Why police need facial recognition
A COMPLEX SHIFT IN PUBLIC SAFETY METHODS IS CALLING FOR US TO LOOK TO THE FUTURE
In light of the recent public discourse over race and policing, some of the world’s largest technology companies — including Microsoft, Amazon and IBM — announced they will no longer provide facial recognition technologies to law enforcement agencies.
even before the death of George Floyd and the important discussion over equality in the justice sector that came as a result, police chiefs in Canada were urgently reviewing their use of Clearview AI, a facial recognition tool that canvasses online images for potential matches to suspects or victims’ images. That urgency was driven by media reports speculating that the company’s software may legally infringe on privacy rights. The cause has been taken up by five provincial privacy commissioners and their federal counterpart. Parliamentarians have also committed to studying this issue.
Many police leaders are rightly asking questions about how such technologies ended up in their services without going through appropriate procurement channels or having clear policies around the use of facial recognition and other artificial intelligence applications. yet the presumption by some commentators that our police agencies aren’t interested in privacy rights, similar to the actions being taken by autocratic regimes, is an unfounded leap.
The reality is that our police services are dealing with a complex shift in public safety challenges that have come about in the digital age. At the same time, the critical laws governing investigations still resemble those of the 19th century and the funding models and tools we afford to our police haven’t radically evolved from those used in the 20th century.
Crime statistics in Canada show a decline in traditional crimes such as burglary and vandalism. However, this doesn’t tell the full story: there is immense growth in cybercrime, such as phishing schemes, as well as cyber-enabled crimes, such as human trafficking and terrorist incitement.
Such crimes tend to victimize vulnerable populations at a greater rate. This includes children, seniors, low-income persons, immigrants and those suffering from mental health difficulties. As Canadians have seen, certain vulnerable populations of a specific ethnocultural profile have been victimized at greater rates. This has been most prominently documented through the volume of missing and murdered Indigenous women and girls.
One of the fastest growing area of crime in the digital age is online child sexual exploitation. The advent of the Internet has emboldened offenders and is linked to domestic and international human trafficking. It’s a crime that’s global in nature and affects children as young as newborns. It haunts its victims well beyond the memories of the painful acts, as the content is easily reproduced. And the perpetrators are protected by the security and anonymity afforded by social media sites, dark-net markets, video-streaming services and online payment systems.
Statistics Canada reported a 233 per cent increase in online child sexual exploitation cases from 2006 to 2016. The Canadian Centre for Child Protection has detected over 13.5 million images and videos of children being sexually abused online in the past three years.
One of the most overlooked areas of this dark reality are the pains taken by the officers who investigate these heinous crimes. They have to look at millions of images, hours after hours of videos of children being sexually abused, to try to help identify victims and suspects. Much of this work is sifting through horrific content that has been previously reviewed by other officers in other jurisdictions to ensure there are no new victims or suspects that can be identified. Many of these officers face severe bouts of post-traumatic stress disorder as a result.
It isn’t a giant leap to believe that with the knowledge of the volume of content officers are exposed to, the risks of severe harm to vulnerable children and the ability for AI applications to assist in investigations, Canadians would want our investigators to utilize technologies that help identify victims and suspects, while reducing harm to themselves. There may be similar parallels in other emerging areas of crime in the digital age where simply hiring more officers, the most popular strategy of politicians to address a growing public safety challenge, may not make a dent in the issue.
This isn’t to suggest that police should have unfettered access to technologies that bump up against civil liberties or serve visible minorities poorly. There are important questions about the effectiveness of these technologies, as independent studies have shown that facial recognition software producers a greater number of errors when analyzing persons of colour. The greatest contributor to this issue is that the training data that the AI algorithms learn from are predominantly made up of white males. This embeds a structural bias in the software itself. Technology vendors must prioritize providing tools that serve the whole population, without a wide margin of error based on race, if they want such tools to be seen as trustworthy.
Police leaders and their oversight bodies must do a better job of highlighting the changing public-safety landscape and how they will balance the rights of all citizens, regardless of their race, while performing their societal obligation to keep communities safe, including online.
They will have to build trust with citizens by being transparent about what technological capabilities they would like to pursue. More importantly, they must be upfront about how they will evaluate new technologies and what data will power them. They should attempt to identify potential biases that can result from such technology and work to mitigate them. As opposed to an adversarial relationship with privacy advocates, our police agencies should work in concert with them and other stakeholders like victims’ advocates, community organizations committed to racial equality and police unions to find reasonable solutions.
In the absence of a thoughtful policy framework and technologies without biases that are informed by the modern challenges of public safety and built on the principles of transparency and trust, there is a serious risk. Reactive policies, such as bans on facial recognition and other AI applications for policing, will leave vulnerable populations like children, visible minorities and others at greater risk than is necessary.
TOOLS GIVEN POLICE HAVEN’T RADICALLY EVOLVED FROM THOSE OF THE 20TH CENTURY.
Neil Desai is an executive with Magnet Forensics and a senior fellow with the Munk School of Global Affairs and Public Policy at the University of Toronto and the Centre for International Governance Innovation. He previously served in senior roles with the Government of Canada.