Criminal Law, Artificial Intelligence

The problem with probabilistic genotyping of DNA

By Staff

There is a concern that a new DNA testing tool being used by Ontario’s Centre of Forensic Sciences (CFS) will gain credibility with jurors it’s not yet earned, says Toronto criminal lawyer Jill Presser, who will be a guest panellist during the Criminal Lawyers’ Association (CLA) Fall Conference.

Presser, principal of Presser Barristers, will be among three panellists discussing emerging developments in DNA technology during the 47th annual conference, scheduled for Nov. 15- 16 at the Toronto Marriott Downtown Eaton Centre Hotel, 525 Bay St.

Billed as Canada’s largest criminal law conference, the CLA event is expected to draw more than 700 attendees, including defence lawyers, judiciary, educators, and media.

Presser will be joined by fellow panellists Jack Laird of the Ministry of the Solicitor General, and Justice David Rose of the Ontario Court of Justice to discuss emerging developments in DNA technology— specifically transference and trace evidence.

Rose is to discuss the importance of technological literacy within the justice system and the need for forensic evidence to be handled with care, she says.

Laird, who is the CFS section head — biology, is to speak extensively on the centre’s use of STRmix — a probabilistic genotyping DNA testing tool.

Presser will also address the use of STRmix from a criminal defence perspective.

Probabilistic genotyping uses statistical rules of probability and mathematical algorithms to test for DNA profiles in instances where a sample is very small, degraded, or includes a mixture of multiple individuals’ DNA, she says.

Using this artificial intelligence (AI), scientists arrive at a “likelihood ratio” that outlines the probability of a suspect’s DNA matching a sample taken from a crime scene.

“There is great confidence in science’s ability to test a bodily fluid from a single source at a crime scene and arrive at meaningful evidence when compared to a sample from a suspect,” Presser says.

“We’ve now arrived at a place where science is trying to make determinations from more varied types of crime scenes. What we’re seeing, for example, is a swab taken from a solid object that arrives at the lab involving fewer cells, or cells that may have been degraded, or comes from more than one source.”

Using algorithms and rules of probability, the AI tool compares and contrasts two hypotheses — that the suspect’s DNA is included in the sample versus that individual’s DNA not being included, she says.

“A CFS expert comes to court in instances where you are not dealing with the gold standard of DNA, and testifies that, ‘Our algorithm tells us, based on the sample we have, that it is more probable that the accused was a contributor to this sample than not.’ Or they may say it’s less likely,” Presser says.

“Either way, they will give you a mathematical ratio that supports the basis for their conclusion,” she says.

There are a number of concerns from a criminal defence perspective with the use of probabilistic genotyping, Presser says, adding that one of the things it does is posit a conclusion about the accused.

“So automatically, when you are developing a ratio, you are beginning with an assumption of guilt. In a justice system that is premised on the assumption of innocence, that’s a problem.”

Another difficulty is that the tool is so “fiendishly” difficult to understand that it’s a challenge to “really appreciate” what it means in terms of reliability and validity, Presser says.

Another shortcoming is such AI tools rely on a comparison of two hypotheses to arrive at a likelihood ratio, she says.

Presser says when statistical probabilities are used to compare and contrast any two hypotheses, “no matter how absurd,” one is always going to be reported as more probable than the other.

“So nothing about the likelihood ratio emerges out the other side to address how objectively likely the hypothesis is,” she says. “Because it expresses a relationship between two hypotheses, there isn’t necessarily a relationship to reality or proof beyond a reasonable doubt.

“There is a great danger with this evidence because it’s difficult for jurors to understand, and I think the ratio that pops out — when expressed by someone with the great authority of a scientist testifying as an expert — is most likely to be understood as the likelihood of the defendant’s guilt,” Presser says.

“Because people are now so familiar with the gold standard of DNA evidence, this new kind of testing enjoys credibility that it shouldn’t be entitled to. It gets some of the halo of the more solid DNA evidence.”

To Read More Jill Presser Posts Click Here