Convicted by software? Not so fast, says California lawmaker
Democratic Rep. Mark Takano questions law enforcement’s use of proprietary forensic algorithms and specialized software in criminal trials
Amid a national debate over bias and discrimination in the criminal justice system, Rep. Mark Takano, D-Calif., is pushing legislation that he says will level the playing field for criminal defendants whose lives may depend on the conclusions of forensic software that are entered as evidence in court.
Crime scene technology is not new, but until recently its accuracy has depended largely on human analysis. Now, bolstered by the belief that the use of technology limits the risk of human error, law enforcement agencies are increasingly turning to artificial intelligence-driven tools such as facial recognition, probabilistic DNA analysis and fingerprint matching to aid in investigations.
But criminal justice reform advocates say forensic technology can be faulty too and can also reflect the biases of the software engineers and scientists who created it. Facial recognition technology, for instance, has been proven less accurate when deployed against women and people of color.
The only way to know whether a program’s findings are accurate enough to be used as evidence in court, they say, is to examine its source code for bias and inaccuracy. But the source code is not always available.
That’s because the companies that sell such software to police and law enforcement organizations often claim in court that their programs are trade secrets, and therefore cannot be examined or tested by lawyers for defendants.
Trade secrets claims argue that a company will suffer financially from publicly releasing proprietary information, such as source code. Judges have typically sided with the companies.
Examine the algorithms
Takano’s legislation, introduced last September, would change the Federal Rules of Evidence to bar judges from preventing defense attorneys from examining software programs that generate evidence being used against their clients.
“We’re trying to ensure that defendants have access to source code and other information necessary to exercise their rights to confront and challenge evidence against them,” Takano told CQ Roll Call. “That is a fundamental core constitutional principle that can’t be eroded by the profit-making interests of a firm.”
Takano’s bill would also task the National Institute of Standards and Technology, known as NIST, with establishing a testing framework for forensic algorithms that federal law enforcement agencies would have to follow when seeking to determine the efficacy of different types of software.
“We’re not saying the science is wrong, and I’m not qualified individually to say that the algorithms work or don’t work,” Takano said. “We need to be able to set some relevant standards, and NIST itself needs to provide some guideposts to judges, to prosecutors and to the defense to guide them through what standards should be in place to understand how this technology works.”
In 2016, a man named Martell Chubbs was convicted of the 1977 murder of a woman in Long Beach, California, and sentenced to more than seven years in prison. The case, which remained cold for more than three decades until police linked Chubbs to a DNA sample from the crime scene, rested on the analysis of software called TrueAllele, sold by a Pittsburgh-based company called Cybergenetics.
TrueAllele’s algorithm “unmixes” DNA samples taken from a crime scene that could belong to numerous individuals, allowing police to compare the unmixed samples against a possible match, such as one belonging to a suspect. TrueAllele uses statistics and probability to sift through the DNA samples to narrow the universe of possible matches.
The software is used by crime labs around the country and has helped exonerate the falsely accused in addition to convicting those found guilty.
In Chubbs’ case, the software concluded that the DNA sample from the crime scene was 1.62 quintillion times more likely to belong to him than someone else.
Chubbs’ attorneys questioned how the software could be sure and sought access to the source code. But TrueAllele entered a trade secrets claim and won. Cybergenetics has entered trade secrets claims in 10 cases where TrueAllele’s findings were used as evidence, said Mark Perlin, who co-founded Cybergenetics and serves as chief scientific and executive officer. The company has yet to lose, he said.
Perlin told CQ Roll Call that using TrueAllele can result in “accurate, objective answers to questions in hundreds of thousands of cases.”
“You overcome the usual failures of DNA mixture interpretation that have reigned for 20 years, where the usual answer, even a simple two-person mixture, is ‘I don’t know, it’s inconclusive,’” Perlin said. “And suddenly the key evidence that can result in convicting a serial rapist, or exonerating an innocent man who has been in jail for 40 years, is lost because the [genetic] answer is inconclusive.”
TrueAllele has undergone significant testing, Perlin says. He believes lawyers seeking access to its source code are trying a legal ploy designed to force Cybergenetics to withdraw from the case because they are unable to disprove the science behind its conclusions. (The company provides defense attorneys access to the software, not the source code.)
“The goal is that the expert and the evidence will just disappear, because the expert will withdraw from the case,” Perlin said.
But forensic algorithms, including probabilistic DNA genotyping software manufactured by other companies and by law enforcement agencies themselves, have been shown to be inaccurate. Legal experts who favor Takano’s bill say it can be impossible to gauge accuracy without source code.
Trade secrets defense
Trade secrets claims can be legitimate because a company may be concerned about commercial theft of its algorithms, according to Rebecca Wexler, a professor of law at the University of California, Berkeley.
“Other times, it might be an illegitimate interest, where they’re trying to hide behind intellectual property to shield their tools from scrutiny and challenge,” she told CQ Roll Call.
Wexler, who aided Takano’s staff in writing his legislation, said the current rules of evidence in criminal court do not account for the rise of forensic software, or for the entrance of trade secrets claims.
“Prior to the introduction of the automated algorithm into forensic science, trade secrets were not a basis to prevent criminal defendants from getting access to the evidence to which they would otherwise be entitled,” Wexler said. “But because these algorithms are being developed by private companies, those companies have an interest in asserting their intellectual property rights.”
At a time when protesters are calling for an end to police access to technology that may discriminate against people of color, Takano says tools already in use must be shown to be accurate beyond doubt.
“It’s wrong to conclude that technology is going to erase or remove the racial biases that have existed in the criminal justice system because of analog or human lab analysis,” he said.
But Perlin said Takano’s bill would stifle innovation and ultimately curb technological advancements in the criminal justice system, including those designed to make justice less subjective.
“This bill doesn’t help judges, innocent people, defenders, prosecutors or police. It doesn’t help protect society. It doesn’t help science,” Perlin said. “It seems to only help two groups: guilty criminals and NIST. So I don’t know what the purpose of it is, unless that’s who you really want to help.”
Jinitzail Hernández contributed to this report.