I've heard that the FBI's computers return a list of the top five or so likely matches, and then a human compares likely suspects' prints to the sample. This does not leave me at all confident in fingerprint analysis.
Does anyone know of any controlled studies that took latent prints and sent them in for a match? It would be interesting to see the results.
If the typical latent print is partial or smudged, it is mighty difficult to claim that the partial print is unique. Sir Francis Galton claimed that 1-in-64 billion prints is a duplicate. Thirty-six bits represents 68 Billion numbers. If we only have half as much data (18 bits, or half of a fingerprint), we can only represent 260,000 numbers. In a nation of 260 million people, the partial print would typically match 1000 suspects (assuming there is no duplicate data in that half print). Even worse, if the print is slightly, but no noticably smudged, the sample might not even match the finger it came from; thus all 1000 suspects are innocent.
Finally, finger prints are completely useless if they were at the scene for an innocent reason. Just because my prints were on the kitchen knife, doesn't mean I stabbed my wife.
This reminds me of a study on DNA. Researchers took DNA samples from 50 people. They sent the samples to a lab to be matched up. The results came back, and the lab incorrectly matched two samples. If the foresics experts have a failure rate of 4%, it doesn't matter if everyone has unique DNA (which they don't). Even if the foresics experts were perfect, and everyone did have unique DNA, we still do not know if the section of DNA that we sample is unique for each person (i.e. does it match other family members or persons of similar cultures).