Report from Law Commission of Ontario examined the AI-powered tool for analysing DNA samples
Statutory changes, prosecutorial guidelines and training programs are among the “urgently needed reforms” necessary to regulate the use of artificial-intelligence-driven DNA-analysis technology in the criminal justice system, says a new report from the Law Commission of Ontario.
The report studied the use of probabilistic genotyping, an AI tool used to analyse DNA samples collected by police for criminal prosecutions. Probabilistic genotyping examines trace DNA samples, which are samples so small they only give an incomplete or fragmentary DNA profile, or contain mixtures of more than one person’s DNA, says Kate Robertson, one of the report’s authors.
The technique compares two different hypotheses – for example, whether the defendant was a contributor to the sample or was not a contributor to the sample – and generates an estimate of which is more probable, says Robertson.
“We've seen time and again that forensic science and expert evidence is a particular danger zone when it comes to miscarriages of justice,” she says. “And because there is a number of frailties in the [probabilistic genotyping] DNA method, there is a danger that this type of evidence may cause justice system participants to believe that there's seemingly infallible or reliable evidence against somebody who's actually factually innocent.”
“These tools cannot stand alone as evidence of probability of guilt, or the probability in the real world that the defendant actually contributed to the DNA.”
The report’s authors are calling on policymakers to prescribe limits on the admission of evidence produced with probabilistic genotyping and to codify its “presumptive inadmissibility in the absence of strict scrutiny.” To increase transparency and accountability, the authors say Canada must modernize the system of oversight governing how DNA information is used, shared and retained in police investigations and criminal proceedings.
The report also calls for new regulations to require transparency of the AI used for criminal justice purposes, “including openness of source code and source code review.”
“Unreliable, inaccurate or biased [probabilistic genotyping] DNA evidence that goes unchallenged or is uncritically received in criminal courtrooms endangers Canada's justice system and the vulnerable communities who are affected by it,” says Robertson.
The report is the fourth instalment of the LCO’s project looking at the use of AI, automated decision making and algorithms in the Canadian justice system.
Along with Robertson, Jill Presser was also a primary author of the report. Robertson is a criminal and regulatory litigator at Markson Law, and Presser is a lawyer focused on criminal law, digital rights, privacy, surveillance and technology, and was recently appointed a Judge of Ontario’s Superior Court of Justice.
In examining cases in which probabilistic genotyping was used, Robertson and Presser were unable to find any in which it was apparent that the court had been alerted to the reliability factors of this type of evidence. In one case from Quebec, the defendant sought disclosure on the forensic laboratory’s use of the technique and was denied, says Robertson.
“Which is a very concerning fact to know, given how important due-process safeguards – like access to disclosure – are in guarding against wrongful convictions.”
Traditional methods of analysing DNA evidence are viewed by the public as a “gold standard,” says Robertson. The danger is that probabilistic genotyping is given undue weight without adequate safeguards in place to establish its reliability, she says.
“There is still an overarching and unique issue associated with [probabilistic genotyping] DNA evidence in that it is inordinately complex in terms of the analysis that creates this evidence.”
“And it can be very difficult, even for the experts that are tasked with processing these analysis techniques, as well as the lawyers who are involved in the cases, as well as the judges and jury members who are required to analyse the evidence.”