Courts have also missed plenty of mistakes from well-intentioned, conscientious scientists, too. In fact, these may be even more common—and harder to catch. Studies show that crime lab fiber, paint, and body fluid analyses, for example, may consistently have error rates of 10 percent or higher. The error rate in fingerprint analysis is possibly between 1 percent and 4 percent. And bite mark evidence is notoriously unreliable though still widely used.

Most of the forensic evidence in criminal trials comes from government crime labs. We need to demand unbiased, reliable results from forensic scientists even if they are paid by the same government that pays the police and prosecutors.

It's not enough to weed out the incompetent scientists. We need to begin to monitor even the good ones. One major barrier to improving forensic evidence in criminal trials is that in most jurisdictions, the state has a monopoly on experts. Crime lab analysts and medical examiners (and to a lesser extent DNA technicians) typically work for the government and are generally seen as part of the prosecution's "team," much like the police and investigators. Yes, science is science, and it would be nice to believe that scientists will always get at the truth no matter whom they report to. But studies have consistently shown that even conscientious scientists can be affected by cognitive bias.

Experts who slant their testimony to favor the prosecution are difficult to overcome.

Most jurors aren't aware of any of these biases; in fact, most give enormous weight to expert witnesses. Even out-and-out frauds like West and Shaibani can persuade jurors if they're presented in court as reputable experts, appear likeable, and can testify with conviction. A study of the first 86 DNA exonerations garnered by the Innocence Project estimated that faulty forensic science played a role in more than 50 percent of the wrongful convictions.

The linked story suggests several proposed reforms to reduce the risk of a wrongful conviction due to sloppy science.