Busting crime with digital 'wanted' posters

-- ST ILLUSTRATION ADAM LEE
-- ST ILLUSTRATION ADAM LEE

DNA found at a crime scene can be used to work out what the suspect could look like

It is well known that in criminal court cases, the testimony of eyewitnesses can be flawed. As the United States Supreme Court noted in one case: "The annals of criminal law are rife with instances of mistaken identification."

A study published in the Columbia Law Review in 2008 reported that of the 200 convicted criminals proved innocent through the pro bono efforts of activist lawyers, 158 had been found guilty because of mistaken eyewitness testimony.

A common practice in crime investigations is for police artists to sketch "wanted" posters based on eyewitness descriptions. Now, however, a better understanding of DNA markers, better software and higher computing power may deliver more accurate "wanted" posters.

In January, for the first time in the US, police in Columbia, South Carolina, investigating an unsolved double murder released a digital picture of a suspect not based on eyewitness accounts.

Instead, it was a software-generated sketch that began with genomic data derived from the suspect's DNA picked up at the scene where a young mother and her three-year-old child were cruelly murdered in 2011.

What this means is that any DNA the police may collect at a crime scene can now be used to work out what the suspect likely looks like.

This technology is called DNA phenotyping, where "phenotype" means a description of one's externally visible traits such as height, weight, hair and eye colour.

The technology focuses on the genes or portions of one's DNA that contribute to physical traits such as skin, hair and eye colours, gait, left-handedness and so on.

From this, the next step is based on the idea that people with the same geographic ancestry - that is, their ancestors came from East, South or South-east Asia, Northern or Southern Europe, and so on - are more likely to share certain genes that contribute to their phenotypes.

In the next step, face recognition software is used to make statistical estimates of how a suspect's facial features and complexion might look like, given his ancestry. This software works against a background of data inherent in three-dimensional mugshots of known persons with the same ancestry, usually those already stored in databases of criminals with their DNA and other personal details.

So, despite its name, the technique does not actually involve isolating the genes responsible for particular traits to directly generate digital pictures of human faces. Instead, these digital "wanted" posters are statistical predictions of what a suspect's face might look like, based on his geographical ancestry gathered from his DNA.

Critics argue that "geographic ancestry" is just a decoy for "race", so this technology is merely a tool for racial profiling.

Whatever your DNA says your geographical ancestry might be cannot tell your race accurately, they argue. They point to a 2003 study where individuals classified as white, brown and black were found to have, on average, 72 per cent, 68 per cent and 63 per cent Caucasian genetic ancestry, in that order.

Conversely, they had 19 per cent, 21 per cent and 27 per cent African genetic ancestry, on average, in that order also. So individuals called whites have "black genes" while those called blacks have "white genes", it would seem.

Likewise, a 2005 study reported that individuals with over 50 per cent African ancestry by DNA had very fair skin colour. In such studies, besides subjective self-identification of race, skin darkness is objectively measured using a technique called skin reflectometry.

So it seems that genetic markers used to impute geographic ancestry may not map squarely onto those that encode for skin colour or other racial markers, like nose and eye shapes.

Since a person inherits only half his DNA from either parent, that means half of the genetic information from each parent is absent in any individual. Aggregate that down the generations and ancestry tracing will necessarily and always be incomplete - at the individual or micro-level.

But what critics ignore is that such differences wash out when very huge numbers of people are considered together statistically - at the population or macro level.

A study of 17 populations published in the American Journal of Physical Anthropology in 2009 found that DNA markers accurately assigned geographic ancestries in 98 per cent of cases.

Currently, the technology is reliable enough when used to infer complexion or facial features for crime investigation purposes, but cannot yet meet the much higher standards of evidence required in court.

But in terms of helping law enforcement agencies focus their investigations better, this seems like a useful tool. With ever larger databases, better software and higher computing power, the technology's precision will rise.

This means the guilty are more likely to be apprehended faster while the innocent are more likely to be excluded from investigations.

Britain allows the use of DNA phenotyping to infer ethnicity as one of its five races: white European, Afro-Caribbean, Indian, South-east Asian and Middle Eastern. Critics say it is this kind of categorising that exacerbates racial profiling by police.

But since it does not isolate specifically only those traits of minorities who may have been traditionally targeted, there is little cause for concern, supporters would argue.

That is, we ought to be worried if the technology sussed out only brown-eyed, dark-skinned folk, say, because that group would then be disproportionately vulnerable.

But the technology is not like that. For instance, blue and brown eyes as well as red and brown hair can be predicted with 95 per cent accuracy. But green eyes or black and blonde hair are hits-and-misses. So the technology is not particularly sensitive or especially specific for minority facial features.

This technology makes no value judgments, so it is not prone to racial biases as much as eyewitness testimony could be.

The police in Columbia, South Carolina, have not found the killer in that four-year-old crime. But this improvement on artist sketches of suspects ought to be welcome.

andyho@sph.com.sg