Reforming Forensics

In the aftermath of the Madrid train bombings on March 11^th, 2004, the FBI quickly joined the Spanish police force to mount an international investigation. Within days of the attacks, in which 191 people were killed and 2,000 more injured, Bureau investigators found an abandoned van near the crime scene, containing a bag filled with bomb-making equipment and covered in fingerprints.

FBI fingerprint experts digitized the prints, ran them through a computerized search of their database and matched them to those of 15 people, including lawyer Brandon Mayfield, a Muslim convert who lived in Portland, Oregon, with his Egyptian-born wife. Mayfield was promptly arrested and imprisoned, only to be released two weeks later, after Spanish police said they had identified an Algerian named Daoud Ouhnane as the mastermind of the attacks.

Mayfield’s is far from being an isolated case: Since 1989, nonprofit legal organization The Innocence Project has documented the cases of 347 people who were wrongly convicted of crimes and then later exonerated (usually on the basis of DNA evidence), often after spending long periods of time behind bars.

In the David Kopf Lecture on Neuroethics at the 46^th annual meeting of the Society for Neuroscience in San Diego in November, Thomas Albright of the Salk Institute for Biological Sciences argued that it is time to reform forensic science to ensure that such miscarriages of justice do not continue to occur. (see Will Neuroscience and Law Collide?)

Tools aren’t the issue

“Forensic science gives us a feeling that we have a leg up on the bad guys, and that with sophisticated tools, we’ll be able to figure out who the culprit is in any given situation,” said Albright. “But this idea stands in sharp contrast to a number of very notable and egregious failures in recent years.”

Forensics currently lacks rigor, he says, and is quite often poorly rooted in science, being plagued by bad data analysis practices and poor use of statistics. What’s more, many forensic scientists seem to have little understanding of how human vision and memory work.

Albright described fingerprinting and eyewitness identification of suspects as processes on which the criminal justice system is heavily based but which are prone to cognitive biases that can influence decision-making in ways that can pervert the course of justice.

Fingerprint experts make mistakes because of three types of cognitive bias, Albright said. One is uncertainty, which arises because of the way the brain’s visual system works. The brain interprets visual stimuli on the basis of very limited information; it “fills in the gaps” to predict what we are seeing. This involves drawing on our past experiences, bringing another bias into play.

“Vision is plagued by noise from many natural sources, such as the poor optics of the eye, dim lighting, and distracting features in the scene, so we’re faced with a lot of uncertainty about what we’re actually looking at,” said Albright. “Our biases pull everything together and help us to fill in the blanks about what is likely to be out there given our prior experience of the world.”

These biases can cause us to perceive meaningful patterns that aren’t there and, sometimes, to see things that actually do not exist. A good example of this is pareidolia, or our propensity to see non-existent faces in objects.

It was because of such biases that Brandon Mayfield was misidentified as the perpetrator of the Madrid train bombings. Mayfield’s print was in the FBI’s database because of his military service, and was not an identical match to the prints left on the bag of detonators. Even so, as a recent Department of Justice review revealed, the FBI used expanded surveillance powers under the Patriot Act to collect information about him and monitor his activities. Some of his biographical information—such as his legal representation of an alleged terrorist—confirmed the FBI’s biases and convinced them that he was guilty.

To minimize these biases, fingerprint experts should analyse prints without having knowledge of the case, Albright said.

Seeing is believing?

As the pioneering work of psychologist Elizabeth Loftus showed, eyewitness testimonies are also subject to bias and memory errors. In a classic 1974 study, Loftus showed film footage of car accidents to 150 undergraduate volunteers and then asked them questions about the events that they had seen in the films, including one about how fast the cars were going. She found that those asked “About how fast were the cars going when they smashed into each other?” gave consistently higher estimates than those asked the same question using the verbs collided, bumped, or contacted instead of smashed.

Our recollections of events are not perfectly accurate; furthermore, memories can be influenced significantly by the way questions are asked under cross-examination.

Even though our memories can be so easily contaminated, we tend to have a huge sense of confidence in how accurate they are. Every time we recollect a memory, we introduce new errors, according to our expectations and biases, and store this altered memory. Even as this process corrupts our memories, our confidence in them increases.

Nevertheless, courts of law place great emphasis on eyewitness testimonies and suspect identification: About three-quarters of the 347 exonerations documented by The Innocence Project involved convictions based largely on the accounts of eyewitnesses.

“If the goal is truth, then misinformed biases and over-confidence are the hidden enemies,” said Albright. “As the old saying goes, seeing is believing. That may be true some of the time, but neither seeing nor believing is equivalent to the truth.”

Albright adds that the procedure for creating the photo arrays shown to eyewitnesses to identify suspects can be inherently biased. The choice of non-suspect photos can influence the outcome of the process, for example when none of them resemble the description of the perpetrator, making the suspect seem more conspicuous.

Albright is developing a less-biased approach of creating photo line-ups, so that at least some of the fillers resemble the description of the perpetrator to a greater or lesser extent. He recently co-chaired a National Academy of Science committee that published a report outlining recommendations on how to improve the accuracy of eyewitness identification.

“As a society we’re very averse to sending innocent people to jail, so this has become a high priority that the public is very concerned about,” said Albright. “There is an unprecedented opportunity for science to intervene and correct this.”