The Tide of Prints
Thousands of times each day across the United States a police officer books a suspect, stops a suspicious character near a crime, or pulls over a speeder, and takes his fingerprints. Or he pulls “latent” prints from an object at a crime scene. He zips the prints to a central fingerprint database, and gets an answer back immediately. The prints reveal what the suspect’s false name and identification concealed: He has outstanding warrants, or a rap sheet as long as the arm of the law. Or worse: He’s an escaped felon, armed and dangerous. That information will enable the police and courts to hold him-or save the life of that lone cop, who might otherwise be ambushed.
That’s the fantasy, anyway. Now, the reality: When police pull prints, they may indeed be able to make a speedy identification-if they can match the prints in their own city’s or state’s databases. Every state except West Virginia, where the Federal Bureau of Investigation’s own identification operations are located, has a computerized “automated fingerprint identification system,” a.k.a. AFIS. Police in one of the West Coast states connected to the Western Identification Network may be able to find a match among neighboring states’ databases. But as law enforcers never tire of observing, crime doesn’t honor borders. If it can’t match a print in its own or its neighbors’ files, a state must mail it to the mother of all fingerprint repositories-the FBI’s 227-million-card (that’s right, card) collection-and wait a month for a response.
On July 30, this logjam is set to be dynamited. On that date, after 10 years’ development and decades of false starts and dead ends, the FBI’s Integrated Automated Fingerprint Identification System (IAFIS), far and away the world’s largest and most complex such system, is due to come online. IAFIS promises to turn around criminal print searches in two hours and civil searches in 24. Even before the system starts up, however, its built-in limitations are becoming evident-and are threatening to grievously undermine its value for some of the crime-busters who will depend upon it most. And already the possibility looms that this centralized system, the product of an epic development saga, may soon be superseded, either by new “biometric” identification technologies (see sidebars “Beyond Fingerprints,” and “Secret Handshakes”) or by dispersed PCs that can do the same job at a fraction of the cost.
Storing and identifying fingerprints has been central to the FBI’s mission almost from the start. When J. Edgar Hoover became director and set out to expand the bureau in 1924, Congress directed that it take over and integrate the nation’s scattered print files.
The dream of instant identification via fingerprints was born just 10 years later, when the FBI tried to automate its fingerprint files by recording ridge counts (one of the most basic fingerprint measures) on punch cards. But 1934’s analog computers could only sort the cards into general classes, not match individual prints. The effort was soon abandoned-except in the movies. Hollywood, which hyped the bureau’s abilities long before The X-Files, recycled footage of G-men punching in fingerprint searches for decades.
Nevertheless, the bureau continued to be revered as the pace-setter in the arcane science, and art, of fingerprint identification. It refined the elaborate turn-of-the-century Henry System for classifying prints according to numerical values derived from their patterns of loops, arches and whorls, and built a massive mechanical beltway for transporting its millions of cards. But the FBI became the captive of its own early successes, anchored to obsolete approaches while the rest of the world went digital.
Clear back in the 1960s, says Peter Higgins, a former FBI deputy assistant director for information-services engineering, “J. Edgar Hoover decided it was about time the bureau started doing something with computers-and do it in six months or less.” In 1967 it launched the National Criminal Information Center (NCIC) and began developing digital fingerprint readers. When the first operating models arrived 10 years later, the bureau proceeded to scan in its millions of “ten-print” fingerprint cards. But search capability lagged, and the FBI struggled through the 1980s to increase that capacity while automating document flow and transport. The bureau continued to rely on a hybrid paper/online system, much of it still in use today, based on 1970s technology that was obsolete when it was installed.
The typical sequence: A local police department mails its fingerprint cards to its state’s AFIS, which, failing to make a match, mails it on to the FBI. There, one of the bureau’s nearly 800 expert fingerprint examiners determines the print’s Henry classification. File prints of that classification are called up and a computer compares the mystery print’s minutiae-points where ridges end or branch-against theirs. The computer proposes a candidate match, which one examiner checks and another verifies.
Under this labor-intensive regime, it takes the bureau about 33 days to turn around a criminal fingerprint search. The delay is even longer at the other end-the police and other agencies that supply the prints of criminals, government job applicants and other subjects to be entered in the national database. A mid-1990s FBI survey found it takes them on average 118 days to process and mail in prints of newly arrested suspects. Paper shuffling eats up time, and agencies may not give priority to supplying a system they know will be slow to respond anyway. Not only will a suspect be released long before his prints show up in the system, he’ll have ample time to, say, buy a gun or get hired by a daycare center.
One result of this double logjam: In the first half of 1998, police around the country unwittingly released more than 5,000 fugitives because fingerprint IDs didn’t come back in time.
Even as the bureau struggled in the 1980s to meet soaring identification demands, the private sector and local governments leapt ahead. Several companies capitalized on the FBI’s early innovations to build truly automated AFIS equipment. As Kevin Wilkinson, a special assistant in the FBI’s Congressional Affairs Office who worked in the Identification Division in the 1970s, recalls, “the technology leapfrogged in a massive way” over the bureau’s own. States and cities, exasperated with the FBI’s slow progress, bought commercial AFIS rigs. Some have progressed to fully automatic “lights out” fingerprint transmission and searching, so called because no eyes need watch until the computer spits out the result.
With the FBI’s identification division showing signs of terminal obsolescence, an outcry arose among those who most rely on it: the cops. The conduit for their dissatisfaction was the FBI’s NCIC Advisory Policy Board, which represents police and prosecutors nationwide. The board tendered a plan in 1989, drafted by Joseph Bonino, the Los Angeles police department’s identification-division commander, that set the bureau on its current path.
The Policy Board report urged a fresh start; a proposed update of the existing hybrid system would throw good money after bad. Since the states had already proven that AFIS works, the FBI should go straight to it without a superfluous trial. Most important, it should dump the Henry classifications, which the bureau had honed into such an exquisite old-tech tool; with computing power now available to sort prints according to general features and extract and search minutiae directly, Henry’s day had passed.
Another question: just whose search system is this? IAFIS will be overwhelmingly weighted toward the ten-print searches that police and courts care most about-background checks run on prints taken from the people they stop, book and sentence. But it may not do as much for the smaller cohort of investigators trying to solve crimes with latent prints. Nowadays, detectives typically don’t dust for prints after burglaries; their states’ crime labs are too overloaded to trace them. If they expect a change under IAFIS, warns Ed German, the Army Crime Lab’s senior special agent, “many police departments are going to be very surprised.”
IAFIS promises to match only 635 latent crime-scene prints a day, even as it conducts 60,000-plus ten-print searches. This was no oversight, but a deliberate allocation of resources. The Advisory Policy Board urged against building a large latent-search capability into the national system because most such searches involve local offenders. And the 635 promised latent searches will consume much more computing power than all the ten-print searches combined. Because a ten-print’s origins are known-say, “white female left index finger”-the search engine can concentrate on the pertinent section of its database. And because ten-prints are crisp and complete, it can compare them with the digital equivalent of a quick glance. Latent prints are often smudged and incomplete, with no indication of which fingers they represent, or the race or gender of the people who left them. And since each search can penetrate only about 30 percent of the IAFIS database, the system must make multiple searches to match a single latent.
Fingers to the Wind
And so, on the eve of its long-awaited debut, the world’s largest computerized ID system is still a work in progress. Some backers and would-be users may be surprised and disappointed at its limitations. And questions of obsolescence already loom.
For instance, given the huge investment it’s now making in fingerprint-based information technology, the bureau has to be watching a bit nervously as other biometric technologies ease into the picture. Higgins sees DNA matching and iris scans, and perhaps even facial recognition, as promising fingerprint alternatives. Bonino and German both urge that CJIS add automated palm-print searches, which are already used in Europe. “Half of all prints left at crime scenes are palms,” notes Bonino.
Even if fingerprinting isn’t replaced by these recognition techniques, the FBI’s investment in large-scale, centralized data technology could be rapidly superseded by new computing techniques. German sees the emergence of what he calls “a new AFIS paradigm”-a decentralized identification network, somewhat like the Internet. In fact, the PC revolution is just now reaching the fingerprint field. Mainframe AFIS systems still run to hundreds of thousands of dollars. But last June the upstart Phoenix Group in Pittsburg, Kansas, began offering an “AFIX Tracker” that, according to its president, Derald Caudle, searches minutiae points in all latent/ten-print combinations, holds 30,000 print records (enough for a small city)-and operates on a Pentium II Windows machine with an 11.5-gigabyte hard drive. The price: $17,000 for the software, or $950 a month to rent the whole rig. German’s Army lab took the first two.
German predicts the paradigm will really pop after this November, when the FBI will release Lockheed Martin’s minutiae-extraction software into the public domain. “It’ll be like phone-company deregulation. A lot of software companies will spring into AFIS.” Where only about 600 larger cities and agencies now have it, individual police precincts will get their own, German believes. Initially, at least, that should create much more demand for the FBI’s IAFIS, as the hub of a fast-expanding network. But what about when all those little AFISes learn to swap prints and searches directly, on the Internet search-engine model?
As prices drop, German foresees, “You could build your own AFIS.” Imagine. Instead of tracking each other down by name via Hotbot or AltaVista, we’ll trace each other’s fingerprints. Who was that woman at the party last night? Pull a print off her wineglass. In the near future, thanks to new technology, look for gloves to become very, very fashionable
Geoffrey Hinton tells us why he’s now scared of the tech he helped build
“I have suddenly switched my views on whether these things are going to be more intelligent than us.”
Meet the people who use Notion to plan their whole lives
The workplace tool’s appeal extends far beyond organizing work projects. Many users find it’s just as useful for managing their free time.
Learning to code isn’t enough
Historically, learn-to-code efforts have provided opportunities for the few, but new efforts are aiming to be inclusive.
Deep learning pioneer Geoffrey Hinton has quit Google
Hinton will be speaking at EmTech Digital on Wednesday.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.