United States to Madrid, Spain to participate in the bombings. Unlikely, but certainly possible. Even a very small false positive rate raises a reasonable doubt.
Especially since the Mayfield case, there has been more questioning of the scientific basis of fingerprint identification both by authorities such as the FBI and the National Research Council as well as in popular culture. The TV show Numb3rs, discussed in the previous post The Magical Mathematics of Numb3rs, features an episode, probably inspired by the Mayfield case, in which a man is wrongly convicted due to an error in fingerprint identification.
In their book The Numbers Behind NUMB3RS, mathematicians Gary Lorden and Keith Devlin have a chapter questioning some of the mathematical basis of fingerprint identification. DNA profiling is seemingly based on detailed rigorous scientific studies of the frequency of the various genetic markers used in the DNA tests.
Comparable studies seem to be lacking where fingerprints are concerned, hence the studies of fingerprint identification that the FBI is now performing and publishing. The comparison between DNA profiling and fingerprinting has led to questions about the accuracy of fingerprint identification.
It may also be the case that questions about the scientific basis of fingerprints may be a way of marketing DNA profiling as a more “scientific” and reliable replacement for now “old fashioned” (“legacy” in the parlance of the software industry — usually meaning it works and the market is saturated so we need to sell a new replacement technology) fingerprint identification.
The uncritical acceptance of fingerprint identification for over eighty years, without apparently performing adequate rigorous studies of the accuracy, illustrates the enormous hypnotic power of mathematics and science in our culture. The popular image of mathematics and science is that they give exact, black and white answers.
“Scientific” tests give reliable yes/no answers. The Madrid bomber was Brandon Mayfield. The bomber was not Mayfield. There are no false positive or false negative error rates. Two plus two is four, not 3.999 plus or minus 0.012. Yet, this is very rarely the case in the real world. In fact, one should almost always demand to know the error rates of numbers and be suspicious of numbers quoted without error rates or other qualifications.
Bradford T. Ulery(a), R. Austin Hicklin (a), JoAnn Buscaglia(b),1, and Maria Antonia Roberts(c)
(a) Noblis, 3150 Fairview Park Drive, Falls Church, VA 22042;
(b) Counterterrorism and Forensic Science Research Unit, Federal Bureau of Investigation
Laboratory Division, 2501 Investigation Parkway, Quantico, VA 22135;
(c) Latent Print Support Unit, Federal Bureau of Investigation Laboratory
Division, 2501 Investigation Parkway, Quantico, VA 22135
Edited by Stephen E. Fienberg, Carnegie Mellon University, Pittsburgh, PA, and approved March 31, 2011 (received for review December 16, 2010)
Proceedings of the National Academy of Sciences (PNAS)
April 25, 2011
The fingerprint image is from Wikimedia Commons and is in the public domain. Fingerprint Image at Wikimedia Commons
About the Author
John F. McGowan, Ph.D. solves problems using mathematics and mathematical software, including developing video compression and speech recognition technologies. He has extensive experience developing software in C, C++, Visual Basic, Mathematica, MATLAB, and many other programming languages. He is probably best known for his AVI Overview, an Internet FAQ (Frequently Asked Questions) on the Microsoft AVI (Audio Video Interleave) file format. He has worked as a contractor at NASA Ames Research Center involved in the research and development of image and video processing algorithms and technology. He has published articles on the origin and evolution of life, the exploration of Mars (anticipating the discovery of methane on Mars), and cheap access to space. He has a Ph.D. in physics from the University of Illinois at Urbana-Champaign and a B.S. in physics from the California Institute of Technology (Caltech). He can be reached at email@example.com.