Canadian academic report urges no facial recognition at legislature, reasons why sow confusion

Canada’s Parliamentary Protective Service is thinking about deploying face biometrics around the country’s legislature.

As part of that process, the organization commissioned a report titled, ‘Face Recognition Technology for the Protection of Canada’s Parliamentary Precinct and Parliament Hill? Potential Risks and Considerations.’ The paper was completed in April, and published to the SSRN platform this month.

Authored by four academics from Toronto Metropolitan University’s Leadership Lab, the report was composed following a series of interviews with professors and advocates including Surveillance Technology Oversight Project’s Albert Fox Cahn and the Canadian Civil Liberties Association’s Brenda McPhail. Representatives of law enforcement and technology labs are conspicuously absent from the acknowledgements, but the researchers did interview staff from the PPS (the closest equivalent to the Secret Service in the U.S.).

The paper concludes that facial recognition reproduces and exacerbates bias, that the powers and immunities granted to members of parliament to allow them to do their jobs need to be considered, and that the technology can erode privacy rights. The use of facial recognition around Parliament Hill would raise numerous questions around risk to legally enshrined rights, according to the report.

Measures taken if the technology is adopted should include considerations of the principles of necessity, proportionality, effectiveness, and minimum impairment, along with the completion of impact assessments, and the public should be consulted.

A section on ‘Accuracy issues and other considerations regarding the use of FRT’ cites NIST’s late-2019 study with a generalization about the near-200 algorithms tested. Lead study author Patrick Grother warned specifically against generalizations about the state of the industry, when deployments are always of a particular algorithm.

A research paper reviewing of the state of the art of facial recognition is cited as evidence that the technology is not accurate enough. The 2020 paper does not use the phrase “very error-prone” which the Leadership Lab paper does, nor does it draw any conclusions about accuracy at all, except that 3D facial data can improve it.

It does refer to a 99.86 percent accuracy rate in a 2017 test attempting to match faces against the Labeled Faces in the Wild dataset, as well as to enormous gains in accuracy in recent years.

“2D approaches reached some degree of maturity and reported very high rates of recognition,” the 2020 paper states.

The Canadian report goes on to review the proliferation of surveillance cameras around Canada’s Parliament Buildings, how they are used, and the possible ways facial recognition could be deployed to secure the legislature.

The report authors advise against those deployments, for the above reasons, but also a lack of regulation of biometrics in Canada, and because of the risk of function creep.

Article Topics

accuracy  |  biometrics  |  Canada  |  facial recognition  |  law enforcement  |  privacy  |  video surveillance