Reliable computer interpretation can address the scientific need for thorough, objective, and informative analysis of DNA evidence.
Ancient societies used distinguishing marks to identify people and their property. During the T’ang Dynasty, Chinese officials certified documents with handprints.1 Babylonians in the pre-Islamic Sassanid Empire established ownership of lost items through identifying marks.2 In the 19th century, more precise measurements of physical features3 and fingerprints4,5 enabled more accurate human identification and statistical association.6
The 20th century witnessed a flowering of diverse forensic modalities (hair, fiber, glass, ballistics, etc.) that could connect crime scene evidence with a suspect.7 Investigative databases that could solve cold cases were established in the earliest decades for fingerprints in London and New York and evolved into DNA databases by the close of the century.8 Today, the success of DNA identification,9 coupled with outsized “CSI effect” expectations,10 have instilled in the modern world a sense of forensic infallibility, with abiding faith in the power of DNA.
There are many consumers of forensic information. Police investigate crimes, prosecutors present evidence that is weighed by judges and juries, while the public funds the forensic enterprise in order to secure better protection from crime. These societal consumers of crime lab information all assume that: a) the forensic evaluation process is thorough and objective, and b) the full measure of identification information has been accurately extracted from the available evidence. These are reasonable expectations. An incomplete or biased approach that discarded information might be ineffective (not find the right person), incorrect (implicate the wrong person), or unusable (inadmissible in court).
However, current forensic science practice does not always meet these expectations. The National Academy of Sciences (NAS) 2009 report on “strengthening forensic science” identified potential flaws or examiner bias in some methodologies.11 Indeed, comparative bullet lead analysis was found to lack a sound scientific basis and is no longer used.12 Since DNA evidence interpretation had not consistently accounted for natural data variation, the federal Scientific Working Group on DNA Analysis Methods (SWGDAM) issued new 2010 guidelines.13 Recent studies report examination bias in DNA mixture interpretation14 and a million-fold information loss in the human review of mixture data.15,16
A thorough and objective evaluation of evidence that yields all the data’s identification information may be desirable, or even necessary, but is it feasible? Isn’t a “match” between evidence and suspect inherently limited to just comparing the data features of these two items? And isn’t this “match” unavoidably biased, since one of the two items is the suspect? If the items either fully “match” or they don’t, how does one measure the amount of information?